IEEE Novosti

If you grew up in the 1980s or ’90s, you likely remember shaky home video footage, taken with a handheld camcorder, of family gatherings, vacations, and other events.Camcorders combined a camera with a video recorder. They included a rechargeable battery, a slot for a videotape, and a shoulder strap. Most were outfitted with an optical zoom lens and a small, articulating screen—a display mounted on a hinge that could tilt and rotate. The operator could check the screen to view what was being recorded.The user’s natural hand and body movements when filming led to jittery footage. The best way to get a steady shot was to place the camcorder on a tripod or a gimbal: a motorized stabilizer.There were fewer poor-quality recordings after Panasonic introduced its PV-460 VHS camcorder in 1988. It was the first video camera to include an optical image stabilizer, which compensated for movements. Stabilization features are now standard in today’s cameras including ones found in smartphones and drones.The PV-460 camcorder was honored as an IEEE Milestone on 9 July. The dedication ceremony was held in Kadoma, Japan, at the Panasonic Museum, which displays the company’s past products.The IEEE Kansai Section in Japan sponsored the Milestone.“The release of the PV-460 fundamentally transformed personal videography, enriching the way people captured travel, events, and family memories,” section members wrote in support of the Milestone nomination. Their proposal is available here.“Its image stabilization features democratized video creation by dramatically lowering technical barriers, allowing ordinary people to express themselves with newfound creative freedom,” they wrote. “Beyond the home, image stabilization technology found critical applications in specialized fields, contributing to advancements in areas such as educational media and telemedicine.”The history of camcordersBefore the camcorder was invented in 1982, people filming events in the 1970s and early 1980s used two pieces of equipment: a video camera and a separate video cassette recorder (VCR), which were connected by a multipin cable. The camera was about the size of a toaster, and the VCR could be as large as a suitcase. To record, the person operated the camera with one hand and carried the VCR in the other or rested it on a shoulder. The cable transmitted the images from the camera to the cassette.The PV-460 was made possible by several groundbreaking innovations, according to the Milestone proposal, one of which dates back to the 1950s.In 1956 Italian manufacturer Durst released its Automatica, considered one of the first cameras to use automatic exposure technology. By combining a light meter with the camera’s internal mechanical systems, the technology removed the necessity of calculating exposure settings by hand when the lighting shifted or other conditions changed. The innovation enabled amateur photographers to take decent pictures.The next breakthrough technology—autofocus—was invented in 1973 by Norman Stauffer, a manager of research for Honeywell in Littleton, Colo. It uses a sensor, a control system, and a motor to focus on a selected area. The invention led to the development of early electronic autofocus cameras, which eliminated the need for photographers to manually adjust the lens. Stauffer received the 1990 IEEE Masaru Ibuka Consumer Technology Award for his invention.“The release of the PV-460 fundamentally transformed personal videography, enriching the way people captured travel, events, and family memories.” —Milestone sponsorsU.S. inventor Jerome Lemelson is credited with developing technologies that underpinned the camcorder, according to MIT. In the 1950s and ’60s, Lemelson filed several patent applications related to video and audio recording devices. In 1980 he was granted patents related to a portable video camera system. In 1982 JVC and Sony used the technologies to develop what they called the camera/recorder, which became known as a camcorder.Sony released the first handheld camcorder in 1983: the Betamovie BMC-100P. It used the Betamax videocassette format and could record up to 3.5 hours of footage on 1.27-centimeter cassette tape. The operator rested the 2.5-kilogram camcorder on top of a shoulder to shoot footage. It sold for around US $2,000 at the time (roughly $33,400 today). The machine couldn’t rewind or play back tapes; it could only record.Other electronics companies including JVC soon introduced their own models using the VCR format, which eventually replaced Betamax.Over time, camcorders became more compact.But none of the companies could fix the shaky-footage problem.Solving a shaky problemA team at Panasonic led by researcher Mitsuaki Oshima took on the task of image stabilization: detecting and correcting small camera movements, referred to as camera shake, according to the proposal. Oshima, an IEEE life senior member, is now an honorary Fellow at Panasonic.“The movements that needed to be detected and corrected included horizontal, vertical, and rotational motions—specifically pitch, yaw, and roll,” the Milestone sponsors wrote. “Rotational motion, in particular, becomes the dominant factor affecting image stability during high-magnification shooting. Therefore, the development team focused on detecting rotational motion and began developing an angular velocity sensor.”An AVS, essentially a gyroscope, detects how quickly an object is changing its orientation in space.Sensors capable of detecting angular velocity were large and expensive at the time, making them unsuitable for consumer video cameras, the sponsors wrote. What was needed, they said, was a compact and inexpensive version.Oshima and his team built a high-performance, small, low-cost vibration-type gyroscope. The stabilization mechanism included a miniaturized sensor paired with an optical-axis correction mechanism.The mechanism adjusts the lens or image sensor to counteract physical shifting and vibrations, ensuring that the light path remains centered on the sensor—which is crucial for maximizing sharpness and quality, the Milestone sponsors wrote.“The system detects lens displacement caused by camera shake and immediately compensates for it, ensuring stable video footage,” they wrote. “As a result, the effects of camera shake are minimized, allowing users to capture smooth and steady videos with ease.”Without Oshima’s image stabilization technology, the PV-460 wouldn’t have been developed and released in 1988.The technology was patented and broadly licensed by other companies. It has become a standard feature in a variety of imaging applications.Awards and accoladesThe PV-460 gained instant popularity when it debuted in June 1988. It received rave reviews at that year’s Consumer Electronics Show.Panasonic received a 100 Award in 1989 from R&D World magazine for “the development of a VHS camcorder with an antishake mechanism.”Oshima’s research paper, “VHS Camcorder With Electronic Image Stabilizer,” and others are available in the IEEE Xplore Digital Library.To learn more about historical figures in engineering, IEEE Milestones, and IEEE History Center programs and events, check out The Institute’s IEEE Tech History collection. IEEE Spectrum also covers aspects of tech history. Milestone plaque displayThe Milestone plaque is to be displayed on the ground floor of the Panasonic Museum, which is open to the public. The museum is located near the now-shuttered Panasonic research lab where the technology was developed. The plaque reads:“In 1988 the pioneering PV-460 camcorder equipped with image stabilization for enabling smooth and steady video capture was introduced by Panasonic. By pairing a miniaturized vibrating-structure gyroscope sensor with an optical-axis correction mechanism, the PV-460 eliminated the jitter caused by hand motion. Broad international licensing of this patented scheme made it a standard feature in film and digital cameras, smartphones, and related imaging devices.”Selected by the IEEE History Committee and endorsed by the IEEE Board of Directors, IEEE Milestones recognize outstanding technical developments around the world that are at least 25 years old. The Milestone program is administered by the IEEE history and heritage group.

In 2005, Nokia sold its billionth mobile phone, a budget-friendly device that went to a customer in Nigeria. By then, the company, based in Espoo, Finland, was making one of every three cellphones globally.But just nine years later, the mobile-device maker offloaded its entire handset division to Microsoft for pennies on the dollar, compared to what it had been worth at its peak. Nokia had risen from obscurity in the 1990s to become a worldwide cultural phenomenon by the turn of the millennium, its signature devices featured in TV shows and movies, announcing their presence with instantly recognizable Nokia ringtones. As Nokia was becoming comfortable in the spotlight, the smartphone era arrived. And what came next was swift and brutal. But, as revealed in Nokia internal documents recently made public and interviews with key Nokia engineers from that era, the company saw it coming. Within 24 hours of Apple CEO Steve Jobs’s iPhone unveiling in 2007, Nokia was already weighing its options. They’d immediately recognized the threat. However, outrunning it was another matter. What follows is Nokia’s story over 14 years, from 1998 to 2012, as the world’s top cellphone maker—how its devices defined their time, how the tech reshaped what phones could be and do, and how the company’s good fortunes in the handset business came to an end. Nokia Was Once Unbeatable The centerpiece Nokia devices, the ones that people probably think of when they see the words “Nokia phone,” were the 3210 and its cousin, the 3310. TechRadar has called the 3310 “the greatest phone of all time.” Nokia’s 3210 phone, released in 1999, was an inexpensive device aimed at younger users. Colin McPherson/Alamy Released in 1999 and 2000, respectively, the two devices sold more than 280 million units worldwide. Their most innovative hardware feature was the internal antenna—the first mass-market phone without even a stub or retractable aerial. “Consumers had the perception that it could not work well without an external antenna,” said Peter Røpke, a former Nokia senior vice president, in a 2016 interview with Slate. The phones shipped with games, including the legendary Snake, one of the most popular pre-smartphone mobile games—in which a pixelated serpent eats and grows with every morsel consumed.Nokia introduced no small portion of the world to texting. At the time of the 3210 and 3310, the prevailing texting standard was SMS (short message service), which allowed up to 160 characters per message. Nokia appended its own Nokia smart-messaging service to SMS, which allowed the sending of small bitmapped images across an otherwise text-only system. A rich-text messaging system that allowed visual images, audio, and video followed in 2002, leading to a multimedia messaging service (MMS) standard that remains in place today. Nokia also enabled users to easily create and share ringtones on their devices. By 2000, Nokia’s custom-ringtone Composer app had popularized a new, short-form musical medium that the ringtone industry, at its peak, would transform into a billion-dollar marketplace in the United States. Nokia introduced its 1100 phone in 2003 and ultimately sold half a billion units, making it the most popular cellphone in history. Paul Chesne/Donaldson Collection/Getty Images A few years later, Nokia reimagined its mobile handsets, releasing the 1100 in 2003. The 1100 sold a half a billion units, more than any cellphone in history. It remains one of the best-selling consumer products ever. Much of the 1100’s success was due to its price tag—in the neighborhood of US $100, making it at the time Nokia’s most affordable device.Also contributing to the 1100’s popularity were features designed for longevity and tough environments, including dust resistance, nonslip sides for better handling in rainy conditions, and a 400-hour standby battery life. The 1100 introduced a flashlight as well, which the user turned on and off by holding down the “C” key. Where most device makers at the time were worried about camera megapixels and color screens, Nokia had leapfrogged its competition with a back-to-basics phone that could survive the rain, endure unreliable power grids, and light the way home.Apple Launched the iPhone, Nokia Scrambled On 9 January 2007, at the Macworld conference in San Francisco, Steve Jobs made a characteristically bold claim. “Today, Apple is reinventing the phone,” he said, soon pulling one of the first iPhones out of his pocket. Apple CEO Steve Jobs famously launched the iPhone at the Macworld Conference in San Francisco on 9 January 2007. Nokia held a rapid-response meeting to the event the following day. Tony Avelar/AFP/Getty Images Rumors of Apple entering the phone market had swirled since the iPod’s debut in 2001, but nobody had really reckoned with what that might mean. “Executive summary: Apple iPhone is a serious high-end contender,” read a slide from a Nokia internal meeting held the day after Jobs’s keynote. (That slide is now in the company’s online archives, opened to the public last year.)“User interface has been a big strength for Nokia,” it continued. “Nokia needs to develop touch [user interface] to fight back.” Peter Bryer, at the time Nokia’s manager of strategic foresight, was part of that 10 January meeting, and he recalls that Jobs’s announcement wasn’t unexpected. But the iPhone’s extensive reliance on multitouch—save for a single home button on the front—did surprise the team. Nokia was already aware of multitouch technology, Bryer notes. In 2006, the U.S. computer scientist Jeff Han had given a celebrated TED talk about it, demonstrating a multitouch screen, which could sense multiple fingers on the screen at a time, not just one. Bryer remembers his colleague Timo Partanen, then Nokia’s director of market and competitor analysis, getting excited about Han’s demo. In 2006, the NYU research scientist Jeff Han showed off a new multitouch interface technology as part of a popular TED talk. By the end of the decade, multitouch—in which multiple fingers can interact with a touchscreen at once—would play a key role in smartphones from Apple, HTC, and Palm. Steve Jurvetson/Flickr “Timo burst into the room, saying, ‘You’ve got to see this TED video of this guy using multitouch,’” Bryer recalls. “We both thought that was cool and that’s the future. Then I looked at the sponsors of the presenter’s research, and among them were Nokia and Microsoft.”And yet it took Nokia years to develop a phone that used multitouch. “Remember, Nokia is based in Finland,” he says. “It’s very cold in Finland. They wear gloves for six months of the year, including the executives. They didn’t think a device like that would work.” Winter gloves were no obstacle to operating the chunky buttons on Nokia phones, a design priority perhaps stemming from the company’s Finnish culture and headquarters. Erol Gurian/laif/Redux Partanen was also at Nokia’s post-iPhone launch meeting, and recalls that there was little concern in the room. “We felt okay,” he says. “This is yet another competitor launching a great product. But we had no doubt that, if it’s successful, we would do the same. We will launch similar products.” In November 2008, Nokia released the 5800 Xpress Music, a year and a half after Apple had launched its iPhone. Shaun Curry/AFP/Getty Images That similar product ended up being the Nokia 5800 XpressMusic, known as the Tube, released in 2008. “The idea was to focus on streaming videos and television,” Partanen says. “So we made a phone with a similar form factor to the iPhone [that was] optimized for streaming content.” But the 5800 was “delayed, delayed, delayed, delayed,” he says. “It didn’t materialize in the way it was planned. It was released as a watered-down version.”Critics skewered the 5800’s “outdated” feature set and “ancient” S60 operating system, which ran on top of Symbian OS, an open-source mobile platform Nokia had recently acquired. The 5800 sold reasonably well for its time, reaching around 8 million units in its first year alone. But it did not feature multitouch. “I think that started to be the point when everybody realized that, hey, this is by far more difficult than earlier competitive issues we’ve had,” Partanen says.Nokia finally released its first device with multitouch in 2010, three years after Jobs’s splashy iPhone announcement and four years after Han’s TED talk demo.How Android Ate Up the Low-End Market Nokia had long owned the low end of the cellphone market, with its sturdy, no-frills devices suited for that segment. So the years immediately following the iPhone’s launch saw the Finnish firm continue to thrive as it kept turning out simple, rugged devices. As one review of the Nokia 1200—successor to the 1100—put it in October 2007, “This handset chucks away all the fancy features you’ve come to expect on a modern mobile, leaving you with a pared-down feature set that’s easy for tech novices to get their heads around.” Two cellphone users in Nairobi, Kenya in 2013 exchange a payment on a Nokia 1200 phone via the M-Pesa Mobile Money Market, a popular online banking service. Trevor Snapp/Bloomberg/Getty Images The 1200 kept the 1100’s dust-proofing, flashlight, and long-lasting battery, and added features aimed squarely at the developing world. The 1200 was the first to include call-time tracking and a multiuser phone book, allowing owners who planned to lend their device to set up call limits based on time or cost. This feature helped enable what Nokia researchers called kiosks—informal pay-per-call services, in which an enterprising phone subscriber charged neighbors and family members by the minute for use of the device.In 2006, Nokia studied how Ugandans used their Nokia phones in rural and remote areas. An internal company slide deck from the time reveals just how keyed-in Nokia was to its lowest-income users. “Village phone operators are often women,” the slide deck notes. “And there tend to be a lot of children around. (Phones need to suffer considerable abuse from chewing, dust, sweat, etc.)”“A unit of phone time is 60 seconds,” another slide states. “But to avoid accidentally going over that time and incurring extra costs, kiosk operators shorten the unit to 57 seconds, allowing a three-second margin of error. Shared mobile used as phone kiosk must show call time.”Nokia’s familiarity with its market couldn’t protect the company forever, though. Nokia sought out user input around the world for the company’s device designs, including hosting “Open Studio” contests soliciting users’ sketches of their dream cellphone. Shaul Schwarz/Getty Images That’s because the iPhone wasn’t Nokia’s only looming smartphone competitor. In September 2008, the first Android phone went on sale—the HTC Dream, which was also sold as the T-Mobile G1.While the iPhone was aimed mostly at early adopters and affluent users who could afford to drop hundreds of dollars on a new phone, Android phones were, within a couple of years, aiming at the same low-cost, global user base Nokia was selling to.“I think it’s fair to say Android is the one that disrupted the market more for Nokia,” Bryer says. “Most of Nokia’s successful devices were not on the high-end market. But then, when Android came along, it started to fill that lower end and eventually took that market away from us.” An executive from Nokia India in 2010 holds the company’s 5530 Xpress Music and 5230 phones, both of which had touchscreens, although only the 5530 had Wi-Fi. Sam Panthaky/AFP/Getty Images With two emerging competitors in the low end and high end, the Finnish device maker responded with a device that split the difference—and satisfied neither camp.Released in 2009, the Nokia 5230 attempted to be a low-priced, touchscreen (though not multitouch) competitor to both the iPhone and Android. It sold an impressive 150 million units, doing especially well in developing countries. But the 5230 didn’t have Wi-Fi—one of the biggest complaints at the time. In the developing world, Wi-Fi connections were still rare, so the lack of Wi-Fi made some sense. But the rest of the world was not pleased. “We had such a big gap and dominant position,” Bryer says. “Which does maybe create a level of comfort which you should never get.”How Nokia Lost the Smartphone Race By the beginning of the 2010s, Nokia could have still drawn from the company’s labs, which were regularly spinning out new technologies and innovations. However, the Finnish handset maker ultimately failed to turn its R&D into viable new product lines in response to the emerging smartphone threat.Nokia’s predicament had precedent—Kodak, dominant in film photography, had actually invented the digital camera in 1975 but failed to commercialize it before digital imaging made its core business obsolete.“The technology coming from our R&D teams was cutting edge,” says Gordon Murray-Smith, director of services and ecosystems intelligence from 2008 to 2011. He recalls attending annual R&D innovation days that showcased work on self-healing materials and flexible screens, long before those technologies were seen elsewhere. “But why was Nokia not able to commercialize some of that really interesting and innovative activity more than it did?”Nokia desperately needed an injection of life to change its fortunes. The company’s first non-Finnish CEO, Stephen Elop (a Canadian fresh off a two-year stint on Microsoft’s leadership team), did not mince words. In an internal memo from February 2011 that was soon leaked to the media, Elop wrote, “The first iPhone shipped in 2007, and we still don’t have a product that is close to their experience. Android came on the scene just over two years ago, and this week they took our leadership position in smartphone volumes. Unbelievable.” In 2011, Nokia released the N9, a smartphone with a Linux-derived operating system. Within a year, Nokia had pivoted toward its Windows Phone-powered line of Lumia devices. Munshi Ahmed/Bloomberg/Getty Images Elop oversaw the 2011 launch of a Linux-based smartphone, the Nokia N9. The N9 ran on a distribution of Linux called MeeGo. Reviewers at the time praised the new smartphone direction the Finnish phone maker had taken. “Possibly the most beautiful phone ever made,” wrote one reviewer about the N9 for Engadget.But the N9’s accolades did not ultimately carry the day. Nokia announced its Lumia line of phones the same year—a direct pivot away from MeeGo toward the Windows Phone. It would be the last major strategic turn Nokia would take as a cellphone manufacturer. From this point forward, a succession of C-suite decisions all but sealed the fate of Nokia’s iconic line of phones. In 2013, Microsoft announced its bid to acquire Nokia’s handset operations. After the sale went through the following year, it rebranded the division Microsoft Mobile. But the year after that, Microsoft decided it had made a costly mistake, writing down $7.6 billion—nearly what it paid for Nokia’s handset division—and laying off nearly half of the former Nokia staff it had inherited.In 2016, Microsoft sold its feature phone assets to HMD Global. The latter still sells Nokia-branded phones—budget-friendly devices as well as nostalgia reproductions of models from Nokia’s glory days. What remained was a brand name, some intellectual property, and two decades of hard-won lessons about what it takes to stay on top—and what it costs when you can’t.“When you look at the players in the world of smartphones today, any of those players would struggle ever to achieve 14 consecutive years of being No. 1,” says Murray-Smith. Partanen says there was a downside to Nokia’s mobile-phone dominance. “Often, being the first mover is not necessarily the best position,” he says. “Being a quick follower is the best position.”The company itself ultimately survived, even if the transition wasn’t painless. Nokia’s revenues, which peaked in 2007, fell sharply through the mid-2010s before the company refocused on a decades-old business line—telecom infrastructure—that many had forgotten Nokia was even in. Nokia now ranks among the world’s top three suppliers of 5G network equipment, serving carriers across more than 125 countries, alongside Ericsson and Huawei. Although the company could never quite crack the smartphone, it now plays a key role in providing the network backbone those smartphones run on.

