AI Revolutionizes Wireless Chip Design, Creating Complex Circuits in Minutes

Using deep learning, scientists are breaking barriers in chip design, crafting intricate structures for next-gen wireless systems in record time.

An enlarged image of the chip’s circuitry in Sengupta’s lab at Princeton.

Specialized microchips that manage signals at the cutting edge of wireless technology are astounding works of miniaturization and engineering. However, they are also complicated and expensive to design.

Now, researchers at Princeton Engineering and the Indian Institute of Technology have harnessed artificial intelligence to take a key step toward slashing the time and cost of designing new wireless chips and discovering new functionalities to meet expanding demands for better wireless speed and performance. In an article published Dec. 30 in the journal Nature Communications, the researchers describe their methodology, in which an AI creates complicated electromagnetic structures and associated circuits in microchips based on the design parameters. What used to take weeks of highly skilled work can now be accomplished in hours.

Furthermore, the AI behind the new system has produced strange new designs featuring unusual circuitry patterns. Kaushik Sengupta, the lead researcher, said the designs were unintuitive and unlikely to be developed by a human mind, but they frequently offer marked improvements over even the best standard chips.

"We are coming up with structures that are complex and look random-shaped, and when connected with circuits, they create previously unachievable performance. Humans cannot really understand them, but they can work better," said Sengupta, a professor of electrical and computer engineering and co-director of NextG, Princeton's industry partnership program to develop next-generation communications.

These circuits can be engineered to operate more energy efficiently or to be operable across an enormous frequency range that is not currently possible. Furthermore, the method synthesizes inherently complex structures in minutes, while conventional algorithms may take weeks. The approach is also highly scalable and capable of being adapted to different chip designs, frequency ranges, and manufacturing process nodes, making it versatile for a wide range of applications. In some cases, the new methodology can create structures that are impossible to synthesize with current techniques.

Uday Khankhoje, a co-author and associate professor of electrical engineering at IIT Madras, said the new technique not only delivers efficiency but also promises to unlock new approaches to design challenges that have been beyond engineers' capabilities.

"This work presents a compelling vision of the future," he said. "AI powers not just the acceleration of time-consuming electromagnetic simulations but also enables exploration into a hitherto unexplored design space and delivers stunning high-performance devices that run counter to the usual rules of thumb and human intuition."

Wireless chips combine standard electronic circuits like those in computer chips with electromagnetic structures, including antennas, resonators, signal splitters, combiners, and others. These elements are combined in every circuit block, carefully handcrafted, and co-designed to operate optimally. This method is then scaled to other circuits, subsystems, and systems, making the design process highly complex and time-consuming, particularly for modern, high-performance chips behind applications like wireless communication, autonomous driving, radar, and gesture recognition.

"Classical designs carefully put these circuits and electromagnetic elements together, piece by piece, so that the signal flows in the way we want it to flow in the chip. By changing those structures, we incorporate new properties," Sengupta said. "Before, we had a finite way of doing this, but now the options are much larger."

It can be hard to comprehend the vastness of a wireless chip's design space. The circuitry in an advanced chip is so small, and the geometry so detailed, that the number of possible configurations for a chip exceeds the number of atoms in the universe, Sengupta said. There is no way for a person to understand that level of complexity, so human designers don't try. They build chips from the bottom up, adding components as needed and adjusting the design as they build.

Sengupta said AI approaches the challenge from a different perspective. It views the chip as a single artifact, which can lead to strange but effective arrangements. However, this approach requires robust experimental validation, which the researchers conducted to ensure the designs meet real-world performance expectations. He said humans play a critical role in the AI system, in part because AI can make faulty and inefficient arrangements. It is possible for AI to hallucinate elements that don't work, at least for now. This requires some level of human oversight.

"There are pitfalls that still require human designers to correct," Sengupta said. "The point is not to replace human designers with tools. The point is to enhance productivity with new tools. The human mind is best utilized to create or invent new things, and the more mundane, utilitarian work can be offloaded to these tools."

The researchers have used AI to discover and design complex electromagnetic structures, which are then co-designed with circuits to create broadband amplifiers. Sengupta said future research will involve linking multiple structures and designing entire wireless chips with the AI system. This work paves the way for automating the co-design of chip-scale systems, potentially achieving unprecedented performance and expanding the possibilities for future wireless applications.

"Now that this has shown promise, there is a larger effort to think about more complicated systems and designs," he said. "This is just the tip of the iceberg in terms of what the future holds for the field."

The article, Deep-learning Enabled Generalized Inverse Design of Multi-Port Radio-frequency and Sub-Terahertz Passives and Integrated Circuits, was published on Dec. 30, 2024, in the journal Nature Communications. Besides Sengupta, the authors included Emir Ali Karahan, the lead author and a graduate student at Princeton; Zheng Liu, Zijian Shao, and Jonathan Zhou of Princeton; Aggraj Gupta and Uday Khankhoje of the Indian Institute of Technology Madras. Support for the research was provided in part by the Air Force Office of Scientific Research, the Office of Naval Research, Princeton Research Computing, and the M. S. Chadha Center for Global India at Princeton University.