In Guadalajara, Mexico, many high schools have motivated teachers and talented students with an interest in science, technology, engineering, and mathematics, but they lack access to advanced tools such as robotics laboratories. The resources shortfall limits the students’ opportunities for hands-on learning on cutting-edge applications.

A team from ITESO, Universidad Jesuita de Guadalajara, is working to change that. Through the EPICS in IEEE initiative, a multidisciplinary group of 15 engineering students, faculty advisors, and IEEE Guadalajara Section volunteers developed RoboMeshA. The portable, self-contained educational platform brings robotics and AI experiences into classrooms.

EPICS is administered by IEEE Educational Activities and funded by the IEEE Robotics and Automation Society.

A mobile laboratory

Rather than requiring a school to build a dedicated computer lab or install complex software, RoboMeshA operates as an all-in-one mobile learning network.

“RoboMeshA brings robotics and AI to students who don’t have access to specialized facilities or preinstalled software,” says team member Fernando Vidal Luna, an IEEE student member and a mechatronics engineering major at ITESO.

Students connect directly to the platform from a user-friendly web browser. They can interact with the robot manually or use its control modes to watch it move and detect and avoid obstacles.

“The project combines mechanical design, embedded systems, control engineering, computer vision, and AI into a single robotic system that functions as a mobile learning laboratory,” says faculty advisor Jorge A. Lizarraga.

The team says young students are interested in technology, programming, and robotics but don’t have an opportunity to work with systems that combine mechanics, electronics, software, and control.

“RoboMeshA allows students to see how all these disciplines work together in a tangible and understandable way,” says team member José S. González, who also is studying mechatronics engineering.

The team has built two units and is developing a modular coupling framework to expand the system’s capabilities for research and classroom demonstrations. The structured system design approach connects independent software components while minimizing internal dependencies, enabling four RobotMeshA robots to operate together.

Overcoming design challenges

The team faced significant hurdles while designing the project.

“One key challenge involved the robot’s structural design,” Luna says. “It wasn’t only about making a chassis where all the components fit and the design had sufficient stability, rigidity, and weight distribution. It was also about ensuring that the electronics were protected while still being accessible for maintenance, testing, and modifications.”

“It was also challenging to design a platform that could be used by students with different levels of experience,” González adds.

“When students realize the technology they develop can inspire others and improve lives, engineering becomes far more meaningful.” —Luis Fernando Luque-Vega

The team partnered with the CETI Colomos and Prepa ITESO high schools to validate the platform in classroom settings.

“We wanted the first interactions with the robot to be simple and intuitive,” González says, “such that students could simply power the robot, connect to its network, and begin interacting with it, rather than having to deal with software installation, extensive configuration, or troubleshooting.”

Engineering with social impact

Many of the students who participated were from ITESO’s applied professional projects program. The experience offered them practical training in project management, system integration, and user-centered design.

The team also presented a research paper and a project poster in May at the Engineering Congress of the Jesuit University System.

“Seeing a design move from a digital model to a physical system was invaluable,” González says. “Working with students from different backgrounds taught us to listen to end users and design for their actual needs.”

Project lead Luis Fernando Luque-Vega, an IEEE member, says he’d like the venture to serve as a blueprint for engineering education.

“I hope RoboMeshA is adopted by schools, universities, and IEEE student branches across Mexico and internationally as a model for integrating technical innovation with community engagement,” Luque-Vega says.

By pairing engineering talent with community service, initiatives such as EPICS in IEEE demonstrate how targeted support can turn academic concepts into real-world solutions.

“When students realize the technology they develop can inspire others and improve lives, engineering becomes far more meaningful,” Luque-Vega says.

For more information on service-learning opportunities, visit the EPICS website.

I’ve seen two total solar eclipses and have been duly impressed by what happens as the moon casts its shadow on Earth. But recently I’ve become even more intrigued by a similar phenomenon that doesn’t involve the sun or the moon—something called an asteroid occultation.

That’s what happens when an asteroid orbits around the solar system and blocks the light of a distant star you’re viewing from Earth. Like the moon during a solar eclipse, the asteroid casts a predictable moving shadow on a swath of Earth’s surface—a small silhouette in the dim light bathing us from that one star.

When such a fortuitous alignment occurs, amateur astronomers can discern things about the asteroid that professionals can’t readily measure, even with their giant telescopes on high mountains. That’s because amateurs are nimble: They can be in just the right place at just the right time to measure an asteroid’s fleeting shadow, which could be just a few hundred meters wide and traveling at tens of kilometers per second. With enough observers, they can collectively map that shadow, revealing the asteroid’s shape.

Even folks on a limited budget can do this, because the size of an asteroid you can measure doesn’t scale with the size of your telescope. If the occulted star is relatively bright, you don’t need much of a telescope at all.

How Do You Catch an Asteroid Occultation?

My own efforts along these lines have been with a modest 5.1-inch-aperture (130-millimeter) Newtonian telescope that sells for about US $300. I attach it to a small equatorial mount ($150) that can track the stars by virtue of some added stepper motors driven by an open-source telescope controller called OnStep. (You could save yourself the time, trouble, and expense of all that DIY hacking by purchasing a motorized mount for as little as $300.)

A flasher provides a calibrated time base for light-curve measurements. It relies on a GPS module [top] to provide a high-accuracy pulse once per second, is gated by an Arduino nano [middle] to prevent flashes occurring at the moment of occultation, and is then passed to a LED [bottom].James Provost

I bought an inexpensive astronomy color camera on Amazon for $260 to take images at the video rates required to capture the rapid changes during an occultation. I chose this camera because it has a relatively large sensor, Sony’s IMX585, which provides a large field of view. A monochrome camera would be better for asteroid occultations, but the monochrome version of this camera is harder to come by and more expensive. If you’re looking for a cheaper option, the monochrome ToupTek G3M662M (about $200) would be a good choice, although its sensor is smaller.

Knowing where and when to catch an occultation in your area is of course critical and can be calculated using free PC software found on the International Occultation Timing Association (IOTA) website. If you plan to contribute your observations to IOTA to increase the body of scientific knowledge about asteroids, you will need to calibrate the timing of your images. You can’t just depend on the time stamps your computer adds to the video frames, which can be way off.

For time calibration, many practitioners use a flasher: a red LED driven from the pulse-per-second signal from a GPS receiver. Asteroid observers use such a pulsing LED positioned in front of their telescopes to calibrate the timing of the images they take. With some effort, it’s possible to reduce the uncertainty to just a handful of milliseconds.

The flasher I built uses a GPS module that I had on hand. But I’d recommend you purchase a different one that accepts an external active antenna. HiLetgo’s NEO-7M $12 module might be a good choice—but don’t forget to remove its antenna-coupling capacitor (marked as C2 on the circuit board) if you do attach an active external antenna to it.

How Do You Make a Telescope Flasher?

You can’t let the flasher just blink away every second, though, because its light might stomp on the very signal you’re trying to detect. So alongside the GPS module, my flasher also contains an Arduino Nano, plus two transistors, three resistors, and a switch. I wired these components together so as to drive the LED directly from the pulse-per-second signal coming from the GPS. The signal passes through a transistor controlled by the Arduino so that the flashes can be started and stopped at prescribed times. I can then program the flasher to produce calibrating pulses near the start and end of each recording session, while suppressing the flashing around the occultation itself.

Over time, an asteroid such as Duccio will pass in front of multiple stars [below]. Each time it does, it will block the light from a star [above] for a time that depends on its width along the line of transit. By combining multiple light curves, it is possible to map the shape of the asteroid.James Provost

So far, I’ve managed to record four occultations that have occurred within easy driving distance of my home in North Carolina. The first was quite short, by an asteroid a mere 4 kilometers wide. The star involved was rather dim, so I really had to squint at my laptop screen to see the star momentarily blink out. The star in my second occultation was brighter, and the dimming much longer, so no squinting was required. My third observation tested the limits of my little telescope with a very dim target star, requiring quite long exposures per video frame (about a third of a second). Thankfully, the asteroid was a big one (120 km wide), so the occultation lasted a few seconds, and I could discern it.

The asteroid I targeted last, named Duccio, is about a dozen kilometers wide and orbits in the main asteroid belt between Mars and Jupiter. Its shadow, moving at a clip of some 24 km per second, took about a half second to pass over me. The star this asteroid blocked was bright enough for me to record the event very distinctly at 24 frames per second, providing excellent time resolution.

Asteroid occultations offer a wonderful natural experiment. And unlike a solar eclipse, observable events probably take place near you multiple times each month. So with a little knowledge and the right gear, you can observe them. You just have to wait for the stars—and the asteroids—to align.

This white paper gives distribution engineers and utility teams a practical overview of the substation exit, the short and high-consequence section where station capacity divides into individual feeders. It explains how covered, spacer-supported overhead construction can reduce common contact-driven faults while leaving room for future circuits.

What you will learn about:

• Why a fault near the substation exposes more customers than one farther along the feeder, and why the first spans out of the station carry so much reliability weight.
• How covered conductors and spacer-cable systems differ from bare overhead conductors, and why covered conductor is not treated as touch-safe insulation.
• Which engineering factors shape a sound exit design, including conductor rating, protection coordination, grounding, and structural loading.
• How overhead spacer cable compares with conventional bare overhead and underground shielded cable across footprint, reliability, and lifecycle cost.
• How to move a project from concept to commissioning using staged design gates and a clear performance specification.

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!

If you have kids, they’re probably back in school right now after the summer break. Mine are too.

My kids range from 9 to 20 years old, and lately I’ve been thinking a lot about what their future careers are going to look like.

Some schools are embracing AI in the classroom. Others are banning it completely. I’m not sure either side has figured out the right answer yet—I’m not sure any of us have.

What I do know is that AI is already changing how engineers, and nearly every other knowledge worker, does their job. So I’ve been thinking: What skills do I actually want my kids to develop for an AI-augmented workforce?

1. Learn to write

The more I use AI, the more convinced I am that it’s a multiplier, not a substitute.

Give a great researcher AI tools and they can accelerate their research. Give an experienced software engineer the same tools and they can build at an incredible rate. But put those tools in the hands of someone without foundational knowledge and you often get something else you’re used to seeing way too much of: slop.

We’ve all seen the vibe-coded applications that barely work. Our inboxes are filled with emails that sound suspiciously identical. Teachers are reading papers that sound like they were all written by the same person. That “person” has the last name GPT.

When everyone has access to the same tools, having an actual voice becomes a differentiator.

Ironically, strong writing skills might be MORE important because of AI, not less. Text is still the primary way we communicate with these models, so being able to clearly articulate what you want improves what you get back.

But more importantly, writing teaches you to develop ideas of your own.

AI can help you express your opinion. It shouldn’t manufacture one for you.

2. Learn to speak

Like written communication, speaking skills are now at a premium—but for a different reason.

We live in a strange moment. We’re more digitally connected than ever, while often feeling increasingly isolated from one another. At the same time, we’re seeing renewed interest in conferences, meetups, communities, and in-person experiences.

When human interaction feels scarce, communication becomes more valuable.

So learn how to explain an idea in front of a room. Learn how to disagree without being disagreeable. Learn how to tell a story. Learn how to listen. Learn how to persuade someone.

There’s a timeless book, originally published 90 years ago, that teaches these interpersonal skills, and it’s probably more valuable for engineers than another white paper on how neural networks work: How to Win Friends and Influence People.

An AI can generate a presentation for you, but it can’t convincingly deliver it to a skeptical audience. That takes emotional intelligence.

3. Learn to be bored

This might be the hardest one.

We have engineered boredom almost completely out of our lives. There’s always a podcast to listen to, a notification to check, a video to watch, a feed to scroll, or now an AI to talk to.

Go for a walk without headphones. Eat without looking at a phone. Sit in the car without immediately reaching for something to fill the silence, and let your mind wander.

Because boredom isn’t wasted time. It’s a breeding ground for original ideas.

The danger I worry about isn’t that AI becomes too intelligent, but that we become too willing to outsource the uncomfortable parts of thinking.

The students going back to school today will enter a workforce filled with technology that I couldn’t have imagined when I was their age. I have no idea what the dominant AI model will be by then or even what interacting with a computer will look like.

That’s exactly why I don’t want to optimize their education around today’s tools. I want them to learn the skills that will outlive the tools.

Write clearly. Speak confidently. Think independently.

And every once in a while, embrace boredom.

—Brian

AI Slop Is Changing How Engineers Review Code

Software engineers have entered a new era of code review. The strategies they’re now testing will determine whether AI can actually provide code that’s faster and more reliable when you factor in the review process. If it can, what does that mean for the entry-level engineers who are still learning to code on their own?

Read more here.

U.S. Tech Firms Change Strategies to Hire International Talent

In response to a barrage of actions by the U.S. federal government to limit legal immigration, tech companies are adapting their search for top talent. IEEE Spectrum looked into the responses to proposed changes, like higher fees for H-1B visa applications.

Read more here.

What It Takes to Be an Adaptable Engineer

As AI changes the job market and day-to-day work of engineers, young professionals are often told they need to be adaptable. But what does adaptability actually look like in practice? The skill has different definitions depending on who you ask, but with the right mindset and support from leadership, adaptability can help keep you afloat.

Read more here.

Many youngsters fascinated by exploring outer space dream of becoming an astronaut, but few do. One who had the right stuff is Pedro Duque, who was Spain’s first astronaut. The aeronautics engineer flew aboard the space shuttle Discovery and the International Space Station.

After retiring as an astronaut, he headed Spain’s Ministry of Science, Innovation, and Universities. Today he is the chairman of HispaSat, a Spanish satellite company.

Pedro Duque

Employer

HispaSat in Madrid

Title

President and chairman of the board

Member grade

Honorary member

Alma mater

The Polytechnic University of Madrid

Nearly 30 years after his first mission, Duque is still Spain’s most famous astronaut. Three public schools have been named after him, and he has received numerous awards. This year, the IEEE Board of Directors made him an IEEE honorary member for “contributions to space exploration, leadership in collaborative science and technology programs, and serving as a role model for younger generations.”

He was unable to attend the 24 April ceremony in New York City, but he expressed his gratitude in recorded acceptance remarks in his award presentation shown during the event.

“We engineers of all specialties recognize the leadership of your institute—the largest and most important engineering society in the world,” he says in the video. “What an honor it is to belong now to an organization whose purpose it is to foster technological innovation and excellence for the benefit of humanity.”

Inspired by Apollo 11

Duque says he knew he wanted to be an engineer from a young age. It’s not surprising that he became interested in aeronautics, as his father was an air traffic controller who explained to him how airplanes worked.

In 1969, when he was 6 years old, he was inspired to become an astronaut after watching the Apollo 11 moon landing.

“I didn’t know anyone who wasn’t attracted to space exploration,” after the moon landing, he says, laughing. “It was presented in such an epic manner, with declarations about its impact on society. The landing made us aware that humanity was exploring new places, and I was keen on knowing more about them.”

His dream of becoming an astronaut was unrealistic at the time, he says, because the country had no space program. Spain was ruled by Francisco Franco, who spent little to no money on scientific innovation, Duque says.

That changed after Franco died in 1975. The country transitioned to a democratic constitutional monarchy and began participating in research and development programs, particularly with the European Space Agency (ESA).

Astronaut duties

Duque earned an aeronautical engineering degree in 1986 from the aeronautical and space engineering school at the Polytechnic University of Madrid.

His first job was as an engineer in the flight dynamics group at GMV, a space and technology company based in Madrid. He was a member of the orbit determination group and worked at ESA’s European Space Operations Centre, in Darmstadt, Germany. He helped develop algorithms, orbit computational software, and computer models.

He was also a member of the space agency’s flight control team for the European Remote Sensing-1 satellite, launched in 1991, and the European Retrievable Carrier, launched in 1992 on the space shuttle Atlantis’s STS-46 mission.

In 1990, ESA recruited candidates for its astronaut program—which Duque says rarely happens. He and several colleagues applied.

“Why not?” he says. “What we thought we wanted to be when we were little was now possible.”

After a considerable selection process, Duque was chosen in 1992 to join the agency’s Astronaut Corps. He trained at the European Astronaut Centre and at the Russian Cosmonaut Training Center (now known as the Gagarin Research and Test Cosmonaut Training Center).

In 1994, he served as the prime crew interface coordinator on the ESA–Russian EuroMir space station. He managed communication between the astronauts on the Mir station and the European scientists to help ensure orbital experiments and operations ran smoothly.

NASA selected Duque to be an alternate payload specialist on the ground for space shuttle Columbia’s STS-78 mission in 1996. In that role, he was trained to operate and manage scientific experiments, equipment, and cargo during a crewed mission. He also supported the flight from the ground as a crew interface coordinator.

His first flight into space was in 1998 as a mission specialist representing the ESA on the space shuttle Discovery’s STS-95 mission. He managed in-orbit tasks, experiments, spacewalks, and equipment operations.

He served as a flight engineer in 2003 onboard the Soyuz TMA-3 Cervantes, a joint mission between Russia and Spain to the International Space Station. He operated the station systems, managed daily maintenance tasks, and performed scientific experiments assisted by the mission commander.

Memories of flying with John Glenn

The years of training and preparation to become an astronaut were rigorous, he says, but the experience was fulfilling. Even though he spent only about 10 days on each mission, he says, “they were very rewarding times because all the preparation paid off, and we got the results we needed.”

Some of his favorite memories were viewing Earth during the day and night for the first time, and watching it pass from one phase to the other. Another was seeing the moon flattened to almost a sliver during the few seconds before it set. Experiencing microgravity was a thrill as well, he says.

“I also cherish the companionship of the people who worked alongside me in space and on the ground,” he adds.

One was John H. Glenn, the first American to orbit the Earth. Duque flew with Glenn on Discovery. They talked about the selection criteria for astronauts. Glenn told him those on board the Apollo and Gemini missions were chosen because they were test pilots, who were thought to be most likely to handle problems or failures effectively and proactively.

“The technology in today’s missions is advanced enough that astronauts won’t be needed to handle probable catastrophic equipment failures,” Duque says. “Instead, they will perform experiments and update or fix the devices used in space capsules. But flights to the moon and even Mars will use new types of spacecraft that might not necessarily work as planned, so those astronauts will have to know how to fix them and possibly solve critical problems on their own.

“Also, future astronauts will have to live with others in a confined space for months, and not everyone can do that.”

The price of sudden fame

As Spain’s first astronaut, Duque became a celebrity. Among his honors are Russia’s Order of Friendship and Spain’s Great Cross of Aeronautical Merit and Princess of Asturias Award. He also received NASA’s Space Flight Medal, which is given to an astronaut who flies aboard a U.S. space mission.

Learning to navigate sudden fame and being treated like a celebrity was challenging, Duque says. Like many engineers, he was accustomed to working behind the scenes and out of the public eye.

“Being famous, both in the profession and the public, was quite difficult in the beginning,” he says. “Being a celebrity doesn’t come easily to me, but after so many years, somehow I learned how to deal with it.”

People might assume that an astronaut’s leadership skills come effortlessly, Duque says, but that’s not always the case.

“Everybody gives so much importance to your opinion, and sometimes I was surprised by that,” he says. “Being an astronaut, you tend to have a certain kind of leadership style because it’s what you have done for years without knowing it.”

Minister of science and other leading roles

His leadership style has served him well. After he retired from the ESA in 2018, he was appointed as Spain’s minister of science, a role he held until 2021. He oversaw the government’s policies on scientific research, technological development, innovation, space programs, and higher education. During his term, Spain committed to contributing US $800 million (€701 million) between 2020 and 2026 to the ESA—which at the time was the largest overall investment in the agency’s history.

