Enlarge/ Illustration of a brain inside of a light bulb. (credit: Getty Images)
Researchers claim to have developed a new way to run AI language models more efficiently by eliminating matrix multiplication from the process. This fundamentally redesigns neural network operations that are currently accelerated by GPU chips. The findings, detailed in a recent preprint paper from researchers at the University of California Santa Cruz, UC Davis, LuxiTech, and Soochow University, could have deep implications for the environmental impact and operational costs of AI systems.
Matrix multiplication (often abbreviated to "MatMul") is at the center of most neural network computational tasks today, and GPUs are particularly good at executing the math quickly because they can perform large numbers of multiplication operations in parallel. That ability momentarily made Nvidia the most valuable company in the world last week; the company currently holds an estimated 98 percent market share for data center GPUs, which are commonly used to power AI systems like ChatGPT and Google Gemini.
In the new paper, titled "Scalable MatMul-free Language Modeling," the researchers describe creating a custom 2.7 billion parameter model without using MatMul that features similar performance to conventional large language models (LLMs). They also demonstrate running a 1.3 billion parameter model at 23.8 tokens per second on a GPU that was accelerated by a custom-programmed FPGA chip that uses about 13 watts of power (not counting the GPU's power draw). The implication is that a more efficient FPGA "paves the way for the development of more efficient and hardware-friendly architectures," they write.
Last week at VLSI Symposium, Intel detailed the manufacturing process that will form the foundation of its foundry service for high-performance data center customers. For the same power consumption, the Intel 3 process results in an 18 percent performance gain over the previous process, Intel 4. On the company’s roadmap, Intel 3 is the last to use the fin field-effect transistor (FinFET) structure, which the company pioneered in 2011. But it also includes Intel’s first use of a technology that is essential to its plans long after the FinFET is no longer cutting edge. What’s more, the technology is crucial to the company’s plans to become a foundry and make high-performance chips for other companies.
Called dipole work-function metal, it allows a chip designer to select transistors of several different threshold voltages. Threshold voltage is the level at which a device switches on or off. With the Intel 3 process, a single chip can include devices having any of four tightly-controlled threshold voltages. That’s important because different functions operate best with different threshold voltages. Cache memory, for example, typically demands devices with a high threshold voltage to prevent current leakage that wastes power. While other circuits might need the fastest switching devices, with the lowest threshold voltage.
Threshold voltage is set by the transistor’s gate stack, the layer of metal and insulation that controls the flow of current through the transistor. Historically, “the thickness of the metals determines the threshold voltage,” explains Walid Hafez, vice president of foundry technology development at Intel. “The thicker that work function metal is, the lower the threshold voltage is.” But this dependence on transistor geometry comes with some drawbacks as devices and circuits scale down.
Small deviations in the manufacturing process can alter the volume of the metal in the gate, leading to a somewhat broad range of threshold voltages. And that’s where the Intel 3 process exemplifies the change from Intel making chips only for itself to running as a foundry.
“The way an external foundry operates is very different” from an integrated device manufacturer like Intel was until recently, says Hafez. Foundry customers “need different things… One of those things they need is very tight variation of threshold voltage.”
Intel is different; even without the tight threshold voltage tolerances, it can sell all its parts by steering the best performing ones toward its datacenter business and the lower-performing ones in other market segments.
“A lot of external customers don’t do that,” he says. If a chip doesn’t meet their constraints, they may have to chuck it. “So for Intel 3 to be successful in the foundry space, it has to have those very tight variations.”
Dipoles ever after
Dipole work function materials guarantee the needed control over threshold voltage without worrying about how much room you have in the gate. It’s a proprietary mix of metals and other materials that, despite being only angstroms thick, has a powerful effect on a transistor’s silicon channel.
Intel’s use of dipole work-function materials means the gate surrounding each fin in a FinFET is thinner.Intel
Like the old, thick metal gate, the new mix of materials electrostatically alters the silicon’s band structure to shift the threshold voltage. But it does so by inducing a dipole—a separation of charge—in the thin insulation between it and the silicon.
Because foundry customers were demanding tight control of Intel, it’s likely that competitors TSMC and Samsung already use dipoles in their latest FinFET processes. What exactly such structures are made of is a trade secret, but lanthanum is a component in earlier research, and it was the key ingredient in other research presented by the Belgium-based microelectronics research center, Imec. That research was concerned with how best to build the material around stacks of horizontal silicon ribbons instead of one or two vertical fins.
In these devices, called nanosheets or gate all-around transistors, there are mere nanometers between each ribbon of silicon, so dipoles are a necessity. Samsung has already introduced a nanosheet process, and Intel’s, called 20A, is scheduled for later this year. Introducing dipole work function at Intel 3 helps get 20A and its successor 18A into a more mature state, says Hafez.
Flavors of Intel 3
Dipole work-function was not the only technology behind the 18 percent boost Intel 3 delivers over its predecessor. Among them are more perfectly formed fins, more sharply defined contacts to the transistor, and lower resistance and capacitance in the interconnects. (Hafez details all that here.)
Intel is using the process to build its Xeon 6 CPUs. And the company plans to offer customers three variations on the technology, including one, 3-PT, with 9-micrometer through-silicon-vias for use in 3D stacking. “We expect Intel 3-PT to be the backbone of our foundry processes for some time to come,” says Hafez.
With billions of government funding flowing into hydrogen hubs across the country, the US is navigating a pivotal year for hydrogen production. Incoming final guidelines from the IRS on the definition of clean hydrogen could significantly impact project viability, funding, and deployment.
Securing a competitive edge in this rapidly-evolving market is more critical than ever. Our newly-released whitepaper, US on the Verge of Clean Hydrogen Boom as Dollars Flow, is your essential resource for the latest hydrogen projects, federal initiatives, and prognosis for the year ahead.
Critical Policy and Regulatory Insights: Understand the implications of forthcoming IRS guidelines, qualifying for 45V tax credits, and the Three Pillar approach to green hydrogen, to help you futureproof your hydrogen investments, projects, and strategy for these changes
The Latest Market Forecasts: Review detailed projections for clean hydrogen growth trajectories, up-to-date production costs, and demand predictions, to help you forge a profitable hydrogen strategy with confidence
Timely Competitive Analysis: Explore the major players dominating the market today, and get the latest updates on the Regional Clean Hydrogen Hubs plan, to provide a window into your competitor’s strategies and secure your competitive advantage
WordPress plugins running on as many as 36,000 websites have been backdoored in a supply-chain attack with unknown origins, security researchers said on Monday.
