The chemical computer: an energy-efficient computing alternative

Mathieu Baltussen, a PhD candidate at Radboud University, has successfully developed a prototype of a chemical computer as part of his doctoral research. This innovative approach aims to address the growing energy demands of current computing technologies by leveraging the inherent computational capabilities of complex chemical reaction networks, such as the Formose reaction.

Baltussen's work demonstrates the potential of chemical computers to perform nonlinear classification tasks, model the behavior of complex dynamical systems, and even forecast future environmental changes - all while requiring significantly less energy and data compared to traditional digital computers. This research paves the way for a new class of biomimetic information processing systems that could complement or even replace certain specialized computing tasks in the future.

Watt Matters in AI

Watt Matters in AI is a conference that aims to explore the potential of AI with significantly improved energy efficiency. In the run-up to the conference, IO+ publishes a series of articles that describe the current situation and potential solutions. Tickets to the conference can be found at wattmattersinai.eu.

View Watt Matters in AI

How does the chemical computer work?

The chemical computer diverges from previous attempts to mimic electronic computers using molecules. Instead of trying to force molecules into transistor-like configurations, which has proven difficult, he utilizes the Formose reaction, a self-organizing chemical reaction network believed to be fundamental to the origin of life. By adopting the concept of a "reservoir computer" from neuromorphic computing, Baltussen extracts useful calculations from this complex system. He successfully predicted basic weather patterns by aligning the behavior of *E. coli* bacteria and the Formose reaction, showcasing the potential of this approach.

The chemical reservoir computer is built around the Formose reaction within a continuous stirred tank reactor (CSTR). The output of this reservoir is then measured using an ion mobility mass spectrometer, which identifies the relative abundance of up to 106 different ions. The nonlinear response of the chemical reservoir to various inputs is recorded and converted into a computational output by training a single linear read-out layer. This setup enables the chemical computer to perform multiple parallel, nonlinear classification tasks, model complex dynamic systems, and conduct time series forecasting.

Curbing AI energy use? 3D optical computing offers a solution

Reducing power usage, yet boosting AI chip perfomance, Lumai's 3D optical computing tech promises to curb AI energy use.

Read More

The energy efficiency imperative

While chemical computers are unlikely to replace digital computers entirely, they could serve a similar function to quantum and neuromorphic computing in data centers by handling specialized tasks with significantly less energy and data. Potential future applications extend to nanotechnology and biomedical science, including the development of smart body sensors, though further research and funding are essential to realize these possibilities.

The drive for energy efficiency is a key motivator, as current AI and computing technologies demand enormous amounts of power. The human brain, for instance, uses only 20 watts of electricity while sending 1,000 impulses per second with its 80 billion neurons, showcasing the potential for more efficient computing paradigms.

Christian Mayr, Chair of Highly-Parallel VLSI Systems and Neuromorphic Circuits at Dresden University of Technology, emphasizes that "the brain has already solved many of the computational challenges we face," particularly in terms of energy efficiency. Neuromorphic computing, inspired by the human brain, aims to replicate its computational principles.

Johan Mentink, a physicist at Radboud University, is also researching neuromorphic computers, aiming to drastically reduce the energy consumption of hardware. His project, funded by NWO, explores new algorithms and materials to directly utilize intrinsic noise in computations, rather than trying to circumvent it, to enable large-scale calculations currently considered impractical. Baltussen is scheduled to defend his PhD thesis on September 22, 2025.