Spiking Neural Networks

last updated Wed Jun 17 2026 00:00:00 GMT+0000 (Coordinated Universal Time)
Neuromorphic ComputingEdge AISpiking Neu…

Child of Neuromorphic Computing.

Physics / mechanism

Spiking neural networks (SNNs) encode and transmit information as discrete, asynchronous events (spikes) over time, rather than the continuous-valued activations of conventional ANNs. A neuron integrates incoming weighted spikes and fires only when its membrane potential crosses a threshold, so computation is event-driven and sparse: no spike, no switched capacitance, no energy spent. This is the algorithmic layer that most Neuromorphic Computing hardware is built to run. Differentiated parameters: spike-encoding scheme (rate vs temporal), on-chip learning rule (STDP / surrogate-gradient backprop), synaptic density, and energy per synaptic operation (~10–100 pJ vs ~1 nJ on GPU).

The decade-long blocker was programmability, not physics: training SNNs was hard until surrogate-gradient methods + maturing toolchains made them trainable like standard DNNs (see Low Power Edge Compute belief ledger). That reframed SNNs from a standalone-chip thesis toward an offload/feature-block inside someone else’s SoC.

Where it fits

Companies using

Connected ideas

Frontier (open questions)

Created 2026-06-17 — filled a concept-page gap surfaced by the fabrication audit (companies tagged spiking-neural-networks had no page to resolve to).

Related concepts