Princeton University
Gear-obsessed editors choose every product we review. We may earn commission if you buy from a link. Why Trust Us?
- A new neural network process has designed wireless chips that can outperform existing ones.
- This convolutional neural network analyzes the desired chip properties then designs backward.
- Much of AI news is hype, but this is open access, peer reviewed research in a reputable journal.
Our world runs on computer chips. From the chips that run new cars to the chips that help your phones and computers process information to the microchips that help track lost animals, there’s very few aspects of modern human life that are not touched by chips. As a result, there’s a huge and consistent push to make better and more innovative chips as fast as possible—sometimes by any means necessary. And apparently, that sometimes means taking a portion of the design control out of human hands.
A group of scientists recently explained their process of letting artificial intelligence technology (AI) design and test a more efficient computer chip. The lead author—electrical engineer Kaushik Sengupta of Princeton University’s Sengupta Lab—was recently awarded an IEEE fellowship for his wireless chip and network research. In publicity around this new paper, he has been careful to explain that the goal is to supplement human productivity rather than replace it.
Related Story • Experts Simulated 500 Million Years of Evolution
- • Experts Simulated 500 Million Years of Evolution
By publishing open access, peer reviewed research (in the multidisciplinary journal Nature Communications) instead of silo-ing his findings into a startup’s proprietary lockbox, Sengupta is helping to advance interesting uses of AI technologies like his team’s convolutional neural networks (CNNs). At the same time, there are strong limitations to even groundbreaking uses of AI—in this case, the research team is candid about the fact that human engineers can’t and may never fully understand how these chip designs work. If people can’t understand the chips in order to repair them, they may be... well... disposable.
To build a new kind of chip, the researchers began with a philosophy called bottom up design, or inverse design. In hardware engineering, inverse design starts with the nitty gritty details and desired outcomes of a piece of hardware. From there, researchers work backward in order to package all the resulting pieces into a functional piece of technology.
Computer algorithms don’t require the same linearity or structure that the human brain usually does, so deciding the order or shape of chip components doesn’t matter to AI the same way it does with human engineers. In fact, the researchers say, our reliance on existing templates or form factors for chip design is quite limiting. And even with these limiting tools, engineers need years of training and experience to even begin trying to improve wireless chip design. The right algorithm, they say, could suggest new paradigms in a matter of minutes. From there, engineers could use these paradigms as innovativ starting points for their own ideas.
Sengupta’s team used a convolutional neural network (CNN). “Convoluted” has come to mean overly complicated, but it originally simply meant for something to be folded, twisted, and overlapping. CNN’s name hints at how the program outperforms the human mind in certain specific tasks: at every step, the computer can consider moves that are more foldy and twisty than a human brain can. Our chip designs look very orderly, but the CNN’s designs look chaotic and blobby.
“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. Before, we had a finite way of doing this, but now the options are much larger,” Sengupta said in a Princeton University statement.
Related Story • Could AI Enable Would-Be Bioterrorists?
- • Could AI Enable Would-Be Bioterrorists?
And, he explained, even the best neural network still has problems like hallucinations—when an AI asserts something that is not based in truth or reality. It may also suggest things that are not possible in real life, for some reason. “There are pitfalls that still require human designers to correct,” Sengupta said in the statement. “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.”
Human designers may simply want to choose designs that are more efficient yet still graspable for the human mind. Based on Sengupta’s comments and the open nature of this research, he imagines his CNN process helping engineers have “aha!” moments as they work on new designs. Joining the two skillsets could lead to breakthroughs that can still be understood and, more importantly, retooled or repaired if need be.
Because of its transparency, this research will join the scientific discourse in a way many AI technologies never do. That, alone, is a major benefit. Time will tell if these more inscrutable chip designs end up powering the wireless networks that blanket our world.
Caroline Delbert
Caroline Delbert is a writer, avid reader, and contributing editor at Pop Mech. She's also an enthusiast of just about everything. Her favorite topics include nuclear energy, cosmology, math of everyday things, and the philosophy of it all.
Science
DeepSeek: Open Weight vs. Open Source
Aliens From a Parallel Universe May Be Around Us
Breathing Like This Can Alter Your Consciousness
The Arctic Isn’t the Carbon Sink It Used to
The World’s Oceans Are Warming Up Way Too Fast
This Bacteria Loves To Dine on ‘Forever Chemicals’
Early Dinosaur Fossils Are Likely Inaccessible
Humans Moved to Europe Earlier Than We Thought
