Can brain-inspired hardware reduce AI’s energy footprint?

In a recent podcast appearance, Wilfred van der Wiel, coordinator of HYBRAIN, discussed how insights from the human brain could help address one of artificial intelligence’s biggest challenges: energy consumption.

As AI systems become more capable, their energy demand continues to rise. Rather than relying on ever-larger and faster data centres, current research, such as the one performed by the HYBRAIN project, is exploring an alternative path based on neuromorphic computing, hardware designed to operate more like networks of biological neurons. These systems aim to perform complex tasks while using significantly less power.

At the nanoelectronics laboratories of the University of Twente, research teams are developing analogue, trainable chips whose behaviour can be adjusted after fabrication. Unlike conventional processors, these devices exploit physical properties to carry out computation, reducing the need for repeated data transfers and lengthy processing steps.

One promising application is speech recognition. By processing raw audio signals directly on neuromorphic hardware, it becomes possible to achieve accurate results with simpler digital models and lower energy consumption, opening new opportunities for use in edge devices such as smartphones, vehicles, and medical implants.

The discussion also touched on broader implications. Hardware that can adapt and learn raises important questions around control, responsibility, and ethics. For this reason, the research is accompanied by dialogue with experts in ethics and philosophy, ensuring that societal impacts are considered alongside technical progress.

Neuromorphic computing is not expected to replace digital systems, but to complement them, especially in scenarios where efficiency, autonomy, and real-world deployment are key. As AI hardware continues to evolve, these brain-inspired approaches could play a decisive role in making intelligent technologies more sustainable.