This article is brought to you by X Square Robot.Large language models gave artificial intelligence a working recipe. Pretrain a large model on broad data, and general capability follows. Robotics has no such recipe. Robotics systems have long been assembled from separate perception, planning, and control parts that rarely add up to intelligence a robot can carry from one task to another, or one machine to another. The central problem in embodied AI is to find the equivalent recipe, and the field does not yet agree on what it is.X Square Robot, a Chinese embodied-AI company, has made an unusually explicit bet. It argues that the recipe is an integrated stack, spanning the data a robot learns from, a world model for predicting changes in the physical world, and an action model that brings together perception, planning, reasoning, and decision-making to generate executable robot behavior. The company also believes that the stack should be built and released in the open. X Square Robot shares its vision of bringing robots into real homes.X Square RobotX Square Robot’s embodied AI stackWhat holds the stack together is a small set of principles rather than a single overarching model.The first is that the basic unit of robot data is an interaction, not a trajectory; a demonstration is successful only if it changes the world as intended, not simply because the joints moved. The second is that pretraining should yield usable capability, not just an initialization for later fine-tuning. The third is that behavior should be modeled around physical events rather than fixed slices of time. These principles make the layers interdependent, since the same robot-free data that trains the action model is also structured to feed the world model. It is worth being precise, though. The company describes the world model and the action model as complementary but independent model families that share a code base. Both sit within its broader World Unified Model, which it has presented as an architecture for training vision, language, action, and physical prediction together.Robot learning data: Engineering for quality and cost, not scaleFor the X Square Robot team, one of the biggest constraints on general-purpose robots is the cost and quality of interaction data, not the number of parameters. To address that, the company built its Universal Manipulation Interface (UMI) data collection system, QUANXTA Zero Series. It works by collecting demonstrations from people wearing a rig with dual grippers rather than teleoperating a robot. This approach is not itself new, and builds on established methods for robot-free data capture. What sets it apart are two engineering choices. X Square Robot emphasizes data quality control, recording trajectories and replaying them on a real robot, with only those that actually complete the task counted as valid.X Square RobotThe first is quality control, and it is the most distinctive part. Rather than accepting recorded trajectories as they are, the system runs a closed inspection loop, and its notable step is physical playback. A sample of trajectories is replayed on the real robot, and only those that actually complete the task count as valid. That makes the validity rate a measured quantity rather than an assumption. For example, a gripper that closes a fraction of a second too early still looks like a grasp in the data, yet it has pushed the object away, so it shouldn’t be classified as valid. A smaller clean dataset can be worth more than a larger noisy one.The second choice is how lower-cost human data and scarce robot data are combined. The company pretrains on a large volume of robot-free demonstrations to build general representations, then adds a small amount of real-robot data as an anchor to the specific machine’s dynamics. It reports that this reaches performance comparable to an all-robot dataset at roughly a 20-fold lower cost of collection, driven mainly by how much cheaper the wearable rig is than a teleoperation setup. The resulting dataset is deliberately model-agnostic, formatted to feed both action models and world models. The caveat is that the strongest results are measured on the company’s own robots and data-collection pipelines. Broader independent testing will help confirm and extend these promising results across a wider range of settings.A world model organized around eventsIn developing its world model, called WALL-WM, X Square Robot took a differentiated approach. Most action models predict a fixed-length chunk of motion from the current image and instruction. That is convenient, but it segments behavior into fixed-duration windows, so the boundaries fall where elapsed time dictates rather than where one action ends and the next begins. WALL-WM instead treats an action-grounded semantic event as its unit: a coherent piece of behavior such as reaching, grasping, or placing, something that can be named in language, seen in video, and executed as motion. X Square Robot’s world model, called WALL-WM, treats an action-grounded semantic event as its unit: a coherent piece of behavior such as reaching, grasping, or placing, something that can be named in language, seen in video, and executed as motion.X Square RobotWALL-WM’s design reflects a specific concern about not discarding what large video models already know. To achieve that, a text-to-video model is coupled to a freshly initialized action network that reads from the video features without overwriting them, which preserves the visual prior. From that one process, it offers two modes. An event mode runs in variable-length segments and suits reasoning over long horizons, while a fixed-length mode produces the steady, real-time output a controller needs. That places WALL-WM between mainstream chunk-based action models and pure video world models, keeping the predictive character of a world model while still yielding executable control.In a series of experiments, the company relied on a generalization test that is more specific than most. A model trained on a limited dataset was evaluated on long-horizon tasks in unseen settings and, on the company’s real-robot benchmark, reportedly outscored baselines that had been fine-tuned on related data. That is a meaningful result if it holds. For now, it is measured on the company’s own benchmark. With the code now being released, the broader community will have the opportunity to test, reproduce, and build on them across more settings.A policy that runs before fine-tuning, and action tokens with meaningThe action layer carries two connected ideas. The first is a requirement the company sets for itself with Wall-OSS-0.5, its vision-language-action model: The pretrained model should run on a real robot before any task-specific fine-tuning. The interest is less in the scores than in the design behind them. The model trains three objectives together, namely discrete action tokens, language grounding, and continuous action generation. And it keeps gradients flowing through all of them rather than freezing parts of the network as some rival designs do. It’s also a more strict method, since it reports untuned behavior such as approaching, grasping, and recovering, including on a deformable task held out of training. As part of X Square Robot’s Wall-OSS-0.5 vision-language-action model design, the pretrained model should run on a real robot before any task-specific fine-tuning. X Square RobotThe second idea is the action interface itself, called X-Tokenizer. Most systems that turn continuous motion into discrete tokens produce codes that the language model cannot interpret. X-Tokenizer reframes tokenization as learning a semantic interface, so that the top-level code stands for the intent of a motion while lower-level codes carry finer detail, all aligned with the language model’s own features. A useful consequence is stability. Adding noise to an action barely moves the intent code, which is what lets one tokenizer to be reused across robots without re-tuning. The tokenizer inside the production action model is a related variant of this approach. Together, the two ideas give the action layer something rather powerful: capability that transfers.The future of embodied AI stacksX Square Robot is betting that its unique approach combining three layers, each specialized in solving a key part of the problem, will stand out from other embodied AI stacks. The physical-playback step that grounds data quality is uncommon and sensible. The reframing of world modeling around events, with one backbone serving both reasoning and control, is a genuinely distinct approach. And the pairing of a deployable pretraining standard with a tokenizer designed as a semantic interface gives the action layer unusual coherence. X Square Robot’s valuation has climbed above 20 billion yuan (about US $2.9 billion), suggesting that investors increasingly view data infrastructure, foundation models, and scalable training systems as long-term differentiators in embodied AI.The next phase will bring broader validation. Much of the current evidence comes from X Square’s own robots and benchmarks. With the world model code now being made public, and as the community begins to test, reproduce, and build on the work, the reported capabilities will be tested across more robots, tasks, and settings.X Square Robot’s recent funding rounds reflect similar confidence. The company’s valuation has climbed above 20 billion yuan (about US $2.9 billion), suggesting that investors increasingly view data infrastructure, foundation models, and scalable training systems as long-term differentiators in embodied AI.What’s next for X Square RobotTo learn more about its future plans, the following Q&A with the X Square Robot team further explores the company’s technology, strategy, and vision.What made now the right moment, technically, to commit to this stack? What recently became possible that wasn’t possible a couple of years ago?It is not one breakthrough but several trends maturing together. Foundation models gave us a shared representation across vision, language, and action, so we can model what a robot sees, what it is asked to do, and how its actions change the world in one framework, rather than as separate perception, planning, and control modules. Compute and infrastructure are finally sufficient for large-scale pretraining over long-horizon, multi-embodiment data. Just as importantly, we realized that data, not model size, is the real bottleneck for general robots—what is scarce is diverse, high-quality, reproducible interaction data. And world modeling has become practical. The useful question is no longer how to predict a few seconds of video, but how to understand the ways actions change objects, contacts, and task states. Two years ago these ingredients existed separately. Today they are mature enough to work as one system.“We realized that data, not model size, is the real bottleneck for general robots—what is scarce is diverse, high-quality, reproducible interaction data. And world modeling has become practical.”Your data system captures demonstrations with a wearable VR rig and custom grippers rather than teleoperating robots. What was wrong with standard teleoperation?Teleoperation is built around controlling the robot. It forces the operator to work within the machine’s kinematics, latency, and viewpoint, and the resulting demonstrations are slower, stiffer, and less diverse. We built our system around capturing human skill instead. Manipulation is really about contact, timing, finger coordination, and recovery, not just the path the hand takes, and a wearable rig records those before the behavior is compressed onto one particular robot. It also breaks teleoperation’s expensive scaling law, in which every demonstration needs a robot. People can generate rich data independently of any robot, and the crucial property is that those demonstrations can still be replayed and executed on a physical robot through the model. Mobility is convenient, but that replay is the real point, because it is what lets the same data be reused across different platforms. In X Square Robot’s approach, demonstrations can be replayed and executed on a physical robot through the AI model, allowing the same data to be reused across different platforms.X Square RobotX Square Robot reports that its pipeline has roughly an 85 percent data-validity rate. Why is quality control such an underrated bottleneck?Because errors in robot data are far more expensive than in language data. A small timing or contact error can change what a demonstration means. If a gripper closes a fraction of a second too early, the motion still looks like a grasp, but physically it has pushed the object away. A dataset that mixes failures and accidental successes teaches ambiguity, not skill, because the real unit is the interaction, not the trajectory. So we run automated inspection, kinematic checks, and physical replay, where we play a sample of trajectories back on the real robot and count only the ones that actually complete the task. Data quality sets the ceiling on how good a policy can be. In our experience a smaller, cleaner dataset often beats a much larger, noisier one, which is why we treat quality control as part of the model, not a preprocessing afterthought.The model runs in both “event mode” and “chunk mode.” When does each matter?Both matter, for different reasons. The physical world changes through events—when contact occurs, a grasp forms, or an object slips—not in fixed-frame windows. Event mode concentrates the model’s attention on those moments, and it matters most for long-horizon tasks, like clearing a table, where progress is a sequence of semantic events rather than a smooth stream. It runs in variable-length segments that follow the task rather than a clock. Chunk mode matters for deployment. Real controllers need a stable, real-time interface, and fixed-length chunks integrate cleanly with existing control systems. We organize learning around events in the first place because a fixed window can split one motion in half or merge two together, which turns training into short-horizon pattern matching and weakens the model on long tasks. So the world model’s job is to connect event-level understanding, which is where the reasoning happens, with a fixed-length output a real robot can actually run.Why make “deployable before fine-tuning” the criterion?Pretraining should produce capability, not just a good starting point. If a model is only useful after heavy fine-tuning, then most of the intelligence still lives in the downstream supervision, not in the foundation model. Deployable before fine-tuning is a more honest test of what pretraining actually learned. A well-pretrained robot should already know how to approach, grasp, move, avoid obstacles, and correct itself. Fine-tuning should adapt it to a specific task or robot, not create the ability from nothing. It is also a practical requirement. A robot in a home or a workplace shouldn’t need a brand-new dataset and a new policy every time the task changes, so a foundation model that already carries general skill, and some ability to recover, is the minimum bar for something genuinely useful in the real world.What is the most challenging part of cross-embodiment learning?Robots differ in control frequency, delay, compliance, sensing precision, and contact dynamics, so the same instruction can require different action decompositions and recovery strategies, and a behavior that works on one arm cannot simply be copied to another. Cross-embodiment learning needs an intermediate abstraction, lower than language but higher than joint angles: how you approach an object, how you make contact, how you apply force, and how you recover from a mistake. When we say cross-embodiment, the main capability we mean is multi-embodiment generalization: transferring across robots, training on many embodiments at once, and adapting to different kinematics. Human-to-robot transfer and other techniques are specific approaches to that goal.“A robot in a home or workplace shouldn’t need a new dataset and policy every time the task changes. A useful foundation model should already carry general skills and the ability to recover.”What would you most like to see other researchers attempt to reproduce or stress-test?Three things, above all. Whether event-level representations really generalize beyond our own datasets, across more tasks, scenes, objects, embodiments, and failure conditions. Whether pretraining stays effective on robots the model never saw during training, or whether its capability is still too tightly coupled to what it has already seen. And whether real-robot evaluation can become a shared language for the field, so that we compare not just success rates but the reasons systems fail, where an instruction was misread, where perception broke down, or where recovery fell short. Robotics has been driven too often by impressive demonstrations, and real progress comes from results that are reproducible and diagnosable.What capability is still missing before robots become dependable in homes?Benchmarks measure competence, like whether a model can finish a task. Homes demand reliability, safe and consistent operation over time in a place that changes every day, with objects moving, instructions that are vague, and people interrupting. The missing piece is not a higher one-time success rate: it is robust recovery. A dependable home robot has to know when it is uncertain, when to slow down, when to ask for help, and how to bring the world back to a safe state after it drops something or misunderstands a request. In a real home, failure recovery matters more than raw success, because the home does not reset itself. Homes also demand careful personalization, learning a household’s routines and preferences over time, with safety and trust as first principles. That combination, not any single skill, separates a capable demonstration from a robot people can live with. X Square Robot’s approach is that, in a real home, failure recovery matters more than raw success, because the home does not reset itself and it demands careful personalization, with safety and trust as first principles. X Square RobotHow do the open-source components fit into X Square Robot’s World Unified Model direction?We see these releases as layers of the World Unified Model direction rather than isolated projects. Wall-OSS-0.5, the action model, asks whether an open vision-language-action model can gain directly measurable capability from large-scale pretraining, so it is the capability layer. WALL-WM, the world model, asks how a robot should understand change in the world, shifting from fixed windows to event-level modeling, so it is the representation layer. The data system supplies the interaction data that both of them learn from. Together they form a loop in which models produce capability, world models organize understanding, and the open-source community drives reproduction and improvement. World Unified Model is the broader architecture those layers support, bringing vision, language, action, and physical prediction together. We are releasing these pieces openly because embodied intelligence cannot be solved by one organization; it needs many embodiments, many real tasks, and broad feedback, and the long-term goal is a stack that keeps learning and ultimately moves robots from laboratory demonstrations toward reliable everyday use.

A practical educational guide to common and uncommon VHF propagation modes, covering thephysics, range implications, and real-world behaviors engineers need to understand.What Attendees will LearnWhy “line of sight” fails as a practical VHF planning model.How refraction, reflection, diffraction, and scattering deliver or destroy signals where geometry alone cannot predict.How tropospheric refraction extends the VHF radio horizon roughly one-third beyond optical line of sight.How temperature inversions form ducts that can carry VHF signals over 1,500 km.How sporadic E, meteor burst, and EME propagate VHF signals across hundreds to thousands of kilometers.What frequency limits, distance ranges, and environmental triggers apply to each propagation mode.How to apply this knowledge to link budgeting, interference prediction, and contingency planning.Download this free whitepaper now!

The computing community recently lost one of its enduring voices: IEEE Fellow Peter G. Neumann. The renowned computer scientist and respected risk analyst died on 17 May at the age of 93.For almost 70 years, Neumann shaped the computing field through his pioneering work on risks, system dependability, security, and fault tolerance with rare intellectual depth and unwavering ethical clarity.Five of those decades were spent as a principal scientist at SRI International in Menlo Park, Calif., where he worked until his death. A detailed narrative of his work, life, and mentoring is available on his SRI web page, where he chronicled his journey.He possessed a rare ability to identify systemic vulnerabilities long before they became widely recognized. He cautioned that interconnected systems, if poorly designed or insufficiently scrutinized, could fail and become targets for exploitation. He insisted innovation always must be accompanied by responsibility, reliability, and a clear understanding of the risks involved.With the widespread adoption of computing, information technology, artificial intelligence, and autonomous systems, Neumann’s insights have become more relevant.From Harvard to Bell LabsNeumann was born on 21 September 1932 in New York City. After graduating from high school, he pursued a degree in mathematics at Harvard, where he had a conversation that shaped his approach to research, according to the Association for Computing Machinery (ACM). In November 1952 he had a two-hour breakfast meeting with Albert Einstein, at which they discussed the importance of simplicity in design.Neumann was among the first generation of Harvard students to program computers and, remarkably for that era, enjoyed exclusive access to the computing systems.After earning his bachelor’s degree in 1954, he continued his education at Harvard, earning a master’s degree in 1955. In 1958 he moved to Germany to become a doctoral student at the Technical University of Darmstadt as part of the Fulbright program, which provides funding for U.S. citizens to study or teach abroad. He earned his doctorate in 1960.After returning to the United States, he joined Bell Labs in Murray Hill, N.J., where he worked on error-correcting codes and survivable communications. He also pursued a second Ph.D. in applied mathematics and science at Harvard, achieving that goal in 1961.Four years later, he was assigned to work on Multics, which became an influential operating system that shaped modern secure computing architectures. Multics was a mainframe time-sharing system designed to serve the diverse needs of multiple users simultaneously. Neumann designed its filing system, which featured hierarchical directories, access control lists, and dynamically paged virtual memory segments. He also played a key role in the design of its input/output system.In 1970 he left Bell Labs to join SRI.Technical contributions at SRINeumann made several seminal and foundational technical contributions while at SRI, including the following:Provably Secure Operating System. The PSOS project he worked on advanced formal methods in operating systems and computer security. The project demonstrated that security could be designed within the initial plan rather than retrofitted.Election integrity and voting systems. He outlined vulnerabilities in electronic systems and advocated for transparency, verifiability, and public accountability.Systems-level risk thinking. He broadened the concept of computer security to encompass human factors, governance, policy failures, social consequences, organizational negligence, and misuse of automation. His system-level perspective now fuels debates on AI governance and digital trust.Intrusion-detection systems. With his colleague Dorothy E. Denning, a security expert, he helped develop an intrusion-detection expert system (IDES), laying the groundwork for modern cyberdefenses.CHERI. He promoted hardware-assisted secure computing: technology that now influences next-generation processors. The Capability Hardware-Enhanced RISC Instructions (CHERI) architecture project, which Neumann led, is now being commercialized by an international, nonprofit alliance.His contributions are united by a simple but profound principle: Security should be foundational, not incidental. Neumann argued that security must be embedded into system architecture from the start—not patched after deployment.ACM’s Risks ForumNeumann’s other enduring contribution was the creation and stewardship of the ACM Risks Forum, formally known as the Forum on Risks to the Public in Computers and Related Systems. For decades, it was one of the most respected online arenas for critical reflection on computing failures, vulnerabilities, security breaches, unintended consequences, and emerging technological threats. He transformed the forum into a scholarly archive of cautionary lessons in computing failures and risks.In 1985 he started documenting how technological systems fail when complexity exceeds understanding and when society places blind trust in automation. He then moderated the community for 41 years, leaving his position in April, weeks before his passing.In 1995 he published Computer-Related Risks, a book that serves as a case-driven guide to how computer systems fail and why. It is still relevant in an era defined by AI, growing cyberthreats, and our deep digital dependence.Intellectual rigor with grace and humilityNeumann viewed computing not as an abstract technical pursuit but as a profoundly human enterprise carrying societal responsibilities. He was thoughtfully skeptical, questioned assumptions, and challenged complacency. His observations often anticipated challenges years before they became mainstream concerns.He exemplified high scholarship ideals and was intellectually honest and ethically steadfast. He had been a frequent critic of lax attitudes the industry has maintained toward both computer security and individual digital privacy. He warned against the industry’s tendency to repeat mistakes.Neumann’s signature contribution was not technical but a stance. He insisted, against industry custom, that recurring computer failures were not unfortunate accidents but rather were predictable consequences of how systems were built and sold.He was fundamentally an optimist about what can be done with research and was a pessimist about corporations.Security is not merely a technical patch, he said, but is a systemic property requiring sound design, governance, and human judgment. He consistently warned that uncontrolled complexity is itself a source of risk.His signature contribution was not technical but a stance. He insisted, against industry custom, that recurring computer failures were not unfortunate accidents but rather were predictable consequences of how systems were built and sold.Honors and recognitionsNeumann was honored with a number of honors including the Electronic Privacy Information Center’s 2018 Lifetime Achievement Award, the Computing Research Association’s 2013 Distinguished Service Award, and ACM’s 2005 Special Interest Group on Security, Audit, and Control Outstanding Contributions Award.In addition to being an IEEE Fellow, he was a Fellow of ACM, the American Association for the Advancement of Science, and SRI. In 2012 he was inducted into the Cyber Security Hall of Fame.An enduring legacyNeumann’s greatest legacy is not necessarily his inventions but his way of thinking. His longtime interest was the risk ecology of computing—the business, technological, social, political, and personal risks that computing has created, along with its tremendous benefits in each of those spheres. He left us a timely lesson: Innovation must be accompanied by responsibility, foresight, and care.Neumann was “one of the last of the old guard and a pointer to the future,” observed IEEE Life Fellow Whitfield Diffie, who helped invent public key cryptography. Highlighting both the significance and enduring relevance of Neumann’s work, a tribute by blogger Phoenix AMTD aptly said: “He spent 70 years cataloging how computers fail. We spent 70 years not listening. Maybe now we will.”Let’s honor Peter G. Neumann not merely by remembering his advice but by following it.

An examination of how satellite vulnerabilities, modern wideband waveforms, and automatic link establishment are driving renewed military and government investment in HF communications.What Attendees will LearnWhy HF (High Frequency) declined — and what has changed — How satellites overtook HF for global communications from the 1970s onward, and why growing awareness of satellite vulnerabilities to anti-satellite weapons, jamming, solar storms, and coverage gaps is reviving interest in skywave propagation as a resilient alternative.How the ionosphere enables and limits global HF communication — Understand the roles of the D, E, and F ionospheric layers in refracting and absorbing signals, the concepts of maximum usable frequency (MUF) and lowest usable frequency (LUF), and how sunspot number, solar flux index, and A/K geomagnetic indices are used to quantify and predict propagation conditions.How automatic link establishment transforms HF operability — Trace the evolution from proprietary first-generation ALE through interoperable second- and third-generation standards to fourth-generation wideband ALE, which automates frequency selection, link setup, and adaptation to changing channel conditions — removing the dependency on highly skilled operators.How wideband HF is closing the throughput.Download this free whitepaper now!

Working in isolation, especially for leaders, is rapidly becoming an outmoded idea. The modern era is defined by rapid technological advancements and increasingly complex, collaborative global challenges. In this environment, leadership can no longer be approached as an individual pursuit.Instead, leadership must be a collaborative effort in which knowledge, responsibility, and innovation are continuously exchanged across teams, roles, and areas of expertise. Success depends on the ability to foster connection, leverage diverse perspectives, and work collectively toward shared outcomes.The shift is especially important in science, technology, engineering, and mathematics fields.IEEE is bringing together emerging professionals and established experts and leaders at the inaugural IEEE International Leadership Conference to address the need for cross-generational knowledge-sharing and to equip professionals with tools for collaborative leadership. Honoring Expertise, Accelerating Potential is the theme of the ILC, scheduled for 3 and 4 October in Budapest.The conference is expected to focus on how leaders can share information across roles, adapt to rapid technological advancements, and build stronger, more connected professional communities. Through discussions, panels, and interactive sessions, attendees can examine how collaboration across experience levels and disciplines can strengthen decision-making and foment innovation.“There are several factors driving this shift [in leadership], including accelerating technological development cycles, the need to build public trust, and the large percentage of the STEM workforce approaching retirement,” says Vickie Ozburn, conference cochair. “Progress in STEM now depends less on individual brilliance and more on the ability to transfer knowledge, adapt, and make decisions that integrate technical expertise with ethical and social considerations.”From hierarchies to shared leadershipInstead of traditional corporate models rooted in hierarchy and individual advancement, a more dynamic framework is taking shape, one that views leadership as a shared ecosystem built on mentorship, continuous learning, and intentional knowledge transfer.It means recognizing that professional development is no longer a one-directional flow of experience from senior professionals to newcomers. Instead, it thrives as a multidirectional exchange. When emerging professionals, mid-career managers, and seasoned experts including retirees are brought together, the result is not only richer dialogue but also more resilient and well-informed decision-making. A cross-generational dialogue enables organizations to honor what has worked, critically assess what has failed, and thoughtfully shape what needs to evolve.Bridging experience to drive future leadershipHoward Wolfman, cochair of the IEEE ILC, underscores the importance of historical perspective in leadership development, invoking George Santayana’s enduring insight: “Those who cannot remember the past are condemned to repeat it.”“In STEM especially, this principle carries significant weight,” says Wolfman, an IEEE life senior member and the founder and principal of Lumispec Consulting, in Northbrook, Ill. “Technological innovation doesn’t happen all of a sudden; it builds on decades of research, lessons learned, and accumulated knowledge. When leaders actively connect insights from across experience levels, they gain a more complete understanding of both opportunity and risk.”That perspective reinforces the need for greater collaboration across roles and experience levels, ensuring that knowledge is not lost and is continuously built upon and applied in new ways. In this way, leadership development becomes a continuous, interconnected process rather than a series of isolated stages.STEM careers are no longer defined by linear progression but by evolving contributions, in which each phase adds value to the field’s broader advancement.What the changes mean for leaders todayAdopting a new leadership paradigm requires a shift in mindset across all levels. For senior leaders, success is defined not only by what they have built but also by the people they mentor and the knowledge they pass forward. Their legacy lies in enabling future leaders to succeed.For emerging young professionals, innovation becomes more informed and impactful when it is grounded in historical context and informed by those who have already navigated similar challenges.“Technological innovation doesn’t happen all of a sudden; it builds on decades of research, lessons learned, and accumulated knowledge. When leaders actively connect insights from across experience levels, they gain a more complete understanding of both opportunity and risk.”—Howard Wolfman, cochair of the IEEE International Leadership ConferenceFor organizations, cross-generational collaboration should be recognized as a strategic advantage, not merely an aspiration. Creating environments where knowledge flows freely and diverse perspectives are actively integrated is essential for long-term success.The evolution reframes the distinction between management and leadership.“A leader does the right thing, and a manager does things right,” Wolfman says. As the environment continues to shift, doing the right thing increasingly depends on drawing insights from across generations and experiences.Building future-ready leadership pipelinesTo build leadership pipelines capable of sustaining innovation and trust, organizations must begin asking more intentional questions:How do we create systems where knowledge sharing is continuous rather than episodic?How do we elevate emerging voices earlier in their careers?How do we ensure that experienced professionals remain engaged and valued contributors?How do we design leadership development as a collaborative, inclusive process rather than a competitive one?Ultimately, leadership cannot be tied solely to titles or tenure. It is about contributing to a continuum in which each generation strengthens the next.The IEEE ILC attendees are likely to leave the event with new insights and with a transformed perspective: Leadership is not about waiting for advancement or recognition; it is about engaging in an exchange of knowledge, responsibility, and vision, where the strength of the whole depends on the contributions of every generation.Registration for the conference opens soon.

A semitrailer that helps propel itself entered commercial road testing in late May, when a power-train kit developed by Nivalis Energy Europe, headquartered in Luxembourg with engineering operations in Germany, was fitted to a trailer supplied by the Amsterdam-based TIP Group. The self-powered trailer was handed over to the German transport operator Sommer for use in its working fleet. The Nivalis Powered Trailer Kit centers on an electric axle codeveloped with the running-gear specialist BPW, based in Wiehl, Germany. The axle, rated at 50 kilowatts-peak, is capable of both propulsion assistance and regenerative braking. It draws on a 60-kilowatt-hour, 400-volt lithium-ion battery pack charged from three sources: the axle itself during braking and deceleration, a full-rooftop array of photovoltaic panels generating up to 3.7 kilowatts-peak, and a 32-ampere, three-phase AC grid connection available during parking stops. The driver’s only window into the system is a small display readable from the cab’s side mirror that shows the system status and battery charge level. Nothing about the trailer’s handling or licensing requirements changes. The partners project savings of up to 7,000 liters of diesel per trailer per year, which is enough to keep about 19 tonnes of carbon dioxide out of the air. These figures are based on a trailer running 100,000 kilometers annually at payloads between 20 and 24 tonnes, on a mix of long-haul and hub-to-hub routes.Pavel Gilman, vice president of sales and marketing at Nivalis, breaks down where those savings come from: roughly 30 to 35 percent from the electric axle during braking and deceleration, 11 to 15 percent from the rooftop solar panels, and the remainder (roughly half) from grid charging during parking stops. The pilot is planned to run for more than a year, spanning multiple seasons. The retrofit cost has not been disclosed, and the pilot is running on a single trailer. But the underlying assumptions are now on the table and they represent a specific, high-utilization use case (meaning a truck that’s almost always on the move, filled to capacity with freight) not a universal one. Across Europe and North America, a growing number of companies have concluded that electrifying the trailer, rather than replacing the tractor unit, may be the fastest and most cost-effective path to decarbonizing long-haul freight. A new battery-electric heavy truck carries a high upfront cost and demands charging infrastructure that most freight corridors do not yet reliably provide. A retrofit kit fitted to an existing trailer is meant to sidestep both problems. The question the industry has been working to answer is whether the energy harvested from regenerative braking, rooftop solar, and grid charging in short bursts when the vehicle is parked for loading and unloading is enough to produce savings that recover the kit’s cost in a reasonable time frame. Several companies now believe the answer is yes, and they are accumulating field data to prove it—though not all of them are going about it the same way. Trailer industry places its bets The competitive landscape has taken shape most visibly in Germany. Trailer Dynamics, an Aachen-based company, has conducted field tests with BMW Logistics, DB Schenker, Duvenbeck, and Volkswagen Konzernlogistik, reporting average fuel savings of around 40 percent for diesel tractor combinations, substantially higher than the up to 18 percent reduction implied by the Nivalis projection. The difference traces directly to battery size, but Trailer Dynamics frames the choice as an economic question rather than an architectural one. “The discussion should not start with battery size, but with the economics of the transport operation,” the company said in response to written questions. “There is no single battery capacity that is universally right for every fleet.” Trailer Dynamics’s modular system offers three configurations ranging from 187 to 551 kWh, sized to match route profile, annual mileage, payload, and charging access. The M300 version, whose designation reflects the capacity of its 300-kWh lithium iron phosphate battery supplied by the Chinese battery manufacturer CATL, adds approximately 4 tonnes to the trailer, roughly three times the one to 1.4 tonnes added to a trailer by the Nivalis system. Both companies’ systems would extend the range of a battery-electric tractor by reducing the energy demand on the tractor’s motor. But Trailer Dynamics explicitly targets that use case, claiming its self-propelled trailer yields combined ranges of up to 850 km—enough to eliminate intermediate charging stops on many long-haul routes. Nivalis has not published range extension figures for electric tractor combinations, and its smaller battery and peak lower output suggest the effect would be more modest. That higher energy-storage capability widens the addressable market for Trailer Dynamics considerably and helps explain the investment flowing into the self-propelled trailer space. In November 2025, the European Investment Bank extended a €25 million loan to the company, backed by the European Union’s InvestEU program, to support commercialization. Trailer Dynamics says it plans to begin industrial-scale production in 2028, with adoption expected to accelerate as European carbon-dioxide reduction requirements tighten toward 2030. ZF, the German automotive supplier, entered the space with its TrailTrax system, using an electric axle rated at up to 210 kW continuous power. ZF claims that between onboard battery storage and energy recovered via regenerative braking, the self-propelled trailer system yields up to 16 percent in energy and carbon-dioxide savings when combined with a truck powered by an internal combustion engine. The company also says TrailTrax can reduce carbon-dioxide emissions by as much as 40 percent with opportunistic plug-in charging. Trailer manufacturers Kässbohrer and Krone have adopted the platform, as has BPW—the same running-gear specialist codeveloping the Nivalis axle. In North America, Range Energy is developing a system with up to 300 kWh of onboard energy capacity, compatible with diesel, battery-electric, and hydrogen fuel cell tractors. Range, which has announced a partnership with ZF to help drive the development and adoption of the Range eTrailer System within the North American commercial trucking industry, is now equipping its trailers with ZF’s AxTrax 2 e-axle for battery-powered propulsion. Range Energy has a separate pilot agreement with DB Schenker, the German logistics company that is also among the European operators that tested the Trailer Dynamics system. Range and DB Schenker say they plan to deploy a powered trailer in commercial trucking operations in North America, with first deliveries scheduled for later this year. The breadth of activity across continents reflects a field that has moved well past the question of whether powered trailers work. The argument now is about which architecture works best and at what cost. What the field does not yet have is a common standard for measuring and reporting savings. The figures from various pilots—an average of 40 percent from Trailer Dynamics, up to 18 percent implied by the Nivalis projection—reflect different routes, loads, seasons, and battery sizes. In some cases, they represent short validation runs rather than sustained operational data. Fleet operators evaluating competing systems are working with numbers that are difficult to interpret and impossible to rank against one another. Both architectures reduce available payload, but by very different margins. The M300’s roughly 4-tonne addition dwarfs the one-to-1.4-tonne addition of the Nivalis system. Trailer Dynamics argues the weight penalty is largely academic in practice, because more than 90 percent of trailer movements are constrained by cargo volume before they approach legal weight limits. Under current European regulations, both systems reduce payload on a one-for-one basis. Frameworks under discussion would change that. New rules could allow up to 4 extra tonnes for electric trucks, with proposals to extend the provision to electric trailers. If amended, the payload effect would turn positive for both systems. Until then, every kilogram of kit is a kilogram unavailable for freight.Small versus large battery systemsThe choice between large-battery and small-battery powered trailers is a bet on which cost will fall faster: battery pack prices or the cost of grid-charging infrastructure. A large-battery system delivers higher savings but requires reliable charging access across the operating cycle. If infrastructure buildout stalls—as it has repeatedly in heavy-duty transport—operators face the same dependency problem that has slowed battery-electric truck adoption. The Nivalis architecture hedges against that risk: Its 32-A connection requires only a standard industrial outlet, and the solar array and regenerative braking handle significant energy input without infrastructure at all. Gilman frames the design philosophy in terms of the industry it serves. “Logistics lives with low margins,” he said. “We are focused on the product which fits the industry technically and financially. It overcomes the capital expenditure hurdle and maximizes financial benefit by adding sources of energy which are symbiotic to each other.” And because Nivalis’s axle is comparatively light, he says, operators won’t be forced to reduce payload. Trailer Dynamics sees it differently. “Long-haul transport will increasingly move toward depot-based and destination-based charging models,” says Michael W. Nimtsch, the company’s managing director. “The question is not how small a battery can be made, but how much economic value each additional kilowatt-hour can generate over the life of the vehicle.” On solar and regenerative recovery, Nimtsch argues that both are useful complements to stored battery energy rather than substitutes for it.“Compared with the daily energy demand of a long-haul truck, solar generation remains relatively modest,” he says. The Nivalis energy breakdown supports that view in relative terms: Grid charging contributes the largest share of projected savings, regenerative braking second, and solar third. That hierarchy means performance depends more on charging access during dwell time than the multisource framing might suggest, even if that access requires only a standard industrial outlet. Trailer Dynamics prices its system between €145,000 and €195,000 and targets a payback period of no more than five years. Nivalis targets five to six years at current costs, falling to three to four years as volumes grow. Asked exactly what the price tag says, the company declined to answer. The minimum annual savings needed, Gilman said, is between €5,000 and €6,000 per trailer. Until someone publishes a full year of results from a trailer running in normal commercial rotation, fleet operators cannot answer the two questions that actually drive adoption: What does this cost, and when does it pay back?

