ACADEMIA

There is currently a search for new materials to build computer microchips that are more energy-efficient and brain-like. However, no theory can guide this effort on a solid foundation. A theory for non-digital computers is necessary to take into account continuous and analog signals, physical effects at the nanoscale, and the fact that the devices created are often not identical. The paper published by Herbert Jaeger, Beatriz Noheda, and Wilfred G. van der Wiel is the first attempt to provide a sketch of what such a theory for neuromorphic computers might look like.

According to Herbert Jaeger, who is a professor of computing in cognitive materials at the University of Groningen in the Netherlands, there needs to be a solid theory behind the engineering of new microchips. Currently, computers rely on stable switches, usually transistors, that can be either on or off, making them logical machines with programming based on logical reasoning. However, the miniaturization of transistors, which has been the key to making computers more powerful, is reaching its physical limit, which is why scientists are now looking for new materials that can produce more versatile switches capable of using more values than just 0 or 1.

Jaeger is a member of the Groningen Cognitive Systems and Materials Center (CogniGron) which is devoted to creating neuromorphic (brain-like) computers. CogniGron brings together scientists with differing approaches, including experimental materials scientists, mathematical theorists, and computer science and AI specialists. Working closely with materials scientists has given Jaeger insight into the challenges they face when developing new computational materials. It has also made him aware of a dangerous pitfall: there is no established theory for the use of non-digital physical effects in computing systems.

Our brain functions differently than a logical system. Although we can reason logically, this is only a small aspect of what our brain can do. The majority of the time, our brain must figure out how to perform simple tasks such as lifting a cup or waving to a colleague. Jaeger explains that "a lot of the information-processing that our brain does is this non-logical stuff, which is continuous and dynamic. It is difficult to formalize this in a digital computer." Additionally, our brain can function despite external factors such as fluctuations in blood pressure, external temperature, and hormone balance. So, how can we create a computer that is both versatile and robust? Jaeger believes that "the brain is proof of principle that it can be done", and is optimistic that it can.

The brain is a source of inspiration for materials scientists who aim to produce materials that mimic the behavior of neurons. Scientists might create materials that oscillate, show bursts of activity, and resemble how neurons work. However, the field is missing a crucial piece of information: even neuroscientists don't fully understand how the brain works. The lack of a theory for neuromorphic computers is a problem, but the field doesn't seem to acknowledge this. In a recent paper, Jaeger, Noheda, and van der Wiel proposed a theory for non-digital computers. The theory suggests that instead of using stable 0/1 switches, non-digital computers should work with continuous, analog signals. It should also account for the various non-standard nanoscale physical effects that materials scientists are studying.

Neuromorphic computing devices made from new materials are difficult to construct, and if you make a hundred of them, they will not all be identical. This is similar to how our neurons are not all the same. Additionally, these devices are often brittle and sensitive to temperature. Therefore, any theory for neuromorphic computing should consider such characteristics. Importantly, a theory supporting neuromorphic computing will not be a single theory but will consist of many sub-theories, just like digital computer theory, which is a layered system of connected sub-theories. To create a theoretical description of neuromorphic computers, experimental materials scientists and formal theoretical modelers must collaborate closely. Computer scientists must be aware of the physics of all these new materials, and materials scientists should be familiar with the fundamental concepts in computing.

The University of Groningen established CogniGron to bridge the gap between materials science, neuroscience, computing science, and engineering. The aim is to bring together these different groups to work collaboratively. Jaeger, one of the researchers at CogniGron, explains that everyone has their blind spots and the biggest gap in their knowledge is the lack of a foundational theory for neuromorphic computing. To overcome this, their paper provides a first attempt at highlighting how such a theory could be formulated and how a common language can be created.