So far, five plugins are known to be affected in the campaign, which was active as recently as Monday morning, researchers from security firm Wordfence reported. Over the past week, unknown threat actors have added malicious functions to updates available for the plugins on WordPress.org, the official site for the open source WordPress CMS software. When installed, the updates automatically create an attacker-controlled administrative account that provides full control over the compromised site. The updates also add content designed to goose search results.
Poisoning the well
“The injected malicious code is not very sophisticated or heavily obfuscated and contains comments throughout making it easy to follow,” the researchers wrote. “The earliest injection appears to date back to June 21st, 2024, and the threat actor was still actively making updates to plugins as recently as 5 hours ago.”
Enlarge/ Michael Jackson in concert, 1986. Sony Music owns a large portion of publishing rights to Jackson's music. (credit: Getty Images)
Universal Music Group, Sony Music, and Warner Records have sued AI music-synthesis companies Udio and Suno for allegedly committing mass copyright infringement by using recordings owned by the labels to train music-generating AI models, reports Reuters. Udio and Suno can generate novel song recordings based on text-based descriptions of music (i.e., "a dubstep song about Linus Torvalds").
The lawsuits, filed in federal courts in New York and Massachusetts, claim that the AI companies' use of copyrighted material to train their systems could lead to AI-generated music that directly competes with and potentially devalues the work of human artists.
Like other generative AI models, both Udio and Suno (which we coveredseparately in April) rely on a broad selection of existing human-created artworks that teach a neural network the relationship between words in a written prompt and styles of music. The record labels correctly note that these companies have been deliberately vague about the sources of their training data.
When it comes to motorsports, the need for speed isn’t only on the racetrack. Engineers who support race teams also need to work at a breakneck pace to fix problems, and that’s something Aakhilesh Singhania relishes.
Singhania is a senior applications engineer at Bosch Engineering, in Novi, Mich. He develops and supports electronic control systems for hybrid race cars, which feature combustion engines and battery-powered electric motors.
Aakhilesh Singhania
Employer:
Bosch Engineering
Occupation:
Senior applications engineer
Education:
Bachelor’s degree in mechanical engineering, Manipal Institute of Technology, India; master’s degree in automotive engineering, University of Michigan, Ann Arbor
His vehicles compete in two iconic endurance races: the Rolex 24 at Daytona in Daytona Beach, Fla., and the 24 Hours of Le Mans in France. He splits his time between refining the underlying technology and providing trackside support on competition day. Given the relentless pace of the racing calendar and the intense time pressure when cars are on the track, the job is high octane. But Singhania says he wouldn’t have it any other way.
“I’ve done jobs where the work gets repetitive and mundane,” he says. “Here, I’m constantly challenged. Every second counts, and you have to be very quick at making decisions.”
An Early Interest in Motorsports
Growing up in Kolkata, India, Singhania picked up a fascination with automobiles from his father, a car enthusiast.
In 2010, when Singhania began his mechanical engineering studies at India’s Manipal Institute of Technology, he got involved in the Formula Studentprogram, an international engineering competition that challenges teams of university students to design, build, and drive a small race car. The cars typically weigh less than 250 kilograms and can have an engine no larger than 710 cubic centimeters.
“It really hooked me,” he says. “I devoted a lot of my spare time to the program, and the experience really motivated me to dive further into motorsports.”
One incident in particular shaped Singhania’s career trajectory. In 2013, he was leading Manipal’s Formula Student team and was one of the drivers for a competition in Germany. When he tried to start the vehicle, smoke poured out of the battery, and the team had to pull out of the race.
“I asked myself what I could have done differently,” he says. “It was my lack of knowledge of the electrical system of the car that was the problem.” So, he decided to get more experience and education.
Learning About Automotive Electronics
After graduating in 2014, Singhania began working on engine development for Indian car manufacturer Tata Motors in Pune. In 2016, determined to fill the gaps in his knowledge about automotive electronics, he left India to begin a master’s degree program in automotive engineering at the University of Michigan in Ann Arbor.
He took courses in battery management, hybrid controls, and control-system theory, parlaying this background into an internship with Bosch in 2017. After graduation in 2018, he joined Bosch full-time as a calibration engineer, developing technology for hybrid and electric vehicles.
Transitioning into motorsports required perseverance, Singhania says. He became friendly with the Bosch team that worked on electronics for race cars. Then in 2020 he got his big break.
That year, the U.S.-based International Motor Sports Association and the France-based Automobile Club de l’Ouest created standardized rules to allow the same hybrid race cars to compete in both the Sportscar Championship in North America, host of the famous Daytona race, and the global World Endurance Championship, host of Le Mans.
The Bosch motorsports team began preparing a proposal to provide the standardized hybrid system. Singhania, whose job already included creating simulations of how vehicles could be electrified, volunteered to help.
“I’m constantly challenged. Every second counts, and you have to be very quick at making decisions.”
The competition organizers selected Bosch as lead developer of the hybrid system that would be provided to all teams. Bosch engineers would also be required to test the hardware they supplied to each team to ensure none had an advantage.
“The performance of all our parts in all the cars has to fall within 1 percent of each other,” Singhania says.
After Bosch won the contract, Singhania officially became a motorsports calibration engineer, responsible for tweaking the software to fit the idiosyncrasies of each vehicle.
In 2022 he stepped up to his current role: developing software for the hybrid control unit (HCU), which is essentially the brains of the vehicle. The HCU helps coordinate all of the different subsystems such as the engine, battery, and electric motor and is responsible for balancing power requirements among these different components to maximize performance and lifetime.
Bosch’s engineers also designed software known as an equity model, which runs on the HCU. It is based on historical data collected from the operation of the hybrid systems’ various components, and controls their performance in real time to ensure all the teams’ hardware operates at the same level.
In addition, Singhania creates simulations of the race cars, which are used to better understand how the different components interact and how altering their configuration would affect performance.