Toshio Fukuda has been blazing trails for most of his career. He is considered to be one of the most prolific scholars in robotics, writing more than 2,000 research papers and authoring several books on the field. He’s an influential figure thanks to his pioneering work developing biomedical robotic systems, industrial robots, micro-nano robotics, mechatronics, and AI-driven automation.Fukuda launched one of the first robotics conferences, the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). It is still popular almost 40 years later.Toshio FukudaEmployerEgypt-Japan University of Science and Technology, in Alexandria TitleProfessor and vice president of research Member gradeLife Fellow Alma matersWaseda University, in Tokyo; University of Tokyo An IEEE Life Fellow, he is a professor emeritus in the department of micro-nano systems engineering and a visiting professor at Nagoya University, in Japan, where he taught for nearly 25 years. Currently, he is a vice president of research at the Egypt-Japan University of Science and Technology, in Alexandria, Egypt.Within IEEE, Fukuda has held top volunteer positions including the organization’s highest office: He served as IEEE president in 2020, becoming the first person of Asian descent to hold the role.He’s a former program director of Japan’s Moonshot program, which by 2050 intends to develop advanced AI robots.Born in Japan, Fukuda has been recognized by the country for his contributions to science with two of its highest awards: the Medal of Honor with a purple ribbon in 2015 and the Order of the Sacred Treasure in 2022.IEEE honored him with this year’s Richard M. Emberson Award for “distinguished service advancing the technical objectives of IEEE, especially in the area of robotics.” The IEEE Board-level award is sponsored by the IEEE Technical Activities Board. Fukuda received the award on 24 April at a ceremony in New York City.As a former IEEE president who has served as a master of ceremonies at several of the organization’s major award events, Fukuda noted that he is more accustomed to bestowing awards than receiving them.“It’s very interesting to be on the receiving end,” he says.The journey into robotics researchAs a teenager, Fukuda spent his summer breaks teaching himself how to build things including transistor radios and steam engines.“It was very nice to have a hands-on hobby and make these kinds of things myself,” he says. His experimentation led him to study engineering.He earned a bachelor’s degree in engineering in 1971 from Waseda University, in Tokyo. He says one of his professors there—Ichiro Kato, regarded as the father of Japanese robotics research—was a good mentor who made a positive impact.Fukuda’s research interests were robotics and mechatronics, a field that combines robotics, electronics, computer science, and control systems.He went on to earn a master’s degree and a doctorate in science from the University of Tokyo, in 1971 and 1977. During those years, he also attended Yale, where he conducted research on advanced control theory in 1973.He reflects fondly on his time at Yale: “It was a very nice environment and a kind of free-thinking atmosphere. It motivated me to study more.”“IEEE doesn’t care who you are, what you do, what country you are from, or whether you are male or female. IEEE accepts people who have energy and passion.”While at Yale, Fukuda served as an assistant to his advisor—which led him to consider a career in academia, he says, because he enjoyed the freedom that research work afforded him.But he realized that such freedom comes with a price. University researchers are expected to raise the money that funds their work. He compares researchers to small-business owners who have to bring in money to keep their enterprise afloat.That realization led him to select robotics as his field because he intended to develop technologies useful to industry, he says.After earning his doctorate, he returned to Japan in 1977 to work as a research scientist at the government’s Mechanical Engineering Laboratory, later renamed the National Institute of Advanced Industrial Science and Technology, in Tsukuba.“There was a lot of research going on at the lab, including practical robotics and theory,” he says.He left Japan in 1979 to become a visiting research fellow at the University of Stuttgart, in Germany. During his year there, he studied systems, software problems, and related topics.He returned to Japan and was hired as an associate professor of mechanical engineering at the Tokyo University of Science. He conducted research into practical uses for robots by visiting industrial plants. He decided to develop robots that inspect industrial equipment such as those used in assembly plants, oil refineries, and power stations—places that “can be hostile environments for humans,” he says.His work drew interest from chemical, oil, and utility companies.“I got a lot of money from them for this very practical application, which funded my research,” he says, laughing.Developing popular robotic systemsFukuda grew tired of making those robots, he says, so he switched to creating ones for scientific applications. He developed many techniques, but he probably is best known for his modular, cellular robotic systems (CEBOTs), which he introduced in 1985.He has described how CEBOTs work in numerous papers published in the IEEE Xplore Digital Library.The CEBOT system is composed of a number of autonomous robotic cells that stick together like interlocking Lego plastic bricks, he says.Each cell is a fundamental modular unit that has a function. When a simple task is given, the system can analyze it and generate the structure of the cellular manipulator. The cells connect to and detach from each other through connection mechanisms and cooperate mutually, creating complex structures and configurations.“You start developing from the component-wise to the cell-wise to a small functional unit—and then you come up with clusters that make bigger systems. We can make a society of robot beings like that,” he explained in his oral history published on the Engineering and Technology History Wiki. “It’s a distributed robotic system, a self-organized robotic system, and also an evolutionary robotic system.“It’s also a fault-tolerant robot system because if something is wrong, you just remove those things and make a new one. You keep the system working. That’s a great thing.”Today CEBOTs are used for a variety of tasks such as delivering medication in hospitals, assisting with planting crops, and transporting products in distribution centers. Check out IEEE Spectrum’s Robots Guide for news from the world of robotics.In 1989 Fukuda joined Nagoya University as a professor of mechanical engineering and micro-nano systems engineering. During his 24-year career there, he was director of the university’s Center for Micro-Nano Mechatronics. He developed a long list of technologies at the university, including many for medical applications. He also conducted groundbreaking research into intelligent robotic systems and micro- and nano-robotics.Another technology he is known for is brachiation robots, which he helped develop in 1988. He calls them monkey robots because they’re based on the pendulum-like movement of monkeys swinging from tree to tree. The gravity-based locomotion enables continuous movement.Brachiation robots now are inspecting high-voltage transmission towers and bridges, searching damaged buildings for survivors, and performing maintenance on pipelines and cables.Fukuda retired from the university in 2013 and was named professor emeritus.He didn’t stay retired for long, though. He next held a teaching appointment at Meijo University, in Nagoya, until he left in 2022 to join the Egypt-Japan University.A prominent volunteerHe joined IEEE in 1980 at the encouragement of one of his research advisors, Professor Fumio Harashima, now an IEEE Life Fellow. After attending conferences and reading the organization’s publications, Fukuda says, he looked forward to becoming more involved.“I wanted to know how to organize a conference and how to edit a paper for one of its Transactions,” he says. “I wanted to know what was going on from inside the organization, not just the outside.”In 1988 he was the founding chair and organizer of IROS, in Tokyo. The conference had 330 attendees that year, and was supported by Harashima. Today it is one of the largest and most prestigious conferences on the topic, attracting more than 9,000 people annually. Out of 120,000 conferences, it was the only conference in the Nature Index database for this year, Fukuda says.In 1996 he and other members launched IEEE Transactions on Mechatronics.He was the founding president of the IEEE Nanotechnology Council, which was established in 2002. He is considered a pioneer in nanotechnology research, particularly regarding how it relates to robotics.Over the years, he has held numerous volunteer positions on IEEE editorial boards and committees.He was the 1998–1999 president of the IEEE Robotics and Automation Society, becoming the first non-U.S. member to hold the title.He was director of IEEE Division X (2001–2002 and 2017–2018), which covers intelligent systems, biological engineering, robotics, control systems, and photonic technologies. He served as the 2013–2014 director of IEEE Region 10 (Asia-Pacific).As the 2020 IEEE president, Fukuda saw the organization through the early part of the COVID-19 pandemic. Because of travel restrictions, he realized IEEE should change how it offered its in-person services, specifically educational programs. He encouraged IEEE Educational Activities to develop an online learning platform. The IEEE Learning Network started with just three courses and now offers nearly 2,000 courses, webinars, and learning materials.An award-winning memberThe Emberson Award joins a slew of other recognitions Fukuda has received from IEEE. They include several from the IEEE Robotics and Automation Society: a 2004 Pioneer Award, a 2009 Saridis Leadership Award, and the 2011 Harashima Award for Innovative Technologies. He is also a recipient of the Board-level 2010 IEEE Robotics and Automation Technical Field Award.He says he feels strongly that IEEE should be a diverse organization that is welcoming to all. As IEEE president, he led efforts to devise a diversity, equity, and inclusion program. Several policies, procedures, and bylaws were revised to give members a safe, inclusive place for discourse.“It’s important for IEEE to make everyone feel comfortable,” he says. “DEI programs are important. All people should be equal. IEEE doesn’t care who you are, what you do, what country you are from, or whether you are male or female. IEEE accepts people who have energy and passion.“It accepted me, from the Far East. That’s why I like it.”You can learn more about Fukuda and his career from the oral history conducted by the IEEE History Center.

A practical educational guide to common and uncommon VHF propagation modes, covering the physics, range implications, and real-world behaviors engineers need to understand.What Attendees will Learn1. Why “line of sight” fails as a practical VHF planning model.2. How refraction, reflection, diffraction, and scattering deliver or destroy signals where geometry alone cannot predict.3. How tropospheric refraction extends the VHF radio horizon roughly one-third beyond optical line of sight.4. How temperature inversions form ducts that can carry VHF signals over 1,500 km.5. How sporadic E, meteor burst, and EME propagate VHF signals across hundreds to thousands of kilometers.6. What frequency limits, distance ranges, and environmental triggers apply to each propagation mode.7. How to apply this knowledge to link budgeting, interference prediction, and contingency planning.Download this free whitepaper now!

Many IEEE members who collect historical engineering artifacts often offer them to the IEEE History and Heritage group, which includes the IEEE History Center, to display. To bring these artifacts to the public, the group created the IEEE Global Museum, which curates traveling exhibits for display at conferences and in libraries, universities, and other venues.The program educates people about how technological progress has unfolded over generations, and how engineers and researchers build on past achievements to benefit humanity.Curating the exhibits has been rewarding, says Daniel Jon Mitchell, director of the group’s heritage programs.“People tell me that they are genuinely moved by having history and artifacts explained to them in an accessible, intelligible way,” Mitchell says. “When people are moved and emotionally affected by what you’re doing, they’re going to remember that. And I think that’s part of the power of what we’re doing.”The most recent traveling exhibit was on display in April in New York City during the IEEE Honors Ceremony, which celebrates engineering pioneers who have developed technologies that changed how people connect with the world. Attendees explored the Microchips That Shook the World exhibit, which drew inspiration from IEEE Spectrum’s Chip Hall of Fame. The exhibit conveys the roles integrated circuits play in fields such as signal processing, audio engineering, and telecommunications. The Commodore 64, one of the artifacts on display, stirred up treasured childhood memories for guests who had used the home computer.Other exhibits have focused on early radio inventions and power and communications technologies.The Global Museum works with IEEE societies to mark their anniversaries by interpreting and displaying pertinent items.A tribute to radio pioneer Edwin Howard ArmstrongThe idea of a traveling museum came to fruition in 2024 after Alexander Magoun, IEEE’s outreach historian, connected with Mike Molnar. The IEEE associate member owns one of six superheterodyne radio prototypes developed by Edwin Howard Armstrong, who probably is best known for inventing the FM radio system. Armstrong received the first IEEE Medal of Honor in 1917.The radio converts incoming frequencies into a fixed, lower intermediate one using a local oscillator and a frequency mixer. The technology paved the way for modern electronic communications devices. The prototype became the focal point of the Global Museum’s flagship Unseen Signals: E. Howard Armstrong’s Radio Revolution exhibit, which celebrates the inventor’s life and his impact on the broadcasting industry and wireless communications.“The radio prototype is one of the most incredible pieces that we could put on display,” Mitchell says. He and Magoun sourced other artifacts including an Audion used in Armstrong’s experiments on wireless signal amplification; a selection of consumer products that attempted to cash in on radio’s popularity, including a flour sifter and laxatives; and a Motorola Walkie-Talkie from the Korean War. They were from museums or private collectors along the East Coast of the United States.“Aside from [Guglielmo] Marconi, Armstrong is the most significant contributor to the history of radio,” Mitchell says. “The exhibit is not only a biography but also a story of the cultural and political implications his work had.”Visitors can play 15 short clips of past radio broadcasts covering politics, religion, sports, or another topic.The Armstrong exhibit was unveiled in 2024 at the National Museum of Industrial History in Bethlehem, Pa.The 93-square-meter exhibit is still traveling around the United States. It is on display until 15 August at the Pavek Museum, in St. Louis Park, Minn.From 21 November until 9 May 2027, it is scheduled to be at the Museum of Innovation and Science in Schenectady, N.Y. Entry to the museum is free for IEEE members with a digital membership card.Collaborating with IEEE societiesThe IEEE History and Heritage group collaborates with IEEE societies to create exhibits for special events. In 2024 Mitchell curated an exhibit to celebrate the 75th anniversary of the IEEE Vehicular Technology Society and its 100th Vehicular Technology Conference. The Our Mobile World exhibit was launched at the conference, held in October in Washington, D.C.“The society’s leadership helped me focus attention on key developments that meant a lot to its members,” Mitchell says.“The IEEE Global Museum wants to present exhibits that connect with its audiences, whether these are IEEE members or the public,” he says. “Just knowing what was important historically doesn’t mean that this will resonate, so I really appreciated the insight.”The exhibit’s artifacts included a Motorola DynaTac “brick” cellphone, a CB radio from the 1980s, and one of the earliest handheld GPS receivers. Visitors played an interactive game to test their knowledge spanning a century of wireless technology, motor vehicles, and mobile communication inventions.Mitchell worked this year with the IEEE Dielectrics and Electrical Insulation Society to launch a virtual exhibit, Powering Up, which is available on the Global Museum website. It provides an overview of high-voltage power engineering, and it highlights the roles that manufacturers General Electric and Westinghouse played in making long-distance, high-voltage transmission of electrical power possible. Videos and photos of impulse generators and tests are featured in the exhibit. Nvidia CEO and cofounder Jensen Huang, who received the 2026 IEEE Medal of Honor, exploring the Microchips That Shook the World exhibit.IEEE Conferences, Events & ExperiencesOne photo shows lightning arcing between high-voltage generators. Others show the impulse generators used at the 1939 World’s Fair in New York City, demonstrations of artificial lightning, and U.S. President Ronald Reagan visiting GE’s high-voltage laboratory in Pittsfield, Mass.The history of microchipsThe Unseen Signals exhibit was created for large venues, but the Microchips That Shook the World exhibit was designed to be displayed in different spaces, Mitchell says. Artifacts are premounted to ensure easy setup, and they’re encased in glass because many are rare.Microchips are crucial for signal processing, audio engineering, and telecommunications, making them a point of interest despite their small size, Mitchell says. One rare artifact on display is the Kodak KAF-1300 image sensor. Invented in 1986, it was used in one of the earliest digital cameras made for photojournalists.The KAF-1300’s image sensor chip “is credited with bringing digital cameras out of the laboratory,” Mitchell says. “Only around 500 were produced.”Visitors can understand how transistors work, he says, by pressing buttons to turn them on and off.“There are billions of transistors in modern microchips,” he notes, “and you can combine them in a way that performs logical functions.”Unseen Signals, one of two identical exhibits, was curated by Mitchell and Stephen Cass, IEEE Spectrum’s special projects editor, with help from several Spectrum colleagues. Together, they served as on-site docents for guests at the IEEE Honors Ceremony.The display also featured a preview of IEEE’s immersive “Inside the Microchip” video project, which delves beneath the silicon surface of Nvidia’s NV20 chip, using forensic photography and computer-generated renderings. The video, to be released this year, aims to teach middle school students about the microchips that are inside their gaming devices.The exhibit was on display at the IEEE Electronic Components and Technology Conference, held in May in Orlando, Fla. Later this year, members will be able to visit it at the Computer History Museum in Mountain View, Calif., and the University of Waterloo, in Ontario, Canada.The IEEE Global Museum is made possible thanks to donations to the IEEE Foundation.

The rapid expansion of artificial intelligence infrastructure is typically framed as an energy problem. Data centers are projected to consume a growing share of global electricity demand: The International Energy Agency estimates they could account for 3 to 4 percent of total global consumption within this decade.Utilities are already adjusting long-term forecasts to accommodate anticipated growth from hyperscale facilities and high-density compute clusters.This framing captures scale. It misses behavior.The emerging issue is not simply how much power large-scale compute systems consume, but how increasingly dense and synchronized computational workloads are beginning to alter the operating characteristics of the electrical grid itself through increasingly unpredictable demand that varies rapidly in both time and location, creating new operational challenges for grid operators.AI’s Capricious Energy NeedsTraditional grid planning assumes relatively predictable demand behavior. Industrial, commercial, and residential loads generally follow established profiles that can be forecast with reasonable accuracy. Even substantial demand growth has historically been manageable through reserve planning, transmission upgrades, and demand management programs.Large-scale compute infrastructure introduces a different class of electrical load. Training—the computational task of making AI models—tends to be highly synchronized across clusters of GPUs, TPUs, and specialized accelerators operating in parallel, computationally dense, and relatively scheduled. Inference—the process of actually using those models—is generally more distributed and user-driven, making demand less predictable both in time and location. Both differ materially from traditional industrial demand profiles, though for different reasons. Unlike many conventional industrial processes, these workloads can ramp rapidly depending on model training cycles, distributed compute coordination, and workload scheduling strategies.From the perspective of the grid, this is not simply higher demand. It is more abrupt demand. High-density compute workloads can produce substantial step changes in electricity consumption over extremely short intervals, including rapid fluctuations occurring within milliseconds. Data-center operators are already deploying mitigation technologies, including batteries, power-conditioning systems, and supercapacitors. Collectively, however, data centers’ rapid load changes can place additional stress on backup-generation reserves, systems that adjust supply as demand changes, frequency-control mechanisms that maintain grid stability, and local transmission infrastructure.Compute-related variability differs from the intermittency introduced through renewable energy integration. Wind and solar variability originate primarily on the supply side and is tied to environmental conditions. Compute-related variability emerges on the demand side, driven by workload synchronization, scheduling behavior, and computational intensity. The interaction between increasingly dynamic supply and demand conditions introduces additional uncertainty into forecasting, reserve management, congestion planning, and balancing operations.Research organizations including the National Renewable Energy Laboratory have emphasized the growing complexity associated with integrating highly dynamic resources into modern grid operations.Location, Location, LocationThe issue becomes more significant when compute activity is geographically concentrated. Large-scale data centers tend to cluster in regions with favorable conditions such as fiber connectivity, access to markets, tax incentives, and historically low electricity costs. Northern Virginia, often referred to as Data Center Alley, remains the most prominent example. The region hosts the world’s largest concentration of data centers and carries a substantial share of global internet traffic.Utilities operating in these regions have already identified data-center growth as a primary driver of future load expansion. Virginia-based electricity supplier Dominion Energy, for example, has repeatedly highlighted hyperscale demand growth in its integrated resource planning documents. Virginia has seen one of the largest data center buildouts worldwide. Here, Amazon Web Services and Iron Mountain data centers dominate the landscape in Manassas, Va. Nathan Howard/Bloomberg/Getty ImagesA sudden increase in electricity consumption within a constrained geographic area can stress substations, transmission corridors, and local balancing operations even if the broader grid maintains sufficient aggregate capacity. This creates localized reliability challenges that are not always visible through system-wide demand metrics alone.Thermal management systems further intensify these effects. Cooling infrastructure in high-density compute facilities must respond dynamically to changing workloads. As processing intensity rises, cooling demand rises as well, often nonlinearly. This coupling between compute and thermal systems means that fluctuations in workload can propagate through multiple layers of facility power consumption simultaneously.High-density compute clusters may also introduce power-quality concerns at the local level. Large concentrations of accelerators, switching power supplies, and high-frequency compute equipment can generate harmonics and nonlinear load behavior that place additional stress on distribution infrastructure. While modern facilities incorporate mitigation technologies, the scale and concentration of next-generation compute facilities may require utilities and operators to revisit assumptions surrounding localized power conditioning, harmonics management, and infrastructure resilience. These conditions can also contribute to short-duration electrical transients that place additional stress on localized infrastructure and power-conditioning systems.Regulations Need UpdatingPart of the challenge is that many existing regulatory and operational frameworks were designed around relatively stable industrial demand profiles. Large rapidly fluctuating loads have historically been constrained because abrupt cycling can complicate balancing operations, increase stress on transmission equipment, and reduce predictability in system operations. High-density compute clusters do not fit neatly within those assumptions.This creates pressure for both operational adaptation and regulatory reassessment.Demand-response mechanisms may allow certain compute workloads to be shifted or curtailed during periods of system stress. Data-center operators are exploring flexible scheduling, battery storage, and behind-the-meter generation. Grid operators, meanwhile, are evaluating planning frameworks and interconnection approaches for increasingly large flexible loads.The Electric Reliability Council of Texas (ERCOT), for example, has publicly acknowledged the growing implications of large flexible loads, including data centers, for long-term grid planning and operational stability. Interconnection queues across the United States continue to expand significantly, reflecting mounting pressure on both generation and transmission infrastructure. Grid expansion timelines, however, are measured in years rather than quarters.This creates a structural mismatch. Compute infrastructure can scale rapidly. Electrical infrastructure generally cannot.The broader implication is that large-scale compute infrastructure is not simply another industrial load category. It represents a shift in the temporal and spatial characteristics of electricity demand itself.Framing the issue solely in terms of aggregate energy consumption risks overlooking these second-order operational effects. Capacity expansion alone does not fully address rapid ramping behavior, synchronization, localized congestion, transient instability, reserve compression, or increasingly demanding load-following requirements.The challenge is not just how much electricity these systems consume. It is how they are beginning to change the operating conditions of the grid itself. The call is not to slow AI development but to recognize that hyperscale computing represents a new category of electrical demand. As AI infrastructure continues to scale, planning frameworks may need to account not only for total energy consumption but also for demand volatility, synchronization effects, and geographic concentration. Grid resilience will increasingly depend on understanding how these facilities consume power, not simply how much power they consume.

This article is crossposted from IEEE Spectrum’s careers newsletter. Sign up now to get insider tips, expert advice, and practical strategies, written in partnership with tech career development company Parsity and delivered to your inbox for free!You want to become a senior developer. A CTO, maybe. Start your own company, perhaps. Or maybe you just want to land your first role in tech.You will not get there from raw engineering skill alone.There’s a skill that’s quietly essential to technical leadership and yet consistently overlooked: public speaking.If you’re anything like I used to be, you’re already listing reasons not to. “I got into this to code, not to give presentations.” “I don’t want to lead.” “I’m too junior to speak about anything.” No, no, and no again. There’s a ceiling on the return from technical skill alone.I was terrified of public speaking for the first three years of my career. I wanted to hide behind code, and for the most part it worked. I did my job and did it well.Then I joined a startup where hiding wasn’t an option. The whole company was five people. I was one of two developers. I had to form opinions on our technical direction and defend them, and the CTO told me directly that I needed to speak up more.A few things happened once I did. I took more pride in my work. I said some cringe-worthy stuff, lived through the mini-anxiety attacks, and got better. To my own disbelief, I’m now an engineering manager whose job is largely speaking to groups of developers and leading presentations, online and in person.Here’s why this is worth your time:Leadership. Communicating ideas clearly, influencing decisions, and aligning your team are core leadership functions, and they matter more the further you climb.Visibility. Speaking lets you show your expertise, build a reputation, and connect with people who open doors to better roles.Durability. As automation absorbs more routine technical work, skills rooted in human interaction and judgment are far harder to replace.The good news is you can build this deliberately, in low-stakes steps.Record yourself. Use a screen-recording tool to walk through your work, explain a concept, or narrate your code. You can edit, re-record, and over-think it as much as you want. That’s the point. It gets you comfortable on camera before the stakes are real.Volunteer for demos. Next time you ship a feature or fix a bug, ask your manager for a short time slot to walk the team through it. No format for that on your team? Suggest a monthly lunch-and-learn and kick it off with a 15-minute lightning talk on something you know.Start small—really small. If your anxiety is spiking, don’t jump into the deep end. In your next meeting, ask one question. Write it down beforehand if you have to. Then be the first to break the awkward silence when someone else asks one. Developers are a famously quiet bunch, so it doesn’t take much to stand out.The further you grow, the more you’ll be expected to hold opinions and voice them publicly. So start now. Record yourself, ask questions, get uncomfortable, and notice that it gets easier every time you do it.—BrianWar Taught this Ukrainian Entrepreneur the Value of ResilienceSalome Mikadze-Struk built her tech company Movadex as an undergraduate student at the height of the COVID-19 pandemic—then kept it running during the outbreak of war in her native Ukraine. Now, she’s channeling what she learned into mentoring tech founders and speaking about the importance of resilience as AI upends the software industry. Read more here. IEEE Rolls Out Large Language Models Virtual Training CourseLLMs are now part of many engineers’ daily workflow, and the demand for technical expertise in implementing and securing the models is rising. But to build tools that work consistently, developers must have a strong understanding of the core principles that govern how the models work. IEEE is now offering a five-course program to teach how to use LLMs effectively, starting with the fundamental engineering behind the technology. Read more here. Make an Origami Circuit BoardTwo researchers at the City University of Hong Kong developed a method to make a circuit trace by simply bending a piece of paperlike material. With the right ingredients—isopropanol and liquid metal—you can make your own origami circuit board. The researchers also created a toolkit, called LiqMetCraft, with software tools and instructions to make it easy for beginners, whether in papercraft or electronics. Read more here.

