Fig. 1: Event-driven spiking neural network with parallel synaptic updates.
From: Modified spike backpropagation design towards highly parallelable hardware implementation
a ASL-DVS event processing pipeline: Dynamic hand gesture (left) is encoded into spatiotemporal spikes and processed by SNN (middle). Layers (blue/red neurons) encode update pulse trains to a memristor crossbar (right), enabling parallel hardware updates. b Limitations of conventional DNN frameworks: Analog activations with off-chip outer product computation and row/column-wise weight updates. c Proposed SNN parallel framework: Spike train inputs (LIF neurons) with one-step in-memory crossbar updates via voltage pulses, fusing outer product computation with synaptic modulation.