After retiring as an astronaut, Pedro Duque headed Spain’s Ministry of Science, Innovation, and Universities. Today, the IEEE honorary member is the chairman of HispaSat, a Spanish satellite company.Isaac Buj/AP

In 2022, he joined Destinus Spain, a European defense manufacturer that develops technologies for aircraft propulsion and auxiliary systems. He advised its strategic committee.

The following year, Spain’s government appointed him president and chairman of the HispaSat board. Headquartered in Madrid, the satellite operator provides broadcasting and broadband services for Europe, North Africa, and the Americas. The public-private partnership was in its origin a joint initiative between the Spanish government and private telecom companies.

The IEEE honor was a surprise

Duque says he was surprised to learn that IEEE added him to its membership ranks. He was aware that a colleague had nominated him, but he deemed it unlikely the nomination would be supported.

“When I started looking into who its members were, I wondered why they selected me,” he says. “Obviously, electrical engineering is not my branch, but it could have been, because growing up, I was just as interested in telecommunications as I was in aeronautics.

“Most engineers know IEEE for its standards and the work it does in achieving consensus in standards development.”

After discussing with several colleagues who were IEEE members about the significance of the IEEE honorary membership—which is bestowed for a significant achievement and impact on society— Duque feels the award is a significant honor.

“I’m still in the early phase of figuring out how I can contribute to IEEE,” he says, “and what I can do for the many hundreds of thousands of members, all whom have impressive qualities.”

Throughout her career, roboticist Barbara Mazzolai has turned to nature for inspiration. Now she wants to ensure the technology she builds gives back to the environment, too.

After starting her career as a biologist, a chance opportunity saw Mazzolai switch streams to engineering and become an early pioneer of bioinspired robotics. Building on her knowledge of biology’s ability to solve a diverse set of problems, she has developed robots based on octopuses, plant roots, and even seeds. “I’ve always been fascinated by living organisms, [and] by the extraordinary variety of solutions in nature, selected by the evolutionary process,” she says.

Barbara Mazzolai

Employer:

Italian Institute of Technology

Occupation:

Associate director for robotics; director of the Bioinspired Soft Robotics Laboratory

Education:

Master’s degree in biology, University of Pisa; master’s degree in eco-management and audit schemes, Scuola Superiore Sant’Anna; Ph.D. in microsystems engineering, University of Rome Tor Vergata

But Mazzolai, now the associate director for robotics at the Italian Institute of Technology, in Genoa, also believes engineering needs to reckon with its own impact on the natural world. That’s why she is advocating for a new field of research she calls “sustainability robotics.”

In a manifesto published in Nature Machine Intelligence in July, she and her collaborators outline a vision for a new approach to designing robots that’s meant to improve the relationship between nature, humanity, and technology.

“We need to reduce the footprint of our technology,” she says. “It’s really about thinking in a different way to open new possibilities for robotics [and] for society.” In this new mode of thinking, Mazzolai considers sustainability a core component of the design.

A child of nature

Mazzolai traces her fascination with the living world back to her childhood growing up on Italy’s Tuscan coast, close to the port city Livorno. Her father was a public-health inspector and a professional mycologist, and the family spent a lot of time exploring forests and learning about the local fungi and plants.

After toying with the prospect of pursuing art, her other major passion, Mazzolai ultimately decided to enroll at the University of Pisa in 1987 to study biology. She was particularly drawn to marine biology, but shortly before graduating with a master’s degree in 1995, she secured a research position at the Italian National Research Council’s Institute of Biophysics studying the cycles of heavy metals like mercury through both living and nonliving parts of the environment.

This involved collecting and analyzing samples from water, soil, vegetables, and even humans to understand the impact these metals have on health and the environment. She balanced this work with studying environmental management at the Scuola Superiore Sant’Anna, in Pisa, graduating with a master’s degree in 1998.

During that time, however, she learned that the university was recruiting biologists to help design new devices for environmental monitoring. She applied for and got the job in 1999 and began working as a research assistant under renowned bioroboticist Paolo Dario, first developing sensors and then robots meant to monitor air, water, and soil.

Even before entering a doctoral program, Mazzolai was promoted to assistant professor in 2004 and shortly afterward made her first foray into bioinspired robotics. In collaboration with colleagues at Sant’Anna, she helped design a soft robot inspired by the octopus. “We proposed it as a paradigm for launching this idea of soft robotics: demonstrating that [robots] can be soft, but at the same time apply strong force to the environment, like the animal does,” she says.

Back to school

In 2007 Mazzolai enrolled in a Ph.D. in microsystems engineering at Tor Vergata University of Rome, which she balanced with her role at Sant’Anna. She was already relying heavily on microfabrication techniques to develop sensors for her robots, and she was keen to push that part of the field forward.

While robots frequently feature sensors designed for perception, such as tactile or proprioceptive sensors, these systems typically focus on understanding the robot’s position in its environment, she says. “But there are few robots that integrate physical or chemical sensors to really understand the environment they move in,” she adds.

“I’ve always been fascinated by living organisms, [and] by the extraordinary variety of solutions in nature.”

Mazzolai was appointed as a team leader at the Center for Micro-BioRobotics of the Italian Institute of Technology in 2009, where she continued her work on the emerging field of bioinspired robotics. Two years later, she completed her Ph.D. and was promoted to director of the center.

Planting the seeds

Around this time Mazzolai says she became interested in using plants as a model for new kinds of robots, expanding bioinspiration beyond just animals. In particular, she was captivated by the ability of roots to efficiently explore the underground environment, and she imagined machines with the same deftness could have applications in both environmental modeling and precision agriculture.

While many bioinspired robots mimic animals, plants also serve as a muse for Mazzolai. This tendril-like bot can coil around other structures like a vine. Italian Institute of Technology

When she first proposed the idea, colleagues were somewhat skeptical of robots based on seemingly static organisms. But in reality, she says, plants move nonstop through a process known as indeterminate growth. “They really grow for their entire life,” she says. “They adapt their morphology, their behavior to the external environment; they repair, they sense, they communicate.”

Trying to mimic a system that operates on such different principles to conventional robotics required some serious thinking, however. Mazzolai says that working in bioinspired robotics sometimes requires you to have “two separate brains”—one of a biologist and one of an engineer.

The process often involves deep study of the target organism to learn the underlying principles that shape how it operates before trying to engineer a robot capable of mimicking them. “It’s not a copy of natural organisms,” says Mazzolai, because a living organism is both difficult to replicate and has different goals.

In the case of plant roots, what makes them so efficient at exploring the soil is that they reduce friction by growing only at the very fine tip of the structure, while the thicker base of the root remains static. This significantly reduces the amount of energy required to push through the earth compared to that of a more conventional drill, which must push the entire structure from above.

To realize this principle in a robot, her team developed a miniaturized 3D printer that sits at the machine’s tip and feeds thermoplastic filament through a heated nozzle to build a snakelike body behind it. This allows the robot to push through the soil efficiently. The tip also contains sensors that allow it to avoid obstacles and detect nearby nutrients or water.

Making robotics sustainable

After spending so much of her career borrowing from nature, Mazzolai is now eager to return the favor. Many modern technologies, including plastics and car batteries, have been developed with little thought about how they will affect the environment at the end of their life cycles, she says.

She wants to ensure that robotics doesn’t follow the same path. This is the inspiration for what she and collaborators now call sustainability robotics. The approach has three central pillars: ensuring that robots have minimal impact on the environment; that they’re available to people from across the world and all socioeconomic backgrounds; and that they’re “symbiotic,” providing benefits to both humans and nature.

More concretely, Mazzolai would like to incorporate the concept of a life cycle into the design of robots, so that at the end of their useful life these machines can be reused, recycled, or even biodegraded.

While that might sound ambitious, she’s confident that all the ingredients to make it a reality are in place. And it’s a vision that she is certain will inspire future roboticists. “There are younger people who want to really work in this field because this is the future, their future,” she says. Facing the threat of ongoing environmental damage, “they want to develop something that can help.”

This article is brought to you by CoolIT, an Ecolab Company.

Beyond 250 kW a server rack can no longer be cooled by a hybrid approach of liquid and air. At this density a 70/30 liquid-air split leaves 75 kW of air load. The air cooling system needed to move it brings cost and complexity few operators will accept. The answer is near-total heat capture. Liquid takes effectively all the heat, air falls below 1 percent of the load, allowing the server to run fanless.

CoolIT builds these loops today from modular coldplate blocks proven across six generations of fanless designs. Processor thermal design power (TDP) keeps climbing generation over generation. This rising heat load is now cascading into the memory, networking, storage, and power components that once ran comfortably on air.

The heat escaped the chip

For years the story stayed simple. Cool the processor and let air handle the rest. That balance has shifted. As TDP climbs, heat spreads outward from the processor and cascades into the components around it. Memory, networking, storage, and power now run hot enough to demand liquid of their own. Engineers designing the next generation of AI servers face a board where heat capture rises with every launch.

Beyond 250 kW per rack, air cooling becomes the bottleneck. Near-total liquid heat capture enables fanless AI server designs built for the next generation of computing.

New parts, new rules

Unlike processors, which are cooled as flat rectangular packages, these peripherals come in a wide range of shapes, sizes, and mounting requirements, each with its own thermal limits. Some run cooler than the processor case temperature, others run hotter, which leaves them sensitive to a design tuned only for CPUs and GPUs. Operators need purpose-built solutions here, matched to the part rather than stretched across the board.

CoolIT engineers meet this with a deep toolkit. Conductive plates, vapor chambers, heat pipes, and thermal transfer plates move heat from components closer to the liquid path. Riding coldplates enable pluggable components. Each solution stays true to the component it serves.

CoolIT Customer Showcase: How GWDG Cools HPC & AI Systems with CoolIT’s Direct Liquid Cooling CoolIT

One loop, one server

Cooling the parts is one challenge. Uniting them is the real work. Full heat capture means folding every one of these solutions into a single server loop that distributes coolant effectively and remains easy to install. Connection reliability, coolant routing, and the time it takes to assemble the loop at rack integration determine whether a design thrives in production or stalls on the bench. CoolIT builds these loops from proven modular blocks, so operators gain performance and deployment speed within the same solution.

Density forces the decision

Rack power continues to climb toward 1 MW, and the case for liquid grows stronger at every step. A 70/30 split of liquid to air holds comfortably at lower density. Past roughly 250 kW it stops working. The 30 percent left to air becomes a 75 kW load inside a single rack, and moving that much heat demands a parallel air system whose cost and footprint few operators will accept. Adding density only widens the gap.

As rack power continues to climb toward 1 MW, CoolIT’s modeling places full heat capture as the standard server design for flagship rack-scale products through 2028.

The simpler, more efficient answer is to capture the heat in liquid and drop air to less than 1 percent of the total load. True 100 percent remains almost impossible to reach in the strictest sense, so the honest and achievable target is near-total capture. That distinction matters to engineers who value precision, and the direction stays clear either way. Full heat capture moves from a premium option to a mainstream requirement as density rises, and CoolIT’s modeling places it as the standard server design for flagship rack-scale products through 2028.

CoolIT delivers it

CoolIT scales heat capture all the way to 100 percent using modular coldplate building blocks proven across six generations of fanless server designs. Engineering teams are already working on designs for the maximum density racks coming next. As the cascade spreads and racks grow denser, near-total heat capture becomes the design that keeps AI running.

Talk to CoolIT about building a server loop engineered for total heat capture.

Shortwave radios offer a way to connect one location on Earth to practically anywhere else with minimal infrastructure. But these radios come with some drawbacks—a significant one being that, unlike satellite communications, their transmission rates for digital data are typically measured in just hundreds of bits per second.

Peter Bloom

Peter Bloom is the founder of Rhizomatica, a nonprofit that works with remote, indigenous, and off-grid communities around the world to build shortwave and cellular-communication infrastructure.

Peter Bloom is the founder of Rhizomatica, a Philadelphia-based nonprofit that has open-sourced a digital shortwave-radio set called the High-frequency Emergency and Rural Multimedia Exchange System, or HERMES. The set operates in the high-frequency (HF) band from 3 to 30 megahertz, as does Mercury, its digital modem. Rhizomatica staff travel around the globe to remote locations in countries like Bangladesh, Brazil, and Ecuador. Wherever they go, they use HERMES to help connect locals to the rest of the world.

Bloom spoke with IEEE Spectrum about how HERMES brings better data rates and encryption to shortwave radios.

How does HERMES connect remote locations?

Peter Bloom: We use the ionosphere as our satellite—or mirror—which helps us move information, voice, and data over really long distances. We’re using small radios that put out about 20 watts of power, and we can pretty reliably do 400- to 600-kilometer links between two radios. We’re talking about places that are not easy to reach, where it’s not simple to put terrestrial infrastructure.

What can HERMES send that a basic voice radio can’t?

Bloom: HERMES is a software stack—it’s a set of different programs that all work together in order to be able to send data over HF. HERMES allows you to send pretty much any file. Depending on what the file is, whether it’s a photo or an email or a voice memo, it just sends it as a file. It’s like a data pipeline over HF.

Why does sending files and data matter more than just voice?

Bloom: In emergency situations, people send their latitude and longitude over HF to say, “Hey, I’m here at this place.” People need to be able to send data over HF if there’s a manifest, a parts list, telemedicine—here’s what we have, here’s what we need. Instead of trying to read that out over the air, it’s much easier to just send the file. Same with a photo—if we need evidence that an area was logged illegally, we can just have someone send that over HF, rather than spending days getting down the river to get the photo where it needs to go.

Why did you build in encryption that amateur-radio regulations in many countries don’t allow?

Bloom: Encryption [regulations] for ham radio operators are different in each country. So it’s all optional—you turn it on, you turn it off. The reason we built the encryption is that some of the partners we work with are in very sensitive areas and don’t want to be sending out information that can be easily captured and used against them.

How has HERMES made an impact?

Bloom: We’ve been working with artisanal fishers in Bangladesh on a pilot project. There’s 10 or 11 boats that have HERMES systems on them. Pretty soon after we installed those, one of the boats had a mechanical issue in the Bay of Bengal, 100 or 200 kilometers offshore. They were able to send their GPS position and an SOS that they were having trouble. They were able to coordinate the rescue of the crew and the boat. So that was a really cool moment of HERMES in action that we’re super happy about.

This article appears in the October 2026 print issue as “Peter Bloom.”

This White Paper gives data engineers and architects a practical overview of how parallel partitioned reads, write-path optimization, and cloud-native bulk loading reduce large-table replication times, and why replication speed has become a business concern as data volumes grow.

Download this free whitepaper now!

Transitioning from years of working in a senior technical or executive role to a leadership position is one of the most challenging phases of a STEM career. It requires moving away from making decisions on your own to mentoring others, making strategic decisions for the organization, and collaborating with coworkers from different generations.

The shift in mindset is known as “legacy leadership,” a philosophy whereby success is no longer measured by personal achievements but by how effectively a senior leader empowers others.

To help seasoned professionals and senior experts navigate the transition, the inaugural IEEE International Leadership Conference (ILC) will provide attendees practical advice on cultivating collaborations and guiding emerging talent.

“Early in our STEM careers, we measure success by what we have achieved,” says Jeewika Ranaweera, cochair of the IEEE ILC program committee. “Later, we should measure success by what we enable, how many people we mentor, how much knowledge we transfer, and how many doors we open for the next generation.”

The ILC is scheduled for 3 and 4 October in Budapest. Registration is open.

Letting go of the “expert” identity

For decades, seasoned technologists have been valued primarily for their technical expertise. Shifting from that identity can feel uncomfortable, but legacy leadership requires measuring success by different standards. They include a leader’s influence on the staff, the ability to uphold the company’s mission, and empowering others to lead and succeed.

The transition requires leaders to find purpose outside their corporate titles, shifting their focus to the long-term sustainability of their teams, their organization, and the broader technical community.

“A professional legacy is not measured only by what we have achieved but also by sharing our knowledge, experience, and opportunities with others,” says Sudhanshu S. Jamuar, another program committee cochair. “The real transition from expert to a legacy builder happens when we stop asking, ‘What more can I accomplish myself?’ and start asking, ‘How many others can I enable to accomplish more?’”

Neeli Rashmi Prasad, IEEE ILC treasurer and sponsorship cochair, adds that the transition transforms a lifetime of technical work into a platform for future innovation.

“The true value of experience is not in how much knowledge we accumulate but in how intentionally we transfer it,” Prasad says. “When we partner with, mentor, and create space for others to lead, our expertise becomes a foundation for progress far beyond our own careers.”

Moving beyond advice-giving

True knowledge transfer requires coaching rather than advising, and mastering the art of active listening. To build deep trust with early-career colleagues and ensure a seamless transfer of leadership to the next generation, senior leaders must avoid offering unsolicited or outdated anecdotes. Effective mentorship is a collaborative loop in which senior experts contribute hard-won industry wisdom while remaining curious and learning from the fresh, cutting-edge perspectives of talented coworkers.

“Knowledge transfer is most powerful when it is a two-way bridge: experience flows from one generation to the next, while new ideas and perspectives flow back,” says Sudeendra Koushik, president of the IEEE Technology and Engineering Management Society and an ILC keynote speaker. “This approach transforms mentorship from simply passing on information into one where the next generation can question, experiment, innovate, and ultimately surpass what came before them.”

Seasoned professionals should encourage independent, disruptive thinkers rather than carbon copies of themselves, Prasad says.

“Legacy leadership is about building continuity,” she says. “We should not simply prepare the next generation to follow the paths we created; we should give them the confidence, knowledge, and networks to challenge those paths, create new ones, and take technology further than we imagined.”

Designing your next chapter

Leadership does not need to stop when one’s job ends; it can evolve. Seasoned, retired professionals can continue contributing meaningful service through pathways that align with their personal passions. They include:

• Advisory boards: steering corporate, technical, or community organizations.
• Civic engagement: applying engineering methodologies to solve community challenges.
• Volunteerism: mentoring the next generation of grassroots innovators through professional networks including IEEE.

Those pathways offer experienced professionals an opportunity to redefine success, not in terms of position, authority, or personal achievement but in terms of sustained impact.

“Retirement from a job should never mean retirement from purpose,” Koushik says. “Our experience becomes even more valuable when we use it in service of the profession, society, and the generation that follows.”

At the same time, the collaborative continuum relies on a proactive younger generation. Emerging leaders need to take responsibility for building their professional connections, Prasad says, advising: “Be bold, stay curious, and build your network early!”

A space for continuity

The conference is designed to combine a drive to cultivate emerging innovators with a deep reservoir of industry stewardship and strategic perspective. Rather than a single classroom session, the conference will feature a dedicated career-readiness and mentorship track where attendees can explore practical frameworks for building strategic professional networks, connecting with peer advocates, and establishing reciprocal knowledge-sharing opportunities across generations.

By participating, seasoned professionals can ensure their decades of expertise continue to yield dividends for generations to come.

“Our professional legacy is not the technology we build, the titles we earn, or the awards we receive,” Ranaweera says. “It is the knowledge we share, the lives we influence, and the future we help others create.”

You can view the agenda, read speaker biographies, and secure your seat at the ILC online.

This article is brought to you by VicOne.

Robot safety has traditionally asked: Can a machine remain safe when something goes wrong? Physical AI raises a harder question: Can a machine remain safe when an attacker changes what it sees, decides, or does even when nothing appears to have failed?

As AI and robotics continue to advance at an unprecedented pace, modern robots perceive through multimodal sensors, interpret context using AI models, and translate those interpretations into physical action. As they move into dynamic environments, their safety increasingly depends on the integrity of the data guiding their decisions.

That dependence creates risks that conventional safety assessments may not fully capture. Recent research has demonstrated that manipulating what a robot sees, hears, or interprets can influence its behavior without requiring direct control.

Such manipulation can occur anywhere across its complex sensing and decision-making system — a layered attack surface encompassing training pipelines, system infrastructure, and runtime perception.

