hardware · ML

AI Inference Acceleratorshipped

MNIST classification in hardware.

A fully-connected NN with ReLU in Verilog, simulated cycle-accurately and validated against PyTorch.

Custom multiply-accumulate units and a dense ReLU layer in Verilog with fixed-point arithmetic, simulated under Verilator with a C++ testbench. A PyTorch pipeline exports weights, and a test suite compares Verilog inference against PyTorch.