This project implements a pre-trained fully connected neural network (FCNN) for hand-written digits from MNIST dataset. It takes images from MNIST dataset (28x28) and output single digit prediction. All parameters (weights and bias) are pre-trained and hard-coded. User can use wishbone bus to pass in inputs and obtain outputs.
This repository contains a Verilog implementation of the FCNN, targeting the OpenLane flow for the Skywater 130nm process and is an entry for the Google/Efabless MPW-3 shuttle.