Silicon brain: Pattern recognition and neurocomputing

For decades, attempts have been made in the fields of artificial intelligence and computer science to reverse engineer the brain and the neural processes of memory, learning and adaptation. The continued interest in building a physical system that mimics the brain is obvious: even the most advanced computers today cannot perform what a brain does. For instance, the IBM Watson supercomputer is composed of a cluster of 90 servers with a total of 2,880 processors and a total of 6 terabytes of RAM. Still, it is outperformed in many applications by a human brain with an average weight of 3 pounds, consuming less than 20 watts. These computers cannot “sense, perceive, interact and organize” as the neurophysiological human brains do, because they are limited in two fundamental elements: complexity and intelligence.

We are creating building blocks of a silicon brain or a “neurocomputer” from micromechanical oscillator elements. A network of coupled micromechanical oscillators can function as a neurocomputer that possesses oscillatory autocorrelative associative memory. This networked system will allow us to store, recognize, and retrieve complex visual patterns through the corresponding synchronized network states. The neural computing paradigm has been long inspired by the understanding that synchronization of oscillations in the brain is somehow related to the associative memory and learning functions. Our experiments show that the smallest unit of a network, a coupled two-oscillator system, demonstrates all the standard hallmarks of synchronization.