Layer One: Corrupting intelligence at its source

In 2017, BadNets demonstrated that a model could behave normally under most conditions, yet fail in the presence of a specific hidden trigger. In one example, a subtle pattern caused a stop sign to be misclassified as a speed limit sign without affecting the model’s behavior on other inputs.

What began as a classification vulnerability has since evolved into action manipulation.

At NeurIPS 2025, researchers introduced BadVLA a backdoor attack targeting Vision-Language-Action (VLA) models that allow robots to see, interpret instructions, and produce coordinated physical movement. Rather than altering a single label, the attack caused conditional deviations in the robot’s action trajectory when a trigger was present. Without the trigger, the model largely preserved normal task performance, while the backdoor remained effective under task transfers and model fine-tuning.

A related study in 2025, GoBA, showed that ordinary objects such as a coffee mug could serve as a reliable trigger. The researchers reported a 97 percent attack success rate without degrading performance on clean inputs.

A critical safety question today is whether Physical AI models remain within their task and safety boundaries under adversarial conditions.

These studies expose a blind spot in model validation: A model may pass testing yet produce corrupted behavior when a hidden trigger appears in operation.

So a critical safety question today is whether Physical AI models remain within their task and safety boundaries under adversarial conditions. Simulation tools such as NVIDIA Isaac Sim, when paired with VicOne Radeis, can test the effects of manipulated inputs before deployment.

VicOne LAB R7 demonstrates Radeis, a Physical AI safety validator for NVIDIA Isaac Sim that tests how adversarial visual inputs affect robot behavior before deployment.VicOne

Layer Two: System vulnerabilities as gateways to AI control

Even a securely trained model can be subverted if the surrounding system stack is vulnerable.

In September 2025, researchers disclosed UniPwn, a Bluetooth exploit chain affecting quadruped and humanoid robots from a major manufacturer. Hardcoded cryptographic keys allowed traffic decryption, authentication checks were bypassed, and command injection enabled root-level execution. The exploit is also described as “wormable.” A compromised robot could scan nearby units and potentially affect an entire fleet.

VicOne Lab R7’s demo shows how chaining three wireless exploits can trigger uncontrolled robot behavior within 60 seconds, resulting in operational disruption.VicOne

Middleware creates another exposure point. Vulnerabilities in ROS 2 and DDS-based systems can enable arbitrary code execution or abuse unauthenticated topics to deliver malicious commands. With sufficient access, an attacker could override motor commands or replace AI model weights without directly attacking the model architecture.

In this case, the components may still function as designed. What has changed is the trustworthiness of the commands flowing through the system. Vulnerability management can help teams identify known risks before deployment, while continuous monitoring can surface emerging threats.

Layer Three: Manipulating perception and reasoning at runtime

At runtime, manipulating inputs that shape perception or reasoning may require neither firmware modification nor a network breach.

In 2024, RoboPAIR demonstrated how carefully structured prompts could redirect LLM-controlled robots into unsafe trajectories. BadRobot exposed a deeper architectural weakness: in several cases, a robot verbally refused a dangerous command while its motion controller executed the action anyway.

Vision-based manipulation is equally powerful. VLAttack showed that an adversarial patch within the camera’s view could reduce a VLA model’s task success rate to zero. FreezeVLA showed that a single adversarial image could freeze a robot’s decision-making loop, making it unresponsive to subsequent instructions.

Runtime assurance must therefore look beyond whether individual components remain available and assess whether cyber events are beginning to affect physical behavior.

In each case, the camera may still work, the model may still run, and the controller may still respond. Yet the resulting behavior can be unsafe because the robot is acting on manipulated perception or reasoning.

Runtime assurance must therefore look beyond whether individual components remain available and assess whether cyber events are beginning to affect physical behavior. Security event correlation, behavioral-impact assessment, and policy-bounded response supported by edge AI, can help contain the affected path without unnecessarily stopping the entire robot fleet.

From point-in-time safety to lifecycle assurance

The risks across these three layers reveal the missing layer in robot safety assurance: cybersecurity. Functional safety addresses failures and unexpected operating conditions; cybersecurity extends that assurance to deliberate manipulation, including attacks that may leave the underlying system apparently functional.

This requires assurance across the robot’s lifecycle. During design, teams need to understand which cyber risks could invalidate assumptions behind intended behavior. Before deployment, they should test whether realistic attacks can cause a robot to deviate from its task or safety boundaries. In operation, monitoring should identify whether cyber events are beginning to affect behavior, contain the affected path, and preserve safe operation where possible.

VicOne’s lifecycle approach combines AI model and vulnerability scanning, simulation-based validation, and continuous monitoring to help secure robots from development through operation.VicOne

While cybersecurity does not replace functional safety, it helps ensure that Physical AI remains within acceptable boundaries even when what it sees, decides, or does is under attack.

For a deeper look at the cybersecurity risks and defense strategies shaping autonomous robotics, download our whitepaper “Securing the Rise of AI Robots: Cyber Risks, Real-World Threats, and Defense Strategies.”

Learn how single-phase direct liquid cooling manages the rising heat of AI and high-performance computing, and how it compares with two-phase and immersion approaches.

Download this free whitepaper now!

Since about 2020, AI has largely focused on training bigger and better models. Large language models (LLMs) ballooned from millions of parameters to trillions. This proved effective: The largest version of OpenAI’s GPT-3, released in 2020, correctly answered just 43.9 percent of questions on a popular knowledge-and-reasoning benchmark. Just four years later, GPT-4o reached a score of 88.7 percent on the same exam, effectively matching those of human experts.

Advanced AI labs are still training ever larger models, but that training has somewhat receded to the background of the AI conversation. In 2026, inference—the use of trained models to produce code, write essays, or make images of ourselves as elves—has come to the forefront.

“It’s like training is yesterday’s news,” says Matt Kimball, principal data-center analyst at Moor Insights & Strategy. “All that any chief information officer wants to talk about is inference.” Nvidia CEO Jensen Huang, speaking at the company’s GTC 2026 conference, touted this change as the “inflection point of inference.”

Part of what’s caused the shift is very simple: LLMs are becoming useful, so people are using them. On top of that, many models on the market today are reasoning models. In response to a user’s query, they run inference not just once but multiple times, reprompting themselves in a process called chain of thought. Reasoning models generate longer outputs, and models with high reasoning effort can produce up to 20 times as much text as those with low or no effort. Adding even more to the world’s inference workload, the rise of agentic AI has resulted in inference running not just as a real-time response to a user’s query but also around the clock, working autonomously toward a user-defined goal.

Amazon’s Trainium chip was originally designed for AI training. However, Amazon Web Services chose to break up AI inference into two parts, with Trainium running the more computationally complex portion and Cerebras’s wafer-scale engine taking on the more memory-intensive portion.Amazon

The resulting explosion in inference demand has led to unexpected alliances among tech giants. OpenAI and Amazon have deployed chips the size of a dinner plate designed by Cerebras, despite Amazon having its own Trainium chips. Nvidia bought key talent and intellectual property from AI-inference startup Groq in a controversial deal worth US $20 billion. And Anthropic is paying LLM competitor SpaceXAI over a billion dollars per month to lease spare compute.

Although they might seem similar, AI training and AI inference are computationally different. These big moves from tech giants signal that in order to support the inference demand, we’re going to need a very different mix of hardware than experts may have expected even a couple of years ago.

How does AI inference differ from AI training?

An untrained LLM is like a jumble of Scrabble tiles on a table. Instead of single letters, though, the tiles show fragments of words, called tokens. Everything you’d need to write almost anything is present, but nothing makes sense.

Training a model organizes this jumble using a guessing game played at scale. The model is shown real text with the next token hidden and asked to predict what comes next. After each guess, the correct token is revealed and then compared to the prediction, and the difference is used to calculate the model’s accuracy. The game is played not with a single sentence but over billions of passages.

While a real game of Scrabble can be played over a bag of chips and a few drinks, AI training is computationally intense. The model updates its parameters through backpropagation, a process that repeatedly calculates how each of a model’s billions or trillions of parameters should shift to make the next prediction better. This is why tech giants are building larger data centers than ever before.

Eventually the model’s creator decides further training isn’t worth the cost, and the guessing game stops. Backpropagation ends, the parameters are frozen, and the LLM becomes a pretrained model. Fine-tuning—a short training run on smaller, more specialized data—adds final tweaks, and the model is deployed.

Next comes inference. This is the process of using the deployed model, which, now that it’s been trained, has learned to spit out Scrabble tiles—tokens—in a sensible order.

You might think that AI inference is less computationally demanding because the backpropagation calculations used to update parameters are eliminated. But Sudeep Bhoja, founder and CTO of the inference-hardware company d-Matrix, explains that inference adds new challenges.

The models are “autoregressive” in nature. That is, the next output depends on the previous one. “So to generate the next token, you have to read all of the weights and all of the [context] from the previous token,” explains Bhoja. The context includes all of your prompts, all of the LLM’s replies, and all of the files you upload. It’s a lot of data and a lot of processing.

An LLM generates its reply in two phases: prefill and decode. Prefill is the model reading a prompt. It processes every token at once, computing how each token relates to all the others. This operation is called attention, and it’s a defining characteristic of the transformer architecture behind modern LLMs. It allows them to respond to a word in its sentence, paragraph, and larger context rather than on its own. Think of it like arranging Scrabble tiles before you place them in a game. Many players move tiles around to imagine how they connect. Self-attention plays a similar role, though instead of moving physical tiles, each token sends a query to the others and receives a score indicating the token’s relevance.

These queries result in two types of vectors: the keys and values. They are typically placed in a store called the KV cache. This isn’t strictly required, as a model could instead recompute these vectors with each new token it generates. But nearly all LLMs use a KV cache to reduce how much computing they do. The KV cache is stored in memory and becomes a scratchpad to which the LLM can return to understand a conversation, and though it starts small, it can swell to dozens of gigabytes.

Prefill is a problem that can be easily divided up and worked on in parallel. This is why GPUs became the dominant AI accelerator as LLMs surged in popularity. Graphics rasterization (computing the color of every pixel on a screen) is also massively parallel, so GPU architectures were a natural fit.

Next comes decode. Here, the model generates its reply one token at a time. At each step it takes the most recent token, weighs it against everything in the KV cache, uses that information to predict the next token, and adds the new token’s key and value to the cache. Then it repeats in sequence, token by token.

This is where the autoregressive nature of the model works against inference speed. Predicting each token requires reading the entire model from memory, and that model consists of possibly tens to hundreds of gigabytes of parameters (the numbers representing what the model learned in training). Crucially, this is in addition to the memory required to store the KV cache.

As a result, the movement of all this data through memory often requires more bandwidth than inference hardware has available. So at least some of the computing parts of a GPU sit idle as it waits for data. Researchers found that Nvidia H100 GPUs running open-source LLMs sit idle 50 to 80 percent of the time.

Memory’s role in inferencing

Shahriar “Sha” Rabii, former head of silicon engineering at Meta and cofounder of the AI startup Majestic Labs, says idled processors are why many companies that are trying to improve AI-inference performance are laser-focused on memory. “With the GPU-based approach, you end up greatly over-provisioning compute and starved on memory. That’s driving the big [memory] scale out,” he says.

Bhoja’s d-Matrix and Rabii’s Majestic Labs both focus on this memory bottleneck. However, their companies imagine different solutions.

d-Matrix’s second-generation AI accelerator, Raptor, aims to improve inference performance by minimizing the distance between compute and memory. The GPUs in most current AI-inference deployments do this by placing high-bandwidth memory (HBM) around the perimeter of the GPU. Each HBM is a stack of DRAM dies linked together and connected to a superfast interface to the GPU. This is great for training, but for inference, the amount of memory you can stack this way and the bandwidth it can provide leave something to be desired.

d-Matrix’s stacked-die architecture

d-Matrix’s Raptor removes that bottleneck by stacking an AI accelerator on a DRAM die. Instead of stacking memory, d-Matrix stacks memory and compute. Bhoja says this reduces the distance that data must travel to “micrometers instead of millimeters.” Like building a skyscraper, going vertical makes it possible to do more inside the same physical footprint.

Majestic takes the opposite approach. Instead of trying to minimize the length that data must travel between compute and memory, the company is focused on improving the memory interface to accommodate longer wire traces while keeping bandwidth high. Longer wires allow Majestic to connect memory stacks that aren’t directly next to the GPU, removing the space limitation of HBM.

“A memory interface has a very short physical distance it can operate over. In the case of HBM, it’s up to 2 or 3 millimeters. You have this shoreline around the periphery, which is the only place where you can put HBM,” says Rabii.

Majestic claims its memory interface can transmit bits as far as about a meter. That’s achieved with a proprietary copper link and a memory-aggregator chip that coordinates data. “The aggregator is the endpoint for the high-speed interface and a way to fan out to many, many commodity DRAM chips,” says Rabii. Because of this, Majestic can support up to 128 terabytes of DRAM memory in a single server rack—a significant increase over Nvidia’s GB300 NVL72 rack, which has about 20 TB of HBM3E.

Majestic Labs’ memory-aggregation architecture

d-Matrix and Majestic have one thing in common: Instead of HBM, they both use off-the-shelf DRAM. This is the most common type of computer memory in the world; it’s in everything from smartphones to cars. Memory analyst Jim Handy says HBM costs two to three times as much as DRAM. d-Matrix and Majestic chose DRAM in part because of this price advantage. However, the proponents of HBM, which include memory giants like Samsung and SK Hynix, aren’t sitting idle.

HBM4, the latest version of HBM memory, is now in production and will be used by Nvidia’s Vera Rubin GPU, which is expected to ship in the second half of 2026. Hoshik Kim, head of memory-systems research at SK Hynix, says HBM4 “will decisively break the memory bottlenecks constraining AI inference today” by doubling HBM’s maximum memory bandwidth and increasing the amount of HBM memory per stack.

Combining chips for faster inference

The big players—Nvidia and Amazon—are going for an all-chips-on-deck approach. Nvidia’s GPUs and Amazon’s Trainium training accelerators are still great for part of the inference workload: the prefill stage, where all the context keys and values are calculated. But to accelerate decode, the part where new tokens are generated, they are looking to new, memory-centric architectures from smaller players.

In Nvidia’s case, the smaller player was Groq (not to be confused with Grok, the family of LLMs trained by SpaceXAI). Nvidia purchased intellectual property and hired talent from Groq at the end of 2025, and just three months later at the Nvidia’s GTC 2026 conference, Jensen Huang unveiled the Nvidia Groq 3 language-processing unit (LPU). Groq’s architecture relies on memory—in its case, SRAM—built directly into the chip’s architecture.

Unless you’re a chip architect, or a hardcore PC gamer, you probably never give SRAM a thought. SRAM has the benefit of being tightly integrated into a compute chip’s architecture—it’s on the same piece of silicon as the processor—and has the drawback of being less dense and more expensive than DRAM. Most chips include only a few dozen megabytes of SRAM. AI inference, however, has ignited new interest in SRAM as a means of bringing the model weights stored in memory closer to compute.

Ian Buck, vice-president and general manager of hyperscale and high-performance computing at Nvidia, says the LPU has a much different set of priorities than the company’s GPUs. The LPU has far less raw computing power than a standard GPU, but it gains 500 megabytes of on-die SRAM connected directly to its floating-point math units. “The benefit is the memory bandwidth. The LPU has seven times the memory bandwidth of the GPU,” he says.

Between the Rubin GPU and the Groq LPU, prefill and decode can both be accelerated to get the best of both worlds, the theory goes. “We do all the attention math and context processing on the Vera Rubin [GPU] rack,” explains Buck. “For all the expert calculations…the matrix multiplications, we do that part on the LPU.” The company packs 256 LPUs into the Groq 3 LPX, a system the size of a data-center rack.

Nvidia’s two-chip approach to inference

Amazon Web Services (AWS), for its part, struck a deal with Cerebras, to pair the Trainium accelerator with Cerebras’s Wafer-Scale Engine 3 (WSE-3). Cerebras takes a similar approach to Groq, though at a much larger scale. WSE-3 turns an entire silicon wafer into a single chip that contains over 4 trillion transistors. The design doesn’t connect to external memory but instead etches 44 gigabytes of SRAM into each wafer. “We store the [model] weights on the SRAM,” says James Wang, formerly director of product marketing at Cerebras who has since moved to SpaceXAI. “So that’s easily 40 to up to 80 billion parameters that we can support on one chip.”

Amazon plans to use AWS Trainium chips for prefill, and Cerebras for decode. But Cerebras’s chips can also go it alone in inference. WSE-3 was deployed by OpenAI to power GPT-5.3-Codex-Spark, a variant of the company’s coding mode, outputting over 1,000 tokens per second. For comparison, OpenAI’s standard GPT-5.4 deployment outputs 50 to 125 tokens per second.

Amazon Web Services’ two-chip inference strategy 

Cerebras can also tackle prefill without moving the workload to different specialized chips. For this, it networks together multiple WSE-3 chips to form a single pool of memory. Cerebras has demonstrated it can serve models with up to 1T parameters, such as Moonshot AI’s Kimi 2.6, though Wang says “the architecture has no innate limitation in terms of how many parameters it will do.”

Despite these differences in strategy, Nvidia and AWS seem to agree that the future of AI inference will be solved by a systems approach that pools different kinds of chips together to tackle the largest LLMs. Or, as Buck says: “To do modern AI inference, you need all the chips.”

Learning to do more with less (bits)

Nvidia became the world’s most valuable tech company because it designed the world’s most desired GPUs. But not all of the attention is focused on improving AI-inference hardware. AI researchers are also learning how to optimize LLM software and hardware in tandem to make the best use of the memory and compute components.

Most computers store numbers in a 32-bit or 64-bit format. These determine how many bits are available to represent a single number. If too few bits are available, the number can’t be stored without losing information. The quality of an LLM benefits from more-precise number formats, but this creates a problem for inference performance. More-precise numbers aren’t free. The bits that describe them take up more space in memory and require more silicon and energy to compute.

Gilles Backhus, cofounder of the AI-accelerator company Tensordyne, says this creates a tension between model size and number precision. “Would you prefer a model that is size x but runs in 8-bit, or would you prefer a model that is twice the size but runs in 4-bit?” The size of each model will be roughly the same in terms of memory and compute, “but the 4-bit approach gives you twice as many synapses, if you will. And people are figuring out that [the 4-bit approach] is worth it.”

The process of converting an LLM from a more-precise number format to a less-precise format is called quantization, and it’s been in use for several years. However, researchers are finding new ways to quantize models down while retaining a large majority of the model’s quality.

Nvidia recently created a new 4-bit number format, NVFP4, for this purpose. AMD, Intel, and Qualcomm have instead rallied around a competing 4-bit number format called MXFP4 that Nvidia also contributed to developing. “It’s the black art of AI,” says Buck, of Nvidia. When Nvidia quantized DeepSeek-R1 from FP8 to NVFP4, scores on seven major benchmarks degraded by less than one percent while performance improved by three times, the company says.

Quantization is likely just the tip of the spear, as AI researchers and startups are investigating a diversity of opportunities for optimization, some of which could dramatically change the silicon found in AI-inference hardware.

Tensordyne’s unique approach to AI inference combines a logarithmic number format with bespoke hardware in the company’s Napier chip. Tensordyne

Tensordyne is expected to accelerate AI inference with a logarithmic number system that leans on a property of logarithms: The log of A times B equals the log of A plus the log of B. So, storing numbers as their exponents lets the chip add where it would otherwise multiply. That matters in silicon because multiplier circuits draw more power and use more die area than adders do. Tensordyne says its rack-scale hardware, called Napier, can produce up to 1,300 tokens per second per user, and can do so while using less than a tenth as much power as comparable Nvidia hardware.

