Unlike classical computers, which process information in bits, the human brain uses short electrical impulses – known as spikes. What matters is the precise timing of these spikes, which carries the actual information.
This timing-based representation allows the brain to transmit more complex information per signal compared to a digital bit, which can only be “0” or “1.” The intervals between spikes and their patterns carry additional meaning, making this highly efficient for tasks such as detecting patterns or sequences in real time. In technology, these principles are replicated by so-called Spiking Neural Networks (SNNs).
“Our goal is to develop systems that mimic biological structures and processes – using minimal energy while achieving maximum responsiveness,” explains Priv.-Doz. Dr. Bernhard A. Moser, Technology and Innovation Manager at SCCH. “Potential applications include robotics, medical technology, and environmental monitoring.”
Dispelling Common Myths
Despite growing interest, several misconceptions about Neuromorphic Computing (NC) persist – such as the assumption that it requires expensive, specialized hardware, is slower than traditional methods, or lacks industrial maturity.
“We show that many of these assumptions are outdated,” says Dr. Thomas Buchegger, Head of the SAL site in Linz. “NC offers tremendous potential not only in energy efficiency but also in speed. Moreover, it can be effectively implemented on standard hardware platforms – not just specialized analog chips.”
A recent joint study by JKU, SAL, SCCH, and TU Graz has demonstrated NC’s versatility in the medical field. Researchers found a way to collect and process electrocardiogram (ECG) data far more efficiently – reducing data volume by more than 80% without sacrificing analysis quality. This means lower storage requirements while maintaining the same accuracy, opening up significant opportunities for the future of medical diagnostics and beyond.
Milestones of the Research Partnership
- World Record in SNN Inference on Standard FPGAs
Using a newly developed demonstrator, the team has shown that Spiking Neural Networks can run at exceptional speeds on widely available FPGA (Field Programmable Gate Array) systems.“Our method processes over 2.5 million images per second – more than 100 times faster than previous systems using the same hardware and data,” says Assoc. Prof. Dr. Michael Lunglmayr from JKU’s Institute for Signal Processing. “It also operates at low power, achieving more than 3 million images per second per watt.”
This breakthrough illustrates how modern AI can be both highly powerful and energy-efficient – on off-the-shelf hardware.
- Patent Filed
The team has filed a patent for a novel, ultra-energy-efficient data acquisition method based on neuromorphic principles inspired by the human brain. - Austrian Workshop Series and Publications
Through the SNNSys workshop series on Neuromorphic Computing and Spiking Neural Networks, organized in collaboration with TU Graz as part of the AIROV conference, the partners are driving knowledge exchange within Austria’s research community. Findings have been published in leading journals such as the Journal of Neuromorphic Computing and Engineering and presented at top-tier conferences, including the IEEE International Symposium on Circuits and Systems. - International Grand Challenge
In September 2025, at the prestigious IEEE International Conference on Image Processing (ICIP) in Alaska, the research team will host an international challenge under its motto “Low Energy, High Speed.” The competition aims to identify the most efficient and fastest AI solutions.“We’re excited to see the creative and innovative approaches this will inspire,” says Moser. “The strong interest from the scientific community shows the incredible momentum in this field.”
