Our technologies are becoming more biological with each passing year. And here's the latest breakthrough: Caltech engineers have developed an integrated computer chip that can learn to heal its own injuries.
Typically, microchips feature pathways so specialized that a single fault can render the entire thing inoperative. So, in an effort to overcome this lack of resilience, the High-Speed Integrated Circuits laboratory developed a chip which contains more than 100,000 transistors which don’t function all at once.
Scientific American explains:
Rather, the researchers burned vast swaths of transistors out of the chip with a laser, then allowed the systems to recalibrate. As long as the blast did not catch any data caches in its crosshairs, the chip could seek out alternate routes and continue to function. With the help of an application-specific integrated-circuit (ASIC) processor on each chip, the system could "learn" which pathways were broken and adjust accordingly.
If a traditional microchip is comparable to an electric circuit (remove one piece and the entire system collapses), this newtechnology is more similar to a human brain. If one pathway becomes inaccessible, the brain will discover novel ways to relay information. Of course, it is possible to inflict catastrophic damage on a system (be it brain or microchip) from which it cannot recover, but with more than 100,000 methods of delivery, these microchips could prove to be extremely robust.
The self-healing chips are an intriguing step in machine evolution, but they do lack one crucial feature of actual living things: the ability to regenerate over time. While the Caltech microchips can withstand extensive damage and figure out ways to work around it, a section fried by lasers will still be fried years later. Unlike biological tissue, which repairs itself over time, each chip has a limited shelf life.
Interestingly, the work was funded by DARPA. As Jesse Emspak noted in Discovery, “It’s pretty easy to see why the military would want this kind of technology — of course, when the robot rebellion happens and you can’t kill the Roomba, we’ll only have ourselves to blame.”
The team's results appear in the March issue of IEEE Transactions on Microwave Theory and Techniques.
Image: Caltech.