Specialist microchips and circuits that are able to handle signals at the cutting edge of wireless technology are needed for future wireless networks and high-resolution sensing.
Now, a team of researchers led by Kaushik Sengupta, a professor of electrical and computer engineering at Princeton University, has tasked AI with creating novel advanced wireless chips, with some surprising results.
These chips are traditionally difficult and expensive to design, as they require extensive engineering on a miniature scale.
Sengupta explains that the circuitry involved in advanced chips is so small and complex that there are more possible configurations than there are atoms in the universe.
The human brain is incapable of understanding such levels of complexity and so chips are built from the bottom up, adding to existing designs piece by piece until the desired effect is achieved.
This can be extremely painstaking and time-consuming with wireless chips incorporating numerous elements, including standard electronic circuits such as those found in computer chips alongside antennae, resonators, signal splitters, combiners and other electromagnetic structures.
AI takes a different approach to human chip design
Rather than building on existing designs and carefully changing components and arrangements one at a time, the AI approaches each chip as a unique whole, essentially designing it anew.
Sengupta said that this has led to strange, random-looking designs that seem unintuitive and were unlikely to be developed by a human mind.
Nevertheless, they frequently perform better than chips designed by humans, and some have produced results that were previously unobtainable.
The new approach could slash the time and costs involved in developing chips for new and existing wireless technologies, sometimes taking just minutes to synthesise a chip that would otherwise have taken weeks.
In other cases, the new chip designs would previously not have been possible, allowing them to be more energy efficient or making them operable across a huge range of frequencies.
The AI design methods could be deployed through NextG, Princeton’s industry partnership programme to develop next-generation communications, of which Sengupta is co-director.
The process isn’t flawless, however, and still requires human oversight.
The AI currently produces faulty arrangements alongside the more efficient ones and is prone to ‘hallucinating’ elements that do not work.
Sengupta said that the point is not to replace the human mind with AI but to “enhance productivity with new tools”.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