Etched, a startup based in San Jose, Calif., is even designing AI accelerators that translate the transformer architecture used by LLMs directly into silicon. Rather than building general-purpose GPUs, the company is wiring up the connections needed for efficient transformer calculations into its chip, making the chip much less flexible but more efficient for the tasks most performed by current LLMs. The company says its first AI accelerator, Sohu, can run Meta’s Llama 70B model at a stunning 500,000 tokens per second, though this approach also means it won’t be able to run LLMs that move away from a typical transformer architecture.

Whether these ideas will prove fruitful remains to be seen. Etched just shipped their first rack in August. Tensordyne believes its first hardware will be available in 2027. Even so, these startups show how the demand for inference performance is fueling unconventional ideas.

Inference is everyone’s game

The sheer variety of approaches to AI-inference acceleration—stacking compute on memory, extending interfaces from millimeters to meters, using an entire silicon wafer for SRAM, squeezing models into 4 bits—raises a question: Which is going to win, and which is going to lose?

But that’s likely not the right question, experts say. The demand for AI is currently insatiable, and while fears of an AI bubble stalk the industry, it has yet to hamper growth.

On the contrary, Kimball of Moor Insights & Strategy thinks inference could drive intense demand for AI hardware in the long term, because it’s not obvious where that demand will end. “You could add a million agents into your organization,” he says. “These things work 24 hours a day; they don’t go home at five at night like we do.”

If AI inference remains as desirable as Kimball expects, the evolution is likely to follow the same trajectory as the CPU. The CPU didn’t improve along a single axis but instead across multiple fronts simultaneously. Once transistor scaling slowed, chip and system architecture innovations of all kinds proliferated. The list of individual innovations that led to today’s ubiquitous, powerful personal compute could fill dozens of books.

A few decades from now, the history of AI inference innovation will show similar depth.

In 2003, Martin Eberhard, a cofounder of Tesla Motors, decided it was time to start wooing investors. To do that, however, he needed an electric car. So he talked to Alan Cocconi and asked if he could borrow the tZero, the revolutionary and blazingly fast electric roadster that Cocconi had built at AC Propulsion.

Eberhard’s idea was to drive the tZero up and down Sand Hill Road in the heart of Silicon Valley and do demonstrations for curious entrepreneurs and VCs. Eberhard was joined by Tom Gage, Cocconi’s partner at AC Propulsion, on many of the visits. Like Tesla, AC Propulsion was also seeking investors, but to build a considerably more utilitarian EV.

Adapted with permission from The EV Guys: How Caltech Engineers Reinvented the Electric Car, by Charles J. Murray, published by Purdue University Press.

In December 2003, Eberhard also proposed a demo at Buck’s of Woodside, a popular restaurant frequented by tech entrepreneurs. At 5 o’clock on any evening, Buck’s probably had more VCs per square foot than any building in the country. Eberhard’s plan was to “show off what a real electric sports car can do,” he wrote in an email to Buck’s owner, Jamis MacNiven. MacNiven happily obliged.

In some ways, the tZero was a hit. When a VC would ride shotgun in the car with Eberhard at the wheel, Eberhard would implore them to touch the dashboard. As they reached forward, he’d punch the accelerator. As the car accelerated and the g forces piled up, the VC was literally unable to touch the dashboard. That was how powerful the tZero’s acceleration was, Eberhard would say.

Many of the VCs were astounded. Some even questioned whether the car was really electric. Many owned Ferraris or Lamborghinis. They knew sports cars—but this? They could never have imagined it was possible to do this with an electric drivetrain.

On December 13, 2003, Martin Eberhard brought AC Propulsion’s tZero electric roadster [yellow] to Buck’s of Woodside, a popular hangout for entrepreneurs and venture capitalists. Next to the tZero is a Scion xB, which AC Propulsion’s principals thought they could turn into a mass-market electric vehicle.Martin Eberhard

Still, the demo at Buck’s garnered little investor interest—with one exception. Google cofounders Sergey Brin and Larry Page were both at Buck’s that day, and they told Gage they knew an individual whose funds were liquid, as he’d recently sold his stake in a startup. What’s more, this individual liked fast cars.

The man’s name was Elon Musk.

Elon Musk, Meet the tZero

In 2004, it wasn’t apparent to anyone that Elon Musk had a future in the auto industry. He was notable for cofounding PayPal, which he then sold to eBay in 2002 for a whopping US $1.5 billion. He had already launched Space Exploration Technologies Corp., or SpaceX, with the stated goal of paving the way to a sustainable colony on Mars. Musk did love fast cars. He owned a million-dollar silver McLaren F1, one of only 64 road-going F1s in the world, as well as a 400-horsepower BMW M5 sports car and a 1967 XK-E Series 1 Jaguar roadster. But he’d never expressed an interest in building cars or starting an auto company, at least not publicly.

Elon Musk was photographed in 2008 at Tesla’s headquarters, then in San Carlos, Calif., around the time when Tesla’s Roadster was being delivered to its first customers.Patrick Tehan/MediaNews Group/Bay Area News/Getty Images

Still, Musk’s affinity for fast cars made it almost impossible for him to ignore an email from Gage on 21 January 2004. “Sergey Brin and JB Straubel both suggested you might be interested in driving our tZero electric sports car,” Gage wrote. (Straubel was the young Stanford engineering graduate who would later serve as Tesla’s chief technology officer.) “The tZero goes quite well,” the email continued. “We ran it against a Viper last Monday and it won four of five sprints on a 1/8th of a mile track. I lost one because I was carrying a 300-pound cameraman. Do you have time for me to bring it by?”

Musk quickly replied. “Sure, I would really enjoy seeing it. Don’t think it could beat my McLaren (yet) though I’m in town Feb 2nd through 4th.” “Hmm, a McLaren, boy that would be a feather in my cap,” Gage wrote back. “I can have it there on Feb. 4.”

Two of the principals behind AC Propulsion were businessman Tom Gage, left, and engineering genius Alan Cocconi.Left: Tom Gage; Right: Alec Brooks

The emails marked the beginning of Musk’s involvement in electric cars and in the auto industry. Gage drove the car to SpaceX headquarters, a warehouse in El Segundo, Calif., about 30 kilometers southwest of Los Angeles. In Musk’s cubicle in the “office” portion of the warehouse, Gage made his pitch.

There was a void in the market, he said. GM had abandoned the EV1. Toyota, Honda, Ford, and Chrysler were shutting down their electric car programs. California’s zero-emission vehicle (ZEV) rules, which mandated the sale of increasing numbers of vehicles with no tailpipe emissions, had been plundered. But electric vehicle technology, he said, was getting a bad rap. Here was the tZero, an electric car that could take off like a jet. The tZero proved that the technology was readily available. He and Cocconi wanted to use that technology to make an electric car that was useful and practical: the eBox, an electrified Toyota Scion.

After building the tZero roadster, AC Propulsion’s principals pinned their hopes on an electrified version of the Toyota Scion they called the “eBox.” It did not appeal to Elon Musk.Jeff Chiu/AP

For Musk, Gage’s introduction of the eBox was unexpected. He was meeting with Gage because he was interested in the tZero. It was the car’s performance that appealed to Musk.

He wasn’t interested in the eBox. He then drove the tZero and offered to buy it.

The lithium-ion version of the tZero electric roadster could get 515 kilometers on a charge and go from zero to 97 km/hr (60 miles per hour) in 3.6 seconds. Only three tZeros were built and only two survive. Scott Sorbe

Gage told him it wasn’t for sale. Undeterred, Musk offered a quarter million dollars if AC Propulsion would squeeze its lithium-ion battery pack into his Porsche. Gage declined again. AC Propulsion needed money to electrify the Toyota Scion, Gage said.

Musk shook his head. The idea seemed incredible to him. “Who wants to take an ugly $20,000 car and buy it for $65,000?” he asked incredulously, as he later recalled during an interview with Vanity Fair magazine. “I wouldn’t want to drive it. My wife certainly wouldn’t want to drive it.”

Many years later, Musk would tell his biographer Walter Isaacson, “Nobody is going to pay anything near that for something that looks like crap.” Musk believed that the way to start a car company was to build high-priced cars first and then let the technology trickle down to the mainstream. It was a classic Silicon Valley approach.

Alan Cocconi, the engineering whiz behind AC Propulsion, stands next to the company’s legendary tZero electric roadster in a picture taken in the early 2000s. The small yellow wheeled pod on the other side of the car is a trailer with a small gasoline engine that, when connected to the tZero, turned it into a hybrid gas-electric vehicle.Martin Eberhard

In Musk’s mind, it was all very obvious. He liked fast cars. He liked the tZero and believed it “could change the world.” He couldn’t even imagine why Gage was sitting here trying to sell him on the idea of the eBox. “Gage and Cocconi were sort of madcap inventors,” he told Isaacson. “Common sense was not their strong suit.”

Gage concluded that he wasn’t going to convince Musk to invest in AC Propulsion. “Well, if you want to do a sports car, then you should talk to Martin Eberhard,” Gage said. A few weeks later, Gage sent an email to Eberhard introducing him to Musk. “Elon Musk heads up SpaceX, is a car enthusiast,” Gage wrote. “He would be interested in hearing about your activities at Tesla Motors.”

Elon Musk, Meet Tesla Motors

As it happens, Eberhard and Marc Tarpenning, Eberhard’s partner and cofounder at Tesla, had considered contacting Musk even before Gage’s email arrived. They’d known of Musk and appreciated the way he thought. A few years earlier, they saw him speak at a Mars Society conference at Stanford University. Musk had talked about the rather improbable idea of sending mice to Mars. The presentation gave them a window into Musk’s unconventional approach to high-tech entrepreneurism and to life in general.

Eberhard and Musk agreed to meet, and then Eberhard emailed Gage. “Any chance of my borrowing the car for next week?” he wrote. Gage, of course, complied.

Martin Eberhard posed next to an electric motor at Tesla’s San Carlos, Calif., headquarters in 2006. Paul Sakuma/AP

By this time, Tesla was nine months old. It still had just three employees—Eberhard, Tarpenning, and Ian Wright, a New Zealand-born engineer and neighbor of Eberhard’s. The founders were arranging to pay the licensing fee on AC Propulsion’s drivetrain technology. And they were making arrangements to build their first cars using the chassis of the Lotus Elise two-seat roadster. They estimated they needed $6.5 million to go further. And that’s where Musk came in.

The original Tesla Roadster prototype, or “Mule,” was built inside the chassis of a 2002 Lotus Elise. Dylan Stewart/Image of Sport/Sipa/Alamy

Eberhard and Wright flew to Los Angeles on a Friday and met Musk in his cubicle at SpaceX. The meeting was supposed to last a half hour, but Musk’s questions came virtually nonstop, and as the meeting progressed, he repeatedly shouted to his assistant to cancel his next meeting.

Over the following weekend, Musk called Tarpenning to get his input about their financial model. “I just remember responding, responding, and responding,” Tarpenning said, according to a 2015 book by Ashlee Vance.

The Tesla founders were all impressed with Musk. He was unlike any of the VCs they’d met with in the previous months. He was technically astute. He’d earned a bachelor’s degree in physics from the University of Pennsylvania, and in his two-day stint as a Ph.D. student at Stanford, he’d intended to do a dissertation on solid-state capacitors for use in electric cars.

Moreover, he wasn’t averse to risk—at least not intelligent risk. He loved technical challenges, and he loved proving that the impossible was possible. “You’re presenting an electric car company to this person on the other side of the table, and he’s doing something even crazier,” Tarpenning said later. “He’s building rocket ships.”

On the Monday after their first meeting at SpaceX, Eberhard and Tarpenning flew back to Los Angeles. Musk agreed to invest $6.35 million. He would become the biggest shareholder as well as chairman of the company.

Now, Tesla Motors was really in business. All it needed was someone to design and build a groundbreaking electric car.

“All Electric Cars Have Sucked”

No one at AC Propulsion believed that Tesla Motors had even a remote chance of success. The whole idea—building and selling electric vehicles and competing against the giants in Detroit, Japan, and Germany—seemed impossible. Even Toyota, which was having so much success with the hybrid Prius, was not planning to build pure electric cars.

The prospect of starting any kind of auto company was unbelievably daunting. Automotive startups had been the undoing of many ambitious entrepreneurs, including Henry Kaiser, Preston Tucker, and John DeLorean. Such endeavors required mountains of money, connections, and expertise. There were unseen obstacles around every corner. And the people who’d launched Tesla, as smart as they were, were almost certainly unprepared for what lay ahead.

Years later, Musk would contend that their struggles were caused by the fact that Tesla had been founded on “two false premises.” The first was that Tesla’s founders believed they could simply convert an existing gasoline sports car to electric. The second was that they could use the existing AC Propulsion technology with little or no modification. “That turned out to be, in retrospect, staggeringly dumb,” Musk said.

In April or May of 2004, Tesla engineers worked on an early test vehicle, or “mule,” of the Roadster. The group included (clockwise from upper left) mechanical engineer Gene Berdechevsky, in the pink shirt, electrical engineer Phil Cole, and mechanical engineer Dave Lyons, in the dark blue shirt.Martin Eberhard

He also later concluded that their early path almost doomed them. “It ended up being much worse than if we had designed the car from scratch,” he said. But Tesla decided it could not go back and start over. It could only deal with the problem at hand.

Eberhard and Tarpenning were terrified of going into production with the existing analog motor controller and drivetrain electronics, which were unreliable and jittery. If those problems weren’t fixed, they knew their new vehicle would fail, and so would the company.

Another looming issue was the safety of the Tesla lithium-ion battery pack. To test it, Eberhard brought the engineering team to his home, where they dug a pit in his backyard. They took a brick of cells, covered it with a sheet of Plexiglass, and then remotely heated one of the cells with an electrical wire. As they expected, the heated cell burst into flame, setting neighboring cells on fire. The cells went off one at a time—pop, pop, pop. “We had a conflagration,” Eberhard said. “One cell caught fire, and it blasted right through the pack.”

Buyers of sports cars were known to be forgiving. In their quest for performance, they could put up with poor reliability. But the fire hazard was another matter, and one that had the potential to take down the company. Eberhard took the news right to Tesla’s board of directors. “It was my first big oh-shit moment to my board,” he said. “I told them we’ll have a day-to-day schedule stop until we figure this out.”

Working with friends from his Stanford days, Straubel began developing a new pack in his garage. The team acquired 7,000 lithium-ion cells from LG Chem, then constructed battery bricks, each with 69 cells, and tested them with different kinds of liquid-cooling systems. By October 2004, they’d finished a prototype pack and used a crane to lower it into the back of a Lotus Elise sportscar.

A few months later—in January 2005—the team had completed a working prototype of that first car. At the end of the month, they showed it off at a board meeting, and Musk took it for a spin. Impressed by its performance, he invested $9 million more, and Tesla completed a $13 million round of funding. Now the vehicle had a name—Roadster—and a tentative production schedule. The plan was to begin delivering it to customers in early 2006.

Tesla’s struggles with the Roadster were not apparent to the outside world, especially to those invited to the reveal of the Roadster at the Santa Monica Airport in July 2006. By then, the yellow test car, or “mule,” had evolved into two prototypes: a red car and a black car, both of which would be available for drives at the event.

Tesla unveiled the Roadster, its first vehicle, at the Santa Monica, Calif., airport on 19 July, 2006.Glenn Koenig/Los Angeles Times/Getty Images

The prototypes were more advanced than the mule, with more of a production-type design. But Musk and the team didn’t know what to expect at the reveal. The company was just coming out of stealth mode, and at that point, Tesla had received no media coverage. It had no customers, no deposits, and no sales team.

Still, Musk planned a huge party for the unveiling—an “awesome event,” in his words, staged inside the airport’s Barker Hangar. He told his personal assistant to invite 350 guests, including Michael Eisner of Disney, movie producer Richard Donner, actor Ed Begley Jr., California Governor Arnold Schwarzenegger, and many other luminaries. All were told to bring their checkbooks in case they wanted to write a $100,000 check to put a deposit on an electric Roadster. Meanwhile, a Roadster prototype zipped around a makeshift road inside the hangar, out the door, down a runway, and back inside again.

Musk took center stage, telling the audience that they were witnessing the start of a new era in automotive technology, according to CNET’s coverage of the event. “Until today, all electric cars have sucked,” he told the audience. “Electric cars play into the strength of Silicon Valley. A lot of the things inside the car are conventional automobile technology. The magic is the battery technology and the software and the controllers.”

Tesla Hooks Arnold Schwarzenegger, George Clooney

Musk’s message was perfect for such an event, especially for the dozens of reporters who were there to publicize the reemergence of the electric car. They adored the Roadster. It was small, powerful, electric, and above all, cool. It was anti-Detroit—a new kind of car that burned no gasoline and was born in Silicon Valley instead of an antiquated factory in Michigan.

The night was also a financial success for Tesla, with twenty $100,000 checks gathered from prospective buyers. And the momentum continued. A few days after the event, Joe Francis, creator of the adult entertainment franchise Girls Gone Wild, sent an armored truck to Tesla’s San Carlos office to drop off $100,000 in cash. A few days after that, Schwarzenegger put his money down, as did actor George Clooney. Within two weeks, Tesla had presold 127 Roadsters.

California Governor Arnold Schwarzenegger was among the first buyers of the Tesla Roadster on the day the car was officially unveiled, 19 July, 2006.Glenn Koenig/Los Angeles Times/Getty Images

Meanwhile, though, the company’s manufacturing woes continued. The mechanical problems weren’t even the biggest issue. The biggest issue was the supply chain. This was ironic, because some in the media admired Tesla for its global approach. They liked the fact that the battery pack, the motor, the chassis, and the assembly had an international flavor. It was a world car, they thought.

But for Tesla, it was a nightmare. The battery pack was being assembled in Thailand by a manufacturer of barbecue grills. The facility was 3 hours from Bangkok, literally in a jungle where the heat was almost unbearable, and the factory building consisted of a truss roof held up by some steel columns. There were no walls because no one there wanted to work indoors.

And because the pack assembler was inexperienced, Tesla engineers were repeatedly flying to and from Thailand to direct the effort. They would find animal droppings on the battery packs, which were sitting out in the open air all day and all night.

For the umpteenth time, Musk wondered if the company would be able to survive. “We’re doomed if we don’t in-source the battery pack,” he told one of Tesla’s manufacturing engineers, “because we have a supplier in Thailand who is great at making barbecues but not great at battery packs. And the supply chain is so long that it takes six months from when the cells are built to when the battery pack is done and in a car. So that means the capital cost is gigantic because we have to pay for all that inventory and process. And inevitably, there are mistakes in the design or fabrication of the battery pack, and then we have six months’ worth of battery packs that don’t work.”

Never mind that this chaotic approach was central to their plan. Tesla had never been envisioned as an old-fashioned, Detroit-style, vertically integrated manufacturer. From the beginning, it had been a Silicon Valley–type enterprise that would rely on others for the bulk of its manufacturing. Only now, as the fledgling company sent batteries and motors and assembled cars back and forth across two oceans, was its plan beginning to appear untenable. “We had this misguided idea that everything must be cheaper and better if built in Asia,” Straubel later said.

Tesla’s Chances of Success: 10 Percent

From the beginning, Musk had never been optimistic about Tesla’s chance of success. He repeatedly said he thought it was approximately 10 percent. “In 2004, the idea of starting a car company was extremely stupid,” he said. “The idea of starting an electric car company was stupid squared.” As he watched Tesla struggle with its supply chain, his earlier words were starting to look prescient.

The only chance, the engineering team concluded, was to bring the manufacturing of all of the subsystems, such as the battery packs, motors, and inverters, in-house. They disassembled their overseas operations and moved them to California, starting with battery pack manufacturing. Assembly stations were loaded into huge shipping containers, transported back to one of the company’s new facilities on Bing Street in San Carlos, and then reassembled there. It took five and a half months. They also redesigned the battery packs and developed machines for automating their assembly.

