Исследование декомпозиции нейронной сети в системе схемотехнического моделирования Proteus

Abstract

The division of a monolithic neural network into blocks with their implementation on programmable logic within the framework of the Fog computing concept is considered. It is assumed that considering possible reconfiguration the implementation of blocks is performed on programmable logic: field-programmable gate array, FPGA (complex programmable logic device, CPLD), System-on-a-Chip, SoC or System-in-Package, SiP. The article explores such an implementation in the Proteus Design Suite based on ATMega32 microcontrollers. Modeling confirms the efficiency of the developed decomposition method. The research was carried out under the RFBR grant 20-37-90036 (Method of synthesizing neural network recognition devices for implementing the Fog computing mode).