Troubleshooting Problems on Race Day
Technology development is only part of Singhania’s job. On race days, he works as a support engineer, helping troubleshoot problems with the hybrid system as they crop up. Singhania and his colleagues monitor each team’s hardware using computers on Bosch’s race-day trailer, a mobile nerve center hardwired to the organizers’ control center on the race track.
“We are continuously looking at all the telemetry data coming from the hybrid system and analyzing [the system’s] health and performance,” he says.
If the Bosch engineers spot an issue or a team notifies them of a problem, they rush to the pit stall to retrieve a USB stick from the vehicle, which contains detailed data to help them diagnose and fix the issue.
After the race, the Bosch engineers analyze the telemetry data to identify ways to boost the standardized hybrid system’s performance for all the teams. In motorsports, where the difference between winning and losing can come down to fractions of a second, that kind of continual improvement is crucial.
Customers “put lots of money into this program, and they are there to win,” Singhania says.
Breaking Into Motorsports Engineering
Many engineers dream about working in the fast-paced and exciting world of motorsports, but it’s not easy breaking in. The biggest lesson Singhania learned is that if you don’t ask, you don’t get invited.
“Keep pursuing them because nobody’s going to come to you with an offer,” he says. “You have to keep talking to people and be ready when the opportunity presents itself.”
Demonstrating that you have experience contributing to challenging projects is a big help. Many of the engineers Bosch hires have been involved in Formula Student or similar automotive-engineering programs, such as the EcoCAR EV Challenge, says Singhania.
The job isn’t for everyone, though, he says. It’s demanding and requires a lot of travel and working on weekends during race season. But if you thrive under pressure and have a knack for problem solving, there are few more exciting careers.
Imagine it’s 2050 and you’re on a cross-country flight on a new type of airliner, one with no fuel on board. The plane takes off, and you rise above the airport. Instead of climbing to cruising altitude, though, your plane levels out and the engines quiet to a low hum. Is this normal? No one seems to know. Anxious passengers crane their necks to get a better view out their windows. They’re all looking for one thing.
Then it appears: a massive antenna array on the horizon. It’s sending out a powerful beam of electromagnetic radiation pointed at the underside of the plane. After soaking in that energy, the engines power up, and the aircraft continues its climb. Over several minutes, the beam will deliver just enough energy to get you to the next ground antenna located another couple hundred kilometers ahead.
The person next to you audibly exhales. You sit back in your seat and wait for your drink. Old-school EV-range anxiety is nothing next to this.
Electromagnetic waves on the fly
Beamed power for aviation is, I admit, an outrageous notion. If physics doesn’t forbid it, federal regulators or nervous passengers probably will. But compared with other proposals for decarbonizing aviation, is it that crazy?
Batteries, hydrogen, alternative carbon-based fuels—nothing developed so far can store energy as cheaply and densely as fossil fuels, or fully meet the needs of commercial air travel as we know it. So, what if we forgo storing all the energy on board and instead beam it from the ground? Let me sketch what it would take to make this idea fly.
Beamed Power for Aviation
Fly by Microwave: Warm up to a new kind of air travel
For the wireless-power source, engineers would likely choose microwaves because this type of electromagnetic radiation can pass unruffled through clouds and because receivers on planes could absorb it completely, with nearly zero risk to passengers.
To power a moving aircraft, microwave radiation would need to be sent in a tight, steerable beam. This can be done using technology known as a phased array, which is commonly used to direct radar beams. With enough elements spread out sufficiently and all working together, phased arrays can also be configured to focus power on a point a certain distance away, such as the receiving antenna on a plane.
Phased arrays work on the principle of constructive and destructive interference. The radiation from the antenna elements will, of course, overlap. In some directions the radiated waves will interfere destructively and cancel out one another, and in other directions the waves will fall perfectly in phase, adding together constructively. Where the waves overlap constructively, energy radiates in that direction, creating a beam of power that can be steered electronically.
How far we can send energy in a tight beam with a phased array is governed by physics—specifically, by something called the diffraction limit. There’s a simple way to calculate the optimal case for beamed power: D1 D2 > λ R. In this mathematical inequality, D1 and D2 are the diameters of the sending and receiving antennas, λ is the wavelength of the radiation, and R is the distance between those antennas.
Now, let me offer some ballpark numbers to figure out how big the transmitting antenna (D1) must be. The size of the receiving antenna on the aircraft is probably the biggest limiting factor. A medium-size airliner has a wing and body area of about 1,000 square meters, which should provide for the equivalent of a receiving antenna that’s 30 meters wide (D2) built into the underside of the plane.
If physics doesn’t forbid it, federal regulators or nervous passengers probably will.
Next, let’s guess how far we would need to beam the energy. The line of sight to the horizon for someone in an airliner at cruising altitude is about 360 kilometers long, assuming the terrain below is level. But mountains would interfere, plus nobody wants range anxiety, so let’s place our ground antennas every 200 km along the flight path, each beaming energy half of that distance. That is, set R to 100 km.
Finally, assume the microwave wavelength (λ) is 5 centimeters. This provides a happy medium between a wavelength that’s too small to penetrate clouds and one that’s too large to gather back together on a receiving dish. Plugging these numbers into the equation above shows that in this scenario the diameter of the ground antennas (D1) would need to be at least about 170 meters. That’s gigantic, but perhaps not unreasonable. Imagine a series of three or four of these antennas, each the size of a football stadium, spread along the route, say, between LAX and SFO or between AMS and BER.
Power beaming in the real world
While what I’ve described is theoretically possible, in practice engineers have beamed only a fraction of the amount of power needed for an airliner, and they’ve done that only over much shorter distances.
NASA holds the record from an experiment in 1975, when it beamed 30 kilowatts of power over 1.5 km with a dish the size of a house. To achieve this feat, the team used an analog device called a klystron. The geometry of a klystron causes electrons to oscillate in a way that amplifies microwaves of a particular frequency—kind of like how the geometry of a whistle causes air to oscillate and produce a particular pitch.
Klystrons and their cousins, cavity magnetrons (found in ordinary microwave ovens), are quite efficient because of their simplicity. But their properties depend on their precise geometry, so it’s challenging to coordinate many such devices to focus energy into a tight beam.