I started my professional journey as an engineer before moving into product strategy and innovation leadership roles for several global technology organizations. Over the years, I have served as a mentor for a variety of programs including Products That Count’s strategic product management, Women in Product mentorship initiatives, and Alchemist accelerator programs.In 2024 and 2025 I led Walmart’s Women in Product mentorship program. I was responsible for designing and implementing the programs, including managing participant registration, matching mentors with mentees, and establishing clear standards for how they would interact.Yet for much of my own early career, I never really had a mentor.As an individual contributor engineer, I was focused on solving problems, delivering results, and figuring things out independently. I was hesitant to ask for help for fear of being judged for what I didn’t know.Part of that was also temperament. I am naturally introverted.That mindset rewarded me well. It made me self-reliant, resilient, and deeply driven. But it also had limits. Looking back, I now realize that believing I had to navigate everything alone was not always a strength. I sometimes wonder how many opportunities I missed simply because I never asked for help.As I moved into product management and later strategy roles, I began collaborating with larger teams, departments, and organizations. The work itself became more cross-functional and people-centered. Over time, I started recognizing the value of mentorship, sponsorship, and collaborative growth in ways I had not appreciated earlier in my career.I received valuable advice from different people at important moments throughout my career. Some helped me navigate conflict with more clarity. Others helped me communicate my contributions more effectively. And others gave me perspective on how to approach uncertainty, deal with organizational complexity, and avoid burnout.But those moments were not the same as mentorship. They were valuable but infrequent interactions, not sustained relationships. No one consistently guided me through difficult decisions, advocated for me with decision-makers and senior leadership, or actively invested in my long-term growth.My understanding of mentorship changed not as a mentee but as a mentor.A leadership multiplierMentorship is often seen as an act of goodwill: admirable but optional. In reality, effective mentorship can be a competitive advantage for everyone involved.For mentees, it can accelerate career growth, strengthen decision-making, and create access to opportunities that hard work alone does not always unlock.Mentorship strengthens an individual’s leadership skills, empathy, and the ability to develop future talent.For organizations, mentorship builds stronger leadership pipelines, more resilient teams, and healthier cultures of growth and trust.By getting involved, I began to understand that meaningful mentorship is not simply occasional advice or career guidance. At its best, it is an active investment in another person’s growth. It includes advocacy, sponsorship, honest feedback, visibility, and sometimes helping people access opportunities they may not have reached on their own.That is why mentorship should not be treated as kindness or incidental support. It is one of the most practical, hands-on, and personal forms of leadership.Advocacy changes careersAdvice can help someone improve, but advocacy and sponsorship can change the direction of a career.In many organizations, career growth depends not only on talent but also on access to honest feedback, influential networks, and sponsors willing to speak about someone’s potential when opportunities are discussed. Access also includes introductions to people who can recognize the value and impact of a person’s work.Sometimes the difference between advice and true sponsorship is illustrated more clearly through stories rather than through leadership frameworks. In The Devil Wears Prada and its sequel Nigel’s relationship with Andy evolves far beyond workplace advice. In the 2006 movie, he helps her grow professionally, pushes her to envision a more expansive future, and guides her through an unfamiliar industry.In the sequel—set two decades later—his investment in her success continues even though their careers diverge. When Andy (played by Anne Hathaway) is laid off during a difficult job market and struggles to find meaningful opportunities, Nigel (Stanley Tucci) quietly recommends her for a role at his firm. She is arguably overqualified for the position, but Nigel recognizes that it is the right opportunity at the right time. His recommendation helps her transition from a career in the news back into working in fashion. She can regain stability and ultimately rebuild career momentum. Over time, the opportunity becomes a turning point, reshaping her professional trajectory.What makes it meaningful is not just the recommendation itself. It is that Nigel continued paying attention to her career growth over the years, believed in her potential, and supported her when she needed it.That is what meaningful mentorship and sponsorship often look like in practice: not surface-level guidance but genuine investment in someone’s long-term growth and success.When mentors provide that kind of support intentionally, mentorship becomes more than guidance. It becomes a competitive advantage—not only for the mentee but also for the mentor and the organization.Why inclusive mentorship mattersMentorship matters because talent alone does not shape a career. Access is important. In many workplaces, advancement depends not only on capability but on guidance, sponsorship, visibility, and informal knowledge about upcoming job opportunities.Not everyone has equal access to such advantages. Research from McKinsey and Lean In suggests that women often receive less mentorship, sponsorship, and career support than men do, even in organizations that publicly emphasize inclusion and leadership development.When mentorship is left entirely to informal networks, opportunity often becomes uneven. And when it’s left to chance, opportunity also is uneven.That’s why inclusive mentorship matters. It creates a more intentional way to support people who might otherwise be overlooked.What great mentors require“A mentor is someone who allows you to see the hope inside yourself,” Oprah Winfrey once said.Great mentorship is not about having all the answers. It’s about showing up with intention. It means listening closely, being candid, and helping someone grow with more confidence and clarity.The best mentors respect their mentees’ time. They come prepared and listen for what is needed rather than rushing to give advice. They are open about their successes and failures because honesty builds trust faster than polished stories do. Great mentors tailor their guidance to the individual and encourage growth while also creating accountability.Above all, good mentors create a psychologically safe space. They make it easier for mentees to ask difficult questions, test or pitch ideas, and talk openly about issues without fear of being judged. Growth usually starts at that point.Organizations have a role to play as well. If mentorship matters, the program should be visible and supported.That can mean including it in stated expectations of leaders, creating ways to connect mentors and mentees, providing mentorship training, and recognizing outcomes that go beyond performance metrics.It also can mean broadening the understanding of mentorship. Peer mentorship, cross-functional mentorship, and even cross-industry mentorship can play important roles.The leadership gap many organizations ignorePromoting mentorship should not involve forcing artificial relationships or turning an employee’s growth into a line on someone’s to-do list. Organizations ought to promote the idea that leaders should invest in others, helping to build stronger teams, more capable leaders, and more organizational resiliency.At a minimum, organizations should ask mentors whether they helped their mentee grow in their career and whether the mentee became more confident, capable, or prepared as a result of the relationship. Did they help junior employees navigate the organization more effectively? What opportunities did they create or find to give the mentees more visibility? Did they help mentees develop communication, leadership, or decision-making skills?Those questions might be hard to quantify, but they get close to the substance of leadership.Legacy is built through peoplePeople might remember the strategies a leader shaped, the products the leader created, or the financial targets that were hit. Such accomplishments matter, of course. But another part of leadership lasts longer. It lives in the coworkers whose careers were advanced because someone took the time to invest in them.

This article is brought to you by Melbourne Convention Bureau (MCB) supported by Business Events Australia.As artificial intelligence accelerates global demand for compute, a parallel constraint is emerging with equal urgency: energy.From hyperscale data centers to electrified industries, AI is driving a step change in electricity demand. This is not a future challenge, it is a present, system-level issue requiring coordinated action across energy, infrastructure, and engineering disciplines.Around the world, the question is no longer whether AI will scale, but whether energy systems can scale with it.Melbourne, Australia is moving beyond participation to become a globally connected leader helping define how these challenges are addressed.A national challenge with global implicationsAustralia’s ambition to lead in artificial intelligence is sharpening focus on the infrastructure required to support it. Data centers are projected to account for up to 11 percent of the nation’s electricity consumption by 2035, placing increasing pressure on generation, transmission, and system reliability.At the same time, insight from the IEEE Power and Energy Society (PES) highlights that meeting energy demand from AI and digital infrastructure is one of the most significant challenges facing engineers over the next decade.The implications are clear. In addition to computing challenges, AI poses major energy systems challenges.“As artificial intelligence continues to scale globally, the challenge is no longer just computational power, it is the energy systems required to support it” —Professor Thas (Ampalavanapillai) Nirmalathas, University of MelbourneWhy Melbourne is leading on the global stageVictoria has developed one of the most advanced and integrated energy ecosystems in Australia and globally, spanning renewable generation, battery storage, grid modernization, and advanced materials.What distinguishes Melbourne globally is how these capabilities are connected and applied at system scale.The city brings together world class engineering research, a rapidly evolving clean energy sector, advanced digital infrastructure, and strong alignment between government, industry, and academia. This convergence is critical in the AI era, where energy, networks and computing systems must be designed together.Victoria’s coordinated investment across these areas is positioning Melbourne not only as a national leader, but also as a reference point in the global energy system transformation.Engineering the systems behind the AI economyThe challenge ahead is that generating more power won’t be enough, as engineers need to design systems that respond dynamically to new patterns of demand.Three priorities are emerging globally:Aligning data center development with grid capacity and renewable supplyEmbedding flexibility through storage, demand response, and system optimizationBalancing digital growth with decarbonization and long-term reliabilityAddressing these priorities requires engineering expertise to be embedded earlier in planning ensuring energy systems, digital infrastructure, and policy are designed in parallel.Melbourne’s strength lies in its ability to integrate this expertise across research, infrastructure, and real-world application. Melbourne Connect is a University of Melbourne–led innovation precinct, supported by government and industry, designed to bring together research, business and policy to deliver real-world solutions.Atlantic GroupResearch leadership shaping global solutionsAt the centre of this capability is the University of Melbourne, where interdisciplinary research is advancing the systems required to support AI driven energy demand.Through the Melbourne Energy Institute, for example, researchers are examining how energy technologies interact across entire systems from generation and networks through to end use.“As artificial intelligence continues to scale globally, the challenge is no longer just computational power, it is the energy systems required to support it,” says Professor Thas (Ampalavanapillai) Nirmalathas, Dean of the Faculty of Engineering and Information Technology at the University of Melbourne.“This is driving a new level of convergence between digital infrastructure and power systems engineering, where integrated, system level thinking is essential.”Converging energy, networks and AIMelbourne’s leadership is further strengthened by world-class interdisciplinary facilities such as the Smart Grid Lab in the Department of Electrical and Electronic Engineering, which enables real-time simulation of power systems, allowing engineers to test how solar, batteries, electric vehicles and other distributed resources interact within future grids. This supports the design of more resilient, efficient energy systems before they are deployed at scale. Melbourne’s Smart Grid Lab in the Department of Electrical and Electronic Engineering enables real-time simulation of power systems. University of MelbourneThese capabilities will become increasingly important as data centers are integrated into the grid.“AI driven demand is not only increasing computing requirements, but also placing new pressures on underlying energy systems,” says Glen Farivar, Senior Lecturer in Power Electronics at the University of Melbourne. “Designing these systems together is essential to achieving both performance and sustainability outcomes.”This reflects a critical shift. Future infrastructure must be co designed across energy and digital systems, not developed in isolation.A living ecosystem delivering real-world outcomesVictoria’s broader energy ecosystem is translating these insights into practice.Investment in renewable energy, grid infrastructure and storage is enabling higher levels of clean energy while maintaining reliability. Battery deployment is supporting the flexibility needed to manage both renewable variability and growing AI-driven demand.At its core, Melbourne offers an integrated environment where research, industry and government collaborate to solve complex system challenges.Why engineering collaboration mattersSolving the energy demands of the AI era cannot be achieved in isolation.It requires engineers, researchers, utilities, and policymakers to work together earlier and more often. More than ever, engineering collaboration is a critical enabler of future energy systems.Environments that bring together global expertise are becoming essential to how solutions are designed and delivered.“Developing future energy systems that are affordable, sustainable, and resilient is a truly grand challenge” —Professor Pierluigi Mancarella, University of MelbourneIn this context, the University of Melbourne is co-leading, alongside Johns Hopkins University and Imperial College London, one of only seven Global Centres in Climate Change and Clean Energy. Through the Electric Power Innovation for a Carbon Free Society (EPICS) Centre, the University is also the Australian technical lead in advancing future energy systems, with EPICS the only Global Centre focused on future energy infrastructure. The new Electric Power Innovation for a Carbon-Free Society (EPICS) Centre will address challenges in clean energy production and storage.University of Melbourne“Developing future energy systems that are affordable, sustainable, and resilient is a truly grand challenge,” says Professor Pierluigi Mancarella, Chair Professor of Electrical Power Systems at the University of Melbourne and Australian director and international co-director of EPICS.“As electricity grids are increasingly becoming the backbone of future energy systems, optimizing their interactions with other sectors, including AI and digitalization, and fostering interdisciplinary and international collaborations are essential,” he adds.Global conferences as part of the solutionInternational conferences are increasingly recognized as critical platforms for advancing engineering solutions at scale. Melbourne’s ability to convene global expertise is central to its leadership.In 2027, the city will host the IEEE PES Generation Transmission and Distribution (GTD) Asia 2027 Conference and Exposition, bringing together engineers, utilities, researchers and policymakers from across the world to address the challenges shaping the future of power systems. IEEE PES GTD Asia 2027 Melbourne Committee (left to right): Dr. Mehdi Ghazavi Dozein (Monash University), Dr. Glen Farivar & Professor Pierluigi Mancarella (University of Melbourne) , Dr. Mohammad Mohammadi (Australian Energy Market Operator (AEMO)).MCB“Melbourne offers a unique environment where world-class research, industry capability and policy leadership come together,” notes the IEEE PES GTD Asia 2027 Local Organising Committee, which includes Professor Pierluigi Mancarella and Dr. Glen Farivar from the University of Melbourne, as well as Dr. Mehdi Ghazavi Dozein of Monash University and Dr. Mohammad Mohammadi of the Australian Energy Market Operator.“Hosting this event creates an opportunity to advance global collaboration on the systems and technologies required to deliver the energy transition at scale.”These forums enable knowledge exchange, standards development and interdisciplinary collaboration, accelerating progress on complex engineering challenges. Attendees view a digital installation at AIME 2025 at Melbourne Connect.MCBWhy Melbourne, and why nowAs AI, electrification and digital infrastructure converge, the future of global energy systems will depend on the ability of engineers to collaborate and innovate at scale.Melbourne provides a proven platform for that collaboration, combining world-class research, a rapidly evolving energy ecosystem, and the infrastructure to connect global expertise. Melbourne Convention Bureau, IEEE Communications Society, and University of Melbourne Representatives.University of MelbourneFor IEEE members, hosting a conference in Melbourne is more than an event decision.It is an opportunity to engage with a globally connected engineering community and contribute directly to solving one of the most significant challenges facing the profession today.Through the support of the Melbourne Convention Bureau, professionals can access tailored, free support to bid for and deliver international conferences, bringing global expertise together in a city actively shaping the future of energy systems.To explore hosting your next conference in Melbourne, contact the Melbourne Convention Bureau at info@melbournecb.com.

“The lowest-cost place to put AI will be in space, and that will be true within two years, maybe three at the latest,” SpaceX founder Elon Musk told the World Economic Forum in Davos this past January, as his company was preparing to go public.Later that month, SpaceX filed an application with the Federal Communications Commission for an orbital data center constellation of up to 1 million satellites in low Earth orbit, 500 to 2,000 kilometers above Earth. And just three days before the IPO, he discussed some initial design specifications for a new AI-1 satellite data center in a video interview.Musk is prone to hyperbole when it comes to timelines. Full self-driving cars by 2017. First human mission to Mars in 2024. Ten thousand Optimus humanoid robots by the end of 2025. Et cetera. For orbital data centers, which he says will be a cost-effective alternative to terrestrial data centers within three years, the math won’t make sense for several years, if ever.Consider this: There are roughly 14,500 active satellites in orbit. Musk’s Starlink constellation accounts for about two thirds of those. Both the launch cadences and satellite-manufacturing capacity would have to scale up astronomically to deploy a million orbital data center satellites.For context, there have been roughly 7,000 orbital launches in all of human history. To loft 1 million satellites into low Earth orbit on SpaceX’s Starship, which is designed to carry up to 60 satellites per vehicle, would require 16,666 launches exclusively devoted to satellite deployments. Considering that SpaceX launched a record 165 orbital missions in 2025, even at 10 times that cadence, it would take a decade. And how long would it take to build 1 million satellites, given Starlink’s current pace of around 4,000 per year and a generous tenfold increase in capacity? Short of a manufacturing revolution, try 25 years.The reality is that the vision of massive constellations of orbital data centers is nowhere close to being realized.As this month’s cover story, “Why Orbital Data Centers Are So Hard” by Andrew Cavalier of ABI Research, makes clear, the reality is that the vision of massive constellations of orbital data centers is nowhere close to being realized.Dina Genkina, IEEE Spectrum’s computing and hardware editor, put the idea into perspective: “Starcloud (a startup that has applied to the FCC for an 88,000 orbital data center satellite constellation) sent one Nvidia H100 GPU in space so far. Their radiator was too weak to let the chip run at full power.”As Cavalier shows, cooling even a single Nvidia H100 GPU in space is difficult: It draws 700 watts, which will require 1.4 square meters of radiator at 60 °C. A 40-kilowatt rack of servers will need an 80-m² radiator; a 100-megawatt data center will require 2,500 of those radiators. Some astronomers are understandably concerned that a million satellites with giant radiative wings would blot out the stars.So if the economics doesn’t make sense, if the chips are at the mercy of the radiative ravages of space, and if humanity will lose its view of the stars, not to mention increasing the risk of triggering the Kessler syndrome, why are the hyperscalers hyping orbital data centers?Genkina offered the obvious answer: sweet, sweet moolah. “The Elon Musk part of it is honestly genius because he’s got xAI building the data centers, SpaceX sending them to space, and Tesla building solar panels,” Genkina says. “It’s almost like he’s paying himself.”Two Analyst’s Views of SpaceX’s Proposed AI1 Data Center SatelliteMichael Pierce, Principal at Technology Strategy PartnersMusk’s timelines are notoriously overly ambitious, but I think SpaceX’s orbital data centers might reach cost parity with terrestrial data centers in 5 to 10 years. The Starlink laser-link network already exists as the communication backbone for any SpaceX compute constellation, and that infrastructure is what no new entrant can replicate quickly. The chip-agnostic payload design probably reflects their disclosed difficulty securing AI silicon as much as any modularity philosophy. My view is that the only realistic near-term application is a SpaceX mega-constellation for inference. Training workloads likely cannot tolerate the synchronization and latency constraints of a distributed orbital system.Our report analyzed the market from the integrator’s vantage point, but AI1 is what it looks like when one player has assembled all the necessary advantages simultaneously. The question is whether the terrestrial data center industrial base will degrade or improve on economics. I don’t have insight into SpaceX’s internal costs, as opposed to public pricing, on all their components, so it’s hard to say if they’ll completely dominate or not. Even if they are not cost competitive with terrestrial data centers for another 5 to 10 years, it may simply be faster to get new compute that just happens to be in space.Matt Hasan, AI strategist and independent consultantMy initial view is that AI1 does not fundamentally change the rationale for space-based data centers as much as it changes the timeline and scale. The underlying drivers remain the same: escalating AI compute demand, growing power constraints on terrestrial grids, and the desire to colocate energy generation with computation.What AI1 does signal is that the concept is beginning to move from theoretical discussion toward engineering and capital allocation decisions. The announcement adds credibility to the idea that hyperscale computing infrastructure may eventually expand beyond terrestrial constraints rather than simply competing for increasingly scarce grid capacity on Earth.That said, significant economic and technical questions remain. Launch costs, maintenance, hardware replacement cycles, thermal management, latency-sensitive workloads, and overall system economics will ultimately determine whether space-based data centers become a mainstream extension of AI infrastructure or remain a niche capability for specialized applications. The key development is not that these questions have been resolved, but that major industry players now appear willing to invest resources toward answering them.

In the 1970s, American Fireworks, a family-run pyrotechnics company in Hudson, Ohio, used a “home run box” to offer quick and easy fireworks displays for the Cleveland Indians (now the Cleveland Guardians) baseball games.The red wooden crate had metal silos to store the rockets. Each switch on the control panel allowed the operator to set off a different firing sequence. This setup instantly triggered the display whenever a Cleveland batter hit a home run. Before computerized firing systems became common, panels like this represented the state of the art. But they did not eliminate human error. On 15 September 2015, the technician in charge of the Indians’ pyrotechnics accidentally set off the fireworks when the opposing team hit a home run. The embarrassed technician was caught on camera holding his head in his hands. This home run box and control panel [left] were used to launch fireworks during Cleveland Indians games. The rockets were housed in metal silos within the box.Left: Jahna Auerbach/Science History Institute; Right: American FireworksThe Early History of FireworksFireworks are one of the many Song Dynasty inventions that migrated from China through the Middle East and into Europe by way of trade routes. Around 200 B.C.E, the Chinese invented small firecrackers by simply tossing pieces of bamboo into a fire. The air inside the bamboo would expand and crack the wood, and the pop supposedly scared away evil spirits. After the invention of gunpowder—a mixture of sulfur, charcoal, and potassium nitrate—about a thousand years later, some clever person thought to pack the powder into the bamboo tubes and ignite them, launching the first fireworks—and the first rockets—into the sky. John Bate’s popular 1634 book on fireworks described fire wheels [left] and a flying dragon [right], consisting of a dragon-shaped rocket that sped along a rope. SSPL/Getty ImagesBy the Renaissance, specialized schools for pyrotechnics had emerged across Italian city-states, and European craftsmen began creating large spectacles for royal occasions and religious celebrations. In 1634, John Bate published the four-volume series The Mysteries of Nature and Art, the second of which described how to create all manner of fireworks. Woodcut illustrations showed fire wheels (now called pinwheels or Catherine wheels), as well as the more ambitious flying dragon—a rocket shaped like a dragon that emitted sparks while speeding across a rope strung between two buildings.During the 18th and 19th centuries, chemists and alchemists discovered new chemical compounds and isolated new elements that expanded the palette for fireworks. Adding barium nitrate produced green, for example, and strontium nitrate produced red. Chemists also mixed in metal particles to create sparkles.The 1880s saw the introduction of the loud screech or whistle that precedes the exploding boom. Amédée Denisse, a graphic artist by trade and a fireworks hobbyist, discovered that a cardboard tube containing potassium picrate added that satisfying auditory effect to his fireworks display.How Did Fireworks Become a 4th of July Tradition?British colonists brought fireworks to the Americas. In 1608, Captain John Smith set them off to celebrate the founding of Jamestown, Virginia, the first permanent English settlement in what would become the United States. More than a century and a half later, while the Continental Congress was meeting in Philadelphia in July 1776, future U.S. president John Adams speculated in a letter to his wife that Independence Day would be celebrated “with pomp and parade, with shews, games, sports, guns, bells, bonfires and illuminations from one end of this continent to the other.”Although Adams got the day wrong—he mistakenly thought the committee would complete the revisions to the Declaration of Independence by the 2nd of July—he was correct in foreseeing that Independence Day would be celebrated with lots and lots of fireworks. Just a year later, on 5 July 1777, the Pennsylvania Evening Post reported on the grand exhibition of fireworks the previous night, which began and concluded with 13 rockets representing the 13 colonies.It’s safe to say that the United States is still obsessed with fireworks. According to the American Pyrotechnics Association, the country spends about US $3 billion on fireworks each year; it’s also the leading importer of fireworks. As the U.S. gears up to celebrate its 250th birthday this 4th of July, expect to see fireworks displays everywhere, from kids with sparklers running in backyards to ambitious professional displays for huge crowds. Modern fireworks displays like the Macy’s 4th of July celebration in New York City are computer choreographed and controlled. Roy Rochlin/Getty ImagesFireworks today are an engineering marvel. State-of-the-art displays are computer controlled with precise digital timing, often tied to musical accompaniment. Designers can spend weeks choreographing complicated patterns and assigning launch times, shell types, and colors. The completed script is uploaded to an electronic firing system, which consists of the control panel and hundreds or thousands of firing modules that connect to the rockets. It can take days to set up the launch site for a large-scale display that lasts just minutes.For example, last year more than 60 licensed pyrotechnicians worked for 12 days to arrange more than 80,000 shells for the Macy’s 4th of July Fireworks in New York City. Each of the firework shells measured up to 25 centimeters in diameter and weighed more than 13 kilograms—a far cry from their bamboo ancestors. More than 120 kilometers of wire connected the bundles of explosives to twelve computers. All that for a 25-minute display.As much as I unabashedly love fireworks, they’re not for everyone and they do have a downside. The explosions can trigger PTSD for military veterans, and they can also upset animals. Every year, thousands of people are injured by mishandled or damaged fireworks. Known to set off wildfires, fireworks are often banned during droughts. Scientists who’ve studied the environmental impact of fireworks displays have noted their tendency to disperse airborne metallic particles and other harmful particulates. A drone light show over Busan, South Korea, shows a member of the K-pop band BTS.Hwawon Ceci Lee/Anadolu/Getty ImagesPerhaps to counter those drawbacks, or maybe it’s just the next technological evolution in aerial display, companies are now offering drone light shows. Fleets of hundreds or thousands of LED-toting drones can be programmed to hover in the air and fly in formation, forming logos and other designs that are more stable than exploding fireworks.These exquisitely choreographed light shows are truly impressive. And yet I relish the full sensory experience of fireworks, including the booms, the smoke, and the smell. So whether you’re celebrating your country’s birth, Guy Fawkes Day, Saint Sylvester’s Night, New Year’s, Diwali, or simply cheering a home run from your favorite team, I hope you get to enjoy this millennia-old technological marvel.Part of a continuing series looking at historical artifacts that embrace the boundless potential of technology.An abridged version of this article appears in the July 2026 print issue as “Rooting for the Home Team.”ReferencesThe American Pyrotechnics Association is a professional organization that encourages safety in design and use of all types of fireworks, provides industry support, and promotes responsible regulation.Barry Sturman and David Garrioch’s 2023 article “Amateur Science and Innovation in Fireworks in Nineteenth-Century Europe,” in the journal Ambix, provides a detailed history of the development of fireworks. Kathy De Antonis’s 2010 article “Fireworks!” for a publication of the American Chemical Society explains the colors, shapes, and packaging of modern fireworks.If you happen to find yourself in Philadelphia before the end of July, check out the Science History Institute’s exhibit Flash! Bang! Boom! A History of Fireworks, which is part of the U.S. celebrations around the semiquincentennial. The home run box shown in this article is part of the institute’s collections.

He was born into a storm, lightning split the summer sky, in avillage the world had not yet heard of.The midwife called it a bad omen, his mother called it a sign. Your firstlife began in a storm, under open sky.One winter night you ran your hand along a cat’s back, and thedarkness cracked open with sparks.Your mother warned the house could burn.You were already chasing what you learned: Light would return.Your second life came underwater, in the current deep. No light,no air, the river pulling you under,the surface closing above you without a sound, andsomething in you refused to sink or sleep.Your third life came at the dam.The water rose. The wall held you in place.One flash, you turned your body and rose back into air, and leftthe weight of water without a trace.Your fourth life came in stone and dark. Entombed for anight in a mountain chapel,visited by no one. Only silence and the memory of a spark. You calledit an awful experience and left it there, untold.Your fifth life came in fever,nine months cholera held you down,until your father said: Survive, and choose your own ground. You rose.Not from the prayer, but from the promise he made.Your sixth life came in silence, and it stayed.Every sound cut through you, a clock three rooms away,a ringing that would not leave, a noise you learned to bear, until youlived inside that noise and made a home in there.Your seventh life burned on Fifth Avenue, not your body, but your work. Not a thiefof fire, but one who stayed with the blaze.A modern Prometheus, your life’s work turned to ash,“I must begin again,” you said, and turned to new ways.Your eighth life came in the street.No storm. No warning. A taxi struck without a sign. Asudden impact: ribs breaking, breath gone.No diagram this time. Only the body, slow to keep up.The ninth life came on quiet wings.That dove found you in the dark, and your spirit rose. She didnot move. A beam of light fell from above.The life you would not return from, the one you loved.Your mother thought you had nine lives, nine closebrushes with death.Each close call, a lesson. A hand that would lead you out of thedarkness and into the dynamo of eternal light. The world profitsfrom the mystery of your mind,Upon your imagination we stand.

