Noisy In-Memory Recursive Computation with Memristor Crossbars

This paper considers iterative dot-product computation implemented on in-memory memristor crossbar substrates. To address the case where true memristor conductance values may differ from their target values, it introduces a theoretical framework that characterizes the effect of conductance value variations on the final computation. For simple dot-products, the final computation error can be approximated by a Gaussian distribution; the mean and variance values of the corresponding Gaussian distribution are provided. For iterative dot-product computation, recursive expressions are derived for the means and variances of the successive computation outputs. Experiments verify the accuracy of the proposed analysis on both synthetic data and on images processed with memristor-based principal component analysis.