In more recent years, advances in semiconductor technology have allowed a single oscillator to drive a large number of solid-state amplifiers in near-perfect phase coordination. This has allowed microwaves to be focused much more tightly than was possible before, enabling more-precise energy transfer over longer distances.
In 2022, the Auckland-based startup Emrod showed just how promising this semiconductor-enabled approach could be. Inside a cavernous hangar in Germany owned by Airbus, the researchers beamed 550 watts across 36 meters and kept over 95 percent of the energy flowing in a tight beam—far better than could be achieved with analog systems. In 2021, the U.S. Naval Research Laboratory showed that these techniques could handle higher power levels when it sent more than a kilowatt between two ground antennas over a kilometer apart. Other researchers have energized drones in the air, and a few groups even intend to use phased arrays to beam solar power from satellites to Earth.
A rectenna for the ages
So beaming energy to airliners might not be entirely crazy. But please remain seated with your seat belts fastened; there’s some turbulence ahead for this idea. A Boeing 737 aircraft at takeoff requires about 30 megawatts—a thousand times as much power as any power-beaming experiment has demonstrated. Scaling up to this level while keeping our airplanes aerodynamic (and flyable) won’t be easy.
Consider the design of the antenna on the plane, which receives and converts the microwaves to an electric current to power the aircraft. This rectifying antenna, or rectenna, would need to be built onto the underside surfaces of the aircraft with aerodynamics in mind. Power transmission will be maximized when the plane is right above the ground station, but it would be far more limited the rest of the time, when ground stations are far ahead or behind the plane. At those angles, the beam would activate only either the front or rear surfaces of the aircraft, making it especially hard to receive enough power.
With 30 MW blasting onto that small of an area, power density will be an issue. If the aircraft is the size of Boeing 737, the rectenna would have to cram about 25 W into each square centimeter. Because the solid-state elements of the array would be spaced about a half-wavelength—or 2.5 cm—apart, this translates to about 150 W per element—perilously close to the maximum power density of any solid-state power-conversion device. The top mark in the 2016 IEEE/Google Little Box Challenge was about 150 W per cubic inch (less than 10 W per cubic centimeter).
The rectenna will also have to weigh very little and minimize the disturbance to the airflow over the plane. Compromising the geometry of the rectenna for aerodynamic reasons might lower its efficiency. State-of-the art power-transfer efficiencies are only about 30 percent, so the rectenna can’t afford to compromise too much.
A Boeing 737 aircraft at takeoff requires about 30 megawatts—a thousand times as much power as any power-beaming experiment has demonstrated.
And all of this equipment will have to work in an electric field of about 7,000 volts per meter—the strength of the power beam. The electric field inside a microwave oven, which is only about a third as strong, can create a corona discharge, or electric arc, between the tines of a metal fork, so just imagine what might happen inside the electronics of the rectenna.
And speaking of microwave ovens, I should mention that, to keep passengers from cooking in their seats, the windows on any beamed-power airplane would surely need the same wire mesh that’s on the doors of microwave ovens—to keep those sizzling fields outside the plane. Birds, however, won’t have that protection.
Fowl flying through our power beam near the ground might encounter a heating of more than 1,000 watts per square meter—stronger than the sun on a hot day. Up higher, the beam will narrow to a focal point with much more heat. But because that focal point would be moving awfully fast and located higher than birds typically fly, any roasted ducks falling from the sky would be rare in both senses of the word. Ray Simpkin, chief science officer at Emrod, told me it’d take “more than 10 minutes to cook a bird” with Emrod’s relatively low-power system.
Legal challenges would surely come, though, and not just from the National Audubon Society. Thirty megawatts beamed through the air would be about 10 billion times as strong as typical signals at 5-cm wavelengths (a band currently reserved for amateur radio and satellite communications). Even if the transmitter could successfully put 99 percent of the waves into a tight beam, the 1 percent that’s leaked would still be a hundred million times as strong as approved transmissions today.
And remember that aviation regulators make us turn off our cellphones during takeoff to quiet radio noise, so imagine what they’ll say about subjecting an entire plane to electromagnetic radiation that’s substantially stronger than that of a microwave oven. All these problems are surmountable, perhaps, but only with some very good engineers (and lawyers).
Compared with the legal obstacles and the engineering hurdles we’d need to overcome in the air, the challenges of building transmitting arrays on the ground, huge as they would have to be, seem modest. The rub is the staggering number of them that would have to be built. Many flights occur over mountainous terrain, producing a line of sight to the horizon that is less than 100 km. So in real-world terrain we’d need more closely spaced transmitters. And for the one-third of airline miles that occur over oceans, we would presumably have to build floating arrays. Clearly, building out the infrastructure would be an undertaking on the scale of the Eisenhower-era U.S. interstate highway system.
Decarbonizing with the world’s largest microwave
People might be able to find workarounds for many of these issues. If the rectenna is too hard to engineer, for example, perhaps designers will find that they don’t have to turn the microwaves back into electricity—there are precedents for using heat to propel airplanes. A sawtooth flight path—with the plane climbing up as it approaches each emitter station and gliding down after it passes by—could help with the power-density and field-of-view issues, as could flying-wing designs, which have much more room for large rectennas. Perhaps using existing municipal airports or putting ground antennas near solar farms could reduce some of the infrastructure cost. And perhaps researchers will find shortcuts to radically streamline phased-array transmitters. Perhaps, perhaps.
To be sure, beamed power for aviation faces many challenges. But less-fanciful options for decarbonizing aviation have their own problems. Battery-powered planes don’t even come close to meeting the needs of commercial airlines. The best rechargeable batteries have about 5 percent of the effective energy density of jet fuel. At that figure, an all-electric airliner would have to fill its entire fuselage with batteries—no room for passengers, sorry—and it’d still barely make it a tenth as far as an ordinary jet. Given that the best batteries have improved by only threefold in the past three decades, it’s safe to say that batteries won’t power commercial air travel as we know it anytime soon.
Any roasted ducks falling from the sky would be rare in both senses of the word.
