Five Industries That Benefit from Spiking Neural Networks

They are the neural networks of the next generation: spiking neural networks – brain-inspired, blazing fast, and extremely energy-efficient. But can these advantages be leveraged in real-world applications? We’re putting SNNs to the test.

 

Large, larger, largest – that’s the impression you can’t help but get when you follow the debates around Artificial Intelligence. It’s right there in the name: Large Language Models. The idea is that the more interconnected neurons you have, the higher the quality of the output. In some applications, though, those expectations may prove to be a mirage. Wherever hardware imposes tight limits and real-time and efficiency have to align, AI heavyweights quickly become a burden.

There are, however, promising contenders that impress with their lightweight nature: spiking neural networks (SNNs), which emulate the brain’s way of processing data and transmit information in the form of spikes. The days when SNNs were purely an academic exercise are over – it's becoming increasingly clear where spiking neurons have the edge over conventional networks. Here are five industries where spiking neural networks can make a difference.

 

1. Mobile communications

The development of future 6G networks has entered a pivotal phase. Industry stakeholders will repeatedly convene within 3GPP to work out the details until, piece by piece, a standard takes shape. And one thing is already clear: the next generation will be lightning-fast. In terms of latency, 6G is expected to break the 100‑microsecond barrier. At the same time, it aims to be the most energy‑efficient generation of mobile networks to date. Artificial Intelligence will also play a key role.

Extremely fast yet extremely energy‑efficient – this apparent contradiction pushes conventional AI models to their limits. SNNs, by contrast, can meet the stringent demands of 6G signal processing, enabling applications such as channel estimation, interference detection, and joint communication and sensing on end devices. What’s more, in light of crises and disasters, the resilience of critical infrastructure is coming into sharper focus. Here, SNNs can bring one of their most remarkable qualities to bear: they can learn complex patterns and recognize and process them in real time while consuming minimal power. That makes them ideal anomaly detectors, helping to identify and counter equipment failures, unintended interference, or hostile attacks in the wireless network.

 

2. Robotics

Mobile and autonomous robots are expected to tackle complex tasks. A drone is a particularly clear example of what that entails. It must continuously sense its surroundings and navigate unknown environments in real time. But size and weight constrain the onboard energy budget, which affects maximum flight time. If, for instance, a drone is tasked with monitoring a wildfire, a power‑hungry AI can drain the battery and jeopardize the mission.

One application area for SNNs here is navigation. Humans perceive 80 percent of their surroundings through vision. Event‑based cameras follow this model, transmitting information whenever brightness changes. This allows drones to react more quickly to changes in their environment and reduce the risk of accidents. SNNs can also help at the root: in the electric motor. The industry strives for designs that are as compact and efficient as possible, which in turn means minimizing the number of built‑in sensors. Conventional sensorless motor controllers, however, struggle at very low and very high speeds. SNNs can process motor signals directly – without sampling or intermediate buffering – and issue control commands. This promises improved sensorless motor control.

 

3. Satellites

Far above the Earth, satellites are meant to support and strengthen terrestrial wireless networks. To rein in the increasingly data‑driven exchange of information, the use of AI is being ramped up. The challenges, however, are immense. Satellites must be powerful enough to process large volumes of data directly onboard, yet they operate under a tight energy budget. They also generate heat during operation, which must be dissipated under difficult conditions to prevent the systems from overheating.

While conventional processing platforms such as CPUs and GPUs struggle with these constraints, spiking neural networks have the potential to overcome them. Whether in Earth observation, communications, or navigation, SNNs can boost performance in many applications – from filtering and compressing collected data directly on the satellite, to onboard beamforming, all the way to remote sensing. In an ESA research project, using an SNN chip for interference detection and classification consumed up to 100,000 times less energy than a conventional neural network.¹ Other studies found that SNNs reduced image‑compression latency by up to 50 percent.²

 

4. Automotive

Anyone getting into a new car today is no longer driving alone. A host of driver-assistance systems support the driver and automatically take over driving functions in various traffic situations. And that is meant to be just a stepping stone on the way to autonomous driving, which would put the virtual copilot fully in the driver’s seat. But safety must be ensured at all times.

Whether in the city or on the highway, in bright sunshine or heavy rain, surrounded by cyclists or pedestrians, vehicles must continuously turn ever-changing, complex situations captured by camera and radar sensors into decisions. Latency has to be kept to a minimum. If a child suddenly runs into the road, even the slightest delay in swerving can end badly. SNNs have an edge here because they capture temporal dynamics; they are therefore well suited to handling sequences of events. And of course, energy efficiency is becoming increasingly important on the road as well. According to studies, SNNs can cut energy consumption by up to 85 percent compared with conventional methods.³ That helps in the emerging era of electric mobility, where automakers can’t afford power-hungry AI.

 

5. Industry 4.0

On the factory floor, even small missteps or slight deviations from the norm can turn into costly production downtime. That’s why, in areas like predictive maintenance, the goal is to detect looming issues quickly so countermeasures can be taken immediately. It helps to have AI running close to the sensor, processing data locally.

Another example is the packaging of food products. Here, the gas mixture that creates a protective atmosphere around the food is subject to strict regulations to keep products fresh for longer. If a company relies on sampling-based monitoring, the result is lots of rejects and even more waste. An alternative is multi-sensor arrays that analyze the gas mixture during packaging, eject it in real time if something is off, and immediately route off-spec items back into production. Throughput, however, is extreme: Beverage cartons come off the line roughly every 20 to 30 milliseconds. Conventional neural networks are simply too slow for that. SNNs, by contrast, can leverage their strength in anomaly detection to make corrections within microseconds. The result: fewer rejects, even less waste, higher production rates, and the digitalization of the circular economy. This example shows that even when ambitions are big, heavyweight AI isn’t necessarily the best solution.

 

References

1. Eva Lagunas, Flor G. Ortiz, Geoffrey Eappen, Saed Daoud, Wallace A. Martins, Jorge Querol, Symeon Chatzinotas, Nicolas Skatchkovsky, Bipin Rajendran und Osvaldo Simeone (2024), Performance Evaluation of Neuromorphic Hardware for Onboard Satellite Communication Applications.
2. Sayan Kahali, Sounak Dey, Chetan Kadway, Arijit Mukherjee und Arpan Pal (2023), Low-Power Lossless Image Compression on Small Satellite Edge using Spiking Neural Network.
3. Aitor Martinez Seras, Javier Del Ser und Pablo Garcia-Bringas (2023), Efficient Object Detection in Autonomous Driving using Spiking Neural Networks: Performance, Energy Consumption Analysis, and Insights into Open-set Object Discovery.