The AI transforms wireless chip design, reducing time and cost while creating complex circuits with new capabilities for faster, high-performance applications.
Researchers at Princeton Engineering and the Indian Institute of Technology have used artificial intelligence to significantly reduce the time and expense of creating new wireless chips while unlocking new capabilities to meet growing demands for better speed and performance.
In the team’s approach, AI generates complex microchip electromagnetic structures and circuits based on specified design parameters. Tasks that once required weeks of skilled labour can now be completed in hours. The AI has also produced unconventional designs with unexpected circuit patterns.
These circuits can be designed for greater energy efficiency or to operate across a much wider frequency range than is currently achievable. The new method can also generate complex structures in minutes, while traditional algorithms may require weeks. Sometimes, it can create designs that current techniques cannot produce.
This process is extended to other circuits, subsystems, and systems, making the design highly complex and time-consuming, especially for modern, high-performance chips in applications like wireless communication, autonomous driving, radar, and gesture recognition.
The scale of a wireless chip’s design can be challenging to grasp. The circuitry is so tiny, and the geometry so intricate, that the number of possible configurations for a chip surpasses the number of atoms in the universe, according to Sengupta. This complexity is beyond human comprehension, so designers typically build chips from the ground up, adding components and adjusting the design.
The AI tackles the challenge differently, viewing the chip as a unified entity. This approach can result in unusual but practical designs. Sengupta explained that humans remain essential in the process because AI can sometimes generate flawed configurations in addition to efficient ones. For now, AI might create elements that don’t function as intended, which requires human oversight.
The researchers have used AI to identify and design intricate electromagnetic structures co-designed with circuits to create broadband amplifiers. Sengupta mentioned that future research will focus on connecting multiple structures and using the AI system to design entire wireless chips.
Reference: Emir Ali Karahan et al, Deep-learning enabled generalized inverse design of multi-port radio-frequency and sub-terahertz passives and integrated circuits, Nature Communications (2024). DOI: 10.1038/s41467-024-54178-1
