Reconfigurable parallel 3-dimensional convolution cluster engine

A reconfigurable parallel 3-Dimensional (3D) convolution engine for performing 3D Convolution and feature map extraction on an image, the reconfigurable parallel 3D convolution engine comprising:

• a plurality of Convolution Neural Network (CNN) reconfigurable engines configured to perform 3D convolution, in parallel, to process a plurality of feature maps;
• a plurality of diagnostics registers configured to detect at least one Single Event Upset (SEU) faults and Single Event Transition (SET) faults while performing the 3D Convolution and feature map extraction on the image;
• a kernel memory space, present in each instance of CNN reconfigurable engine of the plurality of CNN reconfigurable engines, capable for holding a set of parameters associated to a network layer having each operational instance of the CNN reconfigurable engine of the plurality of CNN reconfigurable engines;
• at least one memory controller, an Input Feature Map Memory (FMM) cluster and an Output FMM cluster, wherein the at least one memory controller, in communication with each of the plurality of CNN reconfigurable engines, is configured to:
• fetch image data retrieved from an external memory,
• transfer the image data into one or more input line buffers, of each FMM bank, present in the Input FMM cluster, in order to extract the plurality of feature maps from the image data;
• and
• transfer convoluted output data, indicative of a 3D Convolution and feature map extraction of the image, to one or more output line buffers, of the Output FMM cluster, which further relay the convoluted output data to the external memory;
• and
• a flat dense computing unit configured to:
• convert each of the plurality of feature maps into 1-Dimensional vector;
• and
• perform dense multiplication on the 1-Dimensional vector by using filter kernels values so as to transfer the convoluted output data to the external memory.