Computer simulation of the biomorphic neuroprocessor

In this paper, numerical simulation was carried out and the operability of the input and output devices of a biomorphic neuroprocessor, built using a logic matrix, was shown using the specialized software SPICE (Simulation Program with Integrated Circuit Emphasis) for simulating large electrical circuits with memristor-diode crossbars. The modes of encoding a binary number into frequency and delay of pulses, simultaneously into frequency and delay of pulses by neuron population were investigated. Also, the mode of decoding pulses into a standard binary code for outputting information from the neuroprocessor was investigated. The simulation results of the associative self-learning mechanism in hardware spiking neural networks based on the memory matrix of a neuroprocessor are presented. The learning rules are LTP (Long-Term Potentiation) for a network of four neurons and STDP (Spike-Time Dependent Plasticity) for a hardware spiking perceptron. For the first time, the generation of a new association (new knowledge) in a composite memristor-diode crossbar according to the STDP rule is demonstrated, in opposite to self-learning in existing hardware neural networks with discrete memristors synapses.