COMPUTATION UNIT AND METHOD FOR PERFORMING COMPUTATIONS FOR DEEP NEURAL NETWORKS

摘要

A computation unit (200) for performing a computation of a neural network layer is disclosed. A number of processing element (PE) units (200) are arranged in an array. First input values are provided in parallel in an input dimension (204) of the array during a first processing cycle, and a second input values are provided in parallel in the input dimension (204) during a second processing cycle. Computations are performed by the PE units (202) based on stored weight values. A first adder unit (210a) coupled to the first set of PE units (202) generates a first sum of results of the computations by the first set of PE units (202) during the first processing cycle, and generates a second sum of results of the computations during the second processing cycle. A first accumulator (208a) coupled to the first adder unit (210a) stores the first sum, and further shifts the first sum to a second accumulator (210b) prior to storing the second sum.