In 2008, with mass production of the Tesla Roadster just getting underway, Elon Musk gave an interview at the company’s headquarters, then in San Carlos in northern California. At the time, Tesla was merely a startup in a precarious position, bleeding cash and grappling with many manufacturing problems.Ryan Anson/Bloomberg/Getty Images

Musk concluded that the key to success was not the design of the car itself but rather the manufacturing. Henry Ford had reached the same conclusion a hundred years earlier. It was “the realization of how important it is to build the machine that builds the machine,” Musk said at the Tesla Annual Shareholder Meeting in 2016. “And how much harder it is to build the manufacturing system that builds the product, than it is to create the product in the first place. You can create a demo version of a product…with a small team in maybe three to six months. But to build the machine that builds the machine takes at least a hundred to a thousand times more resources and difficulty.”

Gradually, Tesla’s idea of letting others do its manufacturing slipped away. Packs were built in San Carlos, and then installed in the Lotus Elise chassis there instead of in England.

“We had control now,” said manufacturing engineer Jason Mendez. “We had all the engineers there. We didn’t have batteries on the water, not from Thailand to England and not from England to here.”

Musk began to talk about a new vision. He called it the gigafactory. Raw materials would enter at one end, and a car would exit at the other end. This was the ultimate in vertical integration, and it sounded a lot like Henry Ford’s vision for the River Rouge plant in Dearborn, Mich., in 1917.

Tesla Motors was becoming an auto manufacturer.

Social media plays a significant, multifaceted role in adolescents’ development, influencing how they communicate, learn, socialize, and express themselves.

The benefits, however, are accompanied by risks that can undermine youngsters’ character as well as their cognitive and social development.

The potential problems include excessive screen time, social media addiction, cyberbullying, misinformation, radicalization, privacy violations, exposure to inappropriate content, sextortion, and doomscrolling.

A recent study published in Nature: Human Behaviour found that adolescents who begin using social media at an early age tend to have significantly lower academic performance. A Mashable article highlights additional issues including effects on mental health, self-harm, addiction to social media, compulsive, repetitive checking, and exposure to pornography and violent material.

Protecting minors has largely fallen to parents, schools, and self-regulation by some social media providers.

But that approach has proven ineffective and inadequate, so some governments and policymakers have stepped in and placed responsibility on social media providers.

Australia’s nationwide ban

Australia was the first country to legislate a nationwide social media ban on children younger than 16—which I wrote about in January for Communications of the ACM.

Enacted in December, the ban initially applied to 10 platforms: Facebook, Instagram, Kick, Reddit, Snapchat, Threads, TikTok, Twitch, X, and YouTube. It excluded messaging, gaming, and nonsocial platforms including Discord, GitHub, Roblox, WhatsApp, YouTube Kids, and educational tools.

The law places the responsibility for enforcement on the platform providers through age-assurance mechanisms, requiring the platforms to take “reasonable steps” to prevent those 15 or younger from creating or holding accounts.

It does not, however, apply to content consumption. Children can view publicly available posts and videos without logging in; they cannot comment or post, according to the law.

The legislation mandates that the user’s age be verified with tools such as government-issued identification, biometric or facial age-estimation tools, behavioral or inference algorithms, and self-declaration with optional checks.

Penalties for noncompliance can reach US $35.6 million.

The 10 platforms subsequently removed nearly 5 million accounts of young users.

Although the ban received widespread support, human rights organizations and digital freedom advisory groups have argued that it limits young people’s freedom of expression and access to useful information. They say the ban might contribute to social isolation and the loss of support networks, particularly among marginalized youth.

Promising early outcomes

The Australian ban is producing positive outcomes, according to a recent Time magazine article, “What the World Should Learn From Australia’s Social Media Law.”

Early findings indicate it has reduced account ownership and social media use among young children. A YouGov survey of Australians found that 61 percent of parents of children age 16 and younger reported positive changes including more face-to-face interaction, greater presence and engagement, and improved parent-child relationships. Three in five Australians surveyed called the ban effective.

The ban has encouraged social media platforms to reconsider their features. Snapchat is moving toward a friends-only experience for 13- to 15-year-olds, for example.

The law is stimulating the development of purpose-built online spaces for children younger than 16 that can support their developmental needs, offering alternatives to mainstream social media.

The longer-term impact could be more significant if “no social media account before age 16” becomes an accepted norm, making it easier for parents and schools to support delayed social media use.

Implementation struggles

Despite the early encouraging outcomes, one study found that online platforms struggle to implement age checks. Many under-16 users in Australia have continued to access platforms with little difficulty, the study said. They children have found workarounds to subvert restrictions, such as using a free VPN to bypass age checks—some of which have questionable data-collection practices.

Seven in 10 children retained their existing accounts on restricted platforms, the study found. Other teens created new accounts using incorrect age information. Some were incentivized to seek unregulated offshore platforms not subject to Australia’s law.

The workarounds prompted Australia to double the maximum fine and warn of court action against tech giants for noncompliance.

Emphasis on age verification

A number of other countries are implementing or considering social media restrictions. They include Brazil, Canada, France, Greece, Indonesia, Norway, Poland, Thailand, Türkiye, and the United Kingdom. The European Union is contemplating its own restrictions.

The countries’ mandates for age verification or age assurance shift the policy focus from whether to verify age to how to do so effectively while protecting user privacy.

An article on think tank New America’s website, “Age Assurance and Verification,” describes some methods:

• Age gating and screening. Users self-attest their age by checking a box or inputting a birth date.
• Age estimation. Several techniques are available, including profiling the user’s online activity and scanning the user’s face.
• Age verification. One way is providing a government-issued identification document. Other approaches include digital identity systems, digital wallets, and third-party verification.

Reliable age verification is technically challenging and raises privacy concerns, as outlined in “The Age-Verification Trap,” written by Cinderpoint consultant Waydell D. Carvalho and published in February in IEEE Spectrum. Carvallo says platforms need to balance age verification with protecting users’ personal information.

IEEE’s contributions

IEEE is working on initiatives to provide a safer online environment for children. To help developers build age-appropriate social media platforms and websites, the IEEE Standards Association (IEEE SA) has published two guidelines.

The IEEE Standard for Online Age Verification (IEEE 2089.1-2024) provides a framework for designing, specifying, evaluating, and deploying verification systems. The standard includes requirements for privacy protection, data security, and information management specific to the age-assurance process. It also provides procedures for verifying a user’s age or age range with a high degree of accuracy.

Based on the 5Rights Foundation’s Principles for Children, the other standard (IEEE 2089-2021) provides practical steps to qualify online products and services for children. It requires systems to present information in an age-appropriate way and to uphold the rights established for youngsters in the U.N. Convention on the Rights of the Child.

IEEE SA also offers an online age-verification-certification program, which assesses systems for compliance with the IEEE 2089.1 standard. The program certifies that organizations implement robust processes before granting access to age-restricted products and services, prioritizing children’s safety, privacy, autonomy, and rights.

As outlined in The Institute article “IEEE Makes Strides to Improve Online Safety for Kids,” certification is based on six key indicators: accuracy, frequency of assurance, counter-fraud measures, authenticity, frequency of authenticity checks, and birth date confidence.

Indonesia used key provisions from the two IEEE guidelines to inform its child-protection regulation, which was signed into law last year.

IEEE’s ethically aligned design framework prioritizes human well-being, transparency, accountability, privacy, and protecting vulnerable populations including children.

Calls for platform reforms

Although social media bans would be globally significant policy responses, deeper structural issues remain largely unaddressed. Platform architecture and features contribute to social media harm.

The focus needs to shift from constraints on account provisioning and content moderation to safer platform design.

Meta in August agreed to pay $17.1 billion to settle a lawsuit brought by U.S. states. The suit said Meta designed its social media to be addictive to children, and the company concealed internal research showing Instagram’s addictive effects on teenagers. As part of the settlement, Meta agreed to implement child-safety measures such as setting daily time limits and disabling Facebook and Instagram push notifications during school hours.

The company still faces other lawsuits that could have far-reaching implications, pressuring other tech companies to design safer social media platforms.

Architecture-driven features such as infinite scrolling, algorithmic recommendations, addictive platform design, data-driven engagement, and personalized advertising to minors are other contributing factors to social media addiction.

IEEE Senior Member Katina Michael, professor at the University of Sydney business school and founding editor in chief of IEEE Transactions on Technology and Society, shared her perspective: “Social media bans may offer a short-term response to growing concerns, but they are not a long-term solution,” she says. “IEEE 2089-2021 advocates for socio-technical systems that are designed to promote human well-being, safety, and flourishing. Rather than relying on prohibition alone, we should focus on better design, building digital platforms that embed ethics, accountability, transparency, and human values from the outset.”

Collective responsibility

Protecting children online would require a combination of policy measures, improved platform design, digital literacy, parental involvement, and cultural change.

Building a safe, secure, and inclusive digital ecosystem that supports adolescents’ cognitive, social, and emotional development would require collaboration among technology companies, platform providers, content creators, parents, educators, policymakers, and young people themselves.

Professional organizations such as IEEE can continue contributing through standards development, education, certification while promoting trustworthy and responsible digital technologies.

This article was updated on 16 September 2026.

This interactive webinar will introduce the different types of AI, address the concerns with AI, share how we IBM are approaching Responsible AI, and offer guidance to students about what they can do - as individuals, and members of their IEEE chapters. Participants will also have the opportunity to to apply the Responsible AI approach to a particular use case - IBM Bob, a software development life cycle agent, and Q&A. This will be an interactive session, so have phones ready to engage!

Register now for this free webinar!

For this year’s IEEE annual election, the IEEE Tellers Committee will recognize the Sections with the highest voter turnout (based on total eligible voters) in each Region with an incentive reward after the election results are accepted by the IEEE Board of Directors.The incentive initiative is managed by the Tellers Committee, which retains full authority over all rules, operations, and decisions regarding the program.

Incentive guidelines

The Section sizes and their respective reward amounts are:

• Large sections consist of more than 1,501 eligible voting members. The top-performing large section in each IEEE region will be rewarded with US $1,000.
  
• Medium sections consist of 501 to 1,500 eligible voting members. Each region’s top-performing medium section will receive $600.
  
• Small sections consist of 500 or fewer eligible voting members. The top-performing small section in each region will receive $400.

To learn more about the incentive program, visit the IEEE annual election website. If you haven’t voted in the 2026 elections, you can learn about the candidates and vote here. Send questions to: elections@ieee.org.

Franklin “Frank” Kuo

Codeveloper of ALOHAnet

Fellow, 91; died 14 April

Kuo helped develop ALOHAnet, a pioneering computer system at the University of Hawaii at Mānoa, in Honolulu. The system went online in 1971 and represented the first public demonstration of a wireless packet data network. It was an inspiration for Robert Metcalfe’s development of Ethernet a couple of years later. In 2020 ALOHAnet was designated as an IEEE Milestone.

Kuo earned bachelor’s, master’s, and doctoral degrees in electrical engineering from the University of Illinois, Urbana-Champaign. After earning his Ph.D. in 1960, he joined Bell Labs in Murray Hill, N.J., where he conducted research in computer communications.

After six years at the company, Kuo left to become a professor of electrical engineering at the University of Hawaii. From 1968 to 1971 he and one of his colleagues, IEEE Life Fellow Norman Abramson, developed ALOHAnet. The network connected computers on Hawaiian islands using ultrahigh-frequency radio, transmitting information over radio waves instead of cables.

ALOHAnet became the foundation for modern networks. Kuo pioneered the concept of a random-access protocol, or sharing a single channel without central coordination—which led to the packet-switching principles that underpin modern Wi-Fi and mobile networks.

Kuo authored or coauthored several books including Computer Communication Networks. Published in 1972, it was one of the earliest textbooks on the subject.

He served as director of the university’s Cosine committee, a project funded by the U.S. National Science Foundation to develop computer engineering courses.

He took a sabbatical from 1975 to 1977 to work at the U.S. Pentagon as director of information systems in the defense secretary’s office. He oversaw computer communications applications used in command, control, and intelligence programs.

During the 1980s and ’90s, he helped develop China’s Internet. In 1982 he joined SRI International (formerly the Stanford Research Institute), in Menlo Park, Calif., as a researcher. He also was a consulting professor in Stanford’s electrical engineering department and taught computer networking at Shanghai Jiao Tong University.

As a UNESCO lecturer in Beijing in 1994, he helped Peking University, Tsinghua University, and the Chinese Academy of Sciences connect to the Internet. He also worked with Tsinghua University to develop CERNET, the country’s first nationwide education and research computer network, which was managed by the Chinese Ministry of Education. For his work, he received an honorary degree from Shanghai Jiao Tong University.

In the mid-1990s, Kuo helped found General Wireless Communications, a developer of mobile phone messaging services and games that was renamed Mtone Wireless.

Muhammad Rezaul Karim

Bell Labs researcher

Life senior member, 86; died 18 May

Karim was a distinguished member of the technical staff at Bell Labs in Murray Hill, N.J. His work was instrumental in the development of modern cellular communications technology.

He joined Bell Labs in 1972 and worked in its mobile telecommunications laboratory as part of the team tasked with creating one of the earliest cellular networks.

In 1975 Illinois Bell Telephone petitioned the U.S. Federal Communications Commission to develop and test a cellular system. The FCC, which now regulates radio, TV, telephone, Internet, satellite, and wireless services, authorized the project in March 1977. Karim and his team helped develop key elements of the technology, including the Bell Labs logic that controlled the cellular system, turning the concept into a working one. They also built radio receivers, transmitters, control systems, and cell-site equipment used in the first trial of the cellular system.

The following year, Bell Labs and Illinois Bell deployed the Advanced Mobile Phone Service system across Chicago, with its switching office located in Oak Park, Ill. The initial test used approximately 100 mobile phones to work through hardware, software, and system-design problems.

A subsequent test in 1979 involved 2,500 mobile users, providing a demonstration of the cellular technology in practice.

The trials in Illinois helped establish the technical foundation for the commercial cellular networks that followed.

Later in his career, Karim worked on the asynchronous transfer mode (ATM) technique, a high-speed networking technology crucial to the transition from traditional telephone networks to broadband and digital ones.

In 2000 he published ATM Networks: Application, Systems, and Design, a textbook that served as a guide for designing and implementing ATM-based services.

Karim received a bachelor’s degree in electrical engineering from the Bangladesh University of Engineering and Technology, in Dhaka. He then earned a master’s degree in EE from the University of Manchester, England, and a Ph.D. in EE from Stevens Institute of Technology, in Hoboken, N.J.

Harry Bostic

Former IEEE Region 4 director

Life senior member, 86; died 18 March

Bostic was an active IEEE volunteer who served as the 1998–1999 director of IEEE Region 4. In 2007 he received a lifetime achievement award from the IEEE Central Indiana Section for “outstanding commitment and dedicated service as regional advisor to the volunteers and members of Region 4 and the Institute.”

He was an engineer for 30 years at U.S. Navy’s avionics facility, a research, development, and manufacturing concern in Indianapolis. He worked on flight control, navigation, and weapons systems there. (The facility closed in 1996.)

Edwin C. Jones Jr.

Professor

Life Fellow, 91; died 10 March

Jones was widely recognized for his contributions to engineering education, curriculum development, and accreditation through decades of service to IEEE, ABET, and the American Society for Engineering Education.

He earned a bachelor’s degree in electrical engineering in 1955 from West Virginia University in Morgantown. The following year he earned a diploma of membership (equivalent to a master’s degree) from Imperial College, London. He went on to serve in the U.S. Army Signal Corps for two years. After his service ended, he studied engineering education at the University of Illinois, Urbana-Champaign, earning a Ph.D. in 1961. Jones then joined the university’s faculty.

The following year, he left Illinois to join Iowa State University, in Ames, as an assistant professor. He was promoted to professor in 1995. Two years later he became associate chair of the electrical and computer engineering department and served in that position until 2001, when he retired and was named professor emeritus. In recognition of his commitment to students, Iowa State established a scholarship in his honor.

In 2006 he accepted a part-time position as an adjunct professor in Minnesota at the University of St. Thomas, in St. Paul. He advised graduate students and helped develop the university’s systems engineering program.

An active IEEE volunteer, he served as 1975–1976 president of the IEEE Education Society. He was a member of the IEEE Educational Activities Board, helping strengthen the relationship among engineering education, professional practice, and accreditation organizations. He received an IEEE Centennial Medal in 1984 and the EAB Meritorious Achievement Award in Accreditation Activities in 1986. The IEEE Education Society later named its Meritorious Service Award in his honor.

Jones was elected a Fellow of ABET in 1986. During his years of service as a program evaluator and leader, he helped advance the quality of engineering education and accreditation programs. ABET recognized him with its Grinter Distinguished Service Award, its highest honor.

Alexander Robert Spitzer

Clinical neurology researcher

Life senior member, 70; died 27 February

Spitzer was a neurologist for 40 years at the Wayne State University School of Medicine, in Detroit, where he also was a director of the electromyography laboratory at Harper University Hospital. The lab studied patients’ brain and spinal cord activity in response to sensory stimuli. The evaluations assessed nerve pathway integrity to help diagnose multiple sclerosis, spinal cord injuries, and other conditions.

After earning his medical degree from the Einstein College of Medicine, in New York City, Spitzer completed a fellowship at the U.S. National Institutes of Health, in Bethesda, Md. He then joined Wayne State as a clinical neurology researcher. His pioneering research in applying neural network analysis to electromyography and clinical neurophysiology resulted in peer-reviewed publications, grants, and several U.S. patents.

He mentored generations of neurologists in electrodiagnostic medicine, a medical specialty that uses nerve-conduction and electromyography tests to evaluate and diagnose muscle and nerve disorders.

In 2020 he founded Mackinac Neurology, a telemedicine-based practice that treated pa­tients virtually during the COVID-19 pandemic.

A longtime IEEE volunteer, he held numerous roles on the IEEE Regional Activities Board, now known as the Member and Geographic Activities Board. He was a member of the IEEE Ethics and Member Conduct and Nominations and Appointments committees, as well as the IEEE Educational Activities and IEEE-USA boards. He served as 1977–1979 director of the IEEE Central Indiana Section.

Donald Leo Dietmeyer

Professor

Life Fellow, 93; died 13 February

Dietmeyer was a professor of electrical and computer engineering for 40 years at the University of Wisconsin-Madison.

He developed a lifelong interest in radio and electronics at high school in Wausau, Wisc., and earned a Ph.D. in electrical engineering in 1959 from the University of Wisconsin. He’d joined the university’s electrical engineering faculty as a professor in 1958 while pursuing his doctorate.

Dietmeyer’s research focused on computer-aided design in the areas of switching theory, hardware description languages, and the decomposition of Boolean functions. His research contributed to the development of automation tools for integrated circuit design.

He worked with Jim Duley, a former student, to pioneer the use of the digital system design language. He wrote the textbook Logic Design of Digital Systems, published in 1978.

In the early 1980s, Dietmeyer worked with researchers to develop ConLan, a language-construction method that combined hardware description languages in one underlying framework.

He served as associate dean of the University of Wisconsin’s electrical and computer engineering department from 1983 to 1995. In 1998 he retired and was named professor emeritus.

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!

I lost a cushy engineering management position at the same time I purchased a business, with a mortgage, three kids, and every other financial obligation of being an adult. It was one of the most stressful stretches of my life. If something similar has happened to you, I feel your pain.

You lose more than the income

For many of us, our work became our identity, so we don’t just think, “I don’t have a job anymore.” We start thinking, “I’m not an engineer anymore.”

