Simulating Self-Recovering Electric Circuits Using Neural Networks

Abstract

Neural circuits and their applications for retrainable circuits have created new avenues in circuit design and fault correction. Electric systems and electronic circuits can experience a number of faults due to faulty connection of wires, misfiring of switches, wear-out of certain components and so on. Unlike a conventional circuit, a neural circuit is not affected by a single-component failure and thus can self-correct the fault. Once a fault occurs, the circuit can be retrained almost instantly to obtain the required output with the remaining healthy circuit once again. In fact, the fidelity is even better sometimes as overfitting is reduced. In this paper, a filter and an inverter were subjected to faults with more than half the transistors removed. The simulated circuits retrained in seconds to obtain the desired characteristics with almost the same accuracy. This paper delineates the various aspects of such a self-recovering electric system and the magnitude of the fault that can be recovered from.