How erwanregy/neural-network-on-an-fpga works
This repository describes a parameterized feedforward neural network implemented in hardware. The principal workflow is input signals entering the network, passing through connected dense layers of neurons, and emerging as outputs when computation is ready. Each neuron multiplies inputs by weights, accumulates the products, applies an activation, and returns to waiting for the next input. The relationships shown are documented in the README; no source excerpts were available to verify further implementation details.
Main entry points: neural_network.sv Neural network · neuron.sv Multiply inputs · dense_layer.sv Dense layer
Components 10
Network composition
- Neural network
sources_1/new/neural_network.sv
Layer computation
- Dense layer
sources_1/new/dense_layer.sv
Neuron computation
- Neuron
sources_1/new/neuron.sv - Multiply inputs
sources_1/new/neuron.sv - Accumulate products
sources_1/new/neuron.sv - Apply activation
sources_1/new/neuron.sv - Wait for input
sources_1/new/neuron.sv
Other components
- Input producer
- Output consumer
- Clock and reset
Connections 10
- Input producer → Neural network: provides inputs
README.md - Clock and reset → Neural network: provides control
README.md - Neural network → Dense layer: instantiates layers
README.md - Dense layer → Neuron: instantiates neurons
README.md - Neuron → Multiply inputs: starts computation
README.md - Multiply inputs → Accumulate products: passes products
README.md - Accumulate products → Apply activation: passes sum
README.md - Apply activation → Wait for input: completes activation
README.md - Wait for input → Multiply inputs: restarts on ready
README.md - Neural network → Output consumer: provides outputs
README.md