An apple tree in winter is still an apple tree. A car parked in a driveway is still a car. You’re still an engineer, the same way you were still one every evening you clocked out, and the same way you’ll be one at the next job. Your skills took years to build and won’t evaporate because a company stopped paying for them. This feeling can be particularly rough if part of your identity was tied to a recognizable employer.

However, this can also be a chance to catch up on what you’ve been putting off: working out, being present with your kids, writing, whatever hobby got shelved for a deadline three years ago.

Once you’ve caught your breath, here are some tips for when the actual work starts:

1. Don’t sprint on day one

I started applying for jobs the next morning after the layoff. It felt productive, but it gave me no time to settle my nervous system or consider what I actually wanted next. I took the first offer that came along, at a company I already knew wasn’t right, and quit after exactly 30 days. Give yourself a few days before deciding anything. Plans built out of desperation rarely work out well.

2. Audit your spending

Go through every subscription and recurring charge and cut what isn’t essential. Every dollar you stop bleeding buys you patience instead of forcing a bad offer out of fear.

3. Build a list wider than LinkedIn

Start with former colleagues and vendors, or businesses with a relationship to your previous employer. They already know you or your company, which gives you the halo effect: Some of the trust from your employer carries over to you automatically.

LinkedIn is table stakes and is the most popular place to find work, but that doesn’t mean your search should stop there. Try Facebook, Instagram, and other social media channels too, where plenty of people who aren’t on LinkedIn might have leads for open roles. I found my first job years ago from a Facebook post, and both roles I landed after being laid off came from startup job boards and recruiters I found entirely outside LinkedIn. Your mileage may vary, but LinkedIn isn’t the only game in town.

Don’t skip your inner circle either: Text your family and friends. Good leads rarely come from someone you know directly, but they do come from someone that person knows. Work this list daily and track who you’ve contacted.

4. Eight hours is a long time

With no work to fill your day, you may default to treating the search like an eight-hour shift. You can’t apply productively for 8 hours straight, and that’s why people burn out. Use a focused morning block for the list in step 3, then spend the rest of your day on activities you’ve been putting off.

5. Catalog your wins before you study interview questions

Cataloging your wins is a higher-leverage move than drilling practice questions this early. Can you explain the most impactful project you worked on in the last year? Probably not.

Most people skip this step, then default to generic answers when a recruiter asks about their last role. Write down specific stories showing leadership, technical ability, and grace under pressure. They’ll come up once you’re in interviews, and cataloging them rebuilds your confidence along the way too. Save the deep prep for once an interview is on the calendar.

Ask yourself daily whether today’s work is generating interest in you, or leading you to someone who might hire you. If not, consider skipping it.

One last thing

A layoff rarely reflects your skills. It’s usually a company protecting revenue—nothing more personal than that. Knowing that doesn’t make it easier, but hopefully this gets you back on your feet faster.

—Brian

IEEE Global Careers Fair: September 23-24

Looking for a job? For the first time, IEEE is taking its Career Fair worldwide. The inaugural IEEE Global Virtual Career Fair runs September 23 (5:00 PM EST) through September 24 (8:00 PM EST), following the sun across every region to connect engineering and technology professionals with employers around the world.

Read more here.

United States Invests in Industry Partnerships for Ph.D. Training

While most engineering Ph.D. grads end up in jobs at commercial companies, academia and industry often operate in their own bubbles. Now, the U.S. National Science Foundation is investing in a program to integrate industry experience into STEM doctoral programs and help bridge the gap. Modeled after similar programs in other countries, students in the I-PhD will spend at least one year working on research at an industry site and receive a combination of funding from the company, NSF, and the university.

Read more here.

AI Efficiency Could Cost Us the Next Generation of Experts

When systems engineer Richard Mitchell designed a digitally-controlled nuclear plant, he made a counterintuitive decision: including manual steps for the human operator that a machine could execute on its own. The strategy was meant to keep the operator sharp, and it’s one that could help address one of the biggest issues facing the workforce today: What happens to human expertise when AI does the work that used to build it?

Read more here.

I’m sitting in a Rivian R1S SUV as it drives itself down the leafy streets of Palo Alto, Calif., through areas crowded with touchstones of tech history. We cruise near the landmark HP Garage, the one-car workshop where Hewlett-Packard, and, arguably, Silicon Valley, was founded in 1939. I skirt Stanford University, where a team led by computer science professor Sebastian Thrun won a US $2 million DARPA Grand Challenge in 2005. The team’s Volkswagen SUV, named Stanley, became the world’s first vehicle to navigate a grueling 212-kilometer Mojave Desert course with no human intervention.

This Rivian might look like any other electric SUV in this affluent town, with its concentration of tech bros, venture capital, and startups. But inside this boxy EV is something special: an Autonomy+ system that will allow owners to enter an address, sit back, and let the vehicle drive to any mapped destination in the U.S. and Canada. This point-to-point system is one of the most advanced semiautonomous-driving systems coming to market. It is also a precursor of the company’s bid to make self-driving cars a reality, for robotaxis and—eventually—for everyday car buyers. After years of incremental advances and frustrating setbacks, self-driving has been swept up in the great AI resurgence, and is now a top priority for investors and global automakers, who envision vast new streams of profits.

So here I am, 21 years after that DARPA challenge, riding shotgun in Stanley’s vastly more advanced descendant. Rivian’s Autonomy+ is intended to operate seamlessly on suburban streets like these, sensing and responding to traffic lights, crosswalks, and stop signs. That point-to-point system is set to debut on Rivian’s all-new R2 SUV by roughly the end of this year, and via over-the-air updates for its newest R1S and R1T models. Rivian says it will charge $49.99 a month, or $2,500 up front, versus Tesla’s $99 per month for its rival system, which is somewhat misleadingly called Full Self-Driving (Supervised), or FSD. Mercedes, meanwhile, plans to charge $3,950 for a three-year subscription for the forthcoming MB.Drive Assist Pro on its CLA-Class EV; that system still requires at least one hand on the steering wheel.

Video released by Rivian shows the company’s R1 SUV being driven on a variety of urban and rural roads, according to the company. Rivian plans to introduce this self-driving system to compete with Tesla’s offering before the end of 2026. Rivian

Impressive as it is, Autonomy+ is only a Level 2+ system in the classification system established by the Society of Automotive Engineers. Level 2+ means that a human driver must pay full attention and be ready to retake control at any moment. Rivian, along with a horde of deep-pocketed rivals, is aggressively working toward more impressive (and potentially lucrative) levels of autonomy. At Level 3, drivers could “check out” behind the wheel for limited periods, to scroll through emails or watch a movie—but not to sleep.

The big race right now is to deliver Level 4 autonomy: A car you could (in theory) dispatch to pick up a pizza, and have it carted home on the heated, unoccupied driver’s seat—or in which you could spend the ride lounging alone in the back seat, enjoying a private slice while reading a newspaper.

At Rivian’s software lab in Palo Alto, Calif., a technician evaluated code for the company’s self-driving system.Jason Henry/Bloomberg/Getty Images

Robotaxis currently roaming select cities in the U.S., China, and the Middle East have proved that driverless, Level 4 autonomy is possible. These cars operate in relatively tiny numbers in a couple of dozen cities, and within the specific constraints of commercial services. Now Rivian and its many rivals—including Tesla, Toyota, Mercedes, Volkswagen, and China’s BYD—are racing to bring that level of self-guided mobility to the masses. Rivian’s strategy combines a suite of cameras, radar, and lidar; a custom set of silicon chips, developed in-house, to process sensor data; and an AI autonomy model running on those chips. With $1.25 billion in backing from Uber, Rivian plans to graduate to a fleet of self-driving, Level 4 robotaxis starting in 2028. Those taxis, along with Rivian’s consumer fleet, will be the literal training wheels for extending Level 4 ability to consumer vehicles.

Meanwhile, millions of connected cars, as they cruise every nook and cranny of the globe, are already sending data to train automakers’ systems. The race is on to funnel those data through fast-improving AI models with “end to end” capability: an AI architecture, powered by deep learning, that processes raw sensor data directly into physical vehicle commands. So equipped, engineers anticipate they’ll be able to solve the tricky edge cases—tangled urban streets, unique geographies, swarms of pedestrians, inclement weather—that skeptics once deemed intractable.

Rivian’s Plan for Level 4 Self-Driving

Despite the company’s high media profile, including a spotlight on RJ Scaringe, its MIT-doctorate founder and chief executive, Rivian holds a relatively tiny slice of the U.S. passenger-vehicle market. It sold just 42,000 vehicles last year across its three models, the adventure-minded R1S SUV and R1T pickup, and the Electric Delivery Van. Tesla sold about 1.6 million units. Toyota, the world’s largest automaker, sold more than 11 million.

The first generation of the Rivian Autonomy Processor, an AI processing chip developed in-house, was tested at Rivian’s Palo Alto, Calif., lab in December, 2025. Jason Henry/Bloomberg/Getty Images

Rivian’s underdog strategy is to leverage software and tech to make itself a serious player. Volkswagen, among the world’s largest automakers, saw enough value there to invest up to $5.8 billion in a joint venture called Rivian and Volkswagen Group Technologies. The joint venture gives Rivian crucial capital for development. It gives Volkswagen access to Rivian’s electrical architecture and to the software for the R2, new-generation Rivian SUV that went on sale in June.

Unlike traditional lidar units, which protrude like a layer cake from the roof of a vehicle, Rivian’s unit on the new R2 SUV is housed in a small, sleek enclosure where the windshield meets the roof.Rivian (2);Jason Henry/Bloomberg/Getty Images

“Rivian developed an architecture so important that VW is spending billions to buy it, as opposed to trying to re-create it themselves,” says Bryan Reimer, a research scientist in MIT’s Center for Transportation and Logistics.

But the joint venture doesn’t give VW access to Rivian’s autonomous tech. In March, that R2 architecture underpinned Rivian’s $1.25 billion deal to supply Uber with up to 50,000 robotaxis. The companies plan to initially deploy 10,000 taxis, beginning in San Francisco and Miami in 2028, before expanding across 25 cities in the U.S., Canada, and Europe.

Rivian’s vulnerabilities include struggles with reliability, along with expensive body repair costs that the company says it strove to reduce for its new R2. As impressive as Rivian’s in-house tech may appear, the company has miles to go to catch up with Tesla, which recently announced it has 1.1 million active users of its FSD system. Toyota is also jumping into the game; its Woven by Toyota subsidiary has partnered with the Alphabet-owned Waymo to develop an autonomy platform for robotaxis and consumer cars.

The lidar unit on a Waymo robotaxi protrudes noticeably from the roof of the vehicle.Andrej Sokolow/picture alliance/Getty Images

Until recently, most observers would have gone all-in on Tesla as the winner of the autonomous race. Elon Musk’s company has begun operating a small test fleet of Model Y robotaxis in three Texas cities and in Florida. Tesla has also begun producing a dedicated autonomous vehicle, the Cybercab robotaxi. But in April, Musk pushed back his timeline for Level 4 autonomy for general consumers: “I’m just guessing here, but probably in the fourth quarter” of 2026, he said. It was the latest in a series of deflating walkbacks from the man who once promised 1 million robotaxis on the road by 2020.

Scaringe, during our drive of his company’s make-or-break R2 SUV at a Utah state park, says that showroom Rivians will start adopting some of its robotaxis’ Level 4 capabilities no later than 2030, perhaps beginning with self-parking functions.

How Self-Driving Systems Are Learning From Humans

Like most autonomous cars, Rivian’s system fuses data from multiple sensors to create a robust picture of a fast-moving environment and its obstacles. Data is fed to a neural network—what Rivian refers to as its “Large Driving Model,” or LDM—that churns through hundreds of trillions of operations per second to interpret and fuse data from cameras, radar, and lidar. That network is end to end, meaning that it processes multiple streams of raw sensor data (such as camera pixels) and outputs driving controls (for steering, braking, and acceleration) through a single data pipeline. More traditional systems coded distinct steps for data collection, feature extraction, prediction, and decision-making.

That proprietary AI driver identifies features in images and point clouds, groups them into objects, and tracks them across frames, time-stamped to the millisecond to account for differing frame rates. The AI thus builds confidence over time, acting on object detections that persist across several frames, rather than, say, slamming the brakes due to a camera blip on a single frame. The virtual driver can then navigate safely even when sensors disagree, by favoring the persistent data. The output— commands for electric motors and other systems—is backed by redundant hardware for by-wire systems such as steering and brakes.

During my demo of Rivian’s point-to-point Autonomy+ system, a company test driver sits behind the wheel. Nick Nguyen, the engineer who directs Rivian’s products and programs related to autonomy, watches from the back seat. Compared to, say, a large language model that writes news or fiction, Nguyen says, an autonomous-driving AI is less subjective and easier to evaluate, so there’s little room for error. “We want cliché. We want boring. Just safe, repeatable driving,” he says.

The Rivian R2 SUV plans to offer a self-driving system by roughly year’s end 2026. The R2 competes with the more urban-oriented Tesla Y.Rivian

From my brief drive, I’d say suburban boredom is achieved in this Rivian R1S. Unlike some modes of Tesla’s Full Self-Driving (Supervised), Rivian’s system drives like a soccer dad, obeying speed limits to the digit, stopping gracefully at traffic lights, and easing over speed bumps like a driver delivering antiques. Yet this robo-driver isn’t timid or tentative. For robotaxi companies in the U.S. and China, these types of routine trips are boosting optimism and investment to dizzying heights. Waymo claims 92 percent fewer fatal or serious-injury accidents than human drivers, based on 220 million miles of autonomous ride data. But the real challenge is how well the higher levels of autonomy will work when they reach consumer cars [see Sidebar, “The Growing Proof That Autonomous Cars Save Lives”].

Rivian’s core LDM currently ingests cloud data from up to 125,000 cars for analysis and validation, which then fine-tunes the model through simulations. Onboard computing is smart enough to trigger recording only for unusual scenarios. Owners have to agree explicitly to data collection beforehand. Updated LDMs will be beamed back to customer cars via monthly over-the-air updates. part of that self-reinforcing data flywheel. It’s part of what Scargine calls the “data flywheel,” the self-improving AI loop that continuously refines the system.

As is true for some of its rivals, Rivian no longer needs to fully rely on an onboard high-definition map or even a cellular link to pinpoint the car for navigational purposes. That strategic shift reduces data demands, and ensures steady driving in urban canyons or tunnels with no connections. Instead, the Rivian recognizes and responds to its surroundings through recognition and repetition, just as a human would do interpreting street signs, following lane markers, being alert to hazards.

The Rivian R2 features 11 high-definition cameras and five radars. It will integrate a lidar unit early next year to lay the groundwork for future autonomy. That miniaturized lidar will integrate smoothly into the R2’s existing roofline, an improvement over the bulky, drag-producing units seen on Waymo Jaguars, and older partially autonomous models. Vidya Rajagopalan, Rivian’s senior vice-president of electrical engineering hardware, says lidar costs have fallen from above $10,000 to a few hundred dollars in under a decade.

Vidya Rajagopalan, Rivian’s senior vice president of electrical engineering hardware, holds a RAP1 AI processor chip.Jason Henry/Bloomberg/Getty Images

A mix of sensors plays up the strengths and diminishes the weaknesses of each, Rajagopalan says. Cameras capture color and texture and can distinguish between objects, but they struggle in darkness and low-contrast lighting. Lidar is unaffected by darkness or blinding sunlight, and senses shapes in three dimensions. This inherent 3D capability makes lidar more reliable for slowing or halting a car for random objects—“a tire in the road, or maybe a large dinosaur,” Nguyen quips. Multiple cameras can further contribute 3D data, after a short delay for processing.

Sensors with 360-degree vision can outperform human senses in key situations. Radar and lidar can spot nighttime pedestrians or animals hundreds of meters down the road, something no human can do. But lidar can be thrown off by dust, fog, and snow. Radar can “see” through rain or snow, but with relatively low spatial resolution.

Why Rivian Ditched Nvidia

To handle the flood of sensor data, Rivian has taken on an ambitious challenge: designing its own custom autonomy chip in-house. The Rivian Autonomy Processor (RAP1) is a 5-nanometer processor that can execute 800 trillion operations per second (TOPS), three times as fast as the Nvidia Jetson Orin chip used in its earlier models. The chip will be built to Rivian’s specs by Taiwan Semiconductor Manufacturing Co. , which also makes custom chips for Tesla.

Rivian’s autonomy module contains two Rivian Autonomy Processors, each capable of 800 trillion operations per second.Rivian

Nvidia’s latest automotive system-on-a-chip, the Drive AGX Thor processor, is being adopted by the likes of BYD, Hyundai, Lucid, Mercedes, Nissan, Volvo, and Xiaomi, along with the Aurora and Waabi autonomous-trucking companies.

On paper, a single AGX Thor chip is slightly faster in terms of TOPS, at 1,000 trillion operations per second. But Rivian combines a pair of chips in each autonomy module, giving it 1,600 TOPS and execution rates around 5 billion pixels of data per second, versus 3.5 billion for Nvidia’s Thor.

Rajagopalan says developing the chip and AI software simultaneously shaved a critical full year from development. Experts say it’s the kind of fast-to-market speed that China has mastered and that legacy automakers are struggling to match. The in-house design allows Rivian to custom-tailor its software to the chip, and vice versa. Nvidia’s general-purpose chip, designed to satisfy multiple customers with various needs, must devote computing power to onboard infotainment, displays, or other systems. Rivian’s chip is designed to run autonomy and nothing but.

During my visit to Rivian’s Silicon Valley campus, Rivian engineers Prasun Raha and Mukund Chavan tutored me on the rapid pace of the company’s autonomy evolution. A cluttered wallboard displays a first-gen architecture that Rivian debuted just five years ago. The initial R1S SUV and R1T pickup used nearly a score of electronic control units (ECUs), the “black boxes” that traditionally control vehicle functions. For its latest R1 models, Rivian reduced the ECU count to seven. The zonal architecture organizes nearly every vehicle function into three zones, hugely consolidating the electronics and simplifying manufacturing. Rivian also leaned into an autonomy trend called “early fusion”: mixing raw, time-and-space-aligned sensor data into a shared view before the neural network acts upon it. In late fusion, each sensor performs solo recognition before it’s combined into a single picture.

The self-driving revolution will really begin when the technology migrates from controlled taxi fleets to consumer cars, giving owners back the precious time they waste on commuting.

Early fusion preserves the richest sensor data for maximum accuracy in self-driving. But it demands the enormous computing power the RAP1 can deliver. Raha says the approach helps the multimodal system degrade gracefully and continue to operate with certainty even if, say, a camera’s lens gets covered with mud.

Together, these elements make up Rivian’s third-generation autonomy platform. Displayed on a test bench, a new Autonomy Compute Module pairs two RAP1 supercomputing chips. The module is eight times as powerful as before but 60 percent smaller, according to the company. Raha says the system was designed expressly to expand Rivians to Level 4 autonomy from today’s Level 2+. RivLink, the automaker’s interconnect technology, can bridge multiple RAP modules to scale processing power. “It lets us build this extensible system, with perhaps two more chips for Level 3 or four for Level 4, depending on how the model scales,” Raha says.

Rivian’s Road Map to Full Autonomy

Rivian’s next planned milestone toward self-driving will be Level 3 autonomy—a hands-off and eyes-off system, but for highways only. (Remember, Tesla’s current FSD is technically a Level 2 system: hands off but not eyes off.) On the freeway, Nguyen points out, drivers would be spared the drudgery of dealing with stop-and-go traffic, allowing them to boost productivity or just goof off.

Some autonomy critics are leery of Level 3, envisioning a limbo zone in which drivers are lulled into a false sense of security when a car drives for long stretches with no human attention required. Ford and GM are among the automakers pivoting toward limited eyes-off functions.

