Neural network chip uses memristors

Ars Technica : The strengthening of synaptic connections in our brains underlies memory, cognition, and other mental processes. A similar strengthening occurs in circuit elements known as memristors, whose electrical resistance is a product of the currents that have flowed through them in the past: The more current that flows through now, the easier currents will flow in the future. Exploiting the similarity to make memristor-based chips that mimic neurons has been a challenge because the best memristors are made from materials that are incompatible with silicon-based chip technology. Now researchers from Stony Brook University and the University of California, Santa Barbara, have solved that problem by identifying just the right mix of aluminum oxide and titanium oxide for making chip-compatible memristors. As the team reports in this week’s Nature, its 3- by 3-pixel memristor-based chip can be programmed to identify letters of the alphabet with increasing accuracy.