NEUROMORPHIC SYSTEM UTILIZING THE INTRINSIC CHARACTERISTICS OF MEMORY CELLS

摘要

The subject of the invention is a neuromorphic system comprising a set of at least one input neuron (503, 504, 507, 508), a set of at least one output neuron (505, 506) and a network of synaptic (502) implemented by a set of at least one variable resistance memory cell. Said synaptic network connects the input neurons (503, 504, 507, 508) to the output neurons (505, 506), the resistance of the memory cells being adjusted by sending writing pulses into the synaptic network ( 530, 531, 532, 533) generated by the input neurons and return pulses (514, 515) generated by the output neurons. The characteristics of the write and return pulses are deduced from the intrinsic characteristics of the memory cells so that the combination of a write pulse (530, 531, 532, 533) with a return pulse (514, 515 ) at the level of a memory cell results in a modification of its resistance according to a previously chosen learning rule.