Chips that mimic the brain's adaptability and efficiency

Our world knows more and more supercomputers. But do not be mistaken: the most efficient – and perhaps more importantly: the most flexible – computing is done in the brain, human or otherwise. So what can we learn from how the brain computes? Charlotte Frenkel, recently recognized by NWO as one of the Netherlands' biggest AI talents and thus officially an ‘AiNed fellow’, intends to find out. This new grant allows Frenkel to start a team to research the possibilities of neuromorphic computer chips, inspired by the neocortex area in the brain.

In the quest to extend the battery life and capabilities of computer chips, researchers are turning to the brain for inspiration. The brain's remarkable efficiency, adaptability, and computing prowess have captivated scientists for years. Now, a groundbreaking project, called SynergAI, led by Charlotte Frenkel aims simultaneously to merge neuroscience with AI research, and to bridge the gap between analog and digital computing, potentially revolutionizing the field of computing and artificial intelligence.

At the heart of Frenkel's research is the development of neuromorphic chips that adapt to meet the demands placed upon them. Unlike conventional digital central processing units (CPUs) that execute instructions sequentially, neuromorphic chips seek to emulate the brain's ability to process information in parallel, to efficiently adapt from little new information, and to provide explainable decisions. This way of working makes neuromorphic chips the perfect match for running low-power and trustworthy intelligent systems .

Digital CPUs, which are the current standard in modern computing, process data using discrete digital signals represented by 0s and 1s – also called bits. In contrast, analog CPUs work with continuous signals, allowing for a more nuanced representation of data. By leveraging the best of both worlds, Frenkel aims to find the "sweet spot" between analog and digital computing, harnessing the power of neural-inspired algorithms while maintaining the robustness of digital systems.

To get to that sweet spot, she is inspired by one of the most striking features of the neocortex area of the brain: it is organized in so-called "minicolumns." These collections of roughly 100 neurons are fundamental to many operations, such as pattern recognition, and are a foundational element of the brain’s ability to plan, reason, and build abstract concepts. Unlocking these computing primitives in silicon-based chips could enable machines to perform tasks with unparalleled efficiency and adaptability.

Frenkel's ambitious vision extends beyond theoretical research. She aims to develop two neuromorphic, silicon prototypes by the end of the project: one that is entirely digital (but heavily inspired by the neocortex and its minicolumns) and one that combines analog and digital computing elements.

"I want to contribute to a future in which computers are smaller and more adaptive. This means they will have a far smaller footprint on our society, in terms of resources but also in terms of capital that is needed to update and replace them," Frenkel envisions. Her research represents a significant milestone in the trajectory toward that future.

The implications of Frenkel's work are far-reaching. By creating neuromorphic chips that mimic the brain's ability to adapt and learn, computing systems could operate more efficiently, requiring fewer resources and reducing the strain on our environment. Moreover, the integration of these chips into for example smart prostheses could lead to improved and easer interaction with the patients, thereby improving the quality of life for individuals with limb loss or disabilities. Furthermore, autonomous drones equipped with the advanced computing power of neuromorphic chips could revolutionize search-and-rescue operations with tiny drones, improving responses to emergencies. Frenkel aims to study both implementations as a testcase.

As Frenkel and her team embark on this groundbreaking research, they aim to revolutionize the sustainability of computing technology. And by unlocking the brain's secrets and porting them into neuromorphic chips, they are paving the way for a future where small and efficient computers better complement our human lives by adapting to us.

The AiNed Fellowship Scholarships are a program component of the AiNed National Growth Fund program. The goal of the AiNed Fellowship Scholarships program is to attract AI talent to Dutch academic research institutions in view of international competition for AI talent.