Implementation of a Neural Network in Chisel

Getting started

1. Set up your environment to run Chisel locally. Setup instructions can be found here.
2. Clone the repo

git clone https://github.com/karthik-r-rao/neural-net-chisel
cd neural-net-chisel

Generate look-up table values for sigmoid activation function

Usage:

python3 ./scripts/generate_sigmoid_lut.py <name of output file> <fixed_point_length> <fraction_length>

Always store the output file in src/test/resources/ for the test files to load and initialize the sigmoid look-up table. Example of a look-up table with 1024 entries of 10 bits each, interpreted as fixed point with 3 integer bits and 7 fraction bits:

python3 ./scripts/generate_sigmoid_lut.py ./src/test/resources/sigmoidlut.txt 10 7

Simulation

To test any module:

sbt "testOnly nntests.<name of test class>"

Test parameters can be changed in the respective test files. Example:

sbt "testOnly nntests.NeuronSpec"

Generate verilog files

To generate verilog files for any module:

sbt "runMain nn.<name of driver class>"

The verilog files will be present in the generated/ folder. Example:

sbt "runMain nn.DriverNeuron"