Neuromorphic computing · spiking neural networks

Neuromorphic and spiking neural networks

Computing with discrete spikes in continuous time, and the hardware built to run it.

Spiking neural networks compute with discrete events in continuous time rather than with dense real-valued activations. A neuron integrates its inputs and emits a spike only when it crosses a threshold, so information is carried in the timing and rate of events. This event-driven style maps naturally onto low-power neuromorphic hardware, where computation happens only when a spike arrives.

The central difficulty is training: the spike is a step function with no useful gradient. The field's answer is the surrogate gradient — a smooth stand-in used only on the backward pass — which makes standard optimisation applicable to networks that still behave as spiking systems at inference. Alongside it sits the practical question of how an analogue signal becomes a spike train in the first place.

The neuromorphic studio implements that encoding path as a deterministic, testable kernel. The Academy lesson runs a delta-modulation analogue-to-spike encoder in the browser and ends at the studio's own sealed benchmark measurement for the same kernel, so the reader sees the encoding, then the evidence that it runs bit-for-bit across five language backends.

Key concepts

The vocabulary of the topic.

Spike train: A sequence of discrete events in time; the carrier of information in a spiking network.
Surrogate gradient: A smooth approximation to the spike's derivative, used only on the backward pass to permit gradient training.
Analogue-to-spike encoding: Turning a continuous signal into events, e.g. by delta modulation — the kernel the studio measures.
Event-driven hardware: Neuromorphic substrates that compute only when a spike occurs, trading dense throughput for energy efficiency.

In the federation

Studios working on this.

Each runs standalone in its own repository and federates its evidence through the platform.

SNN Studio

On this site

Run it, don't take it on faith.

The runnable lessons and sealed measurements this site ships for the topic.

Curated literature · a reading path

Read it at source, in order.

Canonical references for the topic, ordered from the foundations to current work. Every one was verified at source — a DOI resolves through doi.org, a standard through its issuer — so each link goes to the real record.

Start with the foundations

1

Paper Networks of spiking neurons: The third generation of neural network models

W. Maass · 1997 · doi:10.1016/S0893-6080(97)00011-7

The framing of spiking networks as a distinct computational model.

How it works

2

Paper Surrogate Gradient Learning in Spiking Neural Networks

E. O. Neftci, H. Mostafa, F. Zenke · 2019 · doi:10.1109/MSP.2019.2931595

How the non-differentiable spike is trained through a surrogate gradient.
3

Paper Training Deep Spiking Neural Networks Using Backpropagation

J. H. Lee, T. Delbruck, M. Pfeiffer · 2016 · doi:10.3389/fnins.2016.00508

An early demonstration of gradient training through spiking layers.

Put to work

4

Paper Loihi: A Neuromorphic Manycore Processor with On-Chip Learning

M. Davies et al. · 2018 · doi:10.1109/MM.2018.112130359

A representative neuromorphic hardware substrate and its on-chip plasticity.
5

Dataset The Heidelberg Spiking Data Sets for the Systematic Evaluation of Spiking Neural Networks

B. Cramer et al. · 2022 · doi:10.1109/TNNLS.2020.3044364

SHD and SSC — the benchmark spike datasets used to evaluate encoders and classifiers.

Keep going