Hydrogen isn’t much further along, despite early hydrogen-powered flights occurring nearly 40 years ago. And it’s potentially dangerous—enough that some designs for hydrogen planes have included two separate fuselages: one for fuel and one for people to give them more time to get away if the stuff gets explode-y. The same factors that have kept hydrogen cars off the road will probably keep hydrogen planes out of the sky.
Synthetic and biobased jet fuels are probably the most reasonable proposal. They’ll give us aviation just as we know it today, just at a higher cost—perhaps 20 to 50 percent more expensive per ticket. But fuels produced from food crops can be worse for the environment than the fossil fuels they replace, and fuels produced from CO2 and electricity are even less economical. Plus, all combustion fuels could still contribute to contrail formation, which makes up more than half of aviation’s climate impact.
The big problem with the “sane” approach for decarbonizing aviation is that it doesn’t present us with a vision of the future at all. At the very best, we’ll get a more expensive version of the same air travel experience the world has had since the 1970s.
True, beamed power is far less likely to work. But it’s good to examine crazy stuff like this from time to time. Airplanes themselves were a crazy idea when they were first proposed. If we want to clean up the environment and produce a future that actually looks like a future, we might have to take fliers on some unlikely sounding schemes.
Tsunenobu Kimoto, a professor of electronic science and engineering at Kyoto University, literally wrote the book on silicon carbide technology. Fundamentals of Silicon Carbide Technology, published in 2014, covers properties of SiC materials, processing technology, theory, and analysis of practical devices.
Kimoto, whose silicon carbide research has led to better fabrication techniques, improved the quality of wafers and reduced their defects. His innovations, which made silicon carbide semiconductor devices more efficient and more reliable and thus helped make them commercially viable, have had a significant impact on modern technology.
Decades before a Tesla Model 3 rolled off the assembly line with an SiC inverter, a small cadre of researchers, including Kimoto, foresaw the promise of silicon carbide technology. In obscurity they studied it and refined the techniques for fabricating power transistors with characteristics superior to those of the silicon devices then in mainstream use.
Today MOSFETs and other silicon carbide transistors greatly reduce on-state loss and switching losses in power-conversion systems, such as the inverters in an electric vehicle used to convert the battery’s direct current to the alternating current that drives the motor. Lower switching losses make the vehicles more efficient, reducing the size and weight of their power electronics and improving power-train performance. Silicon carbide–based chargers, which convert alternating current to direct current, provide similar improvements in efficiency.
But those tools didn’t just appear. “We had to first develop basic techniques such as how to dope the material to make n-type and p-type semiconductor crystals,” Kimoto says. N-type crystals’ atomic structures are arranged so that electrons, with their negative charges, move freely through the material’s lattice. Conversely, the atomic arrangement of p-type crystals’ contains positively charged holes.
Kimoto’s interest in silicon carbide began when he was working on his Ph.D. at Kyoto University in 1990.
“At that time, few people were working on silicon carbide devices,” he says. “And for those who were, the main target for silicon carbide was blue LED.
“There was hardly any interest in silicon carbide power devices, like MOSFETs and Schottky barrier diodes.”
Kimoto began by studying how SiC might be used as the basis of a blue LED. But then he read B. Jayant Baliga’s 1989 paper “Power Semiconductor Device Figure of Merit for High-Frequency Applications” in IEEE Electron Device Letters, and he attended a presentation by Baliga, the 2014 IEEE Medal of Honor recipient, on the topic.
“I was convinced that silicon carbide was very promising for power devices,” Kimoto says. “The problem was that we had no wafers and no substrate material,” without which it was impossible to fabricate the devices commercially.
In order to get silicon carbide power devices, “researchers like myself had to develop basic technology such as how to dope the material to make p-type and n-type crystals,” he says. “There was also the matter of forming high-quality oxides on silicon carbide.” Silicon dioxide is used in a MOSFET to isolate the gate and prevent electrons from flowing into it.
The first challenge Kimoto tackled was producing pure silicon carbide crystals. He decided to start with carborundum, a form of silicon carbide commonly used as an abrasive. Kimoto took some factory waste materials—small crystals of silicon carbide measuring roughly 5 millimeters by 8 mm—and polished them.
He found he had highly doped n-type crystals. But he realized having only highly doped n-type SiC would be of little use in power applications unless he also could produce lightly doped (high purity) n-type and p-type SiC.
Connecting the two material types creates a depletion region straddling the junction where the n-type and p-type sides meet. In this region, the free, mobile charges are lost because of diffusion and recombination with their opposite charges, and an electric field is established that can be exploited to control the flow of charges across the boundary.
“Silicon carbide is a family with many, many brothers.”
By using an established technique, chemical vapor deposition, Kimoto was able to grow high-purity silicon carbide. The technique grows SiC as a layer on a substrate by introducing gasses into a reaction chamber.
At the time, silicon carbide, gallium nitride, and zinc selenide were all contenders in the race to produce a practical blue LED. Silicon carbide, Kimoto says, had only one advantage: It was relatively easy to make a silicon carbide p-n junction. Creating p-n junctions was still difficult to do with the other two options.
By the early 1990s, it was starting to become clear that SiC wasn’t going to win the blue-LED sweepstakes, however. The inescapable reality of the laws of physics trumped the SiC researchers’ belief that they could somehow overcome the material’s inherent properties. SiC has what is known as an indirect band gap structure, so when charge carriers are injected, the probability of the charges recombining and emitting photons is low, leading to poor efficiency as a light source.
While the blue-LED quest was making headlines, many low-profile advances were being made using SiC for power devices. By 1993, a team led by Kimoto and Hiroyuki Matsunami demonstrated the first 1,100-volt silicon carbide Schottky diodes, which they described in a paper in IEEE Electron Device Letters. The diodes produced by the team and others yielded fast switching that was not possible with silicon diodes.
“With silicon p-n diodes,” Kimoto says, “we need about a half microsecond for switching. But with a silicon carbide, it takes only 10 nanoseconds.”
The ability to switch devices on and off rapidly makes power supplies and inverters more efficient because they waste less energy as heat. Higher efficiency and less heat also permit designs that are smaller and lighter. That’s a big deal for electric vehicles, where less weight means less energy consumption.
Kimoto’s second breakthrough was identifying which form of the silicon carbide material would be most useful for electronics applications.