Rivian’s senior vice-president of autonomy, James Philbin, sees Level 3 as an inevitable stepping-stone to Level 4. The company expects it will initially be limited to highways, not the cut-and-thrust of city traffic. If a driver fails to respond to alerts, the system will slow the vehicle, pull off on a shoulder, or call 911. Rivian has not announced a timeline for introducing limited Level 3 capability.

Navigating a Tricky Liability Shift on the Way to Immense Profits

Ready or not, these much more autonomous systems are coming, a natural evolution of today’s semiautonomous helpers. In developed markets, adoption of showroom cars with partial-to-full automation is projected to jump from 8 percent in 2024 to 28 percent by 2030, according to Morgan Stanley.

“One in four cars sold globally may be equipped with smart-driving technology in five years, versus one in eight cars now,” wrote Tim Hsiao, a Morgan Stanley analyst, in a note posted on the company’s website.

Combining cameras, lidar, and radar gives a self-driving car a better view of people and objects in front of it, according to Rivian. The company expects to release a self-driving system before the end of 2026 that will compete with Tesla’s, which uses cameras alone. Rivian

MIT’s Reimer believes the self-driving revolution will really begin when the technology migrates from controlled taxi fleets to consumer cars, giving owners back the precious time they waste on commuting. If owners could truly send their autonomous car to safely chauffeur children, keep aged parents mobile, or run errands—while owners keep working or playing—the automakers who first help make that happen will enjoy a massive competitive edge, he says. As automakers struggle to convert buyers to subscription models, Reimer believes that self-driving appears to be the one advance for which consumers might actually pay plenty.

But the greatest impediment to that revolution has little to do with technology. Public skepticism over self-driving is rampant; and the fate of fully autonomous testing in New York City is uncertain. Even going from Level 2 to Level 3 might shift legal liability for some accidents from drivers to automakers. But with Tesla still fighting lawsuits over its rudimentary Autopilot systems, those questions aren’t anywhere near settled.

Experts worry that self-driving cars may become as politicized as EVs. Labor unions are pushing back, fearing job losses from taxis to trucking. A crazy quilt of state or local regulations has failed to create coherent industry guidelines. Publicized failures—even ones that don’t result in injuries, such as Waymos driving onto a flooded street or impeding emergency workers—give the industry a black eye. Companies like Tesla and even Waymo, Reimer says, have too often relied on an arrogant “Trust me” approach, resisting regulation and oversight.

Nevertheless, the momentum toward truly self-driving cars, and massive backing from automakers and AI-besotted investors, suggests their time has come. The rest of the journey will depend as much on social and regulatory issues as technical ones, and so Reimer has a bit of advice.

“Do it right, and share all your data,” he says. “Earn the right to scale…. It’s about establishing trust, and developing a framework in which we truly believe these systems can operate as a trusted part of our transportation network.”

This article was updated on 08 September 2026.

Plenty of people remain spooked by autonomous vehicles, or AVs. Some experts and policymakers have cautioned that AVs won’t necessarily make roads safer. When it comes to partial or full autonomy, the picture isn’t entirely clear, in part because there aren’t enough self-driving cars to make meaningful apples-to-apples comparisons.

Yet mounting research suggests that self-driving cars crash significantly less often than people, and with far fewer injuries. Evidence also shows that advanced driver assistance systems (ADAS) and other building blocks of autonomy—some of which are already mandated on every new car—are also reducing occupant and pedestrian injuries and deaths, along with insurance claims.

On the ADAS front, the Insurance Institute for Highway Safety found that automatic emergency braking (AEB) systems that recognize people in front of the car cut pedestrian crashes by 27 percent. Those AEB systems are mandated for all light vehicles in the U.S. by 2029, and more than 90 percent of new models already comply under a voluntary automakers’ agreement. A separate IIHS study found that automated braking greatly reduced rear-end crashes, by 50 percent, and their injuries by 56 percent. The Highway Loss Data Institute found that cars with AEB alone showed a 13 percent drop in property-damage claims. Cars that bundled ADAS features, including automatic braking for pedestrians, adaptive cruise control, and lane-departure warnings, saw claims reductions up to 39 percent.

Move to Level 4 autonomy, and Waymo says its robotaxis have now given 20 million paid rides over 220 million miles, the equivalent of 250 lifetimes of driving. In March, Waymo’s independent study showed 92 percent fewer fatal or serious-injury crashes, a 13-fold reduction versus human drivers in comparable city environments. That included 92 percent fewer pedestrian injuries, 83 percent fewer crashes with airbag deployments, and 82 percent fewer crashes with any injuries whatsoever. That included a 96 percent reduction in injury-causing crashes at intersections, among the deadliest environments for any automobile.

How Does Limited Fair-Weather Data Compare to Traditional Crash Statistics?

A key question is whether Waymo’s robotaxis, currently limited to fair-weather operation in a handful of cities in the U.S., are directly comparable to humans driving a wider variety of roads in much more variable conditions.

The IIHS is looking to dig deeper by cleaning up often-incomplete data. Researchers estimate roughly half of human crashes go unreported, and up to one-third of injury accidents, because drivers hope to avoid insurance price hikes. That potentially skews safety numbers in favor of human drivers. And while Waymo leads the industry in transparency, and robotaxi operators are required to report even the tiniest scrape to the National Highway Traffic Safety Administration (NHTSA), not every company voluntarily reports their total miles driven.

Related: Rivian’s Gambit for Full Autonomy

The IIHS’s latest July study flatly stated that automated cars crash less often than people. But it also sought clarity by creating a more-reliable category of “police-reportable crashes.” It then compared crash rates of human-driven cars against Waymo taxis in San Francisco, Phoenix, Los Angeles, and Austin. Waymo’s Jaguar I-Pace taxis traveled about 50 million driverless miles over the study period, versus 222 billion human miles in the same cities.

In a potential boost for public trust, the study generally supported Waymo’s own findings. Waymo taxis were involved in 68 percent fewer crashes overall than human drivers: 76 percent lower in Phoenix, 71 percent in LA, and 35 percent in San Francisco. A 4 percent higher Waymo rate in Austin may reflect an extremely small sample size. Significantly, Waymo’s injury crashes were still 81 percent lower on a per-mile basis.

The industry and its supporters continue to press the safety advantages of autonomous vehicles that never get drunk, drowsy, or distracted. Yet for this fledgling AV industry, there are still no national performance or safety standards. A crazy quilt of state or local regulations can allow or prohibit their deployment. That balkanized approach makes it harder to compare crash rates, according to the IIHS, which is calling for better federal reporting standards.

A posting on the IIHS website quotes the institute’s director of statistical services, Eric Teoh: “Those are encouraging signs for the future of driverless vehicles.” Teoh, who was also the lead author of the institute’s study, added that “Now we need to get the data-collection system right, so that we can ensure that level of safety continues as these technologies become more prevalent.”

Amazon’s Zoox Gets an Exemption for its Robotaxis

On July 30, in a move seen as fast-tracking the tech’s deployment, NHTSA granted Zoox, a subsidiary of Amazon, the first-ever exemption from certain motor-vehicle safety standards. That will allow commercial operation of Zoox’s toaster-shaped robotaxis, which have no steering wheel or pedals aboard. The agency determined that Zoox’s purpose-built robotaxi “would provide an equivalent level of safety” as a compliant vehicle, thereby satisfying the standard for an exemption.

On that final day of the SAE’s Automated Transportation Symposium, NHTSA also announced a partnership with SAE Industry Technologies to develop the nation’s first performance and competency standards for AVs, via a three-year, $5 million “A2SCEND” consortium.

Some doctors and health professionals are arguing that policymakers need to stop viewing self-driving cars as a tech moonshot but rather as a critical public-health intervention. Jonathan Slotkin, a neurosurgeon, makes a powerful case for the medical and societal benefits of AVs. Researchers at the Johns Hopkins Bloomberg School of Public Health say that highlighting the social value of AVs is critical to driving public trust and adoption.

Consider that roughly 40,000 people in the U.S., including more than 7,000 pedestrians, are killed each year in roadway accidents. About 1.16 million people die in roadway crashes around the world, making them the leading cause of death for children and young adults between the ages of 5 and 29. Cutting that by even 50 percent—let alone the 90 percent reductions suggested by some studies—would save 580,000 lives a year. That social and economic gain would dwarf that of seat-belt adoption or anti–drunk driving campaigns.

This article was updated on 08 September 2026.

Many researchers and students in Kenya, Rwanda, and Uganda struggle to access and publish scientific and technical articles because of financial barriers including publishing fees and subscriptions to research libraries. To help, IEEE has made its Xplore Digital Library more accessible by offering discounts on subscriptions and lowering fees to publish articles. But the number of papers published by technologists in the three nations still lags behind those from other developing countries.

It might be that many researchers haven’t received training in methodology, been instructed on how to write academic articles, or fully understand the process for publishing in scientific journals.

Staff from the IEEE Publication and Information Products group and IEEE volunteers held educational workshops this year in the three countries. The sessions covered the publishing process, IEEE publication outlets, ways to ensure the integrity of research papers, and tips for making better use of IEEE Xplore.

“We want to make sure those in this region are on par with other research communities and ensure they have the support and knowledge they need to make informed publishing decisions,” says Kristopher Zakrzewski, the IEEE area manager for Europe, the Middle East, Africa, and parts of Central Asia. “Our goal is to give them the tools they need to increase visibility and allow them to participate in global conversations in the technology space.”

Workshops on the publishing process

More than 120 participants attended the workshops, which were held in February at the Novotel Nairobi Westlands hotel, the University of Rwanda, and Makerere University, in Kampala, Uganda.

IEEE volunteers who are also authors showed attendees how to prepare, submit, and publish papers. They covered the peer-review process and the benefits of working with IEEE, which publishes about 30 percent of the world’s technical literature on electrical engineering and computer science.

IEEE Senior Member Nelson Ijumba presented at the session in Rwanda. Member Kennedy Ronoh led the Nairobi workshop. Sheila N. Mugala spoke to attendees in Kampala.

“The great thing about these sessions,” Zakrzewski says, “is that each had a local author who presented tips and best practices to ensure that new and returning authors have the information they need to prepare their paper for submission, determine where best to publish their article, and find the right journal or conference that would be the best fit for their research.”

One of the facilitators at the Uganda session was IEEE Senior Member Mayur Kumar Chhipa, head of engineering at the International Business, Science, and Technology University in Kampala and vice chair of the IEEE Uganda Section. The university has about 200 engineering students and about 50 researchers.

More than 100 people attended Chhipa’s session, where he shared practical guidance on conducting literature reviews and identifying high-impact research.

“Researchers in Uganda typically present their paper at an IEEE conference, and that’s it,” he says. “What we’re trying to do is encourage them to take the next step and get their paper published in an IEEE journal.”

He encourages his students to submit a summary of their thesis to an IEEE conference, he says.

“Otherwise,” he says, “their thesis sits in the university’s library or collects dust on a bookshelf.

“When you publish your research, the world knows you are a scholar who has done good work. Having a paper published at a conference or in a journal can help you get into a master’s program globally.”

IEEE Xplore access

Attendees were given an overview of the features of their IEEE Xplore subscription. The digital library contains more than 7 million technical documents from industry-leading journals, conferences, ebooks, and eLearning courses, as well as partner content.

IEEE provides access to the library to more than 50 universities in Kenya through a subscription agreement with the country’s Library and Information Services Consortium, which includes university and public libraries and research institutions. Sixteen universities in Uganda and one institution in Rwanda receive discounted subscriptions.

“It was really important to establish the direct correlation between having access to the technical literature and the publishing output from their university and the region as a whole,” Zakrzewski says.

Many publishing options

The workshops covered publishing options offered by IEEE. That includes both traditional and open access journals, with more than 200 periodicals in total.

There are approximately 180 hybrid journals, which contain a mix of subscription-based and open-access articles, and 30 gold open access journals.

Open access is a publishing model that makes scholarly research and literature freely available online to everyone. Instead of institutions paying for subscriptions, authors or funders typically pay an article processing charge (APC) of between US $2,160 and $2,800 to have their piece published. IEEE offers authors in Kenya a 50 percent discount off the APC rate, and authors from Rwanda and Uganda can publish in IEEE open access journals for free.

The open access program provides authors with greater visibility for their research and enhances discoverability, Zakrzewski says, leading to an increased number of references and citations.

Publishing with IEEE opens additional opportunities including scholarship awards, research assistant job offers, networking opportunities, and speaking engagements.” —IEEE Senior Member Mayur Kumar Chhipa

IEEE Xplore contains more than 200,000 open access articles, he notes. More than 109,000 articles have been published in IEEE Access, a multidisciplinary open access megajournal.

“IEEE supports author choice,” Zakrzewski says. “We really want to make sure that an author has the option to publish the research that will meet any consortium, funder, university, or coauthor requirements—which is why we’re focusing on growing our open access program to complement our traditional publishing program and offer more options to authors.”

The sessions are having an impact

Participants at the Uganda session told Chhipa that they appreciated the IEEE Xplore Digital Library demonstrations and found the guidance on academic publishing valuable.

“Many attendees mentioned that the session helped them better understand how to search for relevant literature, evaluate the quality of research papers, and write stronger manuscripts for publication,” he says.

“I have observed increased interest among students and faculty in using IEEE Xplore as their primary research resource,” he adds. “Researchers are also more aware of ethical publishing practices and are developing stronger research proposals and manuscripts.

“The program contributes to building a stronger research culture by encouraging evidence-based research, international collaboration, and higher-quality publications, which will ultimately enhance the global visibility of research from Uganda and Africa.”

Publishing has its privileges

Chhipa says getting your research published has many benefits, and IEEE staff and members agree.

IEEE and several of its societies offer student grants to help cover the expense of traveling to conferences and presenting papers. The money typically covers airfare and a hotel room. Some grants also pay for conference registration fees, Chhipa says.

Chhipa assists students at his university with writing and submitting research papers to IEEE journals and conferences. Students gain confidence when their paper gets accepted, he says. One who attended the recent IEEE session was informed that his paper was accepted by an IEEE conference—which Chhipa says he was excited about.

He encouraged that student to apply for a travel grant.

“Maybe he’ll get it. Maybe he won’t. But at least he learned how to write a paper, apply for a visa to attend the conference, and book an airplane ticket,” Chhipa says. “It will help him grow personally and professionally.”

Presenting a paper at an IEEE conference can be life-changing, he says.

“It opens additional opportunities,” he says, “including scholarship awards, research assistant job offers, networking opportunities, and speaking engagements. This is how publishing a research paper in the IEEE Xplore Digital Library can directly, positively impact the life of students and scholars from Africa, especially Uganda, Rwanda, and Kenya.”

This article is brought to you by Tsubaki KabelSchlepp.

In modern automated manufacturing, six-axis articulated robots perform high-speed, multidirectional maneuvers under demanding operational cycles. However, as robot arms swivel, rotate, and extend, the electrical cables, fiber optics, and pneumatic hoses supplying them endure severe mechanical stress. Torsional twist, rapid acceleration, and repeated contact with machine structures often lead to premature conductor fatigue, insulation breakdown, and costly unplanned production halts.

To overcome these multi-axis motion challenges, the Tsubaki KabelSchlepp Robotrax System provides a specialized three-dimensional cable carrier engineered specifically for complex robotic motion.

Managing High Tensile Forces With Central Steel Technology

Conventional cable carriers often transfer operational movement stress directly onto internal electrical lines and hoses. The Robotrax system changes this dynamic through a central steel cable that runs through the core of every chain link.

The Robotrax system’s central steel cable absorbs the primary tensile loads and preserves conductor integrity, dramatically extending cable service life.

When robot arms undergo rapid directional shifts and accelerations up to 10 g, this internal steel cable absorbs the primary tensile loads. By isolating electrical and fluid lines from pulling forces, the design preserves conductor integrity and dramatically extends cable service life. Mechanics can easily calibrate and adjust system tension using an integrated clamping piece, ensuring consistent mechanical support throughout long operational cycles.

Spherical Link Design and Modular Cable Routing

The foundation of the Robotrax system lies in its open, single-piece plastic links featuring spherical snap-on connections on both sides. This geometry allows the carrier to flex smoothly across three axes, providing radial rotation of up to ±450 degrees per meter depending on the model size.

To optimize internal organization, carrier links contain up to three distinct chambers. This physical separation prevents signal interference and mechanical abrasion between heavy power lines, sensitive data channels, and fluid hoses. For standard models (R040 through R100), technicians can press cables directly into the carrier without tools, drastically reducing installation and maintenance time. Larger configurations, such as the R140X, incorporate swiveling crossbars with snap locks alongside vertical and horizontal dividers for customized interior partitioning.

​ROBOTRAX System

1. Steel cable for transferring extremely high tensile forces
2. Tension piece for locking the chain links
3. Type with toolless opening swivel crossbars and divider module available
4. Open design
   – Fast cable laying as the cables are simply pressed in
   – Easy checking of all cables
5. Special plastic for long service life
6. Protective covers or heat shields made from different materials are available for different environmental conditions
7. Quick-release bracket for fixing and continuation
8. Strain relief with LineFix clamps
9. Protection against hard impacts, excessive abrasion and premature wear as well as limitation of the bending radius through protector

Active Retraction and Impact Protection

Large robot work envelopes and high-speed motion trajectories can cause loose cable carrier loops to swing and strike the robot body. To eliminate these destructive collisions, Tsubaki KabelSchlepp integrates the Pull Back Unit (PBU).

The PBU serves as an active retraction mechanism that maintains optimal tension on the cable carrier throughout the entire motion cycle. By preventing excess slack and eliminating interfering contours, the PBU minimizes collision risks across complex movement paths. The unit requires zero maintenance on its retraction element and offers standard mounting configurations for leading industrial robot platforms, including KUKA, ABB, and FANUC.

Tsubaki KabelSchlepp’s Pull Back Unit maintains optimal tension on the cable carrier and minimizes collision risks across complex movement paths.

Additionally, external protectors can be retrofitted onto individual chain links. These durable impact shields limit the minimum bending radius to prevent over-flexing while shielding the chain body from severe external abrasion. If wear occurs, technicians simply replace the modular protector rather than the entire cable carrier assembly.

Built for Demanding Industrial Environments

From automotive welding cells to high-speed machining centers, Robotrax systems adapt to severe working conditions through tailored protective accessories:

• Heat Shields: Aluminum-coated textile fiber covers protect against radiated heat, hot weld spatter, and flying sparks.
• Protective Covers: Coated polyester sleeves shield sensitive lines against aggressive cutting fluids, hydraulic oils, paint overspray, and abrasive dust.
• LineFix Strain Relief: Multi-layer clamping devices anchor cables securely at both ends to prevent axial displacement during intense motion.

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!

Asking someone to be your mentor is weird.

Walking up to someone and asking, “Will you be my mentor?” has always seemed to me like the adult version of a kid walking up to another kid at a party and asking, “Will you be my friend?”

What you’re really asking is: “Will you commit some amount of unpaid time to guiding my career for an indefinite period?”

Framed that way, of course some people hesitate to say yes.

But formal mentorship isn’t the only way to benefit from the wisdom of those who came before. I’ve never formally asked anyone to mentor me. And yet I’ve had dozens of unofficial mentors.

The Copy-Paste Method

One way to learn from others is by copying what you observe.

Sometimes this means reading books or blogs from engineers you respect and directly applying their ideas to your work.

I’ve also been fortunate to work alongside some extremely talented engineers, and I shamelessly copied the things they did well.

When I meet one of these engineers, I try to figure out what they’re doing differently: How do they approach a problem? What do they read? How do they communicate in meetings? What do they know that I don’t?

Then I steal whatever seems useful and apply it to my own career.

Great artists steal. Engineers should too.

Curiosity Compounds

Still, just observing has its limits. Asking questions can get you even farther.