Today, you probably asked a question of a large language model, or accepted a connection suggestion on LinkedIn, or watched a recommended video on YouTube, or took a different route to work based on a traffic prediction from Google Maps. In other words, you probably used artificial intelligence. But what you might not know is how much energy that interaction consumed or why. AI requires processing massive amounts of data, which is usually done in large data centers populated by thousands of GPUs capable of executing up to trillions of operations per second. But each of those GPUs achieves that by consuming as much as 1,000 watts apiece. For comparison, if you’ve got a newer smartphone, it probably uses less than 1 W. That kilowatt figure puts GPUs on the same level as vacuum cleaners, dishwashers, and stoves, but with the big difference that data-center processors are operating uninterrupted around the clock.Fundamentally, a lot of this inefficiency is because GPUs are trying to simulate the workings of artificial neural networks using software and billions of transistors, which requires using energy to move massive amounts of data. What’s more, the simulated artificial neurons that make up these networks lack even a fraction of the complex computing behavior of the biological neurons that comprise the most energy-efficient computing system that we know, the human brain.The brain is roughly one million times as energy efficient at many of the comparable tasks we set for AI. To try to approach these efficiencies, a radically different way of computing called neuromorphic engineering is seeking to build electronic components and circuits that act more like the brain’s neurons and the synapses that connect them.Huge amounts of work have gone into making electronics operate more like biological neurons and synapses. Some research has focused on developing new, experimental devices, but they aren’t yet reliable enough to be used in large systems. Other efforts aim to implement neurons and synapses by interconnecting many complementary metal-oxide-semiconductor (CMOS) transistors—the workhorses of digital logic—to simulate a single neuron and synapse. But this approach requires so many transistors (and a few bulky capacitors) that it greatly limits the size of the system that can be constructed, making it unclear how such brain-inspired hardware could ever scale up and compete with state-of-the-art GPUs.But all along there was an artificial neuron and a synapse—each a single device—hiding in plain sight. We found them last year. They were each made possible by an ordinary CMOS transistor—and not even a very good one at that. This is the story of their accidental discovery and their great promise for lowering the environmental footprint of AI.Biological and artificial neuronsModern digital electronics is based on producing and manipulating the ones and zeros of the binary code through the operation of metal-oxide-semiconductor field-effect transistors. MOSFETs have evolved in recent years, but their classic form consists of a piece of silicon that has been doped to contain an excess of either positive (p-type) or negative (n-type) charge carriers. (CMOS logic contains transistors of both types.) The device has two terminals connected to the silicon through regions highly doped with the opposite polarity of the rest of the silicon—the source and the drain. Another terminal, the gate, sits atop the silicon that separates the source from the drain. The gate itself doesn’t connect directly to this silicon, instead resting above a thin layer of insulating dielectric.Notably, there is a fourth terminal that attaches to the bulk of the silicon; think of this bulk terminal as connecting to the underside of the chip. It doesn’t typically get much attention, but it’s very important to our story.When voltage is applied at the gate and the bulk terminal is grounded, charge carriers of the same polarity as the source and drain are attracted to the channel region. In the case of an n-type source and drain, that will be electrons; for p-type it will be holes. The presence of these charges forms a conductive channel that reduces the resistance between the source and the drain by several orders of magnitude, and the device switches on. As the voltage at the gate increases, this physical phenomenon produces a current signal that, when plotted against the gate voltage, rises steadily. This response is ideal for logic gates, converters, multiplexers, memories, and other digital circuits. But it is not a good fit for mimicking the behavior of a neuron.In real neural tissue, brain cells, called neurons, consist of a cell body, a long projection called an axon, and short branching projections called dendrites. The suite of behaviors and computing this collection of components is capable of is rich and broad, but the portion that artificial neural networks hope to copy is this: When the cell body’s voltage is perturbed enough to reach a particular threshold, a self-propagating pulse of voltage, called an action potential, shoots down the axon. The axon terminates in a synapse, an electrochemical connection between the axon and another neuron’s dendrites. The action potential will then temporarily boost the voltage of this next neuron, by an amount that depends on the strength of the synaptic connection. If enough action potentials reach these dendrites in a given time—from this neuron or from others that might also form synapses there—the cell body’s voltage will surpass the threshold and trigger its own action potential.The MOSFET NeuronThe unusual action the authors discovered is understandable if you consider that a MOSFET contains a hidden bipolar-junction transistor.TRANSISTOR BEHAVIORUnder normal operation, with the bulk terminal grounded, increasing voltage at the drain leads to current that increases steadily. When the voltage decreases, current follows the same sloped path. Although some pairs of electrons and holes are created by current crashing into silicon atoms, these are swept away before they can accumulate.NSRAM BEHAVIORAdding resistance to the bulk terminal means these extra holes pile up, increasing the bulk voltage relative to the source. Once that voltage reaches a certain value, the hidden transistor activates, causing current to spike. Current remains high until the drain voltage drops past a certain point. To get closer to the behavior of real neurons, artificial neurons should produce a current spike when a critical voltage threshold is crossed and then quickly relax back to a resting state on their own. This spike needs to be sudden—nonlinear. It should also exhibit some hysteresis; that is, the activation and relaxation voltages should be different from each other to ensure that current flows only for a certain amount of time.What’s wanted from an artificial synapse, the thing that connects two artificial neurons, is less complicated, but equally important. The main thing is that its conductance can be electronically adjustable. The device’s conductive states should increase and decrease in a linear pattern and remain stable over time.No single MOSFET working under the standard operation mechanism can reproduce either of these neural properties. Instead, it’s been done by combining them into complex circuits. Until now, each neuron and each synapse has been implemented by interconnecting dozens and sometimes even hundreds of MOSFETs, which is highly inefficient in terms of area, performance, and cost. To limit the amount of space needed, chips can multiplex their signals, sending them to neurons and synapses serially, but such sequential processing introduces additional delays.Despite these area-and-time penalties on tasks such as audio processing, computer vision, or health monitoring, state-of-the-art brain-inspired microchips have achieved power reductions up to a thousandfold compared with those of GPUs or CPUs on the same task. If we could create neurons and synapses from individual devices that are readily manufacturable instead, we might target more massive implementations while maintaining energy efficiency.Reinventing the MOSFET for AIWorking in our laboratory in 2024, one of my students was measuring a memory circuit that consisted of one transistor and one memristor—a type of nonvolatile memory device first fabricated in 2008. The student’s memristor circuit was built from two-dimensional material atop a silicon microchip containing MOSFETs. The MOSFETs were created in a commercial foundry using fabrication technology called the 180-nanometer node, which was cutting-edge in the year 2000.One day the student forgot to connect the bulk terminal of the transistor. What he observed was a sudden increase in current with high nonlinearity that self-relaxed when the voltage was ramped down (a phenomenon called a hysteresis loop). This was a very promising neuronlike behavior!After a fruitless week of trying to think of an explanation for this behavior, I (Lanza) asked Pazos, then my postdoctoral fellow, to try to observe and control this phenomenon in chips without memristors. This time, we applied pulses of voltage—like the spikes a neuron would produce—instead of the ramped voltage that my student used when he first saw the peculiar behavior.Pazos’s new data helped us understand what was going on. The key was that oft-ignored fourth, or bulk, terminal of a MOSFET. Under ordinary operation, many mobile charge carriers flitting through the channel collide with the silicon atoms, producing free pairs of electrons and holes—a process known as impact ionization. The electric field created by the potential difference between the source and the drain causes these new free electrons to drift toward the positively biased drain and the holes to move toward the bulk terminal, which is usually grounded, removing the charge without any drama.However, when the bulk terminal of the transistor is floating—unconnected as it was in my student’s experiment—the holes produced by impact ionization cannot be driven to the ground. Instead, they accumulate in the bulk of the silicon, increasing its voltage. Then things start to get interesting.It helps here to imagine a MOSFET as two different kinds of transistors occupying the same physical space—the intentionally constructed MOSFET and a hidden, bipolar junction transistor. A bipolar device transmits a current signal across two p-n junctions, in this case the interfaces between the source and the channel region and the channel and the drain. This signal is in proportion to a smaller current at a third terminal in between, called the base. In our experiment, that third terminal is the bulk.To get current flowing through a bipolar transistor, you need a big enough potential difference between the base and one of the other terminals, so that current can get across the p-n junction. Let’s say this “threshold voltage” is 0.7 volts, although the real number depends on device geometry and silicon doping. In our device, that potential difference comes from those holes that were accumulating in the bulk, because it was not connected to ground. Once it reaches the threshold voltage, the device becomes sharply conductive, producing an abrupt increase of current. This sharp current increase eventually falls off once the drain voltage is lowered, because that lowering reduces the rate at which holes are generated in the bulk. The remaining excess holes recombine with stray electrons or leak away, and finally the bulk voltage falls. This cycle of hole accumulation, current spike, and hole removal gives rise to a hysteresis loop, very much like the electrical behavior of a biological neuron as it integrates ionic currents, fires a spike, and relaxes back to its resting voltage.Initially, we observed this behavior only in a few transistors, and the relaxation time was very different for each of them. So, to try to control it better, we adjusted the resistance of the bulk terminal using a second MOSFET. Simply setting that resistance suddenly caused all the transistors to fire at the same voltage with hardly any variability. In other words, we found we could create perfect electronic neuron behavior in a single silicon transistor by controlling the bulk contact resistance. Setting the resistance can be done by doping the silicon during fabrication, but we think the two-transistor cell—where one acts as the bulk resistance—offers much greater versatility because it allows for electronic control.We had to make sure the phenomenon would last, otherwise such a device would be useless. To our delight, every single one of the devices we tested worked over 10 million cycles. Not even one of them failed during our tests.The MOSFET SynapseTo be honest, we were amazed. Dozens of research groups and companies all around the world have spent many millions of U.S. dollars over the past 20 years trying to emulate these neural behaviors using experimental memristor-like devices and other things, with limited success, mainly due to reliability and cost issues. We managed it in the cheapest and most industry-standard device: the MOSFET. This result was so shocking that we decided to confirm it using microchips from a different foundry. It was successful: All the behaviors could be reproduced, and perfect yield was achieved once again.We were happy with the results and had started the process of filing for a patent and writing up our findings for the journal Nature, when our lab made another astonishing discovery: The same kind of MOSFET could act as a synapse, too!Recall that in ordinary operation some electrons crash into silicon atoms to create pairs of electrons and holes. We noticed that at specific values of bulk resistance a significant amount of the charge from this impact ionization would get trapped in the gate dielectric. This trapped charge interferes with the flow of current through the MOSFET, effectively changing the device’s conductance. Importantly, this new conductance is stable and adjustable at will. It was then that we realized the MOSFET could also be used as an electronic synapse.As it was in the neuron transistor, the bulk terminal was the key. A negative bulk-source voltage drives electrons into the dielectric, decreasing conductance. A positive one pushes holes in, increasing it.From neuromorphic device to circuit to systemHere’s how the MOSFET synapse and the MOSFET neuron, together called a neurosynaptic random-access memory, or NSRAM, could work together to achieve a simple neural circuit: Say you had a circuit consisting of three synapse MOSFETs and a neuron MOSFET. The synapses have already been programmed as we’ve described, so that each has a different conductance. Spikes of voltage with different patterns and frequencies are applied to the gate of each of these transistors. What emerges from their drains are spikes of current with amplitudes modulated by the synapses conductance values.The spikes converge at the drain of the neuron MOSFET. With each spike, impact ionization causes charge to build in the bulk of the silicon. Some of it will drain away, but if enough spikes arrive in a short enough period of time, the bulk voltage will reach a value at which the “hidden” transistor triggers a spike of current through the MOSFET. This current would then go on to become the input to other MOSFET synapses, and so on. The behavior is exactly the kind of integrate-and-fire action real neural circuits deliver.The competitive advantage of our single-MOSFET electronic neurons and synapses is straightforward: We can produce with only one or two transistors the electronic signals that today require, at an industrial level, dozens and sometimes even hundreds of components. And moreover, unlike other emerging technologies, our solution is fully compatible with today’s silicon manufacturing lines and exhibits a yield of 100 percent in key figures of merit with near-zero variability.Building functional circuits for brain-inspired computing and AI based on this technology is as exciting as it is laborious. It will require us to improve our computer models to resemble the behavior of both devices more accurately and to do so with computational efficiency. We must also perform accurate circuit- and system-level simulations to validate computing architectures, design peripheral circuitry to drive and convert signals, and undergo multiple fabrication rounds to optimize performance.But all that will be worthwhile, because it could result in brain-inspired microchips for AI with better energy efficiencies than what we have now. These chips will first be a fit for smaller-scale, “edge-AI” tasks, such as bringing greater intelligence to battery-powered systems. But if we can scale up such chips, maybe in the long run they can compete with state-of-the-art GPUs.

In 1839, J.M.W. Turner painted The Fighting Temeraire. The old warship, once a hero of the Battle of Trafalgar in 1805, glides like a ghost across the canvas, towed by a small steam tug belching smoke on its final voyage to the ship-breakers. The image shows a clear moment of change: sail giving way to steam, and with it, a major shift in power. The ship relied on timber, rope, canvas, and Britain’s seafaring towns. The tug depended on coal mines and iron foundries that supplied machine shops in the Midlands. Turner showed the tension of this time, when new technology changed who held power.By Turner’s time, the United States had already defeated Britain’s navy in two wars—one for liberty on land, another for freedom of the seas. The 13 colonies used new technology in creative ways to win their freedom, and by keeping up with innovation, they managed to defend their freedom. Now, as the U.S. celebrates its 250th anniversary, we can ask: What does it really mean for a country to be independent? We tend to focus on how nations and individuals defend freedom but rarely turn that focus to the tools and systems that sustain freedom. Declaring independence is only the beginning: Independence must still be engineered.Forging freedomLong before the first shots were fired at Lexington and Concord in 1775, Britain had drawn the lines of conflict through technology. The Wool Act of 1699 choked colonial textile exports. The Hat Act of 1732 crushed local hat-making. The Iron Act of 1750 forbade finished iron goods. Each statute tightened the knot: Colonial capability existed only at Britain’s discretion. The Boston Tea Party may have been a loud response, but resistance also took subtler, more empowering forms. At a 1769 Virginia ball, more than a hundred women arrived in homespun gowns. Every thread was defiance.When war came, everyday tradespeople pivoted to the fight. Farmers turned plowshares into gun barrels, while clockmakers turned their precision skills to making firing mechanisms. By 1777, two weapons production models had emerged—centralized sites like the Springfield Armory that could produce high-quality guns in large quantities, and household workshops that were more agile and could meet local needs. In parallel, the new nation developed an equally important source of supplies and support: France sent gunpowder and loans and eventually opened a second naval front in 1781, which proved as decisive as any weapon. After the war, the young republic pursued industrial strength with the same resolve it had shown in battle. In 1789, Samuel Slater arrived from England with textile-spinning technology that he’d memorized, sowing the seeds of U.S. manufacturing, whose early growth rested on domestic cotton, slave labor, and copied techniques. By 1816, gun manufacturer Simeon North’s milling machines were producing interchangeable metal parts, allowing the armed forces to cannibalize parts. In 1822, Thomas Blanchard’s copying lathe automated the shaping of gunstocks. In the 1830s, the federal government imposed tariffs that shielded infant industries, fulfilling Alexander Hamilton’s vision for industrial policy: Build capacity first, then compete.At the 1851 Great Exhibition in London, American revolvers and reapers with swappable parts stunned international observers. By the 1860s, land-grant colleges were spreading technical education across the nation. Engineering moved into the mainstream, from niche to national necessity, driving broad, though uneven, prosperity. As the Industrial Revolution bloomed, the early U.S. focus on industrial capacity via farms, factories, and formidable wealth positioned the country to compete with the most advanced industrial powers in the world.The right and responsibility to repairFor nearly two centuries, that ethos endured, with government-guided infrastructure and markets deciding the details. But around the U.S. bicentennial, in 1976, a conviction took hold across party lines. Finance began to outrank fabrication, and Wall Street prioritized futures contracts over companies owning the factories that made up their supply chains. Domestic factories closed or moved offshore, and companies turned to just-in-time manufacturing and shipping, ostensibly as a way to save on costs. Shipbuilding felt this shift as much as any industry. Shipyards closed, and suppliers of specialized castings and components disappeared along with them, as did skilled technical workers who retired without replacement. Now the U.S. Navy struggles to build submarines fast enough to replace its aging fleet. Other changes took hold, among them the idea that the company that builds your tractor or medical equipment could prevent you from fixing it yourself. Invasive “terms of service” prevented customers from reaching for a wrench, instead allowing companies to keep reaching into customers’ pockets. These changes are symptoms of both structural and infrastructural fragility. When we lose the ability to understand and sustain the systems we rely on, we lose control—bit by bit.RELATED: Why We Must Fight for the Right to Repair Our ElectronicsNo nation can build everything alone, of course. From hand-forged muskets to finely printed microchips, the sovereignty etched into our tools demands a prudent calculus: what to make at home, what and with whom to trade. Engineering is how a nation keeps its independence alive. Independence requires both the courage to innovate and the stewardship to maintain what has been built. The American Revolution was itself an act of engineering—daring in vision and deliberate in pairing anvil and alliance. Generations later, can a nation that cannot see its own dependencies, build and maintain its critical tools, or repair what breaks still call itself free?Turner’s Snow Storm—Steam-Boat off a Harbour’s Mouth, completed three years after The Fighting Temeraire, captures this part of the story. Sea and sky dissolve into a churning vortex around the ship. Turner claimed he had himself lashed to the ship’s mast for four hours so that he could paint the sensation of standing inside a system too vast and tangled to comprehend. A nation that loses sight of what it depends on stands there too: lashed to nothing except the churn.

What do an instinct to fix things and the 1999 global panic over whether computers would survive the date change to 2000, known as the Y2K bug, have in common? Both helped shape IEEE Senior Member Ajay Prasad’s career.Prasad is an industry process director at Dassault Systèmes in Detroit. His focus is global oversight of industry process experts specializing in Enovia, a product lifecycle management (PLM) solution and one of the company’s flagship products.Ajay Prasad Employer Dassault Systèmes in DetroitTitle Industry process directorMember grade Senior memberAlma maters Bangalore University, in Bengaluru, India; and the University of Birmingham, EnglandAs a child growing up in Bangalore, India, his curiosity to build real-world solutions was ignited by his father, a mechanical engineer. Prasad’s father often fixed things around the house, including cars and bicycles. His ability to take something broken and return it to working order laid the groundwork for his son’s career in engineering.Prasad was in his final year of undergraduate studies when the Y2K panic hit its peak.“Nobody knew what would happen when the year turned to 2000,” he says, “and it was almost projected like the end of the world was coming.”The phenomenon left him with the desire to fix computer problems, but he wasn’t sure how he would go about it, as he had no background in computer science.As it turned out, computer systems didn’t crash when the 1900s ended. The world did not end on Jan. 1, 2000, and neither did his interest in how computers worked.The consulting pivot that changed his careerPrasad graduated in 2000 with a bachelor’s degree in industrial engineering and management from the RV College of Engineering, in Bengaluru. It was at a time when tech companies were heavily recruiting engineers, regardless of their specialization.“They were mainly looking for problem-solving skills,” Prasad says.His parents expected him to immediately enroll in a master’s degree program, he says, but a job offer from Tata Consultancy Services in Bengaluru to work as an assistant systems engineer trainee changed that plan.“My dad was actually out of town for work when the job offer came in,” he says. “I knew he wanted me to stay in school, but honestly, I was done studying for a while. I wanted to get some work experience.”He accepted the offer, then broke the news to his father. His parents were supportive of his decision, but his dad offered one piece of advice: Keep the idea of an advanced degree in the back of his mind.Several months of working on mainframes helped him understand algorithms and how to code to achieve outcomes, he says, and the more he learned about computer systems, the more he wanted to pursue a computer science career. With a solid engineering foundation, he says, he knew the pivot made sense. But he also wanted the academic credentials to back up his tech skills.Heeding his father’s advice, he paused his career at Tata and enrolled in the master’s degree program in computer science at the University of Birmingham in England. At the time, it was one of the few schools offering the program to students who had no undergraduate computer science degree. When he graduated in 2002, he briefly considered pursuing a Ph.D., but he returned to India and a new role at Tata.Building a global perspectiveAs a systems engineer, he worked on the MatrixOne platform, a PLM software solution that helped manufacturers oversee products from design to launch. He spent a lot of time customizing the MatrixOne software to meet customer needs. The experience gave him insights into the pain points that different users of the platform faced, such as managing complex product data across large teams and keeping track of complicated supply chains.In 2004 Tata transferred him to Minneapolis, where he continued working on the MatrixOne platform.During that time, Dassault acquired MatrixOne and folded it into its existing Enovia product line. He remained involved with the product until he left Tata in 2008. To scratch an entrepreneurial itch, he became a consultant for the product, helping customize the platform for U.S. clients.The move also forced him to make a decision: He needed to choose between settling in the United States or returning to India. Inclement weather made up his mind, he says.“I was heading to my next project across the country, and it was winter,” he says. “During the entire drive, I was trying, unsuccessfully, to outrun a massive snowstorm. I was young, and it was an adventure, but it helped clarify where I wanted to be at that point in my life.”He returned to India in 2010, armed with a more global perspective and expertise with Enovia. As he looked for a job, he focused on a role with the company that owned the platform he’d worked on for years.“Dassault Systèmes has continuously pioneered new technologies and concepts and set benchmarks in the PLM space,” he says. “When an opportunity opened up there for me, I jumped at it.”Instead of a programming role, though, he was hired as an Enovia technical sales specialist, working in Dassault’s Bengaluru location. It was an eye-opening experience, he says.“It put me on the other side of the table: trying to sell software to customers,” he says. “This was the opposite of my experience customizing software after the sale was complete.”The role of technical salesThe position involved both presale and postsale duties. Technical salespeople bring subject-matter expertise that bridges the gap between a product’s functionality and the customer’s needs. The role works directly with the sales team to craft a presentation that showcases the value of the software as a solution.On the postsale side, technical sales professionals work with service teams to customize software solutions to ensure customer goals are met. If needed functionality doesn’t exist, they work with the R&D group to create it. They also offer suggestions to customers on how to improve their processes.When Prasad stepped into his new role, a senior colleague described technical sales as an “exam syndrome” because customers are judging you and your presentation against competitors. The analogy didn’t land well with him.Recalling all his years of formal education, he had a different perspective: “I wanted to think of it more as an opportunity to fully understand a customer’s problem, then solve it better than anybody else could.“Every customer has unique pain points. When I can offer solutions that deliver value, they’ll buy the software.”It’s his belief that the position is best served by professionals with both engineering and computer science backgrounds. He advocates that engineering students consider adding computer science to their studies, and he draws on his own educational experiences to support the position.Combining engineering and computer scienceDassault recognized the value in his approach. In 2015 he was hand-picked to be part of the company’s new Worldwide Enovia Center of Excellence team in Auburn Hills, Mich. As an industry process expert, he was able to put his Enovia expertise into action.He’s now a senior leader managing a global technical sales team. One of his objectives, he says, is advocating to engineers that technical sales is a viable career move.“The moment an engineer hears the word sales, they tend to stop listening,” he says. “They don’t want to be a salesperson in the traditional sense.”That’s too narrow a view, he says, adding: “I think everyone is a salesperson to some degree.”If engineers looked at technical sales differently, they’d see an exciting opportunity, he contends.“In this role, they have the ability to not only develop solutions but also explore the why behind the need for a solution at all,” he says.“As engineers, sometimes we are so focused on engineering concepts and principles that we get bogged down in the details and don’t focus on what the problem really is,” he says. “I learned with technology that even before you try and create a solution, you need to understand the logic of the problem first.”From problems to patentsHis approach has delivered measurable results. He holds one patent and has a second under consideration. His combination of engineering and computer science expertise played a crucial role in each, he says.His first patent, granted in 2023 by the U.S. Patent and Trademark Office, was for his solution to improve product benchmarking for clients with large-scale data management issues. It replaces traditional spreadsheets with powerful databases and a user-friendly interface, ensuring information is up to date, accessible, and shareable.“I think that being part of the IEEE community is a huge value for folks in the engineering space. It’s a great way to collaborate and to understand what’s happening, especially in your local ecosystem.”His second patent, pending with the USPTO, is designed to help customers manage large projects that involve a high volume of engineering design tasks. Instead of relying on ambiguous communication between engineers and project managers, his solution would draw data from the work management system and update the project management dashboard automatically. It would replace guesswork with real-time data.Prasad has authored the peer-reviewed technical paper “Transforming Product Development With a Platform-Based Approach to Product Lifecycle Management,” which was published by SAE International. His writings on the use of data tracking and AI in product lifecycle management have appeared on Engineering.com and in Wavelengths, a monthly publication from the IEEE Southeastern Michigan Section.In February, Dassault marked Prasad’s success by promoting him to worldwide Enovia industry process director. The title reflects a career built on the belief that engineering and computer science are stronger together, and that technical sales is where the combination delivers its greatest value, he says.The value of IEEEPrasad first encountered IEEE at a student branch meeting he attended at Bangalore University in 2000, shortly before graduation. The meeting featured engineers from industry discussing the work they did—which sparked his interest in joining, he says. But with his first job waiting for him, the timing wasn’t right to become active with the organization.It took nearly 25 years, he says, before he felt he had enough spare time and professional experience to contribute actively and meaningfully to IEEE. He joined the Southeastern Michigan Section in 2024, was quickly elevated to senior member, and then took on a leadership role.He was nominated to be conference chair for this year’s Innovative Applications of AI in Industry event. Together with a team of eight, he led the planning and execution of the in-person conference, the first time it was held since the COVID-19 pandemic shelved it.The event explored how AI is permeating practically every aspect of our lives. Speakers came from Amazon, Torc Robotics, academia, and health care.The event was a success, he says, and he hopes to parlay its momentum into a multiday conference in the coming years.As a representative from the section, he served as a technical judge at this year’s Robofest, a competition held in May for students in Grades 4 through 12. Since the annual event’s inception, more than 40,000 students from 35 countries have participated. He says his involvement helps him understand how students use robotics to solve problems.“I think that being part of the IEEE community is a huge value for folks in the engineering space,” he says. “It’s a great way to collaborate and to understand what’s happening, especially in your local ecosystem. There’s always something going on in terms of a conference or a talk where you can listen, gain knowledge, and network. It’s also an invaluable opportunity to discover where you can add value at IEEE.”