“Silicon carbide is a family with many, many brothers,” Kimoto says, noting that more than 100 variants with different silicon-carbon atomic structures exist.
The 6H-type silicon carbide was the default standard phase used by researchers targeting blue LEDs, but Kimoto discovered that the 4H-type has much better properties for power devices, including high electron mobility. Now all silicon carbide power devices and wafer products are made with the 4H-type.
Silicon carbide power devices in electric vehicles can improve energy efficiency by about 10 percent compared with silicon, Kimoto says. In electric trains, he says, the power required to propel the cars can be cut by 30 percent compared with those using silicon-based power devices.
Challenges remain, he acknowledges. Although silicon carbide power transistors are used in Teslas, other EVs, and electric trains, their performance is still far from ideal because of defects present at the silicon dioxide–SiC interface, he says. The interface defects lower the performance and reliability of MOS-based transistors, so Kimoto and others are working to reduce the defects.
A career sparked by semiconductors
When Kimoto was an only child growing up in Wakayama, Japan, near Osaka, his parents insisted he study medicine, and they expected him to live with them as an adult. His father was a garment factory worker; his mother was a homemaker. His move to Kyoto to study engineering “disappointed them on both counts,” he says.
His interest in engineering was sparked, he recalls, when he was in junior high school, and Japan and the United States were competing for semiconductor industry supremacy.
At Kyoto University, he earned bachelor’s and master’s degrees in electrical engineering, in 1986 and 1988. After graduating, he took a job at Sumitomo Electric Industries’ R&D center in Itami. He worked with silicon-based materials there but wasn’t satisfied with the center’s research opportunities.
He returned to Kyoto University in 1990 to pursue his doctorate. While studying power electronics and high-temperature devices, he also gained an understanding of material defects, breakdown, mobility, and luminescence.
“My experience working at the company was very valuable, but I didn’t want to go back to industry again,” he says. By the time he earned his doctorate in 1996, the university had hired him as a research associate.
He has been there ever since, turning out innovations that have helped make silicon carbide an indispensable part of modern life.
Growing the silicon carbide community at IEEE
Kimoto joined IEEE in the late 1990s. An active volunteer, he has helped grow the worldwide silicon carbide community.
“Now when we hold a silicon carbide conference, more than 1,000 people gather,” he says. “At IEEE conferences like the International Electron Devices Meeting or ISPSD, we always see several well-attended sessions on silicon carbide power devices because more IEEE members pay attention to this field now.”
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
RoboCup 2024: 17–22 July 2024, EINDHOVEN, NETHERLANDS
ICRA@40: 23–26 September 2024, ROTTERDAM, NETHERLANDS
We present Morphy, a novel compliant and morphologically-aware Flying Robot that integrates sensorized flexible joints in its arms, thus enabling resilient collisions at high speeds and the ability to squeeze through openings more narrow than its nominal dimensions.
Morphy represents a new class of soft-flying robots that can facilitate unprecedented resilience through innovations both in the “body” and “brain.” The novel soft body can, in turn, enable new avenues for autonomy. Collisions that previously had to be avoided have now become acceptable risks, while areas that are untraversable for a certain robot size can now be negotiated through self-squeezing. These novel bodily interactions with the environment can give rise to new types of embodied intelligence.
Segments of daily training for robots driven by reinforcement learning. Multiple tests done in advance for friendly service humans. The training includes some extreme tests, please do not imitate!
Here I am, without the ability or equipment (or desire) required to iron anything that I own, and Flexiv’s got robots out there ironing fancy leather car seats.
We unveiled a significant leap forward in perception technology for our humanoid robot GR-1. The newly adapted pure vision solution integrates bird’s eye view, transformer models, and occupancy network for precise and efficient environmental perception.
LimX Dynamics’ humanoid robot CL-1 was launched in Dec 2023. It climbed stairs based on real-time terrain perception, 2 steps per stair. 4 months later, in Apr 2024, the 2nd demo video showcased CL-1 in the same scenario. It advanced to climb the same stair 1 step per stair.
New research from the University of Massachusetts Amherst shows that programming robots to create their own teams and voluntarily wait for their teammates results in faster task completion, with the potential to improve manufacturing, agriculture and warehouse automation.
LASDRA (Large-size Aerial Skeleton with Distributed Rotor Actuation system (ICRA18) is a scalable and modular aerial robot, can assume a very slender, long and dexterous form-factor and very light weight.
We propose augmenting initially passive structures built from simple repeated cells, with novel active units to enable dynamic, shape-changing, and robotic applications. Inspired by metamaterials that can employ mechanisms, we build a framework that allows users to configure cells of this passive structure to allow it to perform complex tasks.
Testing autonomous exploration at the Exyn Office using Spot from Boston Dynamics. In this demo, Spot autonomous explores our flight space while on the hunt for one of our engineers.
Meet Heavy Picker, the strongest robot in bulky waste sorting and an absolute pro at lifting and sorting waste. With skills that would make a concert pianist jealous and a work ethic that never needs coffee breaks, Heavy Picker was on the lookout for new challenges.
AI is the biggest and most consequential business, financial, legal, technological, and cultural story of our time. In this panel, you will hear from the underrepresented community of women scientists who have been leading the AI revolution — from the beginning to now.
On Thursday, Anthropic announced Claude 3.5 Sonnet, its latest AI language model and the first in a new series of "3.5" models that build upon Claude 3, launched in March. Claude 3.5 can compose text, analyze data, and write code. It features a 200,000 token context window and is available now on the Claude website and through an API. Anthropic also introduced Artifacts, a new feature in the Claude interface that shows related work documents in a dedicated window.
So far, people outside of Anthropic seem impressed. "This model is really, really good," wrote independent AI researcher Simon Willison on X. "I think this is the new best overall model (and both faster and half the price of Opus, similar to the GPT-4 Turbo to GPT-4o jump)."
As we've written before, benchmarks for large language models (LLMs) are troublesome because they can be cherry-picked and often do not capture the feel and nuance of using a machine to generate outputs on almost any conceivable topic. But according to Anthropic, Claude 3.5 Sonnet matches or outperforms competitor models like GPT-4o and Gemini 1.5 Pro on certain benchmarks like MMLU (undergraduate level knowledge), GSM8K (grade school math), and HumanEval (coding).
