Designing wireless microchips, essential for cutting-edge technologies like faster internet and autonomous driving, has always been a complex and expensive task.
But now, artificial intelligence (AI) is revolutionizing the process, making it faster, cheaper, and capable of discovering entirely new functionalities.
Researchers at Princeton University and the Indian Institute of Technology (IIT) have developed an AI system that designs advanced wireless chips in just hours—work that would otherwise take skilled engineers weeks to complete.
Their findings, published in Nature Communications, highlight how this AI system creates not only faster and more energy-efficient chips but also strange, unconventional designs that deliver unprecedented performance.
Professor Kaushik Sengupta, who led the research, explains that the AI-generated designs are unintuitive and would be unlikely to emerge from traditional human approaches.
“The structures are complex and look random,” Sengupta said, “but when connected with circuits, they achieve performance levels that were previously impossible.”
Wireless chips combine standard electronic components, like those found in computers, with specialized electromagnetic structures such as antennas and signal splitters. Engineers traditionally assemble these elements step by step, a process that is labor-intensive and limited by human intuition.
The new AI system changes the game by analyzing the chip as a whole rather than as individual parts. This approach opens up a vast design space, offering possibilities beyond what engineers could imagine. The AI creates intricate patterns for circuits and electromagnetic elements that maximize performance, often defying standard design rules.
One major achievement is the creation of broadband amplifiers, which are critical for faster, more reliable wireless communication. By co-designing the electromagnetic and circuit elements, the AI produces chips that are more efficient, can operate across broader frequency ranges, and deliver higher performance than traditional methods.
The geometry of advanced wireless chips is so detailed that the number of possible designs exceeds the number of atoms in the universe, Sengupta said. No human could explore such vast possibilities, but AI can quickly test different configurations to find the most effective solutions.
Interestingly, some of the AI-generated designs feature unusual shapes that don’t follow traditional engineering logic. “Humans cannot really understand them,” Sengupta said, “but they work better.”
The AI system also speeds up the design process significantly. Structures that take conventional algorithms weeks to create can now be generated in minutes. This efficiency could accelerate the development of next-generation technologies, including 5G, radar, and autonomous driving systems.
Despite the AI’s impressive capabilities, it isn’t perfect. Sometimes, the AI produces faulty designs or “hallucinates” elements that don’t work. Human engineers still need to review and refine the AI’s outputs to ensure the chips function correctly.
“The goal isn’t to replace human designers but to enhance their productivity,” Sengupta said. “AI can handle repetitive tasks, allowing humans to focus on creative, innovative work.”
The researchers plan to expand their work by designing entire wireless chips with the AI system. Future efforts will focus on creating more complex systems and linking multiple AI-designed structures to push the boundaries of wireless technology.
“This is just the beginning,” Sengupta said. “We’ve only scratched the surface of what AI can do for wireless chip design. The possibilities are limitless.”
With this breakthrough, AI is set to transform how microchips are designed, unlocking faster, cheaper, and more powerful wireless technologies for the future.
