Effect of Mismatch on the Reliability of ON/OFF-Programmable CNNs

In this paper, the reliability of computation of ON/OFF-programmable cellular nonlinear networks (CNNs) is analyzed. This paper redefines the classical concept of robustness (tolerance to physical imperfections) in ON/OFF-programmable CNNs so that it can be used to predict the computation reliability of a physical realization. Also, a systematic approach to improve robustness is described, and guidelines for optimizing reliability with the given resources are given. Random mismatch is identified as the main source of errors in a physical realization. At the presence of random mismatch, we describe the probability of a cell producing an error with specific inputs and the error probability when processing random data. In addition, we show how the error probability could be defined in case the cell needs to work correctly with all possible input combinations.