IBM claims breakthrough in cognitive semiconductors

Scientists at IBM said that the company has designed  computing cores that to some extent mimic digital neurons and synapses on silicon.

It claimed that the cores successfully mimic parallel processing in human brains without set programming.

The so-called neurosynaptic chips use algorithms and silicon circuitry to produce the equivalent of spiking neurons and synapses in biological systems and has already made two chips that are being tested right now.

IBM said that systems using such chips will learn through events, correlate information, and think, remember and learn the outcomes.

The chips were frabricated in 45 nanometre SOI-CMOS and contain 256 neurons, 262,144 programmable synapses in one core and 65,536 learning synapses in another core.  Big Blue has already demonstrated their use for navigation, vision, pattern recognition, memory by association and classification.

It aims to produce a chip architecture with 10 billion neurons and a hundred trillion synapses, all within a kilowatt of power and taking less than two litres in volume.

IBM is collaborating with a number of universities including Columbia, Cornell, UC Merced and the University of Wisconsin.  The chips were made at its fab in Fishkill, New York State.