This article is brought to you by Capital One.After five years leading natural language understanding and eventually the entire Alexa AI organization at Amazon, Prem Natarajan made a nontraditional move: He became Chief Scientist at a bank. Not just any bank: Capital One, a financial institution serving over 100 million customers, helping everyday Americans manage their financial lives.For Natarajan, a veteran of DARPA-funded research and academia who had watched machine learning evolve from task-specific applications to foundation models, the logic was clear. Some of the most interesting advances in AI research and deployment were shifting from big tech’s horizontal platforms to industry verticals like finance, where the most complex problems aren’t just building models but making AI work under the constraints of real-world customer problems, contextual business knowledge, continuous learning, with an incredibly high bar for accuracy and privacy.That’s also what made Capital One the right place to do it. For decades, the company has been recognized as one of the most data- and analytics-driven financial institutions in the industry. Its business model from the very beginning was built around using data and technology to personalize financial products for customers. A decade ago, Capital One went all in on the cloud and rebuilt its data ecosystem, creating a unified environment for data, compute, and AI and machine learning experimentation. Today, its modern infrastructure, disciplined approach to governance, and deep bench of talent form the foundation that allows it to lead in enterprise AI.Advances in AI research and deployment are shifting from big tech’s horizontal platforms to industry verticals like finance.So, why does a bank need a Chief Scientist? The answer lies in a fundamental misconception about AI in financial services. Most financial institutions still view AI as a technology to deploy – leveraging the latest large language model, deploying it through APIs, and integrating it into existing workflows – rather than a scientific discipline. Capital One is doing something different: building a scientific community and research organization to solve real-world customer problems and invent impactful AI solutions that don’t yet exist.While widely available foundation models can handle general tasks, they can’t yet solve many domain-specific challenges, such as detecting fraud in real-time across billions of transactions, or providing state-of-the-art conversational tools so customers can engage when, how, and where they want to.These challenges of making AI reliable, scalable, and well governed require original research and scientific innovation that is funneled back into the business to create real-world applications to address customer needs.The Constraints That Demand Innovation Prem Natarajan, an IEEE Fellow, is Chief Scientist at Capital One. “If you want to solve really important problems in AI and see your work come to life, this is one of the few places you can do that,” he says.Capital OneBecause banks are dealing with people’s finances, there is an incredibly high bar for getting it right when it comes to AI. Take fraud, for example. Even a minor fraud event can have a devastating impact on certain customers. The best fraud models and platforms can detect and help mitigate fraud in the time it takes someone to tap their card, which is table stakes for protecting customers and their financial information with accuracy and speed. Looking at these types of challenges, Capital One and Natarajan saw that serving millions of customers meant solving AI problems at a scale and complexity that many enterprises don’t encounter. These same constraints create a unique research environment.At Capital One, the approach to building AI is to provide value to customers in ways never possible before, improving their financial lives and meeting them where they are with services they actually need. That focus, combined with massive scale and world-class risk management requirements, makes the scientific problems both harder and just as consequential as those found in most big tech labs.Advancing AI Through “Destination-Back Thinking”Capital One’s approach to AI research and innovation starts with what Natarajan calls “destination-back thinking.” Rather than asking what’s possible with current technology, the team envisions the customer experience they want to deliver – perhaps a car buyer who works long days and can only research the options at 10 p.m., or a customer facing an unexpected expense who needs immediate, personalized guidance – and then works backward to identify the scientific breakthroughs required to get there.“You’re thinking back from where you’re providing incredibly valuable services,” Natarajan explains. “Once you have that vision clearly, you work back and say, what are the gaps? What are the things we need to invent?” This ensures that when problems are solved, the impact is essentially guaranteed, because the team has already identified what will make a tangible difference in customers’ lives.But methodology alone isn’t enough. Capital One’s nearly 15-year bet on cloud-first architecture created something rare in financial services: a unified data and compute ecosystem that can support the kind of scientific experimentation typically seen in big tech research labs. As the only major U.S. bank to go all-in on public cloud infrastructure, Capital One eliminated the legacy systems that can constrain AI research at most financial institutions. This modern tech stack enables rapid iteration, large-scale model training, and what Natarajan calls “continuous learning,” systems that improve after deployment rather than degrading over time. This unique approach to infrastructure is a critical component in making new categories of research possible.Agentic AI: From Research to ProductionThe research agenda manifests in systems already serving customers. Early last year, Capital One launched what may be the first fully agentic AI customer service experience built entirely in-house by a bank: a car buying tool that takes actions on behalf of customers based on their requests, not just answers questions. Behind it lies extensive research into multi-agentic AI reasoning systems that can navigate real-time data, business knowledge, constraints, and guardrails, with various agents that can work together to accomplish complex tasks.Capital One has launched a fully agentic AI customer service experience powered by extensive research into multi-agentic reasoning systems that can navigate real-time data.The team is also working on solving things like tokenization challenges, protecting sensitive data while enabling model training. To accelerate this cutting-edge work, Capital One has established partnerships with Columbia University, the University of Southern California, and the University of Illinois, and became the only bank funding NSF’s national AI research centers in 2025, investing millions in initiatives that span mental health, materials discovery, science, technology, engineering, and mathematics education, human-AI collaboration, and drug development.In the spring of 2026, the company hosted its inaugural AI Symposium to deepen connections and foster insight-sharing between the scientific AI community, leading AI labs, startups, and its own technology, science, and AI leaders and partners.Building a World-Class AI OrganizationCapital One is building the next generation of AI talent. Join the team inventing impactful AI solutions to shape the future of finance. Learn more at https://capitalone.science/External validation suggests the strategy is working. Evident AI ranked Capital One as the leading bank in AI talent and a global leader in AI innovation for three consecutive years, noting the bank accounted for 38 percent of all AI patents filed by the top 50 financial institutions. Capital One was also recognized by IFI Insights as the only financial institution among the top U.S. patent leaders in agentic and generative AI in 2025, alongside the likes of Google, NVIDIA, DeepMind, IBM, Microsoft, Intel, Adobe and Samsung. Capital One’s AI team – which has experience from leading AI labs and top universities – represents expertise rarely found outside Silicon Valley.But recruitment requires a mission. “If you want to solve really important problems in AI and see your work come to life, this is one of the few places you can do that,” Natarajan says. The pitch is consistent: Capital One isn’t just optimizing algorithms for niche financial applications like high frequency trading, it’s using science to enhance financial experiences for over 100 million everyday Americans, expanding engagement and real-time insights, personalization, and access to their personal finances and products like never before.Capital One was recognized as the only financial institution among the top U.S. patent leaders in agentic and generative AI in 2025, alongside the likes of Google, NVIDIA, DeepMind, and Microsoft.The frontiers Natarajan is most excited about – agentic AI systems that can dramatically improve performance by reframing how problems are solved, and domain-specific reasoning that understands contextual and financial nuance – represent the next phase of innovation. “By just casting the problem in an agentic framework, you can actually get way more performance” from the same underlying models, he explains.It’s this kind of applied research, like translating general capabilities into production systems for millions of customers, that defines the Chief Scientist’s mandate. When recruiting talent to his AI team, a group comparable only to the most sophisticated tech companies in caliber, Natarajan frames the opportunity around a mission. He invokes Steve Jobs’ famous challenge to John Sculley: “Do you want to spend the rest of your life selling sugared water, or do you want to change the world?” For Natarajan, the parallel is clear. Building AI systems that transform financial services for millions of everyday Americans – that’s changing the world. And it requires the kind of scientific rigor that only a Chief Scientist can lead.

In the mid-noughties, when music by the Killers and Franz Ferdinand blared out of every pub and nightclub I passed, I spent my days and nights struggling through a Ph.D. in applied mathematics. My research focused on simulating how special light waves interact in liquid crystals and using simple equations to approximate and understand those interactions. When I look back at my thesis now, liquid crystal technology is old hat, and I imagine my work could be completed with AI assistance in a matter of days—maybe hours. But the same cannot be said for the work of the pure mathematics Ph.D. students with whom I shared a cramped office at the University of Edinburgh. At the time, I felt sorry for these colleagues, who day after day sat at their desks, seemingly tearing their hair out and making no progress. (Though I was struggling too, I was at least always making some headway.) When we finished and went our separate ways, some hadn’t even published a paper.Now, in hindsight, I finally understand why they toiled for years on abstract mathematical problems that only a handful of people in the world care about. It wasn’t arrogance, as I thought at the time; they weren’t trying to prove their superior intelligence by being the first to solve a seemingly intractable mathematical problem. It wasn’t even a form of masochism (which was my second guess)—penance for some imagined inadequacy. I realized they derived joy, satisfaction, and meaning from the long journey toward understanding.“Sometimes, understanding just strikes you as being very beautiful.” —Jeremy Avigad, Carnegie Mellon University“Sometimes, understanding just strikes you as being very beautiful. Sometimes it’s a feeling of accomplishment, like completing a marathon,” muses Carnegie Mellon University mathematician Jeremy Avigad. “But it’s not quite either of those: It’s just a wonderful feeling when you’ve been thinking long and hard about something complex, difficult, and then—all of a sudden—it just comes together.”This feeling has driven mathematicians throughout history. Likewise, the way mathematicians pursue that feeling has changed little over the centuries. They notice or imagine links, patterns, or properties in numbers, shapes, or logical structures. From this, they write conjectures—unproven statements of their speculation. They or other mathematicians then use logical reasoning and the tools of mathematics in often creative ways to prove or disprove those conjectures. Finally, yet other mathematicians verify (or challenge) the proofs.Invariably, this process requires a whole heap of thinking time. “I went to a pure maths camp with classes where we would sit with hard maths problems for half an hour and no one would say anything—everyone was just thinking,” says Krystal Maughan, a mathematician and computer scientist about to get her Ph.D. at the University of Vermont. “But then we would work together and kind of tease out the problem.”This is the age-old joy of math in action. But today’s AI systems are starting to make inroads into bypassing this slow, deliberative process. Taking this trend to its logical conclusion, what happens if AI makes the mathematician’s struggle completely unnecessary? Might AI even sideline humanity completely?AI’s Growing Role in MathematicsFor decades, computation has accelerated mathematical progress. This began 50 years ago, when mathematicians used a computer to prove the four-color theorem, which asks whether any map can be colored using no more than four colors, with no adjacent regions sharing the same color. The answer is yes, and the computer proved it, controversially, by checking 1,936 cases in a way no human could realistically verify.Yet throughout this computational era, even in proofs relying on massive computational resources, the role of the human mathematician has remained central. Humans propose conjectures, guided by intuition. They devise strategies to prove them, guided by creativity and experience. And humans verify whether those proofs are correct.Now AI is challenging the status quo. In just a few years, large language models (LLMs) have evolved from “stochastic parrots,” capable of little more than regurgitating basic mathematics scraped from the internet, into advanced mathematical reasoning machines.Last summer, systems from Google DeepMind and OpenAI reached a level equivalent to the world’s most mathematically gifted high school students, achieving gold-medal status at the International Mathematical Olympiad. In this annual competition, contestants must solve six notoriously difficult problems from various areas of mathematics.Earlier this year, Google DeepMind’s experimental AI system Aletheia achieved an even more significant milestone when it autonomously produced publishable Ph.D.-level research results. While the work itself is obscure mathematically—calculating structure constants in arithmetic geometry—the significance lies in the complex reasoning it displayed in tackling an unsolved mathematical problem. And more recently, a new general-purpose AI system from OpenAI disproved an important conjecture in combinatorial geometry. This result would have been worthy of publication in a major mathematics journal if humans had been the authors, and top mathematicians hailed the feat as a milestone for AI in mathematics, demonstrating independent, original, and sophisticated thinking.Another shift has come from combining LLMs with mathematical tools known as proof assistants, which have been around for more than a decade. These systems—such as Isabelle, Lean, and Rocq—are specialized programming languages that check mathematical proofs step-by-step, verifying their logical correctness. Traditionally, mathematicians have had to translate their theorems and proofs into this machine-readable format by hand, a laborious process known as formalization. Now, LLMs are starting to remove this bottleneck, automating the translation of informal proofs into formal code that proof assistants can verify.From Human Proof to Formal ProofEuclid’s famous proof that there are infinitely many prime numbers appears very different when formalized in Lean, a proof assistant. Human mathematicians routinely skip steps and rely on shared understanding; formalization makes every assumption and inference explicit so a computer can verify the proof.HUMAN PROOF We want to show that for every natural number n, there’s a prime p that is at least n. Consider the smallest prime factor of n! + 1. Call it p. It is obviously prime. To show p is at least n, assume, for contradiction, that it is not.p then clearly divides n!, so it also divides (n! + 1) − n! = 1. But this is impossible: p is prime, and 1 has no prime divisors. So p is at least n.LEAN PROOF /- Euclid’s theorem on the **infinitude of primes**. Here given in the form: for every `n`, there exists a prime number `p ≥ n`. -/theorem exists_infinite_primes (n : ℕ) : ∃ p, n ≤ p ∧ Prime p :=1let p := minFac (n ! + 1)have f1 : n ! + 1 ≠ 1 := ne_of_gt <| succ_lt_succ <| factorial_pos _2have pp : Prime p := minFac_prime f1have np : n ≤ p := le_of_not_ge fun h =>have h1 : p ∣ n ! := dvd_factorial (minFac_pos _) h3have h2 : p ∣ 1 := (Nat.dvd_add_iff_right h1).2 (minFac_dvd _) pp.not_dvd_one h2 ⟨p, np, pp⟩❶ Definitions must be explicit. The proof formally defines p as the smallest prime factor of n! + 1 before it can use that quantity.❷ Formal proofs build on earlier formal proofs. Here Lean invokes a previously verified theorem showing that p is prime.❸ Hidden logical steps become explicit. A human mathematician can write that p “clearly” divides 1. Lean requires the proof to invoke a formal theorem about divisibility and show exactly why that conclusion follows.With technical assistance from Sidharth HariharanVersions of such systems, sometimes called reasoning agents, are becoming highly sophisticated. In February, for example, the AI company Math, Inc. used its aspirationally named reasoning agent Gauss to formalize a proof that had earned the mathematician Maryna Viazovska, of EPFL, in Switzerland, a Fields Medal in 2022. Gauss first helped human mathematicians complete the formalization of Viazovska’s solution to the 8-dimensional sphere-packing problem in a matter of days, and then autonomously formalized the more complicated 24-dimensional case in just two weeks.Such achievements suggest that AI is already capable of handling some mathematical tasks long considered uniquely human. As the technology advances, more of the day-to-day work of human mathematicians is likely to become fair game for AI.Mathematicians Debate AI’s Role in DiscoveryHuman mathematicians could become “priests to oracles.” —Yang-Hui He, London Institute for Mathematical SciencesIn September 2025, I attended the 12th Heidelberg Laureate Forum—an annual conference that brings hundreds of young mathematicians and computer scientists together with their intellectual idols. AI dominated the conversation and, from the get-go, tension was in the air.Speakers described a future in which superhuman AI mathematicians transcend human knowledge and capabilities: forming conjectures, searching solution spaces, proving conjectures, and finally verifying the proofs and generalizing the results, all without human involvement. If this future comes to pass, Yang-Hui He of the London Institute for Mathematical Sciences memorably declared, human mathematicians could become “priests to oracles.”While such startling predictions were being voiced on stage, my gaze was drawn to the audience. Frowning, fidgeting, and exchanging furtive glances—the crowd’s unease was palpable. Trill White, a student at Australia’s Deakin University, later recalled sitting in that hall and thinking: “ ‘That’s devastating. What will people have to contribute to mathematics? Will it become something that no one understands?’ I did get a sense that this is going to change everything.”“We certainly started realizing AI has the potential to replace us.” —Jessica Randall, Google Developer GroupsJessica Randall, a South African mathematician for Google Developer Groups, says she sensed a collective existential dread rising among the young mathematicians. “I could feel everyone was worried, because they hadn’t thought that far ahead,” she says. “It was like a big bombshell that hit us, and we certainly started realizing AI has the potential to replace us.”Some established mathematicians, including He, seem comfortable with AI taking on tasks that are currently the preserve of human mathematicians. That’s because they just want to know the answers to the biggest questions in mathematics—such as the six remaining Millennium Prize Problems—even if AI does it all. “A lot of mathematicians are pragmatic and just want to understand. They would sell their soul for the solution to a problem,” jokes Avigad. “Whatever it takes, right?”But this “just want to know” camp is by no means the only faction: Most mathematicians do not hope or expect AI to replace them entirely. Instead, two broad alternatives are emerging. The first is a human-centric aspiration that prioritizes human understanding of mathematics and treats AI as a tool, much like a calculator. The second is a collaborative “teamwork makes the dream work” vision, where humans and AI work together to tackle problems neither could solve alone.The Human Role in MathematicsNumbers are “a way of bringing us to agreement.” —Akshay Venkatesh, Princeton UniversityFields Medalist and Princeton mathematician Akshay Venkatesh has been thinking about this topic from the human-centric viewpoint for years. In 2022, he used his Fields Medal Symposium to implore the mathematics community to deeply consider what AI might mean for the practice of mathematics. At the time, the idea that AI could replace mathematicians seemed far-fetched. Now, he says, “we’re reaching the point where, for at least some tasks with abstract mathematical reasoning, computers are becoming competitive with humans.”For Venkatesh, the question is not just what computers can do, but what mathematics is for. “Sometimes I think when we use numbers, it’s not so much that we are describing phenomena that are intrinsically numerical, but that we can all agree exactly what the numbers mean,” he says. “It’s a way of bringing us to agreement.”Mathematician and machine learning expert Maia Fraser, of the University of Ottawa, shares this sentiment. She says the joy she derives from mathematics is something distinctly human that integrates the subconscious and conscious mind. She describes starting with an intuitive sense that a certain thing should be true and gradually bringing out something that she can express in a rigorous proof. Communicating and sharing these deep-born thoughts is “a form of collective intelligence that is something beautiful about the human spirit,” she says.By these arguments, an AI proof of a mathematical conjecture that has stubbornly resisted human efforts would be useful only if comprehensible to humans. “That the statement can be proved by AI is already useful information,” concedes Fraser. “But then it’s still an open problem to come up with an elegant, beautiful human proof.” Even if no such proof exists, she says, searching for it “is still a valuable endeavor.”AI and the Future of Mathematical CollaborationA more collaborative approach to AI in mathematics comes from Terence Tao, who first competed in the math Olympiad at the age of 10. In 1986, 1987, and 1988, he won bronze, silver, and gold medals, respectively, making him the youngest winner of each of the three medals in Olympiad history. Now a Fields Medalist and professor at the University of California, Los Angeles, he has earned a reputation as one of the most gifted mathematicians alive.Unlike some of his peers, Tao is neither dismissive of AI nor fearful. Instead, he sees it as the catalyst for a fundamental shift in the discipline—a transition toward what he calls “big mathematics.” He envisions a future of large-scale, decentralized collaborations between humans and machines, where complex mathematical tasks can be diced and sliced, with humans claiming the creative parts and AI doing the lion’s share of the technical grunt work.Three Futures for AI in Mathematics AI as a toolAI as a partnerAI as an oracleRole of AIAssistantCollaboratorAutonomous researcherWhat matters most?Human understandingShared discoveryAnswersAlready, Tao is experimenting with this concept, working on problems alongside scores of online collaborators, some using AI tools. “A hundred years ago, almost every mathematics paper was single author,” he says. “But now I collaborate with people I’ve never met—and maybe in the future, I won’t even know if they are AI or real people.”The key to Tao’s vision is uniquely mathematical: formalization. When a proof is translated into code and checked step-by-step by proof assistants, it removes any chance of human error or dishonesty. This approach changes how collaboration works, because trust is established through verification rather than reputation or rapport. An idea from an unknown researcher or even an amateur can be taken seriously if it has a formal proof.“If it wasn’t for this formal verification layer, opening projects up without any safeguards would just be a disaster,” adds Tao. “But in math, we can completely check and verify outputs, and this really filters out a lot of the rubbish.”The Risks of AI in MathematicsFrom the young researchers at the Heidelberg Laureate Forum to some of the biggest names in the field, mathematicians all seem to agree on one point: AI has the potential to transform their discipline. But there’s far less consensus on what that transformation will mean in practice.Some worry about the accessibility of AI tools. Traditionally, mathematicians have required little more than intuition, training, and a pen and paper to advance their field. If this slow, deliberative process is no longer valued by society, and particularly by research funders, then mathematics could become an elitist activity, only practiced by select organizations that can afford to work with proprietary AI models.Another concern is motivation. As AI systems take on more of the work, the incentive to engage deeply with difficult problems may weaken. Princeton’s Venkatesh says that the long human process of formulating and understanding a proof may be hard to justify, not just to funders, but even to mathematicians themselves. “There have been times where I’ve spent years thinking about something, and I’ve slowly struggled to understand it,” he says. “If your computer can do large chunks of that for you, will you have the motivation to spend that time?”That concern extends to the next generation. If students can use AI to jump straight to answers, they most likely will. But every time they skip the struggle, they miss an opportunity to build the foundations of their own unique intuition. Over time, some worry, the next generation of mathematicians may suffer from a form of intellectual atrophy, unable to think outside the AI box that trained them.In response to such fears, the mathematics community is taking action. Individuals are writing essays, organizing workshops, and debating in journals, while institutions and community groups are developing guidelines for how AI should be used in research and publication. Indeed, mathematicians are applying the same rigor and curiosity that they use every day to reckon with the challenges of AI. Taken together, these efforts reflect a broad effort to try to retain control over the direction of mathematics in the era of AI.So, is AI sucking the soul out of math? In one way, it is doing the opposite. It is forcing mathematicians to confront deep questions about what mathematics is, why they have devoted their lives to it, and the purpose math serves in society. At the same time, though, it is reshaping the practice of mathematics in a way that may be difficult to reverse.“Mathematics makes me a better problem solver at normal problems, because it frames my mind to think in a very logical, rational way,” says Randall, who noted the existential dread at the Heidelberg Forum. “It helps with every aspect of my life.” As AI transforms mathematics, many researchers wonder whether future mathematicians will be able to say the same.

When considering the 1960s sitcoms Bewitched and I Dream of Jeannie, both of which featured women with supernatural powers navigating life with mortals, most people wouldn’t connect them with pursuing an engineering career. But Karen Panetta did. The sitcoms’ main characters—Samantha Stevens, a witch; and Jeannie, a genie—were “strong, empowered female leads using magic,” Panetta says, and they inspired her to become an engineer, as it was like sorcery to her.Panetta, an IEEE Fellow, is dean of graduate education at the Tufts University engineering school, in Medford, Mass., outside of Boston.Karen PanettaEmployer Tufts University, in Medford, Mass.Title Dean of the engineering school’s graduate educationMember grade IEEE FellowAlma maters Boston University and Northeastern University in BostonLike Samantha and Jeannie, Panetta has made magic happen, such as when she helped to invent the first CPU digital-twin simulator. Digital twins are computer simulation programs that track and adjust the operations of a physical device in detail. Her simulator has been adapted for several industrial uses, including by NASA to help design spacecraft.Panetta also mentors young women to encourage them to pursue a STEM career through the Nerd Girls program she launched at Tufts in 2000. Engineering undergraduate students work on technology for socially conscious projects such as environmental cleanup, renewable energy, and the development of assistive devices to improve mobility for people with disabilities.Panetta received this year’s IEEE Mildred Dresselhaus Medal for “contributions to computer vision and simulation algorithms, and for leadership in developing programs to promote STEM careers.” The award, sponsored by Google, was presented at the IEEE Honors Ceremony on 24 April in New York City.Receiving the medal is particularly special to Panetta, she says, because she knew its namesake: Mildred Dresselhaus, an IEEE Life Fellow who pioneered the study of carbon nanostructures at a time when researching physical and material properties of commonplace atoms was unpopular. She was a MIT professor of physics and electrical engineering, and died in 2017.Panetta nominated Dresselhaus for the IEEE Medal of Honor, which she received in 2015.“Millie was a rock star,” Panetta says. “I can’t think of another medal that really encapsulates her spirit and what I’ve dedicated my life to.”Finding a creative outlet in engineeringAs a child growing up in Boston, Panetta built trapdoors and other features in her treehouse, she says.“I also explored fashion and sewed my own clothes,” she adds. “I wasn’t very successful, but I was very creative.”She was a top performer in math and science classes in high school, so her father encouraged her to pursue civil engineering.“I didn’t know what an engineer was, and my father, who was a mechanic working on heavy construction equipment, only knew about civil engineers,” Panetta says. “I started taking computer programming classes at school, but knowing how to type on a keyboard and make a software program wasn’t good enough for me. I wanted to know what was inside the box.”Her thirst for knowledge inspired her to pursue a bachelor’s degree in computer engineering at Boston University.“My father was very disappointed that I didn’t pick civil engineering,” she says, laughing.She commuted to school, and she struggled to find study groups for her classes, so she joined IEEE to connect with peers.She became active in the university’s student branch, organizing events including the IEEE Student Professional Awareness Conference, which helps students learn practical career skills including résumé building, interviewing, and networking. She organized a SPAC for her branch, and IEEE Life Senior Member Jim Watson volunteered to speak at the event. It changed her life, she says.Watson was the director of commercial and industrial marketing at Ohio Edison in Akron, where he worked for 36 years.“He flew to Boston to speak at our event, but fewer than 20 students attended. I was embarrassed,” Panetta says. But Watson told her the important lesson was that she showed up and organized the event.“He said I would be successful because of that,” she says. “He didn’t care about the attendees’ grade point averages, only that we were professional enough to organize the talk.“That encouragement was the first time anyone outside of my family ever told me that I would succeed, so it was reaffirming. To this day, I still use some of the techniques that I learned in his presentation in my own classroom to teach students.”Panetta graduated in 1986. Her IEEE membership helped her get hired for her first dream job: a diagnostic engineer at Digital Equipment Corp.While attending the IEEE Computer Society’s annual symposium on very large-scale integration in Boston, she handed her résumé to a DEC representative, who hired her to work in Hudson, Mass.While working full time, Panetta attended Northeastern University, in Boston, as a part-time graduate student. She earned a master’s degree in electrical engineering in 1988.Developing the first CPU digital twinIn the early 1990s, Panetta was assigned to work with Ernst Ulrich, one of DEC’s most respected consulting engineers, she says. He was developing a new CPU using millions of CMOS transistors.“I thought, ‘Wow, what a great opportunity,’” she says, “not realizing they assigned it to me because no one else wanted to work with him, as he set rigorous standards, expecting those who worked with him to think outside of the box and hold their own to bullet-proof new concepts.”Panetta and Ulrich wanted the ability to test the CPU while still designing the hardware and software. That way, both would be ready to use at the same time. Typically, the hardware was developed before the software was written.“We decided that we were going to simulate the machine to see how it was going to run—which was unheard of,” she says.During a meeting with the company’s top engineers, Panetta shared her idea for an algorithm that could accomplish the team’s goal. She was met with silence.“It’s going to be the engineers who better society because we know how to work together. We’ve proven that IEEE members know how to work across geographic boundaries, ethnic boundaries, and gender boundaries. And that’s a good model for the world.”“I thought to myself, ‘Did I just say something stupid?’” she says. “But then, the top engineer looked at me and said, ‘I have been doing this for 50 years, and you, a kid just out of school, comes up with this [solution] like it’s obvious.’”Her idea became the basis for the digital twin simulator. It used behavioral models to run software on a CPU simulation. The software passes information through the system, she says, just like it would pass information through wires or interconnects.“We did successfully have a complete model of millions of transistors,” Panetta says. “I efficiently simulated hundreds of thousands of experiments and ran the software on this simulated model so that we knew exactly how it was going to perform on the real machine. That had never been done before.”Her groundbreaking work led to a promotion: from computer analyst to principal software engineer.When she began managing a team and hiring staff members, Panetta noticed the younger employees knew the theory but didn’t have the technical skills to hit the ground running, she says.“It took the company two years to train somebody before they could really contribute technically to a team,” she says. She decided she wanted to help prepare students for jobs in industry.In 1995 she was accepted into DEC’s Engineers and Education program, in which full-time employees who wanted to teach could take a leave of absence to complete a degree while still being paid. Participants were then placed in academic institutions for two-year stints to help students bridge the gap between classroom theory and real-world problem-solving.After earning a Ph.D. in electrical engineering from Northeastern in 1994, Panetta began her teaching assignment at Tufts. After one year, she left her job at DEC to join the university as its first female electrical engineering professor. At the time, the department had only one female undergraduate EE student.“I showed up to work dressed in an all-pink suit,” she says, laughing. “Other professors looked at me like I didn’t belong there because I looked different.”She didn’t let that stand in the way of reaching her goals: preparing the next generation of students for jobs and mentoring young women who were interested in becoming engineers but who felt they wouldn’t be accepted and therefore couldn’t pursue a career in the field.Launching the Nerd Girls programWhen Panetta began teaching, she noticed that students weren’t getting any hands-on engineering experience, so in 1996 she created an internship program. It was the precursor to Nerd Girls.At the time, she was consulting for NASA’s data visualization and animation lab in Langley, Va., translating complex information into a user-friendly animated form. The programs visualized Earth’s atmosphere and identified pollutants, their origins, and their effects on people and the environment.Panetta needed a larger team to help conduct the research, so she asked her undergraduate students if they wanted to participate.“Female students flocked to me because they could relate to the work I was doing, loved how their skills could benefit humanity, and didn’t see me as the classic nerd professor with no life,” Panetta said in a 2008 interview with The Institute about the program. “Eventually, the girls outnumbered the boys.”“The research project ended up winning awards,” she added. “Tufts couldn’t believe that undergrads had a hand in it. That’s when things really turned around.”Nerd Girls officially launched at Tufts in 2000 as a class where students work closely with industry on engineering projects. Examples have included building a solar-powered car, developing a battery for the last functioning twin lighthouse in the United States, and creating devices to help people train service animals.“Everyone who has participated in the program graduated with a bachelor’s degree,” Panetta says. “I’m also very proud that 98 percent of participants pursue a graduate degree within three years of earning their bachelor’s.”The program is open to all students, regardless of gender.Creating a community at IEEEPanetta became an active IEEE volunteer in 2004 after meeting Arthur Winston, the IEEE president at the time. Winston, an IEEE Life Fellow, was an electrical engineering professor at Tufts. He helped found the Gordon Institute, a leadership-focused engineering school at the university.“I sat next to him on a bus, and he invited me to attend the IEEE Boston Section meetings,” she says.Panetta eventually was elected by the section as a member-at-large—which allowed her to attend conferences and other events.To help spread the word about the Nerd Girls program throughout IEEE, Winston connected Panetta to Mary Ellen Randall, who was chair of IEEE Women in Engineering at the time. Randall is the current IEEE president and CEO. Panetta joined IEEE WIE and was elected as its 2007–2009 chair.In that position, she worked with Randall and Leah Jamieson, the 2007 IEEE president, to hire more staff to support the program and launch its magazine.“At that time, we didn’t have any way to connect to members or tell the stories of women in technology,” Panetta says. “I wanted people to read the stories of women from around the globe and how they overcame adversity. So I launched the IEEE Women in Engineering Magazine in 2007.”Panetta serves as the award-winning publication’s editor in chief, and she is a member of several other IEEE societies and committees.IEEE is helping to change the world for the better, she says.“It’s going to be the engineers who better society,” she says, “because we know how to work together.“We’ve proven that IEEE members know how to work across geographic boundaries, ethnic boundaries, and gender boundaries. And that’s a good model for the world.”

