Next-gen automotive semiconductors are critical for level 5 driverless cars
Michigan Engineers are leading hardware efforts to enable the AI required for full autonomy.
5 minutes
In “Autonomy from the inside out,” Michigan Engineer magazine delves into these innovations, why they’re important, and the limitations of today’s technology.
The driverless robotaxis chauffeuring around San Francisco, and advanced driver assistance features on more than half of new vehicles sold this year show just how far autonomous vehicle technology has come in the past decade. But to fulfill the promise of fully autonomous cars that drive themselves anywhere under any conditions, experts say we need a new generation of semiconductors.
“Cars have been seen as computers on wheels for a while, but to achieve full autonomy, they need to be more like traveling data centers,” said Valeria Bertacco, the Mary Lou Dorf Collegiate Professor of Computer Science and Engineering. “To make that leap, the auto industry is going to need new materials, architectures, systems and manufacturing processes for chips that are faster, cheaper, lower-power and more durable.”
University of Michigan researchers are working with global industry leaders to reimage AV computing systems and innovate big on their tiniest components. The effort is supported by $10 million from the state of Michigan to the Michigan Semiconductor Talent and Technology for Automotive Research (mstar) initiative, which also includes imec, KLA, the Michigan Economic Development Corporation, Washtenaw Community College and General Motors.
Behind the Data:
Automation boosts safety
Automatic emergency braking systems cut rear-end collisions by 50%. Waymo’s robotaxis have been involved in 81% fewer injury-causing crashes than an average human driver on the same routes in San Francisco and Phoenix, peer-reviewed research has shown.
Automated vehicles (AVs) require immense computational power that grows exponentially with each level of autonomy. With each new generation, the capabilities of sensors will grow, leading to even more data. “This is a constrained, embedded system,” said Reetuparna Das, associate professor of computer science and engineering. “The need for more efficient AI hardware is very well understood. It’s a billion-dollar market.”
Shrinking transistors have led to the “exponential growth in computing power in our lifetimes,” said Becky Peterson, associate professor of electrical and computer engineering and director of the U-M Lurie Nanofabrication Facility. But they’re reaching their size limits. The technology that got the industry where it is today is unlikely to get it to full autonomy.
Last year, semiconductor industry representatives toured the Mcity Test Facility and looked to the horizon with many of the solutions being explored at U-M. “It’s great to see people coming together to address these aspects,” Michael Sun, who leads the automotive business development unit at Taiwan-based TSMC, said. “I’ve been working in this area for a long time, and I think there’s a lot of momentum right now.”
Solutions from Michigan Engineering and mstar
The deep learning network models in today’s AVs loosely mimic the structure of a biological brain, but they’re bombarded with a continuous stream of unfiltered information, requiring a gargantuan amount of processing power. U-M researchers are developing a new approach based on a tungsten oxide memristor they’ve been advancing for at least a decade. It detects contrast, motion and sudden events, much like our brain and eyes work together to filter and focus our attention. They’re testing a more efficient processor called a neuromorphic chip and a different type of algorithm called a spiking neural network. “Neuromorphic sensors don’t capture frames like conventional cameras do. Instead, they detect change in each pixel independently,” said Wei Lu, the James R. Mellor Professor of Engineering in mechanical engineering.
While traditional “system on a chip” architecture involves printing all the components on a single piece of silicon, AVs’ computational needs push its limit in terms of physical size and complexity. The chiplet approach involves smaller, modular components that could be mixed and matched on a circuit board to build more tailored, durable systems. U-M researchers are developing a more robust chiplet communication protocol that can operate for years in a moving vehicle. “Chiplets would end up having to do a lot of talking to each other in vehicles because they’re solving big problems,” said Mike Flynn, the Fawwaz T. Ulaby Collegiate Professor of Electrical and Computer Engineering. “Not only does their communication need to be robust, it also needs to be efficient in terms of energy and bandwidth. We’re trying to make it high-speed and low-power too.”
“You could probably get at least 1,000 times the energy performance from the materials that we investigate and design,” said John Heron, an associate professor of materials science and engineering.
His focus is on a special class of materials with unique—and linked—magnetic and electronic properties that integrate computation and memory into the same device. Magnetoelectric materials could take advantage of quantum effects to store and process data.
Because it’s still important for now that innovations are compatible with current manufacturing, several U-M researchers are working to stack materials on top of silicon. Peterson is adding transistors made with an ultra-thin layer such as zinc tin oxide on top of the first silicon layer to achieve very low power, high performance processors and memory. Ageeth Bol, professor of chemistry and materials science and engineering, is going vertical with 2D slivers of molybdenum disulfide.