Alexander V. Boyko, Alexander V. Bakhshiev, Anton M. Korsakov

Peter the Great St. Petersburg Polytechnic University, Computer Vision Systems LLC, St. Petersburg, Russian state scientific center for robotics and technical cybernetics (RTC)

In this paper, a variant of the compartmental spiking neuron model hardware implementation on operational amplifiers is proposed. The relevance of the work is due to the growing need for both hardware implementations of neural network solutions in general and the need to develop adaptive abilities of networks, primarily to changing environmental conditions. One of the promising directions is the implementation of spike neural networks, in which the main functional element is not a neuron, but a compartment of the neuron membrane. The hardware implementation of such neuron models on a discrete element base should make it possible to facilitate experimental research in this area. The proposed solution is based on the CSNM neuron model. The paper considers the existing approaches to the hardware implementation of neuron models and selects the implementation approach on operational amplifiers. Schematics of each compartment of the implemented neuron model have been developed. Test experiments and comparison with a mathematical model were carried out, the results of which allowed us to state that the implementation reproduces the required time characteristics of signal conversion processes in a neuron quite accurately. The proposed implementation makes it possible to flexibly change the structure of the dendritic and synaptic apparatus of a neuron, and it is convenient to interpret signals for comparison with a mathematical model. The disadvantage of the proposed solution is low energy efficiency, however, for research purposes, this aspect is not critical at this stage.

spike neural networks, deep learning, neuromorphic systems, spike neuron, compartmental neuron model, machine learning, operational amplifier, hardware implementation