EQUIPMENT AND METHODS FOR SIGNAL PROCESSING IN NEURAL NETWORKS

Abstract

Several aspects of the invention relate to a device and method for processing signals in neural networks. The device typically comprises a computational circuit configured to perform convolution operations, wherein the computational circuit has multiple input rows, and an activator buffer with multiple buffer segments coupled in turn to multiple input rows of the computational circuit. In some respects, each of the many buffer segments comprises a first multiplexer with multiple multiplexer inputs, and each of the many multiplexer inputs of one of the first multiplexers on one of the many buffer segments coupled to the data output of the activator buffer on one of the other many buffer segments.