Caltech researchers have demonstrated a complex integrated circuit that survives substantial damage by reconfiguring the way it processes information.
The chip does not physically repair flaws; it uses a second processor to come up with new ways to perform a task in spite of the damage. The chip can also be programmed to prioritize energy savings or speed. Ali Hajimiri, the Caltech professor of electrical engineering who led the work, says chips that tune their own performance on the fly could also perform better under ordinary circumstances.
Self-healing circuits could be resilient to manufacturing flaws, and they could withstand damage caused by high temperatures or the deterioration that comes with aging. That could mean more robust military communications equipment and portable consumer electronics that can take a beating.
Hajimiri’s group is the first to demonstrate this kind of capability in a complex integrated circuit—in this case a power amplifier, a type of circuit that processes signal transmission in cell phones and other telecommunication devices. The self-healing chip consists of 100,000 transistors, several types of sensors, and an additional embedded processor that monitors the circuit’s performance and runs algorithms to assess how it can be improved.
In work published this month in the journal IEEE Transactions on Microwave Theory and Techniques, the Caltech group showed that circuits equipped with the self-healing system continue to work even after the circuit has been repeatedly blasted with a laser to knock out about half the transistors. It takes just tens of milliseconds to adjust to the damage. A circuit that wasn’t subjected to this attack was able to consume 50 percent less power than an ordinary circuit by reconfiguring itself for maximum efficiency.
The secondary processor that makes these results possible monitors the circuit by running a program that analyzes sensor data about temperature, voltage, current, power, and more. It can be programmed to optimize these parameters for a particular outcome—for example, to maximize the purity or power of the signal produced by the amplifier. The program then figures out how to change the circuit to best achieve that goal. It is possible to change the voltage applied to particular transistors in the circuit, or to change the way signals are routed through it so as to avoid a damaged area. Hajimiri says the circuit has about 250,000 possible states.
Hajimiri says it should be possible to apply this concept to any kind of circuit, no matter the function. In the power amplifier demonstration, the self-healing system doesn’t take up any extra area because the secondary processor is positioned underneath.
The concept could free chip designers from having to make sure that circuits can withstand rare events like temperature extremes, voltage fluctuations, or interference. The ability to do so usually comes at a cost of performance.
“You can design a chip that will run in these worst-case scenarios, but most of the time it’s not the worst case, and you could be running faster or with less power most of the time,” says Subhasish Mitra, a professor of computer science at Stanford University, who was not involved with the work. As silicon transistors are ever more aggressively miniaturized, says Mitra, manufacturers will need circuit designers to provide more reliability.
“Until recently, the economics discouraged this kind of design,” says Thomas H. Lee, who heads the Stanford Microwave Integrated Circuits Laboratory. “But it’s getting a lot harder to do a good job of manufacturing chips, and I think embedded repair systems will become common.”
These weird virtual creatures evolve their bodies to solve problems
They show how intelligence and body plans are closely linked—and could unlock AI for robots.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
Chinese hackers disguised themselves as Iran to target Israel
But they left a few clues that gave them away.
DeepMind says it will release the structure of every protein known to science
The company has already used its protein-folding AI, AlphaFold, to generate structures for the human proteome, as well as yeast, fruit flies, mice, and more.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.