Binarized Leaky Integrate and Fire (LIF) neuron supports binary [0/1] inputs and [-1/1] binarized weights. Inputs are multiplied by weights and accumulated on the internal membrane. Membrane is exponentially decaying with every clock cycle. Once membrane value (potential) reaches threshold, neuron spikes and membrane value is decreased.
membrane += inputs * weights
membrane *= decay_factor
membrane -= threshold if membrane > threshold
spike = 1 if membrane > threshold
While reset is held high, input values are assigned to weights After reset cycle neuron is active. The only output of the neuron is the binary spike!
| # | Input | Output | Bidirectional |
|---|---|---|---|
| 0 | input1 / weight1 | spike | |
| 1 | input2 / weight2 | ||
| 2 | input3 / weight3 | ||
| 3 | input4 / weight4 | ||
| 4 | input5 / weight5 | ||
| 5 | input6 / weight6 | ||
| 6 | input7 / weight7 | ||
| 7 | input8 / weight8 |