71 4-Neuron LIF Spiking Neural Network
71 : 4-Neuron LIF Spiking Neural Network
- Author: Ilim Stankulov
- Description: A simple 4-neuron spiking neural network implementing the LIF model
- GitHub repository
- Open in 3D viewer
- Clock: 10000000 Hz
How it works
This project implements the Leaky Integrate-and-Fire model in a simple 4-neuron spiking neural network chip designed for TinyTapeout GF180nm
Every neuron has a membrane potential and an incoming current. Each clock cycle, the neuron receives input current from neurons that fired in the previous cycle and leaks a fraction of its current voltage (1/8 of the current voltage) according to the following equation:
V[t+1] = V[t] - V[t]/8 + I[t]
Once the voltage inside a neuron reaches 64, it fires a spike and resets its membrane potential to 0.
Every neuron's spike feeds back into other neurons through a weight matrix. If the connection weights are positive, they push the voltage towards the threshold, if they are negative, they pull the voltage away from the threshold.
How to test
-
Hold rst_n low for at least 3 clock cycles, then release. All membrane potentials and spike outputs clear to zero.
-
Assert ui[2] high (mode_prog = 1). Shift in 128 bits MSB-first by toggling ui[1] (s_clk_en) while driving ui[0] (s_data) with each bit. After 128 bits, uo[4] (prog_done) goes high confirming the weight matrix is fully loaded. De-assert ui[2].
-
Assert ena = 1. Spike outputs appear on uo[3:0], one bit per neuron. Behavior depends entirely on the programmed weights — zero weights produce silence, positive weights produce excitation, negative weights produce inhibition.
External hardware
No external hardware is required.
IO
| # | Input | Output | Bidirectional |
|---|---|---|---|
| 0 | s_data — serial weight bit input | spike_0 — neuron 0 spike output | |
| 1 | s_clk_en — serial clock enable | spike_1 — neuron 1 spike output | |
| 2 | mode_prog — 1=program weights, 0=run | spike_2 — neuron 2 spike output | |
| 3 | spike_3 — neuron 3 spike output | ||
| 4 | prog_done — all 128 weight bits loaded | ||
| 5 | |||
| 6 | |||
| 7 |