Design and evaluation of single-electron associative memory circuit

Abstract

Recently, neural networks (NN) have attracted much attention because they have many advantages such as parallel-processing and noise-harnessing functions. An NN is composed of many neurons, which operate in parallel and each one operates stochastically. Although each neuron is sensitive to noise and behaves randomly, a network based on the neurons can operate with high reliability. Moreover, it is known that the network exhibits various complicated functions. In this study, we attempt to mimic such useful functions of the NN by nanoelectronic circuits – i.e. single-electron (SE) circuits. The SE circuit also operates stochastically, like a biological neuron, because of the probabilistic electron tunneling phenomenon. Therefore, it can be used for implementing the stochastic neuron operation. In this work, we propose a method for hardware NN implementation on the SE circuit and confirm that it can perform the associative memory function by computer simulation. First, we simulate the circuit operation without noise. This simulation demonstrates that our circuit can store some patterns and recall them from a given noisy or related pattern. However, it fails to recall some input patterns. In the next step, we simulate thermal noise in the circuit. As a result, we can improve the failure rate in recall and increase circuit performance.