What could you do if you could make a circuit trace by just bending a piece of paper? How about bridging modern technologies and traditional handicrafts while providing opportunities for learning skills in both.As part of our interdisciplinary research into digital craftsmanship at the MEI Lab at the School of Creative Media, City University of Hong Kong, we came across research that demonstrated how to impregnate paperlike material (technically a “nonwoven textile”) with the kind of liquid metal used to make conductive ink. Initially, the impregnated material is nonconductive because an insulating oxide layer forms that encapsulates microscopic droplets of the liquid metal. However, applying pressure via shaped molds will crack open the insulating layer, allowing neighboring particles to merge, and thus creating conducting regions in the shape of the mold.Both of us were introduced as children to origami and kirigami (similar to origami, except that cutting is allowed in addition to folding). We, along with our colleagues, decided to see if those traditional techniques could be used on the new material to eliminate the need for molds. Our goal was to allow crafters to make hybrid papercraft creations that contained easily integrated elements such as LEDs and motors.In particular, we were interested in the possibility of combining the separate stages of creating a papercraft object and adding electrical conductors. Previous approaches to creating electrified papercraft objects relied on adding a separate flexible conductor—such as adhesive copper tape—to the paper. This increases the effort required and runs the risk of creating open circuits as the conductive material conforms to the object’s shape. Isopropanol and a gallium-indium liquid material are used to impregnate a paperlike material that is 55 percent polyester and 45 percent cellulose. Electronic components such as LEDs and motors are held in place with masking tape. James ProvostOur first step was to see if the pressures involved in bending and cutting alone would be sufficient to create conductive traces. We became frequent visitors to our university’s materials science and engineering department to fabricate samples and then to borrow equipment to characterize their behavior. We soon confirmed that the pressures involved in folding and cutting—ranging from 2.5 to 100 megapascals—were enough to create conductive traces. We also confirmed that normal handling of the paper didn’t accidentally create conductive paths.We made a number of changes to the original method for creating the impregnated paper. For example, instead of immersing the paper in a mixture of isopropanol and liquid metal, we used an airbrush to spray the mixture onto the paper. That allowed us to vary how much was deposited on the paper and to use cardboard stencils to mask some areas from being impregnated, allowing folding and cutting in those regions without creating unwanted conductive traces. We also experimented with the ratios of isopropanol and liquid metal.We became frequent visitors to our university’s materials science and engineering department.After optimizing the mixing ratios and amount applied via airbrush, we were left with a material that reliably conducts with a resistance of 23.18 ohms per centimeter for cut edges and 4.4 Ω/cm for folded edges. The folded edges retain their conductivity even if later flattened out, and the conductivity is the same on either side of the paper. We estimate the combined cost of the paper and liquid metal (available from many online vendors) is about US $1.80 to make a 10- by 10-cm piece.The next step was attaching electronic components to the traces. To make the connections more flexible, we cut down the rigid leads of LEDs and attached conductive thread to the stumps. We then held the threads in place using masking tape. Similarly, we connected conductive thread to the terminals of a power supply.As our goal was to use this material educationally, we now needed to make it easy for a beginner—whether in papercraft or electronics—to try it out. We created a toolkit, dubbed LiqMetCraft. This consists of all the required materials, plus a browser-based software tool that lets the user select or create designs and then gives guidance on physical construction.We created three versions of LiqMetCraft. The first is based on Chinese papercraft in which a piece of paper is folded into a fanlike segment and then cut to create a radially symmetric design. We provided circles of paper with a doughnot-shape impregnated region, with an untreated region that created a gap in the donut. We attached positive and negative terminals to either side of the gap. The user could specify in the software how many times they wanted to fold the disk and then draw potential cuts, receiving immediate feedback on what the unfolded disk would look like, as well as guidance on how to place LEDs. To make our paper sample, isopropanol and liquid metal are mixed in specific ratios while being cooled by an ice bath. Sonic waves are used to ensure the liquid metal breaks up into microscopic droplets. The mixture is then applied via airbrush, while stencils prevent some areas being covered for different papercraft templates. James ProvostThe second version of LiqMetCraft was based on origami. We supplied rectangular pieces of paper with two conductive regions separated by a border down the middle. The software tool provided templates for 12 origami designs, with step-by-step instructions for folding them. Once the project was completed, the user could add components, such as a motor, by taping them to the folds.The final version supported 3D paper model making. In this case, the initial paper supplied was a rectangle with an untreated rectangular central area. By cutting this paper in half and then further cutting the halves into patterns separated by a spacer, the user could make various self-standing models. The software allowed the user to draw a pattern on screen, and then have a cutting machine produce a template for cutting the impregnated paper.We had 42 participants, evenly divided into three groups, try out the different versions. All found it easy to use, and we were pleasantly surprised that some participants moved beyond the supplied designs to their own creations.For full details of the current process, see our open access LiqMetCraft research paper published in CHI ‘26: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. In the future, we plan to try different substrates for the impregnating solution, as well as explore further types of papercraft, such as pop-up books. We’re also interested in developing ways to use the material to support inputs as well as outputs by constructing switches and potentiometers directly out of the material. Imagine traditional papercraft creations becoming interactive devices!This article appears in the July 2026 print issue.

SummaryRFIC design is a complex “dark art” that limits progress in wireless technologies like 5G, autonomous vehicles, and satellite communications.Princeton researchers use reinforcement learning and inverse design to rapidly create RFICs from scratch.Diffusion models rapidly generate novel or human-interpretable RF layouts, achieving record performance and drastically reducing design time.Future progress needs large, shared chip design datasets and open ecosystems so AI can learn universal electromagnetic and circuit behaviors.Take a moment and try to imagine your life without the wireless advances of the past three decades.Have you lost your luggage? What a shame AirTags have not been invented. The airline representative has promised to call with updates, so settle in for a long wait by the kitchen telephone, because there are no affordable cellphones. You’ll be stuck listening to whatever is on the radio while you wait, because there are no streaming services. That’s not even to speak of all the movie plots that would have been ruined.This is just a tiny sliver of how wireless technology makes itself felt in your day-to-day existence. The effects it has had on supply chains, infrastructure, and how the economy runs have been world-altering.None of it would be possible without the radio-frequency integrated circuits that allow all our devices to unobtrusively send and receive information.Now imagine what the further evolution of this technology will bring: Wide-spread autonomous vehicles, quantum communications, 6G mobile service and satellite communications. Continued momentum will depend on newer and more advanced versions of today’s RF chips.But there’s the rub. Whereas the design of most of the world’s computing chips has been standardized into its own science, RF design has remained stubbornly in the realm of art. A dark art, even, that is mastered only through years of experience. As any sorcerer will tell you, the dark arts keep their own schedule. And that schedule is impeding progress not just in RF chip design but in every other technology that depends on it.About seven years ago, in the wake of AlphaGo’s victory over world Go champion Lee Sedol, my students at Princeton and I began to wonder: Could AI be taught this art as well? Recent successes suggest that, to a large extent, it can. Over the last few years, our group and other leaders in the field have started to develop machine-learning-driven algorithmic methods for designing RFICs. Some of the resulting chips look more like modern art than circuit layouts. Yet in many cases, the physical prototypes bested state-of-the art circuits in terms of performance. The real achievement, however, is that it took the AI orders of magnitude less time to conceive a working design than it would a human designer.This is not about one or two RF chips. AI-enabled design could be the future of all RF design, and maybe much more.The Dark Art of RFIC DesignSo why do these chips all have to be crafted by hand? Why aren’t RFICs designed with an algorithmic synthesis process, much as CPUs and GPUs are?The design of RFICs is an exercise in engineering across multiple physical domains. Maxwell’s equations, operating across different spatial and temporal scales, govern how electromagnetic fields interact with active and passive devices that must be carefully codesigned for the chip to function. Alongside these are the laws of thermodynamics, which determine how heat is generated and removed during operation, as well as the mechanics of thermal expansion and contraction that dictate how reliably the chip and its packaging survive temperature changes.AI Could Short-Circuit RFIC Design The design of a radio-frequency integrated circuit requires human intuition and multiple, often-repeated optimization steps. The hope is that through an understanding of Maxwell’s Equations, an AI can be taught to short-circuit this process and quickly produce a design.Simultaneously accounting for all the physical constraints these impose makes the design space almost impossibly large. Every decision involves complex priorities that often compete with one another, preventing the optimization of any of them.To better understand the issue, let’s walk through the steps involved, after which you’ll better understand why a single new chip design takes years and tens to hundreds of millions of dollars. Most of the area of radio-frequency integrated circuits is dominated by complex electromagnetic structures. Human-designed RFICs, like this broadband power amplifier [1], start with templates and follow a symmetric, understandable pattern. But freed from the constraints of human-designed templates and the need for humans to even understand the rationale of electromagnetic structures, power amplifier ICs [2–5] and low-noise amplifiers [6] can take on truly wild-looking yet efficient designs. SENGUPTA LABLet’s say you’re an engineer assigned to design a new 28-gigahertz power amplifier for a 5G-millimeter-wave handset. (This is the type of RFIC that boosts the 5G signals on your phone and transmits them to the antenna where they can be picked up by a distant base station). Where do you start?RFIC design has some features in common with house building. Just as the blueprint for a house dictates the number of bedrooms and bathrooms to be built and the hallways connecting them, the blueprint for an RFIC—called the architecture—establishes the kinds of elements the RFIC needs to fulfill its intended function. Instead of rooms, the architecture includes, for example, the number of stages of amplification your power amplifier needs. Instead of hallways, it shows the paths that signals must take to get through those stages.The blueprint for RFICs is actually mostly hallway; passive elements, like inductors and transmission lines, take up far more real estate than active elements like transistors.Here’s why. As you have probably experienced yourself, a typical CPU’s transistors overheat when faced with operating frequencies of just a few gigahertz. The frequencies RFICs can operate at are higher by an order of magnitude—28 and 39 GHz for 5G signals, 26.5 to 40 GHz and even higher for satellite communications, and 77 GHz for automotive radar. Under this onslaught, a CPU’s transistors would fail.RFIC transistors avoid this fate because these chips cleverly manage the signal’s energy with careful electromagnetic design. This takes the form of byzantine networks of metal elements that dominate the chip’s real estate. These structures are geometrically regular, often symmetrical, and so intricately constructed they sometimes resemble lacelike filigree. But while they may look decorative, they are essential to the chip’s functioning.Electrically speaking, these “hallways” work more like the chip’s plumbing. Like plumbing, this extensive labyrinth of passives confines electromagnetic energy only to the places it should be traveling around the chip.The major challenge in RFIC design is putting all these elements together to ensure they work, just as constructing a house from its blueprints demands exact specs for load-bearing beams, pipes, and external walls. On an RFIC, the architecture needs to be realized with physically fabricable transistors and passive components that are connected just so, to permit the signal to travel through the chip and be processed. The way these devices are connected locally is what we call the circuit’s topology.The RFIC Design ProcessTo make that power amplifier, then, your first step is to identify a candidate circuit template: The combination of structures that will meet the goals of a particular architecture with a specific circuit topology. Over the years, researchers have eased your burden by developing reusable design templates for specific functions. For example, templates suggest how many amplification stages a circuit needs (because sometimes, combining the output of two smaller amplifiers will result in better bandwidth and efficiency than you would get from a single larger one). And they suggest what the general configuration of the passive structures should be. Today there is an extensive library of such templates.However, these can’t simply be used off-the-shelf, because each comes with trade-offs. Some have better gain at the expense of stability; some better bandwidth at the expense of efficiency; still others are more energy efficient at the expense of output power, and so on. There is rarely a clear best choice.To arrive at the “sweet spot” where all these different parameters are balanced into optimal harmony, designers will typically lay out several different versions of the circuit, using intuitions and methods they have picked up in their years of training.The challenge is that the decision around the architecture, circuit topology, or the electromagnetic passives cannot be done separately. One decision influences the others. So, designing an RF circuit can often feel like trying to fit an oversized carpet into too small a room—press down one corner, and another pops up.At microwave and millimeter-wave frequencies, even the smallest misstep is the difference between a chip that works and one that doesn’t, and any number of things can go wrong. For example, when an electromagnetic wave encounters a transistor—or any other component —the path it travels must be properly “matched” to what comes next. If it isn’t, some of the energy reflects backward instead of flowing forward. Imagine trying to connect a high-pressure fire hose directly to a narrow garden hose. Without the right adapter, water will splash backward at the junction. Very little will make it through. In electronics, this is called the impedance-matching problem.To prevent those reflections, engineers design special transitions, essentially microscopic adapters, that smooth the handoff between components. On a chip, these adapters can be surprisingly intricate. They don’t just pass the signal along; they can also split it, combine it, or distribute it across multiple paths with carefully controlled timing and strength.Once you’ve done the architecture, plumbing, and everything in between comes the moment of truth. Have all the choices you have navigated through the enormous design space resulted in an RFIC that meets its specifications? If the specifications are not met, you will have to go back, either redoing the topology or the entire architecture, and repeat the whole process. So get ready for months of time- and resource-heavy simulation and iteration. Perhaps you now see why, for decades, a core belief has persisted in the RFIC community: “RF design is an art.” It was said that only an experienced designer—with an artisanal understanding of how the pieces make up the whole—could master the subtleties of analog and RF design. Unfortunately, this entrenched notion has long held back algorithmic innovations in the field just when we need them most. Traditional, artisanal RFIC design is hitting its limits as the complexity of these systems inexorably grows.AI for RFIC DesignWhile RFIC designers continued their battle against their “oversized carpet” problem, a series of interesting developments emerged in allied disciplines. Across a range of other previously intractable problems like protein folding and climate modeling, AI has been able to successfully navigate multidimensional complex spaces. This gave us the incentive to look deeper into AI for RF. After all, the combinatorial complexity of protein folding is not that different from the nature of the design space in our domain.We were not the first to think of using artificial intelligence to speed up parts of RFIC design. Researchers had previously trained machine learning algorithms on circuit templates in the hope of speeding up the normal optimization processes. While this approach was undoubtedly faster than humans at optimizing templates, it still relied fundamentally on libraries of existing designs invented by humans.Training an AI to Design a Chip A machine learning system learns to do end-to-end RFIC design like other AIs learned to play such games as Go. Essentially, it turns the process into a game, learning from the results of its own efforts.We didn’t want that. We wanted to break free from the restrictions of prefabricated topologies. Because while a designer’s experience and hard-won heuristics are crucial to building a working design, they also place fundamental limits on it. Furthermore, such an approach would necessarily require simulation steps as part of the optimization cycle, and even the fastest simulations use a lot of computing resources. Worse still, in many advanced cases, such as for broadband designs, there are no existing templates.But if we didn’t start with templates, where could we start?The goal here was to allow algorithms to determine—entirely from scratch—every parameter for architecture, constituent circuits, and electromagnetic passives. This approach differs fundamentally from conventional optimization, which is limited to determining the parameters—like transistor dimensions and passive component geometries—that optimize structures originally devised by humans.In our new approach, the architecture begins essentially from nothing and is progressively assembled through successive iterations. The system explores the design space by generating myriad candidate circuit combinations and mapping the resulting performance trade-offs as it navigates this landscape. Because the process is not biased by prior human design choices, it can produce completely novel circuit topologies that look markedly different from those created by human designers.In some ways, the approach echoes AI systems such as AlphaGo Zero, which achieved superhuman performance not because it was trained on games played by humans but because it explored the rules by playing against itself. Similarly, our algorithm develops new circuit architectures by exploring and evaluating its own design strategies. In so doing, it learns to understand circuits, electromagnetics, and the close codesign they need to achieve the end-to-end design of RFIC.Inverse Design for RFICsTo realize this capability, we proceeded in two stages. First, we developed a reinforcement-learning (RL) framework that determines the optimal system architecture, circuit topology, device parameters, and even the properties of the electromagnetic interfaces that connect different circuit elements. In this stage, the algorithm effectively defines how signals should propagate and interact across the system.The algorithm trains very similarly to how a computer learns to play a game. If you let it play enough times, it can learn to play better by observing the relationship between the actions it took and the score it achieves. In a similar way, the RL agent here learns to design effective circuits by playing with a set of combinations, and over time, it can map the space between the circuit performance to its architecture, topology, and parameters. This training takes a few days to a week, but once trained, the agent can design circuits very quicklyThe next step was to determine the physical structure of the IC’s electromagnetics—the plumbing—that can create the desired properties of the passive elements, which are characterized by a set of metrics called scattering parameters. These measure if a signal entering a component actually moves forward—or is reflecting backward, being wasted, as in our previous example with the fire hose and the garden hose.Deriving the structure from the desired scattering parameters is an example of an approach called inverse design, which appears across many areas of engineering. In structural engineering, for example, one might collaborate with an architect on a physical goal—such as creating large interior spaces with high ceilings—and then determine the arrangement of arches or buttresses that can support it.Generative AI for Electromagnetic NetworksBut RF integrated crcuits pose a particular challenge for inverse design: The process must account simultaneously for circuit behavior and the electromagnetic responses of the interconnects and passive elements that link them together. But it has to figure that out without doing a lot of artisanal iterating.So we replaced our RF circuit simulator with an AI-based emulator. This AI model can predict the behavior of electromagnetic fields going through any structure—even totally arbitrary two-dimensional shapes—without having to compute the underlying physics from scratch, as simulation tools do. It would predict the solution of Maxwell’s equations and tell you the scattering parameters for any structure you showed it, without actually doing the math. With such an AI in hand, what a time-consuming electromagnetic solver normally takes minutes or hours to accomplish is reduced to milliseconds.We chose to build our emulator around a convolutional neural network—a machine learning model that has been remarkably successful for image processing. Such networks can extract spatial features from any structure, and it turns out that the image of a structure contains a lot of spatial information that can accurately predict its electromagnetic performance. Then we trained it on a vast number of random pixelated structures whose scattering parameters had been labeled.Once we had our inverse-design RL and suitable AI emulator, we essentially had an end-to-end AI designer. So we asked it to design us a power amplifier.Unconventional RF ArchitecturesIn 2023, we published this proof of concept—a power amplifier targeting the millimeter-wave band, specifically spanning 30 to 100 GHz, which covers most of the relevant 5G and radar frequencies. The final design achieved the best combination of wide bandwidth, output power, and efficiency then reported for a silicon-based power amplifier—meaning it could amplify a large amount of data across a wide swath of frequencies—while maintaining record efficiency.The structure of the IC’s electromagnetic pathways was unlike anything any human would ever consider. Since the AI is not trained on human designs, the layout that emerged looked more like an arbitrary pattern or perhaps a QR code than the regular symmetrical structures we are used to seeing.One unexpected insight revealed by this prototype, and our research generally, is that there’s no evidence that the templates we’ve historically relied on are even close to optimal for modern design goals. It’s not that a human designer can never come up with a better design. But with the removal of the templates and the time to synthesize cycle upon cycle of optimized circuits, it is now clear that AI-driven synthesis could break traditional design barriers and push the limits of RFIC capabilities.Our 5G amplifier had only one input port and one output port. Adding more inputs and outputs to a design is not straightforward. Every port electromagnetically couples to every other port, so the scattering parameters quickly add up. Two ports give you four scattering parameters. Four ports, 16 scattering parameters. The math gets ugly fast. Could our model keep up?We next trained our model on larger classes of electromagnetic structures with many input and output ports. In 2024, we published work showing that multiport integrated circuits are no problem for these AI algorithms either. Where previously multiport electromagnetic simulation required days or weeks of toil, this model evolved new structures in minutes. Since then, a plethora of work in the space by research communities across the globe have demonstrated the power of inverse design in RFIC.Combining the reinforcement learning framework with the inverse design, we now had the ability to create an RFIC from specifications all the way to a fabrication-ready layout. We’ve so far shown this is true for RFICs ranging from low-noise amplifiers to subterahertz and broadband power amplifiers. The hope is that this will work just as well for other circuits.Making AI Designs InterpretableOur goal was to make RFIC design better and easier, but we didn’t want to make it beyond human understanding. Chip testing and debugging is a long, arduous process, sometimes even more so than design. Engineers often prefer ICs to have interpretable structures, so that if a problem crops up, they can understand how the chip works well enough to debug it.To create structures that are more interpretable, we turned to diffusion models, which you may know from their remarkable ability to generate realistic images from text prompts.AI-driven synthesis could break traditional design barriers and push the limits of RFIC capabilities. Imagine you go to your favorite image-generation engine and ask it to create a painting of the sky in the style of Picasso, Van Gogh, or Michelangelo. You will get images that capture the essence of their brushstrokes, their use of colors, and their framing. All are pictures of the sky nonetheless, but in different styles.Electromagnetic design is similar in that multiple structures can have very similar electromagnetic responses. Instead of using text input, we used scattering parameters as our input, and the electromagnetic structure of an RFIC chip as our output. As part of the inputs to the diffusion model, we created a dial that sets the spatial frequency of the final structure. By turning the dial, a designer can direct the model to synthesize structures with low (classical-looking and interpretable), medium (mazelike structures), or high (pixelated or arbitrarily-shaped) spatial frequency.From prompts to output, the entire process took about 6 minutes. With this diffusion model, algorithms can now both discover novel architectures and accelerate the creation of conventional, so-called classical ones.All an RFIC designer needs to do is specify virtually any valid set of scattering parameters. As long as they are physically realizable under Maxwell’s equations, the model pops out a corresponding structure as if it were a vending machine.The Future of AI-Driven RFIC DesignThe results of our investigations have drawn the attention of the RF community. The traditional bottom-up design process is clearly beginning to reverse.But there are still questions: How generalizable are these methods? Can they consistently deliver truly high performance? Can we get to a place where AI produces designs that maximize every conceivable trade-off, holistically optimizing every parameter to its most ideal physical state? We want to take this strategy beyond RFIC design and invent other kinds of circuits that are different from anything humans have ever done.These are exciting and ambitious prospects, but we are not there yet. AI can hallucinate a design that creates bad circuits that don’t work. This means verification methods need to remain under human oversight. And, while hallucinations are rare, it would still be good to reduce their occurrence.History suggests that meeting these dreams of the future will take much more data than we’ve been using. Before the creation of the ImageNet repository—a repository of 14 million varied, human-annotated images—image-recognition models didn’t function well in the real world. The datasets they had been trained on were too tiny to be effective. ImageNet’s massive amounts of training data ushered in a revolution that led to AI that can generalize and recognize images in the wild. The rest was history.If the goal for RFIC and analog design is a universal foundational model—something that learns the governing laws of electromagnetics and circuit behavior—then we also need data.The good news is that this data is plentiful. Around the world, countless engineers at companies and academic labs simulate nearly identical RF circuits and passive structures every day. The bad news is that it’s all locked away behind nondisclosure agreements.Open ecosystems have propelled other areas, and we think the RFIC community should do the same. There had been some movement toward this. Natcast, the operator of the U.S. CHIPS and Science Act’s R&D program, would have bolstered shared infrastructure and innovation for the next generation of wireless, sensing, and defense technologies. Unfortunately, both the organization and the program it ran specifically for machine learning and RFICs have been closed.But the momentum Natcast’s effort sparked hasn’t died out. Building on our early work, groups across the community have already demonstrated remarkable advances. AI-driven IC design is part of a much broader technological shift. From biology and materials science to automotive and aerospace engineering, AI is reshaping how complex systems are conceived and optimized. Deeper collaboration between AI researchers and chip designers will unlock the field’s full potential. It’s by no means a foregone conclusion, but if we get this right, this genie won’t stay in its bottle.