Tech companies have been caught up in a race to build the biggest large language models (LLMs). In April, for example, Meta announced the 400-billion-parameter Llama 3, which contains twice the number of parameters—or variables that determine how the model responds to queries—than OpenAI’s original ChatGPT model from 2022. Although not confirmed, GPT-4 is estimated to have about 1.8 trillion parameters.
In the last few months, however, some of the largest tech companies, including Apple and Microsoft, have introduced small language models (SLMs). These models are a fraction of the size of their LLM counterparts and yet, on many benchmarks, can match or even outperform them in text generation.
On 10 June, at Apple’s Worldwide Developers Conference, the company announced its “Apple Intelligence” models, which have around 3 billion parameters. And in late April, Microsoft released its Phi-3 family of SLMs, featuring models housing between 3.8 billion and 14 billion parameters.
OpenAI’s CEO Sam Altman believes we’re at the end of the era of giant models.
In a series of tests, the smallest of Microsoft’s series of models, Phi-3-mini, rivalled OpenAI’s GPT-3.5 (175 billion parameters), which powers the free version of ChatGPT, and outperformed Google’s Gemma (7 billion parameters). The tests evaluated how well a model understands language by prompting it with questions about mathematics, philosophy, law, and more. What’s more interesting, Microsoft’s Phi-3-small, with 7 billion parameters, fared remarkably better than GPT-3.5 in many of these benchmarks.
Aaron Mueller, who researches language models at Northeastern University in Boston, isn’t surprised SLMs can go toe-to-toe with LLMs in select functions. He says that’s because scaling the number of parameters isn’t the only way to improve a model’s performance: Training it on higher-quality data can yield similar results too.
Microsoft’s Phi models were trained on fine-tuned “textbook-quality” data, says Mueller, which have a more consistent style that’s easier to learn from than the highly diverse text from across the Internet that LLMs typically rely on. Similarly, Apple trained its SLMs exclusively on richer and more complex datasets.
The rise of SLMs comes at a time when the performance gap between LLMs is quickly narrowing and tech companies look to deviate from standard scaling laws and explore other avenues for performance upgrades. At an event in April, OpenAI’s CEO Sam Altman said he believes we’re at the end of the era of giant models. “We’ll make them better in other ways.”
Because SLMs don’t consume nearly as much energy as LLMs, they can also run locally on devices like smartphones and laptops (instead of in the cloud) to preserve data privacy and personalize them to each person. In March, Google rolled out Gemini Nano to the company’s Pixel line of smartphones. The SLM can summarize audio recordings and produce smart replies to conversations without an Internet connection. Apple is expected to follow suit later this year.
More importantly, SLMs can democratize access to language models, says Mueller. So far, AI development has been concentrated into the hands of a couple of large companies that can afford to deploy high-end infrastructure, while other, smaller operations and labs have been forced to license them for hefty fees.
Since SLMs can be easily trained on more affordable hardware, says Mueller, they’re more accessible to those with modest resources and yet still capable enough for specific applications.
In addition, while researchers agree there’s still a lot of work ahead to overcome hallucinations, carefully curated SLMs bring them a step closer toward building responsible AI that is also interpretable, which would potentially allow researchers to debug specific LLM issues and fix them at the source.
For researchers like Alex Warstadt, a computer science researcher at ETH Zürich, SLMs could also offer new, fascinating insights into a longstanding scientific question: How children acquire their first language. Warstadt, alongside a group of researchers including Northeastern’s Mueller, organizes BabyLM, a challenge in which participants optimize language model training on small data.
Not only could SLMs potentially unlock new secrets of human cognition, but they also help improve generative AI. By the time a child turns 13, they’re exposed to about 100 million words and better than chatbots at language, with access to only 0.01 percent of the data. While no one knows what makes humans so much more efficient, says Warstadt, “reverse engineering efficient human-like learning at small scales could lead to huge improvements when scaled up to LLM scales.”
Enlarge/ Ford Mustang Mach E electric vehicles are offered for sale at a dealership on June 05, 2024 in Chicago, Illinois. (credit: Scott Olson / Getty Images)
CDK Global touts itself as an all-in-one software-as-a-service solution that is "trusted by nearly 15,000 dealer locations." One connection, over an always-on VPN to CDK's data centers, gives a dealership customer relationship management (CRM) software, financing, inventory, and more back-office tools.
That all-in-one nature explains why people trying to buy cars, and especially those trying to sell them, have had a rough couple of days. CDK's services have been down, due to what the firm describes as a "cyber incident." CDK shut down most of its systems Wednesday, June 19, then told dealerships that evening that it restored some services. CDK told dealers today, June 20, that it had "experienced an additional cyber incident late in the evening on June 19," and shut down systems again.
"At this time, we do not have an estimated time frame for resolution and therefore our dealers' systems will not be available at a minimum on Thursday, June 20th," CDK's told customers.
As the world’s 5G rollout continues with its predictable fits and starts, one unlikely 5G competitor is now coming into view: Wi-Fi. Private 5G networks—in which a person or company sets up their own facility-wide cellular network—are today finding applications where Wi-Fi (the IEEE 802.11 wireless connectivity standard) was once the only viable game in town. This month, the Newbury, England-based telecom company Vodafone is releasing a Raspberry Pi-based private 5G base station that it is now being aimed at developers, who might then jump-start a wave of private 5G innovation.
“The Raspberry Pi is the most affordable CPU[-based] computer that you can get,” says Santiago Tenorio, network architecture director at Vodafone. “Which means that what we build, in essence, has a similar bill of materials as a good quality Wi-Fi router.”
“In a Raspberry Pi—in this case, a Raspberry Pi 4 is what we used—then you can be sure you can run that anywhere, because it’s the tiniest processor that you can have,” Tenorio says.
Santiago Tenorio holds one of Lime Microsystems’ private 5G base station kits.Vodafone
Private 5G for drones and bakeries
There are a range of reasons, Tenorio says, why someone might want their own private 5G network. At the moment, the scenarios mostly concern companies and organizations—although individual expert users could still be drawn to, for instance, 5G’s relatively low latency and network flexibility.
Tenorio highlighted security and mobility as two big selling points for private 5G.
