Automation of RF Chip Design AI-Supported Development of Advanced Wireless Chips

Related Vendor

Wuxi InfiMotion Technology Co., Ltd.

A US consortium around Princeton University is relying on artificial intelligence to develop RF chips for wireless communication faster, cheaper, and smarter.

The development of high-frequency chips for wireless applications is on the verge of a fundamental transformation: a research consortium led by Princeton University is working on AI-supported design methods that aim to massively reduce time, costs, and demand for skilled workers. The project is funded with nearly 10 million US dollars by the National Semiconductor Technology Center (NSTC), a consortium of state and industry led by Natcast.

The goal of the project is to automate the previously highly manual development of RFICs (Radio Frequency Integrated Circuits)—the chips that enable communication in devices for 5G/6G, satellite communication, autonomous driving, and connected medical technology. Unlike digital circuits, whose design processes are largely automated, RF chips have so far been largely "handcrafted."

The development of these specialized wireless chips is extremely expensive and requires special skills, explains project leader Prof. Kaushik Sengupta, Professor of Electrical and Computer Engineering at Princeton. The development of new generations of semiconductors is primarily based on intuition and experience. This is complicated by physical conditions arising from both the advancing miniaturization and the required properties of radio technology—such as overlays, nonlinear effects, or environmental disturbances. As a result, the development of new chips in this area is particularly challenging, complex, and costly.

Artificial intelligence could help here. "When you reach a point where the manual, labor-intensive aspects of design can be automated and you can discover new architectures or new functions, a great opportunity presents itself," says Sengupta. Not only could this drastically accelerate routine processes, but by freeing up resources and offering new perspectives, it could also highlight new ways that are functionally superior to previous methods.

The methods used combine, among other things, reinforcement learning—known from strategic AI applications like Go—with diffusion models, which recently caused a stir in chemistry with the development of designer proteins. These methods are intended to discover novel architectures that leave traditional design rules behind.

As early as 2022, Sengupta's doctoral students Karahan and Liu presented an AI-optimized design at the IEEE International Microwave Symposium— and promptly won the top prize. A year later, a Best Paper Award followed in the IEEE Journal of Solid-State Circuits. These successes were reason enough for Natcast to select Princeton as one of three leading teams for the AIDRFIC program.

In addition to Princeton, the University of Southern California, Drexel University, Northeastern University, as well as industry giants like Keysight, RTX, Cadence, and Qualcomm are involved. The latter also participate in an advisory capacity, as do Texas Instruments, Nokia Bell Labs, and Ericsson.

The long-term vision: A new toolchain for automated RF design that unleashes creativity, shortens development times, and empowers new market participants. Especially in the context of a growing skills shortage in RF development, this could be a decisive factor for the industry's future viability.

"We are on the brink of a paradigm shift," says Sengupta. "If we let AI think backwards from the problem, instead of using classic top-down methods, possibilities open up that were previously beyond our imagination." (sg)

Secure and Compliant Authentication in Laboratories

Secure & Contactless User Authentication at GMP/GLP