I’ve asked managers how they approached difficult conversations, and I’ve asked engineers what their process was for solving problems I thought were impossible.

If someone seems unusually knowledgeable: “What are you reading right now?” If I respect someone’s work: “What’s something you think I could do better?”

These aren’t profound questions. They don’t need to be. You get one useful piece of information, apply it, and move on.

And if you don’t work around exceptional engineers, you can still do this. The only real requirement is curiosity. When you encounter something you don’t understand, make it a rule to investigate instead of moving past it.

You don’t need one person willing to guide your career. You need a collection of people who know things you don’t.

Pay attention to them. Ask questions. And shamelessly copy the good parts.

Ask me!

If you have a career question you’re struggling with, like an upcoming decision, a problem at work, an interview, whatever—submit it here: https://docs.google.com/forms/d/e/1FAIpQLSdj_2BZIhrGF__7BCLH33zJ9NMv8C7Vsg9NNusASrYj7-9Idw/viewform. You can include your name or remain anonymous.

I’ll be reading through them and answering some in future articles. Consider it mentorship without the awkward “will you be my mentor?” conversation.

—Brian

ICYMI: The Institute June 2026 issue

IEEE members have a wealth of experience and knowledge to draw from. In the most recent issue of The Institute, several members share their career advice for engineers, from engineers. You can also learn about other IEEE programs and courses.

Read more here.

In the 1960s NASA began developing a system of reusable space shuttles to make its work more efficient and to reduce costs. The shuttles could launch like rockets, maneuver in Earth’s orbit, and land like airplanes. They also could carry large satellites to and from orbit.

Like other types of transportation, machinery eventually breaks down, and parts need to be replaced or fixed. And the cargo being carried to and from Earth has to be moved to its final destination. To complete such tasks, Spar Aerospace (now part of MDA Space) of Brampton, Ont., Canada, and the National Research Council in Ottawa developed a robotic arm, the Shuttle Remote Manipulator System. The project was a joint venture between the U.S. and Canadian governments.

Known as Canadarms, the robotic tools attached to shuttles’ exteriors. They allowed astronauts to handle and transfer tools, satellites, and other payloads. Inspections of the shuttle and repairs could be completed using the robots.

The system was first deployed in 1981 aboard Columbia’s second flight. Canadarm was used for 30 years on five shuttles and on the International Space Station.

The robotic arm was dedicated on 19 June as the 300th IEEE Milestone. The ceremony was held at MDA Space headquarters. The IEEE Toronto Section sponsored the nomination.

“It is appropriate that the 300th Milestone is the Canadarm,” says Michael Geselowitz, senior director of the IEEE History and Heritage group. “The technology spans aerospace, robotics, and computing fields of interest. It involves international cooperation between the United States and Canada, and it shows how IEEE and its members are at the cutting edge of many frontiers of science and technology.”

International collaboration for space exploration

Seeking to collaborate with other countries on the reusable spacecraft, NASA invited Canada to participate in 1969. It took some time for the country’s officials to determine what technology it could contribute. They learned of a robot that loaded and replaced spent fuel bundles in Canada’s deuterium uranium nuclear reactors, according to the Milestone webpage. That robot, developed by DSMA-Atcon (also now part of MDA Space), inspired what would become the Canadarm.

A proposal was submitted in 1974 to design and build the Shuttle Remote Manipulator System. The robotic arm would unload the contents of the space shuttle’s payload bay. NASA approved the project, and development began in 1975.

Canada had no space agency at the time, so the country’s National Research Council coordinated the organizations that collaborated on the project. Spar Aerospace led the subcontractor team, which included DMSA-Atcon, CAE, and the Canadian subsidiary of RCA Corp. Engineers from the University of Toronto’s Institute for Aerospace Studies contributed to the project.

Building an arm for zero gravity

NASA had strict requirements for the robot: The arm had to be lightweight and small enough to fit on the shuttle, as detailed in an article published by the University of Toronto. It also had to move forward and backward, up and down, left and right, and rotate along three perpendicular axes (known as six degrees of freedom).

To achieve all that, engineer Peter Carlisle Hughes designed the robot with two shoulder joints, one elbow, and three rotating wrists.

“Each joint had six degrees of freedom, and the arm had six links so that it could grab anything from any angle and move it anywhere,” Hughes said in the article. The IEEE life member worked at the Institute for Aerospace Studies.

“This milestone is a reminder of the privilege we all have at MDA Space—as engineers, designers, builders, operators—to build technology that shapes history.” —Holly Johnson, MDA Space vice president

The arm was 50 meters long and weighed 400 kilograms. It was made of materials that could withstand outer space’s harsh environment: titanium, stainless steel, and graphite epoxy. The arm was so lightweight that it couldn’t support itself under Earth’s gravity, so it lay on air bearings on the lab floor at Spar’s Brampton headquarters.

CAE engineers, including IEEE Life Member David A. Weston, designed the display and control panel as well as the hand controllers astronauts would use to monitor and operate the robot.

Because the robotic arm was meant to work in zero gravity, a room that simulated a weightless environment was built to test it. A computer-based simulation facility was constructed in Spar’s headquarters to evaluate its controllability using two simulation models, according to the University of Toronto. RIGID, an early computer simulation model, tested every part of the arm except for its flexible properties. ASAD, which stood for “all singing, all dancing,” examined the arm’s movements, ensuring the joints operated correctly. Both were created by Hughes and Spar engineer Andrew A. Goldenberg, who is now a professor emeritus at the University of Toronto.

The facility was also used to train astronauts on how to use Canadarm.

It took five years for the first Canadarm to be completed. In February 1981, it was presented to NASA at the Kennedy Space Center in Cape Canaveral, Fla., and deployed that November.

Lift off into space

Astronaut Stephen Robinson is anchored to a foot restraint on the extended Canadarm2 attached to the International Space Station during an extravehicular activity he conducted in 2005.NASA

The Canadarm was attached to the outside of the shuttle. Astronauts were able to monitor the arm’s movements through a live video feed provided by cameras installed on the wrist and elbow joints, according to the Milestone webpage. Using a hand controller and monitors located in the shuttle’s flight deck, astronauts handled and transferred tools, satellites, and other payloads weighing up 266,000 kilograms using minimal electricity.

NASA ordered four more systems. In 2001, Canadarm2 was attached to the International Space Station and used to help build the orbiting laboratory. It is a permanent part of the station, still completing maintenance tasks and moving supplies.

During the course of the 30-year shuttle program, the arms performed successfully and achieved the flight’s mission.

The original Canadarm took its final flight in July 2011 aboard the Atlantis shuttle.

Celebrating IEEE’s 300th Milestone

The IEEE Milestone dedication ceremony was held at MDA Space’s headquarters in Toronto, where the division that developed the Canadarm was located. The event brought together IEEE leaders and many of the engineers who helped develop the robotic system. Jill Gostin, the 2026 IEEE president‑elect, gave the opening remarks at the ceremony. She emphasized that the Milestone was not only celebrating the technology but also “the engineers, builders, programmers, and visionaries who believed technology could expand human possibility and who dared to push the boundaries of what humanity could achieve beyond Earth.”

To commemorate the achievement, Holly Johnson, vice president of MDA Robotics and Space Operations, and IEEE Life Senior Member David Michelson, chair of the IEEE Communications Society’s Communications History Committee, unveiled a bronze plaque that honored the technology. Michelson was the Milestone’s proposer.

“This milestone is a reminder of the privilege we all have at MDA Space—as engineers, designers, builders, operators—to build technology that shapes history,” Johnson said. “That same pioneering spirit that drove our team in those early days of space exploration now propels us into a new era as we work to build the infrastructure for the moon and beyond.”

The plaque, which was placed at MDA Space headquarters, reads:

In 1981 NASA first deployed a Shuttle Remote Manipulator System aboard the Space Shuttle. Developed by Spar Aerospace (now MDA Space) and the National Research Council of Canada, the Canadarm allowed astronauts to safely and reliably manipulate and transfer heavy payloads outside of the Shuttle, and to conduct inspections and repairs. This robotic system played a key role in the Shuttle and International Space Station programs, and revolutionized human spaceflight.

Reviewed by the IEEE History Committee and approved 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.

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.

A little over a decade ago, I led the controls design for a first-of-its-kind full digital-control system for a U.S. nuclear plant. It was, on paper, a beautiful machine—engineered to run itself the way a modern airliner does, with operators watching over a system that rarely needed them. And we made a decision that, to an efficiency-minded observer, looked backward: We deliberately left manual steps inside sequences the system could execute on its own.

We were solving a specific problem. An operator who only ever supervises automation slowly stops being an operator. The hands go cold. The mental model of what the plant is actually doing gets fuzzy. Then comes the day the automation hands control back. It’s always the worst day, because automation only quits when it’s confused or in trouble. But by then, you have a person in the chair who hasn’t truly operated the thing in years. The manual steps were there to keep the human current. It was inefficient by design, on purpose.

That plant, as it happened, was never built. It was shelved amid the politics and economics that surround nuclear power in this country, for reasons that had nothing to do with the engineering. But the design instinct outlived the project, and I’ve come to believe it’s the most useful idea I can offer to the argument now consuming every boardroom: What happens to human expertise when AI does the work that used to build it?

AI Is Disrupting the Engineering Career Ladder

The data has gotten hard to wave away. A Harvard University working paper covering some 65 million workers at more than 280,000 U.S. firms found that after companies adopted generative AI, junior employment fell roughly 9 percent within six quarters relative to nonadopters, while senior employment kept right on growing. A Stanford analysis of ADP payroll records points the same way: The youngest workers in the most AI-exposed occupations lost ground after late 2022 while their more-experienced colleagues held theirs. The Stanford researchers found that the losses concentrate where AI automates the work; where it merely augments, junior employment holds steady or rises.

The causal story is still contested, and honesty requires saying so. Researchers at the New York Fed attribute much of the rise in young-graduate unemployment not to AI but to remote work, arguing that firms are reluctant to hire inexperienced people whom they cannot train and mentor at a distance. But notice what the explanations share. Whether a model is absorbing the formative work or distance is severing the mentorship around it, both describe the same broken mechanism: the apprenticeship channel through which expertise passes from senior to junior. Either way, “entry-level” has quietly come to mean “three years of experience required.”

Strip away the noise and you’re left with one deceptively simple problem: You cannot become a senior engineer without first being a junior one. Expertise is not downloaded. It is earned through failed builds, dead-end debugging sessions, and the “why on earth did that work” moments that a capable AI will now happily spare the newcomer. Spare them enough of those and you produce a cohort that can supervise a model on paper but never developed the gut sense to know when the model is confidently, catastrophically wrong.

Most of the commentary stops at the diagnosis, or reaches for policy solutions that treat the loss of junior jobs as an economic problem. Yet it’s also an engineering problem, and safety-critical fields have already spent decades learning how to solve it.

Aviation’s Lessons About the Automation Paradox

My own career started at the sharp end of automation. My first job out of school was verifying and validating the software in the digital jet-engine controller that decides, faster than any pilot could, how a fighter plane’s engine responds. Even then, in the late 1980s, the central tension was visible: The machine outperforms the human in routine cases, but the human is all that stands between the aircraft and disaster in the cases the machine didn’t anticipate. This tension is known as the automation paradox, in which increasingly capable automation gives human operators less practice, while leaving them only the most difficult situations.

Aviation learned, repeatedly and expensively, what happens when human skills atrophy inside that gap. The canonical example is Air France flight 447, which fell into the Atlantic in 2009. The proximate cause was mundane. Iced-over airspeed sensors fed the autopilot bad data, and it did what it is designed to do: It disconnected and handed control of the airplane back to the crew. What followed was not a hardware failure. It was a competence failure. A recoverable situation became an unrecoverable one because the pilots, conditioned by thousands of hours of watching the automation fly, could not read a high-altitude aerodynamic stall and hand-fly their way out of it. The airplane was working. The training the automation had quietly eroded was not.

The industry’s response is instructive, and it’s the same move we made in that nuclear control room. It did not rip out the autopilot. It built deliberate manual practice back in. In 2017 the FAA issued Safety Alert for Operators 17007, “Manual Flight Operations Proficiency,” declaring that “manual flight is the foundation upon which other technical flying skills are built.” The alert formally recognized skill decay as a hazard in its own right. Some airlines amended their procedures to encourage hand-flying both the initial climb and initial descent in benign conditions, knowingly trading a sliver of fuel efficiency to keep the crew’s raw flying skills alive. That trade is the whole point. A perfectly optimized system that produces incompetent operators is not optimized at all. It has simply moved its failure mode somewhere the spreadsheet can’t see it.

Manual Gates Could Preserve Engineering Skills

Put the aviation lesson and the nuclear instinct side by side and they point to one design pattern we now need in AI-augmented work: the deliberate “manual gate.”

A manual gate is a point in a workflow where a human takes the controls, not because it is the fastest way to get the task done, and not only as a safety interlock, but specifically to exercise and preserve a skill that would otherwise decay. The distinguishing feature is that it is chosen. You decide, as a matter of design, which competencies your organization must keep alive in human beings because those are the ones you will need on the bad day. Then you engineer the friction required to keep them warm.

Picture how this might work on a software team that leans on AI for most of its code. The team places a manual gate around the skill it can least afford to lose: debugging. When a defect surfaces in a critical module, the assigned engineer—deliberately, often a junior one—must first reproduce the failure, trace it to root cause, and write an automated test that captures the bug, all with the AI assistant switched off. Only after the engineer commits to a diagnosis does the model come back on, to propose the fix, generate alternatives, and sweep the code base for similar bugs. The engineer then compares their diagnosis against the model’s. When the two disagree, that’s the design working, surfacing the disagreement before the bad day instead of during it.

This approach reframes the junior engineer entirely. The instinct today is to let AI do the entry-level work because it is faster and cheaper. But some of that work is not overhead to be eliminated. It is the training apparatus of your future senior staff, and you should protect it the way you’d protect any other piece of critical infrastructure. It may not be efficient this quarter, but dismantling it quietly mortgages your capability a decade out.

Why Companies Must Keep Training Junior Engineers

None of this is free, and pretending otherwise would insult the people who have to sign the budgets. A deliberate manual gate is, by construction, less efficient in the near term than full automation. Keeping juniors doing formative work and running the manual sequences costs something now to protect something later.

That’s a hard sell in a market that judges most leaders on quarterly results. A hired executive who carries “unnecessary” humans that AI could replace will hear about it from the board long before the payoff arrives. The math only works for someone insulated from that pressure: a founder with control, a private company, an institution with a genuinely long horizon, or a regulator willing to require workers to demonstrate their skills regularly, as pilots must. Which means the organizations most likely to preserve their own expertise are the ones structurally able to spend short-term margin on long-term capability; everyone else will need that outside push.

So here is the argument, in one line: Deliberate inefficiency is not waste. In safety-critical engineering we have always known it as insurance, and we buy it on purpose. As AI takes over the work where expertise is forged, the smart move is not to resist the automation. It is to keep our hands on the controls by design—so that when the automation fails, as it always eventually does, there is still someone in the chair who knows how to fly.

Across IEEE, our strength lies not only in the excellence of our individual communities but also in our ability to bring them together around shared problems that demand interdisciplinary solutions. Our mission as a public charity—to advance technology for the benefit of humanity—is becoming an increasingly powerful differentiator. It is more than a statement of principle; it is a strategic advantage. When engineers and technologists serve with purpose and lead with heart, they strengthen the future of our profession and demonstrate why IEEE is uniquely positioned to lead at the intersection of technology and societal impact.

IEEE Humanitarian Technologies is a consortium of programs and initiatives—supported by a global network of volunteers and technical professionals—working together to apply technology to solve the world’s most pressing problems. These include Empower a Billion Lives, EPICSinIEEE, MOVE, IEEE REACH, IEEE SIGHT, IEEE Smart Village, and IEEE Tech4Good. These programs embody our mission in action. They are not simply charitable activities; they are strategic assets that help IEEE lead globally, innovate boldly, and remain essential to technical professionals at every stage of their careers. While deeply human in purpose, humanitarian technologies are fundamentally engineering challenges, demanding the full depth of engineering rigor and realized through disciplined, deeply technical work.

Cultivating Technical Leaders

IEEE Humanitarian Technologies sits at the intersection of engineering excellence, societal need, and global opportunity. Its programs allow our members to show the world that engineering and technology are forces for good, capable of addressing urgent challenges with precision, creativity, and compassion. These programs do more than inspire; they strengthen the technical ecosystem that underpins IEEE’s leadership.

Bringing together experts from power and energy, communications, computing, robotics, biomedical engineering, and many other domains to address real-world problems, these interdisciplinary intersections are where breakthroughs emerge. When engineers and technologists collaborate with the right humanitarian frameworks across sectors and cultures, they illuminate new constraints, design pathways, and opportunities that traditional project environments rarely reveal. This is how humanitarian technologies help shape the future of engineering itself.

These efforts also illustrate a broader opportunity for IEEE. By identifying critical challenges that can be addressed only through collaboration across disciplines, IEEE can mobilize the power of its global community toward solving problems around the world. In doing so, we strengthen both our impact on society and the value we provide to members, partners, and future generations.

These programs also build the leadership capacity our profession needs. Engineers working in humanitarian contexts learn to navigate ambiguity, engage diverse stakeholders, manage constraints, and design for environments where failure has real human consequences. They develop systems thinking, ethical reasoning, and cross‑cultural fluency—competencies increasingly essential in a world where technology and society are deeply intertwined. They also learn to transition from R&D to implementation by engineering the support, manufacturing, and delivery systems that make solutions viable in specific countries, all while balancing competing requirements. In doing so, humanitarian programs equip professionals with the capabilities that define modern technical practice.

Humanitarian technologies also help prepare the future technical workforce. Students and young professionals increasingly seek meaningful, high‑impact work. By engaging in purpose‑driven projects, they can discover their own capacity to grow, strengthen their technical skills, and become the leaders and problem‑solvers who will guide our profession forward.

Purpose Inspires Engagement

Our members feel this deeply. Engagement research shows that members increasingly cited “giving back to my profession and the world community” as a reason for joining the organization and renewing their membership. Those with higher membership grades identify “participation in humanitarian technology efforts” as one of the most satisfying experiences IEEE offers. These are not just data points; they are also signals of what our community values and what it expects IEEE to champion.

Younger generations amplify this even more. Millennials view IEEE through a global lens, prioritizing “humanitarian impact” and “large-scale collaboration.” One millennial member shared that teaching robotics to children in under-resourced communities transformed them into a deeply engaged member. Gen Z members emphasize inclusivity, environmental responsibility, and purpose-driven engineering, recommending that IEEE offer humanitarian-based challenges and competitions to increase engagement.

These findings reveal something powerful: Humanitarian programs are not only meaningful; they also are magnetic. They attract younger engineers, keep them engaged, and help them build a professional identity rooted in purpose and impact. They also create loyalty and develop the leadership pipeline IEEE needs for the decades ahead.

These programs also strengthen our brand. Members across segments describe IEEE as an organization that works hard to make real changes in the world. That perception is not just flattering, it is strategic. It positions IEEE as a global leader in responsible innovation that can be trusted to guide technology for the public good, catalyzing innovation that benefits society at scale.

As we look ahead, IEEE has an opportunity to become the world’s leading convening force for developing interdisciplinary technology solutions to solve humanity’s most important challenges. Our future relevance will be defined not only by the technologies we advance but also by the problems we choose to help solve.

Read more powerful stories about how technology is improving lives across global initiatives in the 2025 IEEE Social Impact Report at ieee.org/advancing-technology/building-better-world/social-impact-report.

—MARY ELLEN RANDALL

IEEE president and CEO

Please share your thoughts with me: president@ieee.org.

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 GorrMathWorks

Then, 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/MIT

In 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.