5G telecommunications, according to industry hype when 5G first launched in 2019, was going to be all about buzzy applications like mobile augmented reality and autonomous vehicles. But the surprise plot twist came when replacing home cable internet turned into 5G’s most widely adopted new application.Fixed wireless access (FWA) now serves over 14 million U.S. customers, and contributes 28 percent of worldwide wireless traffic. Fixed wireless access is what the term sounds like: broadband internet delivered over a cellular radio link to a stationary location—no cable, no fiber, no trenching, no satellite broadband antenna pointed at the sky. What makes FWA distinctive is that it repurposes the same towers, spectrum, and 5G infrastructure that was built for mobile devices.One U.S. Federal Communications Commission commissioner has called FWA 5G’s killer app. And that’s true not just in the United States either. Jio, India’s largest carrier, is also one of the world’s largest FWA providers, with over 9 million customers as of last year.Carriers discovered they could repurpose surplus 5G capacity, while also exploiting a usage pattern quirk: mobile traffic starts to drop after 8 p.m., just when home internet usage peaks. The result is broadband, delivered via traditional cellphone towers, at a lower cost than fiber deployment. For these reasons, FWA provides real price competition to cable broadband, while reaching underserved rural and suburban communities.Fixed Wireless Access Repurposes Ambitious 5G InfrastructureFWA is cheaper to deploy than fiber, and for most homes and small businesses, fiber’s gigabit speeds are overkill anyway. And since FWA uses the same wireless networks built for cellular service, FWA works anywhere that receives a steady cellular signal.As cellular networks extend into areas with minimal service, FWA’s coverage map expands with them. In these remote locales, the other main viable broadband alternative typically comes from satellite services like Starlink—which are, compared to FWA, more expensive, with higher delays, and lower bandwidth.While most FWA deployments use currently underused microwave bands, some FWA deployments use electromagnetic spectrum that 5G launched but that mostly failed with mobile users. Millimeter waves operate at frequencies 10 to 40 times higher than 4G’s spectrum, offering high data rates from their wide available bandwidth.However, there are good reasons 5G mobile users today don’t generally use millimeter-wave spectrum. Millimeter waves can’t penetrate buildings. Plus, they lose signal strength within a kilometer or two of the transmitter. Millimeter-wave antennas are also a real drain on cellphone batteries compared to microwave and radio-wave tech.Yet none of these challenges applies to a fixed station with a clear line of sight to a nearby tower. FWA home units (called customer premise equipment or CPEs) outperform 5G handsets by a significant margin. That’s mostly because of hardware. CPEs carry larger, more sensitive antennas than a typical cellphone, paired with more capable transceivers. CPEs also tend to be plugged into wall outlets, making battery concerns a nonissue.Another 5G technology that did not gain traction in mobile wireless is multi-user multiple-input multiple-output (MU-MIMO). A base station with MU-MIMO uses an array of antennas to serve multiple users on the same frequency simultaneously.However, maintaining a MU-MIMO signal involves tracking each user individually—a problem that quickly becomes overwhelming with enough mobile users. FWA is different, however. Static CPEs, with their steadier downlink traffic loads, are an ideal match for MU-MIMO technology.So, FWA internet service not only uses mostly fallow spectrum but also uses 5G spectrum more efficiently than do 5G mobile users—for whom, of course, these 5G technologies were originally designed!How FWA Became 5G’s Surprise Killer AppNot long ago, the high-bandwidth use cases for 5G made for an impressive list: millisecond latency for autonomous vehicles, mobile augmented reality headsets with extensive high-speed data needs, and massive machine connectivity for an expanding internet of things (IoT).These applications have all stalled. Autonomous vehicles pose challenging—and still unsolved—problems unrelated to spectrum allocation. Augmented and virtual reality technologies have yet to create meaningful spikes in bandwidth demand. And the IoT has, to date at least, fragmented across an array of competing standards.Mobile carriers had built dense 5G networks for mobile customers whose needs rarely saturated the network’s capacity. Home broadband usage peaks in the evening hours, precisely when cellular networks are quietest.FWA sits at cellular networks’ crossroads of supply and demand.The Advent of 6G Will Only Expand FWA’s ReachIn December, the telecom standards body, the Third Generation Partnership Project (3GPP), issued its latest 5G specification—Release 20, the final “5G only” update. So, although 6G is still years away (its first specifications are expected in early 2029), engineering decisions that will define 6G are being made today. And FWA is not on the margins of that conversation; FWA is currently considered an established day-one use case.6G wireless technology promises to expand FWA’s reach—not only via spectrum but also via geometry. Instead of following 4G and 5G’s connectivity model—strong signals near towers and weak signals far away—future 6G networks will let homes connect to multiple towers simultaneously, using a technology called distributed MIMO (multiple-input, multiple-output).Where 5G’s version of MIMO (a.k.a. massive MIMO) concentrates user communication with dozens of antennas at a single tower, distributed MIMO uses antennas across multiple base stations and coordinates them to deliver signals to your home from multiple directions simultaneously.The practical result: Because no single tower is responsible for any given connection, the “edge” of a cell network—that outer boundary where signal strength falls off and service degrades—no longer represents a hard limit on who gets well served. A home that would once have been too distant from a tower, or blocked by terrain, could now be within reach of several base stations working together.6G may eventually adopt distributed MIMO technology for mobile users, when synchronization challenges and other signal engineering hurdles are solved and deployed for real-world cellular networks. The jury, as of 2026, is still out on whether the full distributed MIMO problem will be solved once the 6G standards start to be set in place, within three years.As demand for FWA grows, carriers will also deploy increasingly capable millimeter-wave infrastructure for fixed customers first—the stationary CPE use case that millimeter wave best suits. The dense millimeter-wave antenna infrastructure that FWA requires is the same infrastructure that future mobile applications will eventually inherit. AR glasses, AI-powered wearables, and other bandwidth-hungry applications originally promised for 5G are not canceled—they are waiting for the infrastructure to arrive.The pathway to FWA is being prepared at lower frequencies, too. There is growing interest today in the largely unoccupied FR3 band, which spans roughly 7 to 24 gigahertz, situated between crowded low/mid-bands and the much higher millimeter-wave frequencies. Recent field trials by Nokia have demonstrated FR3’s viability for both cellular and FWA applications. FR3 is emerging as one of the more promising near-term frontiers for extending FWA coverage beyond its current footprint.None of this was the plan. No carrier executive in 2020 stood on a stage and announced that 5G’s defining achievement would be delivering living room broadband to rural homes and suburban subdivisions underserved by cable.FWA became 5G’s killer app because the engineering economics made it happen. Surplus wireless capacity met unmet consumer broadband demand, with the physics of a stationary receiver doing the rest.That is not a criticism of the engineers or the carriers. It is simply how technology sometimes advances—sideways, through gaps nobody was trying to fill.But FWA’s model of prioritizing unconnected users may in the end prove to be telecom’s on-ramp to everything else. Fix the digital divide first. Tomorrow’s sci-fi future appears set to follow close behind.

Andrew Ng has serious street cred in artificial intelligence. He pioneered the use of graphics processing units (GPUs) to train deep learning models in the late 2000s with his students at Stanford University, cofounded Google Brain in 2011, and then served for three years as chief scientist for Baidu, where he helped build the Chinese tech giant’s AI group. So when he says he has identified the next big shift in artificial intelligence, people listen. And that’s what he told IEEE Spectrum in an exclusive Q&A. Ng’s current efforts are focused on his company Landing AI, which built a platform called LandingLens to help manufacturers improve visual inspection with computer vision. He has also become something of an evangelist for what he calls the data-centric AI movement, which he says can yield “small data” solutions to big issues in AI, including model efficiency, accuracy, and bias. Andrew Ng on... What’s next for really big models The career advice he didn’t listen to Defining the data-centric AI movement Synthetic data Why Landing AI asks its customers to do the work The great advances in deep learning over the past decade or so have been powered by ever-bigger models crunching ever-bigger amounts of data. Some people argue that that’s an unsustainable trajectory. Do you agree that it can’t go on that way? Andrew Ng: This is a big question. We’ve seen foundation models in NLP [natural language processing]. I’m excited about NLP models getting even bigger, and also about the potential of building foundation models in computer vision. I think there’s lots of signal to still be exploited in video: We have not been able to build foundation models yet for video because of compute bandwidth and the cost of processing video, as opposed to tokenized text. So I think that this engine of scaling up deep learning algorithms, which has been running for something like 15 years now, still has steam in it. Having said that, it only applies to certain problems, and there’s a set of other problems that need small data solutions. When you say you want a foundation model for computer vision, what do you mean by that? Ng: This is a term coined by Percy Liang and some of my friends at Stanford to refer to very large models, trained on very large data sets, that can be tuned for specific applications. For example, GPT-3 is an example of a foundation model [for NLP]. Foundation models offer a lot of promise as a new paradigm in developing machine learning applications, but also challenges in terms of making sure that they’re reasonably fair and free from bias, especially if many of us will be building on top of them. What needs to happen for someone to build a foundation model for video? Ng: I think there is a scalability problem. The compute power needed to process the large volume of images for video is significant, and I think that’s why foundation models have arisen first in NLP. Many researchers are working on this, and I think we’re seeing early signs of such models being developed in computer vision. But I’m confident that if a semiconductor maker gave us 10 times more processor power, we could easily find 10 times more video to build such models for vision. Having said that, a lot of what’s happened over the past decade is that deep learning has happened in consumer-facing companies that have large user bases, sometimes billions of users, and therefore very large data sets. While that paradigm of machine learning has driven a lot of economic value in consumer software, I find that that recipe of scale doesn’t work for other industries. Back to top It’s funny to hear you say that, because your early work was at a consumer-facing company with millions of users. Ng: Over a decade ago, when I proposed starting the Google Brain project to use Google’s compute infrastructure to build very large neural networks, it was a controversial step. One very senior person pulled me aside and warned me that starting Google Brain would be bad for my career. I think he felt that the action couldn’t just be in scaling up, and that I should instead focus on architecture innovation. “In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data. Having 50 thoughtfully engineered examples can be sufficient to explain to the neural network what you want it to learn.” —Andrew Ng, CEO & Founder, Landing AI I remember when my students and I published the first NeurIPS workshop paper advocating using CUDA, a platform for processing on GPUs, for deep learning—a different senior person in AI sat me down and said, “CUDA is really complicated to program. As a programming paradigm, this seems like too much work.” I did manage to convince him; the other person I did not convince. I expect they’re both convinced now. Ng: I think so, yes. Over the past year as I’ve been speaking to people about the data-centric AI movement, I’ve been getting flashbacks to when I was speaking to people about deep learning and scalability 10 or 15 years ago. In the past year, I’ve been getting the same mix of “there’s nothing new here” and “this seems like the wrong direction.” Back to top How do you define data-centric AI, and why do you consider it a movement? Ng: Data-centric AI is the discipline of systematically engineering the data needed to successfully build an AI system. For an AI system, you have to implement some algorithm, say a neural network, in code and then train it on your data set. The dominant paradigm over the last decade was to download the data set while you focus on improving the code. Thanks to that paradigm, over the last decade deep learning networks have improved significantly, to the point where for a lot of applications the code—the neural network architecture—is basically a solved problem. So for many practical applications, it’s now more productive to hold the neural network architecture fixed, and instead find ways to improve the data. When I started speaking about this, there were many practitioners who, completely appropriately, raised their hands and said, “Yes, we’ve been doing this for 20 years.” This is the time to take the things that some individuals have been doing intuitively and make it a systematic engineering discipline. The data-centric AI movement is much bigger than one company or group of researchers. My collaborators and I organized a data-centric AI workshop at NeurIPS, and I was really delighted at the number of authors and presenters that showed up. You often talk about companies or institutions that have only a small amount of data to work with. How can data-centric AI help them? Ng: You hear a lot about vision systems built with millions of images—I once built a face recognition system using 350 million images. Architectures built for hundreds of millions of images don’t work with only 50 images. But it turns out, if you have 50 really good examples, you can build something valuable, like a defect-inspection system. In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data. Having 50 thoughtfully engineered examples can be sufficient to explain to the neural network what you want it to learn. When you talk about training a model with just 50 images, does that really mean you’re taking an existing model that was trained on a very large data set and fine-tuning it? Or do you mean a brand new model that’s designed to learn only from that small data set? Ng: Let me describe what Landing AI does. When doing visual inspection for manufacturers, we often use our own flavor of RetinaNet. It is a pretrained model. Having said that, the pretraining is a small piece of the puzzle. What’s a bigger piece of the puzzle is providing tools that enable the manufacturer to pick the right set of images [to use for fine-tuning] and label them in a consistent way. There’s a very practical problem we’ve seen spanning vision, NLP, and speech, where even human annotators don’t agree on the appropriate label. For big data applications, the common response has been: If the data is noisy, let’s just get a lot of data and the algorithm will average over it. But if you can develop tools that flag where the data’s inconsistent and give you a very targeted way to improve the consistency of the data, that turns out to be a more efficient way to get a high-performing system. “Collecting more data often helps, but if you try to collect more data for everything, that can be a very expensive activity.” —Andrew Ng For example, if you have 10,000 images where 30 images are of one class, and those 30 images are labeled inconsistently, one of the things we do is build tools to draw your attention to the subset of data that’s inconsistent. So you can very quickly relabel those images to be more consistent, and this leads to improvement in performance. Could this focus on high-quality data help with bias in data sets? If you’re able to curate the data more before training? Ng: Very much so. Many researchers have pointed out that biased data is one factor among many leading to biased systems. There have been many thoughtful efforts to engineer the data. At the NeurIPS workshop, Olga Russakovsky gave a really nice talk on this. At the main NeurIPS conference, I also really enjoyed Mary Gray’s presentation, which touched on how data-centric AI is one piece of the solution, but not the entire solution. New tools like Datasheets for Datasets also seem like an important piece of the puzzle. One of the powerful tools that data-centric AI gives us is the ability to engineer a subset of the data. Imagine training a machine-learning system and finding that its performance is okay for most of the data set, but its performance is biased for just a subset of the data. If you try to change the whole neural network architecture to improve the performance on just that subset, it’s quite difficult. But if you can engineer a subset of the data you can address the problem in a much more targeted way. When you talk about engineering the data, what do you mean exactly? Ng: In AI, data cleaning is important, but the way the data has been cleaned has often been in very manual ways. In computer vision, someone may visualize images through a Jupyter notebook and maybe spot the problem, and maybe fix it. But I’m excited about tools that allow you to have a very large data set, tools that draw your attention quickly and efficiently to the subset of data where, say, the labels are noisy. Or to quickly bring your attention to the one class among 100 classes where it would benefit you to collect more data. Collecting more data often helps, but if you try to collect more data for everything, that can be a very expensive activity. For example, I once figured out that a speech-recognition system was performing poorly when there was car noise in the background. Knowing that allowed me to collect more data with car noise in the background, rather than trying to collect more data for everything, which would have been expensive and slow. Back to top What about using synthetic data, is that often a good solution? Ng: I think synthetic data is an important tool in the tool chest of data-centric AI. At the NeurIPS workshop, Anima Anandkumar gave a great talk that touched on synthetic data. I think there are important uses of synthetic data that go beyond just being a preprocessing step for increasing the data set for a learning algorithm. I’d love to see more tools to let developers use synthetic data generation as part of the closed loop of iterative machine learning development. Do you mean that synthetic data would allow you to try the model on more data sets? Ng: Not really. Here’s an example. Let’s say you’re trying to detect defects in a smartphone casing. There are many different types of defects on smartphones. It could be a scratch, a dent, pit marks, discoloration of the material, other types of blemishes. If you train the model and then find through error analysis that it’s doing well overall but it’s performing poorly on pit marks, then synthetic data generation allows you to address the problem in a more targeted way. You could generate more data just for the pit-mark category. “In the consumer software Internet, we could train a handful of machine-learning models to serve a billion users. In manufacturing, you might have 10,000 manufacturers building 10,000 custom AI models.” —Andrew Ng Synthetic data generation is a very powerful tool, but there are many simpler tools that I will often try first. Such as data augmentation, improving labeling consistency, or just asking a factory to collect more data. Back to top To make these issues more concrete, can you walk me through an example? When a company approaches Landing AI and says it has a problem with visual inspection, how do you onboard them and work toward deployment? Ng: When a customer approaches us we usually have a conversation about their inspection problem and look at a few images to verify that the problem is feasible with computer vision. Assuming it is, we ask them to upload the data to the LandingLens platform. We often advise them on the methodology of data-centric AI and help them label the data. One of the foci of Landing AI is to empower manufacturing companies to do the machine learning work themselves. A lot of our work is making sure the software is fast and easy to use. Through the iterative process of machine learning development, we advise customers on things like how to train models on the platform, when and how to improve the labeling of data so the performance of the model improves. Our training and software supports them all the way through deploying the trained model to an edge device in the factory. How do you deal with changing needs? If products change or lighting conditions change in the factory, can the model keep up? Ng: It varies by manufacturer. There is data drift in many contexts. But there are some manufacturers that have been running the same manufacturing line for 20 years now with few changes, so they don’t expect changes in the next five years. Those stable environments make things easier. For other manufacturers, we provide tools to flag when there’s a significant data-drift issue. I find it really important to empower manufacturing customers to correct data, retrain, and update the model. Because if something changes and it’s 3 a.m. in the United States, I want them to be able to adapt their learning algorithm right away to maintain operations. In the consumer software Internet, we could train a handful of machine-learning models to serve a billion users. In manufacturing, you might have 10,000 manufacturers building 10,000 custom AI models. The challenge is, how do you do that without Landing AI having to hire 10,000 machine learning specialists? So you’re saying that to make it scale, you have to empower customers to do a lot of the training and other work. Ng: Yes, exactly! This is an industry-wide problem in AI, not just in manufacturing. Look at health care. Every hospital has its own slightly different format for electronic health records. How can every hospital train its own custom AI model? Expecting every hospital’s IT personnel to invent new neural-network architectures is unrealistic. The only way out of this dilemma is to build tools that empower the customers to build their own models by giving them tools to engineer the data and express their domain knowledge. That’s what Landing AI is executing in computer vision, and the field of AI needs other teams to execute this in other domains. Is there anything else you think it’s important for people to understand about the work you’re doing or the data-centric AI movement? Ng: In the last decade, the biggest shift in AI was a shift to deep learning. I think it’s quite possible that in this decade the biggest shift will be to data-centric AI. With the maturity of today’s neural network architectures, I think for a lot of the practical applications the bottleneck will be whether we can efficiently get the data we need to develop systems that work well. The data-centric AI movement has tremendous energy and momentum across the whole community. I hope more researchers and developers will jump in and work on it. Back to top This article appears in the April 2022 print issue as “Andrew Ng, AI Minimalist.”

The end of Moore’s Law is looming. Engineers and designers can do only so much to miniaturize transistors and pack as many of them as possible into chips. So they’re turning to other approaches to chip design, incorporating technologies like AI into the process.Samsung, for instance, is adding AI to its memory chips to enable processing in memory, thereby saving energy and speeding up machine learning. Speaking of speed, Google’s TPU V4 AI chip has doubled its processing power compared with that of its previous version.But AI holds still more promise and potential for the semiconductor industry. To better understand how AI is set to revolutionize chip design, we spoke with Heather Gorr, senior product manager for MathWorks’ MATLAB platform.How is AI currently being used to design the next generation of chips?Heather Gorr: AI is such an important technology because it’s involved in most parts of the cycle, including the design and manufacturing process. There’s a lot of important applications here, even in the general process engineering where we want to optimize things. I think defect detection is a big one at all phases of the process, especially in manufacturing. But even thinking ahead in the design process, [AI now plays a significant role] when you’re designing the light and the sensors and all the different components. There’s a lot of anomaly detection and fault mitigation that you really want to consider. Heather GorrMathWorksThen, thinking about the logistical modeling that you see in any industry, there is always planned downtime that you want to mitigate; but you also end up having unplanned downtime. So, looking back at that historical data of when you’ve had those moments where maybe it took a bit longer than expected to manufacture something, you can take a look at all of that data and use AI to try to identify the proximate cause or to see something that might jump out even in the processing and design phases. We think of AI oftentimes as a predictive tool, or as a robot doing something, but a lot of times you get a lot of insight from the data through AI.What are the benefits of using AI for chip design?Gorr: Historically, we’ve seen a lot of physics-based modeling, which is a very intensive process. We want to do a reduced order model, where instead of solving such a computationally expensive and extensive model, we can do something a little cheaper. You could create a surrogate model, so to speak, of that physics-based model, use the data, and then do your parameter sweeps, your optimizations, your Monte Carlo simulations using the surrogate model. That takes a lot less time computationally than solving the physics-based equations directly. So, we’re seeing that benefit in many ways, including the efficiency and economy that are the results of iterating quickly on the experiments and the simulations that will really help in the design.So it’s like having a digital twin in a sense?Gorr: Exactly. That’s pretty much what people are doing, where you have the physical system model and the experimental data. Then, in conjunction, you have this other model that you could tweak and tune and try different parameters and experiments that let sweep through all of those different situations and come up with a better design in the end.So, it’s going to be more efficient and, as you said, cheaper?Gorr: Yeah, definitely. Especially in the experimentation and design phases, where you’re trying different things. That’s obviously going to yield dramatic cost savings if you’re actually manufacturing and producing [the chips]. You want to simulate, test, experiment as much as possible without making something using the actual process engineering.We’ve talked about the benefits. How about the drawbacks?Gorr: The [AI-based experimental models] tend to not be as accurate as physics-based models. Of course, that’s why you do many simulations and parameter sweeps. But that’s also the benefit of having that digital twin, where you can keep that in mind—it’s not going to be as accurate as that precise model that we’ve developed over the years.Both chip design and manufacturing are system intensive; you have to consider every little part. And that can be really challenging. It’s a case where you might have models to predict something and different parts of it, but you still need to bring it all together.One of the other things to think about too is that you need the data to build the models. You have to incorporate data from all sorts of different sensors and different sorts of teams, and so that heightens the challenge.How can engineers use AI to better prepare and extract insights from hardware or sensor data?Gorr: We always think about using AI to predict something or do some robot task, but you can use AI to come up with patterns and pick out things you might not have noticed before on your own. People will use AI when they have high-frequency data coming from many different sensors, and a lot of times it’s useful to explore the frequency domain and things like data synchronization or resampling. Those can be really challenging if you’re not sure where to start.One of the things I would say is, use the tools that are available. There’s a vast community of people working on these things, and you can find lots of examples [of applications and techniques] on GitHub or MATLAB Central, where people have shared nice examples, even little apps they’ve created. I think many of us are buried in data and just not sure what to do with it, so definitely take advantage of what’s already out there in the community. You can explore and see what makes sense to you, and bring in that balance of domain knowledge and the insight you get from the tools and AI.What should engineers and designers consider when using AI for chip design?Gorr: Think through what problems you’re trying to solve or what insights you might hope to find, and try to be clear about that. Consider all of the different components, and document and test each of those different parts. Consider all of the people involved, and explain and hand off in a way that is sensible for the whole team.How do you think AI will affect chip designers’ jobs?Gorr: It’s going to free up a lot of human capital for more advanced tasks. We can use AI to reduce waste, to optimize the materials, to optimize the design, but then you still have that human involved whenever it comes to decision-making. I think it’s a great example of people and technology working hand in hand. It’s also an industry where all people involved—even on the manufacturing floor—need to have some level of understanding of what’s happening, so this is a great industry for advancing AI because of how we test things and how we think about them before we put them on the chip.How do you envision the future of AI and chip design?Gorr: It’s very much dependent on that human element—involving people in the process and having that interpretable model. We can do many things with the mathematical minutiae of modeling, but it comes down to how people are using it, how everybody in the process is understanding and applying it. Communication and involvement of people of all skill levels in the process are going to be really important. We’re going to see less of those superprecise predictions and more transparency of information, sharing, and that digital twin—not only using AI but also using our human knowledge and all of the work that many people have done over the years.

Quantum computing is a devilishly complex technology, with many technical hurdles impacting its development. Of these challenges two critical issues stand out: miniaturization and qubit quality.IBM has adopted the superconducting qubit road map of reaching a 1,121-qubit processor by 2023, leading to the expectation that 1,000 qubits with today’s qubit form factor is feasible. However, current approaches will require very large chips (50 millimeters on a side, or larger) at the scale of small wafers, or the use of chiplets on multichip modules. While this approach will work, the aim is to attain a better path toward scalability.Now researchers at MIT have been able to both reduce the size of the qubits and done so in a way that reduces the interference that occurs between neighboring qubits. The MIT researchers have increased the number of superconducting qubits that can be added onto a device by a factor of 100.“We are addressing both qubit miniaturization and quality,” said William Oliver, the director for the Center for Quantum Engineering at MIT. “Unlike conventional transistor scaling, where only the number really matters, for qubits, large numbers are not sufficient, they must also be high-performance. Sacrificing performance for qubit number is not a useful trade in quantum computing. They must go hand in hand.”The key to this big increase in qubit density and reduction of interference comes down to the use of two-dimensional materials, in particular the 2D insulator hexagonal boron nitride (hBN). The MIT researchers demonstrated that a few atomic monolayers of hBN can be stacked to form the insulator in the capacitors of a superconducting qubit.Just like other capacitors, the capacitors in these superconducting circuits take the form of a sandwich in which an insulator material is sandwiched between two metal plates. The big difference for these capacitors is that the superconducting circuits can operate only at extremely low temperatures—less than 0.02 degrees above absolute zero (-273.15 °C). Superconducting qubits are measured at temperatures as low as 20 millikelvin in a dilution refrigerator.Nathan Fiske/MITIn that environment, insulating materials that are available for the job, such as PE-CVD silicon oxide or silicon nitride, have quite a few defects that are too lossy for quantum computing applications. To get around these material shortcomings, most superconducting circuits use what are called coplanar capacitors. In these capacitors, the plates are positioned laterally to one another, rather than on top of one another.As a result, the intrinsic silicon substrate below the plates and to a smaller degree the vacuum above the plates serve as the capacitor dielectric. Intrinsic silicon is chemically pure and therefore has few defects, and the large size dilutes the electric field at the plate interfaces, all of which leads to a low-loss capacitor. The lateral size of each plate in this open-face design ends up being quite large (typically 100 by 100 micrometers) in order to achieve the required capacitance.In an effort to move away from the large lateral configuration, the MIT researchers embarked on a search for an insulator that has very few defects and is compatible with superconducting capacitor plates.“We chose to study hBN because it is the most widely used insulator in 2D material research due to its cleanliness and chemical inertness,” said colead author Joel Wang, a research scientist in the Engineering Quantum Systems group of the MIT Research Laboratory for Electronics. On either side of the hBN, the MIT researchers used the 2D superconducting material, niobium diselenide. One of the trickiest aspects of fabricating the capacitors was working with the niobium diselenide, which oxidizes in seconds when exposed to air, according to Wang. This necessitates that the assembly of the capacitor occur in a glove box filled with argon gas.While this would seemingly complicate the scaling up of the production of these capacitors, Wang doesn’t regard this as a limiting factor.“What determines the quality factor of the capacitor are the two interfaces between the two materials,” said Wang. “Once the sandwich is made, the two interfaces are “sealed” and we don’t see any noticeable degradation over time when exposed to the atmosphere.”This lack of degradation is because around 90 percent of the electric field is contained within the sandwich structure, so the oxidation of the outer surface of the niobium diselenide does not play a significant role anymore. This ultimately makes the capacitor footprint much smaller, and it accounts for the reduction in cross talk between the neighboring qubits.“The main challenge for scaling up the fabrication will be the wafer-scale growth of hBN and 2D superconductors like [niobium diselenide], and how one can do wafer-scale stacking of these films,” added Wang.Wang believes that this research has shown 2D hBN to be a good insulator candidate for superconducting qubits. He says that the groundwork the MIT team has done will serve as a road map for using other hybrid 2D materials to build superconducting circuits.