A commercial storefront business, for instance, might be attracted to the extra security protections that a SIM card can provide compared to password-based wireless network security. Because each SIM card contains its own unique identifier and encryption keys, thereby also enabling a network to be able to recognize and authorize each individual connection, Tenorio says private 5G network security is a considerable selling point.
Plus, Tenorio says, it’s simpler for customers to access. Envisioning a use case of a bakery with its own privately deployed 5G network, he says, “You don’t need a password. You don’t need a conversation [with a clerk behind a counter] or a QR code. You simply walk into the bakery, and you are into the bakery’s network.”
As to mobility, Tenorio suggests one emergency relief and rescue application that might rely on the presence of a nearby 5G station that causes devices in its range to ping.
Setting up a private 5G base station on a drone, Tenorio says, would enable that drone to fly over a disaster area and, via its airborne network, send a challenge signal to all devices in its coverage area to report in. Any device receiving that signal with a compatible SIM card then responds with its unique identification information.
“Then any phone would try to register,” Tenorio says. “And then you would know if there is someone.”
Not only that, but because the ping would be from a device with a SIM card, the private 5G rescue drone in the above scenario could potentially provide crucial information about each individual, just based on the device’s identifier alone. And that user-identifying feature of private 5G isn’t exactly irrelevant to a bakery owner—or to any other commercial customer—either, Tenorio says.
“If you are a bakery,” he says, “You could actually know who your customers are, because anyone walking into the bakery would register on your network and would leave their [International Mobile Subscriber Identity].”
Winning the lag race
According to Christian Wietfeld, professor of electrical engineering at the Technical University of Dortmund in Germany, private 5G networks also bring the possibility of less lag. His team has tested private 5G deployments—although, Wietfeld says, they haven’t yet tested the present Vodafone/Lime Microsystem base station—and have found private 5G to provide reliably better connectivity.
“The additional cost and effort to operate a private 5G network pays off in lower downtimes of production and less delays in delivery of goods,” Wietfeld says. “Also, for safety-critical use cases such as on-campus teleoperated driving, private 5G networks provide the necessary reliability and predictability of performance.”
For Lime Networks, according to the company’s CEO and founder Ebrahim Bushehri, the challenge comes in developing a private 5G base station that maximized versatility and openness to whatever kinds of applications developers could envision—while still being reasonably inexpensive and retaining a low-power envelope.
“The solution had to be ultra-portable and with an optional battery pack which could be mounted on drones and autonomous robots, for remote and tactical deployments, such as emergency response scenarios and temporary events,” Bushehri says.
Meanwhile, the crowdfunding behind the device’s rollout, via the website Crowd Supply, allows both companies to keep tabs on the kinds of applications the developer community is envisioning for this technology, Bushehri says.
“Crowdfunding,” Bushehri says, “Is one of the key indicators of community interest and engagement. Hence the reason for launching the campaign on Crowd Supply to get feedback from early adopters.”
Enlarge/ Ilya Sutskever physically gestures as OpenAI CEO Sam Altman looks on at Tel Aviv University on June 5, 2023. (credit: Getty Images)
On Wednesday, former OpenAI Chief Scientist Ilya Sutskever announced he is forming a new company called Safe Superintelligence, Inc. (SSI) with the goal of safely building "superintelligence," which is a hypothetical form of artificial intelligence that surpasses human intelligence, possibly in the extreme.
"We will pursue safe superintelligence in a straight shot, with one focus, one goal, and one product," wrote Sutskever on X. "We will do it through revolutionary breakthroughs produced by a small cracked team."
Sutskever was a founding member of OpenAI and formerly served as the company's chief scientist. Two others are joining Sutskever at SSI initially: Daniel Levy, who formerly headed the Optimization Team at OpenAI, and Daniel Gross, an AI investor who worked on machine learning projects at Apple between 2013 and 2017. The trio posted a statement on the company's new website.
Insects have long been an inspiration for robots. The insect world is full of things that are tiny, fully autonomous, highly mobile, energy efficient, multimodal, self-repairing, and I could go on and on but you get the idea—insects are both an inspiration and a source of frustration to roboticists because it’s so hard to get robots to have anywhere close to insect capability.
We’re definitely making progress, though. In a paper published last month in IEEE Robotics and Automation Letters, roboticists from Shanghai Jong Tong University demonstrated the most bug-like robotic bug I think I’ve ever seen.
Okay so it may not look the most bug-like, but it can do many very buggy bug things, including crawling, taking off horizontally, flying around (with six degrees of freedom control), hovering, landing, and self-righting if necessary. JT-fly weighs about 35 grams and has a wingspan of 33 centimeters, using four wings at once to fly at up to 5 meters per second and six legs to scurry at 0.3 m/s. Its 380 milliampere-hour battery powers it for an actually somewhat useful 8-ish minutes of flying and about 60 minutes of crawling.
While that amount of endurance may not sound like a lot, robots like these aren’t necessarily intended to be moving continuously. Rather, they move a little bit, find a nice safe perch, and then do some sensing or whatever until you ask them to move to a new spot. Ideally, most of that movement would be crawling, but having the option to fly makes JT-fly exponentially more useful.
Or, potentially more useful, because obviously this is still very much a research project. It does seem like there’s a bunch more optimization that could be done here; for example, JT-fly uses completely separate systems for flying and crawling, with two motors powering the legs and two additional motors powering the wings plus with two wing servos for control. There’s currently a limited amount of onboard autonomy, with an inertial measurement unit, barometer, and wireless communication, but otherwise not much in the way of useful payload.
Insects are both an inspiration and a source of frustration to roboticists because it’s so hard to get robots to have anywhere close to insect capability.
It won’t surprise you to learn that the researchers have disaster relief applications in mind for this robot, suggesting that “after natural disasters such as earthquakes and mudslides, roads and buildings will be severely damaged, and in these scenarios, JT-fly can rely on its flight ability to quickly deploy into the mission area.” One day, robots like these will actually be deployed for disaster relief, and although that day is not today, we’re just a little bit closer than we were before.
Intel’s use of dipole work-function materials means the gate surrounding each fin in a FinFET is thinner.Intel
Santiago Tenorio holds one of Lime Microsystems’ private 5G base station kits.Vodafone
