Courtesy of Dmitri Strukov, UCSB
Electronic components called memristors have enabled a simple computing circuit to learn to perform a task from experience. After processing data during a training phase, the device classified 3-by-3-pixel images as one of three letters of the alphabet, researchers report in the May 7 Nature. The study is a step toward building computers that operate more like the human brain.
A memristor is a circuit element whose resistance depends on the past electrical pulses that surged through it. That’s similar to the way synapses work in the brain. Electrical engineer Dmitri Strukov at the University of California, Santa Barbara and colleagues created an image-classifying circuit made of 24 crisscrossing wires, with memristors slotted into each intersection point.
Future computers packed with memristors and transistors could excel at tasks such as pattern recognition that human brains tackle far more easily than today’s computers.