Princeton University will lead a joint government-industry effort using AI to develop advanced semiconductors for wireless communication and remote sensing. The chips are critical for next-generation wireless networks, satellite communication, self-driving cars and smart health care technologies.
Kaushik Sengupta, professor of electrical and computer engineering at Princeton, will direct the effort. His team will focus on automating the design of microchips for radio-frequency wireless communication, the technology that enables electronic devices to communicate with one another and with the physical world.
A grant of nearly $10 million to fund the work was announced May 28 by the National Semiconductor Technology Center, a public-private consortium managed by the National Center for the Advancement of Semiconductor Technology, or Natcast.
“Fundamental research leading to technology breakthroughs has fueled U.S. innovation for decades,” said Andrea Goldsmith, dean of Princeton’s School of Engineering and Applied Science. “The visionary research proposed by our Princeton team will transform technology in wireless communications and sensing. This project is an outstanding investment by Natcast to ensure U.S. leadership in these important fields.” Goldsmith has served on a number of corporate boards and co-founded two wireless communication companies.
“Embracing AI for radio frequency design is paramount for maintaining the United States’ leadership in technological innovation,” Natcast CEO Deirdre Hanford said in a news release. “Leveraging AI not only accelerates our research capabilities but also ensures the U.S. remains at the cutting edge of communication infrastructure.”
Everything from laptops to car sensors to satellites rely on the transmission of high-speed, low-latency and low-power signals. The chips that handle these signals must keep pace with soaring demand for more data at faster speeds. But designing these specialized wireless chips is extremely expensive and relies on rarified skillsets, according to Sengupta.
“They are fundamentally handcrafted,” Sengupta said. “But if you could get to a point where the manual labor-intensive aspects of design can be automated out and you can start discovering new architectures or new functionality, there lies a window of opportunity.”
While the chips powering today’s computers and data centers benefit from high degrees of automated design, wireless chips currently do not. That’s largely because wireless chips must deal with overlapping forces and uncontrolled environments. Each design stage requires intense involvement from experts across many domains.
Sengupta said this complexity drives long lead times and high costs. It also restricts creative problem solving. “You’re sort of limited to the human imagination,” he said. “It’s a very bottom-up approach.”
That approach typically starts with functional knowledge about circuits and tries to nudge designs toward modern demands. But by leveraging artificial intelligence, Sengupta’s research group has flipped convention on its head, starting from the demands and working backwards to find optimal circuit designs.
Their AI-assisted architectures often arrange components in ways that defy intuition but outperform traditional chips. Graduate students Emir Ali Karahan and Zheng Liu, advised by Sengupta, presented their work at the 2022 IEEE International Microwave Symposium and caught the wireless community’s attention, winning the symposium’s top award. That was a key moment for the Princeton team, Sengupta said, and helped position them as leaders in the space today. A related paper won the 2023 Best Paper Award from IEEE Journal of Solid State Circuits. And more recently, they have continued to push the bounds of what this approach can accomplish.
Mengdi Wang, associate professor of electrical and computer engineering at Princeton, will contribute to the team’s artificial intelligence and machine learning research. She said two main techniques will drive their automation efforts: reinforcement learning, famously good at creating AI that dominates games of strategy, such as Go; and RFdiffusion models, which enabled the Nobel Prize-winning chemistry of designer proteins.
The team will also include experts from the University of Southern California, Drexel University, Northeastern University, and industry partners at RTX, Keysight and Cadence. Senior leadership from Qualcomm, Skyworks, Texas Instruments, Nokia Bell Labs, Ericsson and Maury Microwave will form the advisory board.
The project will aim to harness Sengupta’s approach and develop automated design tools to slash costs, spur creativity and increase competition. The Princeton-led team was one of three teams selected by Natcast for this round of funding; the other two will be led by Keysight Technologies and The University of Texas at Austin.
