R&D

Neuromorphic Computing: Learning from the Brain

A curious fact about the brain's energy consumption opens up a different way of thinking about how computers process information.

Rob Hoeijmakers

29 Oct 2026 • 3 min read

I recently came across a video explaining something I hadn't given much thought to before: the human brain runs on roughly 20 watts of power.

That's remarkably little, considering everything it does. It recognises faces, interprets language, coordinates movement, learns from experience and helps us navigate the world.

Yet when it comes to arithmetic, a cheap pocket calculator can easily outperform us.

I knew, of course, that brains and computers work differently. But what caught my attention was a more fundamental distinction: the way they organise memory and processing.

It turns out there's an entire field of research exploring this difference, and what we might learn from it when designing computers.

It's called neuromorphic computing.

Why conventional computers work differently

Most modern computers are descendants of an architecture developed in the 1940s, commonly associated with mathematician John von Neumann.

In a conventional computer, processing and memory are largely separated. The processor performs calculations, while memory stores the information and instructions it needs.

Data must continually travel between them.

This arrangement has been extraordinarily successful. It has supported decades of technological progress, from personal computers to today's powerful AI systems.

But moving information consumes time and energy. For some workloads, transferring data can actually require more energy than performing the calculations themselves.

This is part of what engineers call the von Neumann bottleneck.

Modern processors use sophisticated techniques to reduce this problem, but data movement remains an important challenge, particularly for artificial intelligence.

And this is where the brain offers an interesting alternative.

What can we learn from the brain?

The brain consists of networks of neurons connected through synapses. These networks process information, but changes in their connections also play an essential role in learning and memory.

Processing and memory are closely intertwined.

Another difference is how neurons communicate.

Biological neurons exchange electrical impulses called spikes. Activity is distributed throughout the network, with individual neurons responding to incoming signals and their own internal state.

Conventional computers can also process information in parallel and respond to events. Neuromorphic researchers are exploring what happens when these principles are combined with memory and processing that are more closely intertwined.

This is the basic idea behind neuromorphic computing.

Some neuromorphic systems use artificial neurons that communicate through spikes, forming what researchers call spiking neural networks. Other designs take inspiration from different aspects of biological information processing.

These are electronic systems, usually built from silicon. They aren't biological brains, and the objective isn't necessarily to recreate human intelligence.

The interesting question is whether borrowing some of the brain's operating principles could make particular forms of computing more efficient.

What is happening in research?

Neuromorphic computing isn't a new idea. Researchers have been experimenting with brain-inspired electronic systems for decades.

But the growing demand for more efficient AI has given the field renewed attention.

Intel, for example, has developed specialised research processors called Loihi. In 2024, it introduced Hala Point, a system using more than a thousand Loihi 2 processors and supporting up to 1.15 billion artificial neurons.

The system was deployed at Sandia National Laboratories to investigate neuromorphic computing at a much larger scale.

This doesn't mean Intel has built an artificial brain. It means researchers now have substantial hardware on which to test these alternative approaches to computation.

Other research groups and specialist companies are exploring different approaches, each inspired by particular aspects of biological information processing.

There are still important limitations. Neuromorphic hardware requires suitable algorithms and software, and it isn't automatically better than conventional processors. Whether it offers an advantage depends heavily on the task.

Where might it lead?

One promising application is AI that can operate efficiently in small devices.

Think of industrial sensors, wearable technology or robots that need to interpret their surroundings continuously, often with limited power available.

Instead of sending information elsewhere for processing, these systems could potentially respond locally, using much less energy for particular tasks.

Researchers are also investigating whether neuromorphic architectures can make it easier for systems to adapt and learn while operating.

These possibilities remain active areas of research. Neuromorphic processors may eventually complement conventional computing rather than replace it.

But that is already an interesting prospect.

A different way of thinking about computers

What started with a curious observation about the brain's energy consumption led me to an entire field of research I knew very little about.

Neuromorphic computing is still developing, with experimental systems, promising applications and plenty of unanswered questions.

It's one of those developments happening a little beyond the everyday technology conversation. And precisely the sort of thing I enjoy discovering.