A spiking neural network (SNN) is an artificial neural network that represents communication between neurons as discrete events called spikes. Because those events occur over time, an SNN can make spike timing and event-driven activity part of its computation—not just the numerical strength of an activation.
How a spiking neural network works
In an SNN, neurons communicate by sending spikes: brief, discrete events at particular points in time. A spike sequence can carry information through both the occurrence of events and their timing. This makes SNNs a natural way to represent temporal patterns or inputs that arrive as events. A 2019 Nature perspective discusses spike-based computation, while a 2022 review in Neural Computation surveys the range of models and methods.
As an Amazon Associate I earn from qualifying purchases.
Threshold-based neuron models
One common model family is integrate-and-fire. In a simplified account, incoming signals change a neuron’s internal state; when that state reaches a threshold, the neuron emits a spike. This is an example, not a universal definition: SNNs use different neuron and synapse models, and their exact dynamics depend on the model chosen.
How SNNs differ from conventional neural networks
Many conventional artificial neural networks are described in terms of numerical activations passed between units. An SNN instead represents communication with discrete spikes and can explicitly use when those spikes occur. The distinction is about the computational representation; it does not by itself establish that one kind of network is better for every task.
#1 Best Overall
| Aspect | Conventional ANN, in a common description | Spiking neural network |
|---|---|---|
| Communication | Numerical activations | Discrete spike events |
| Time | May be implicit or handled through the architecture and input representation | Spike timing is an explicit part of the representation |
| Natural fit | Depends on the task and model | Can naturally represent temporal structure and event-driven input |
These are broad descriptions, not claims that every ANN or SNN is built the same way. A practical comparison should consider the target task, input format, model and training approach, software and hardware support, and measured results on the intended workload.
Where SNNs are studied and used
Research and application areas include computer vision and robotics, as well as brain-machine interfaces, control and navigation, speech recognition, event detection, and classification. These are areas of study and application, not evidence that SNNs are equally mature or widely deployed in each one. A 2022 review indexed by PubMed covers SNN theory, training, frameworks, and applications: Spiking neural networks: a survey.
Are spiking neural networks more energy-efficient?
Energy efficiency is a motivation for neuromorphic computing, but it is not a guaranteed property of every SNN implementation. A 2019 Nature perspective describes the field as promising the possibility of reducing computing platforms’ energy requirements; that framing is a research goal, not proof of a particular saving. A 2022 Nature Computational Science perspective also discusses neuromorphic computing’s opportunities and challenges.
To assess an efficiency claim, look for a comparison on a named workload that identifies the hardware, baseline, and measurement method. Without those details, a general energy multiplier or claim that SNNs are always more efficient is not meaningful.
Rank #3
What to check when evaluating an SNN
Whether an SNN is useful depends on the complete system, not just the network label. Check:
Quick Recap
Best Value
Rank #4
- Input: Is the data naturally temporal or event-based, or does it need to be converted into spikes?
- Model: Which neuron and synapse models are used, and what do their dynamics assume?
- Training: How is the network trained, and does that method suit the task?
- Implementation: Which software and hardware support the chosen model? Framework support and maintenance can change, so verify the current status of any specific tool.
- Evidence: Is there a measured benefit on the workload you care about, with a clear baseline and measurement conditions?
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
