The article, titled “Complex chemical reaction networks for future information processing” and published on 13 March 2024 in Frontiers of Neuroscience, proposes that highly interconnected chemical reaction networks (CRNs), which exploit nonlinear dynamics and self-organisation, could serve as sustainable, brain-inspired, energy-efficient information-processing systems. In the video below, postdoctoral researcher Katja-Sophia Csizi from IBM Research explains the complexity of CRNs and how they could potentially complement or replace CMOS-based computing, though practical implementations and real-world applications remain to be demonstrated.

Key highlights of the paper include:

• CRNs can exploit nonlinearity, self-organization, fading memory, and parallel processing to encode and compute with high dimensionality and dynamical richness, going well beyond simple oscillatory chemical systems.
• They can act as reservoir computing substrates, where only the readout layer is trained while the internal reaction network remains fixed, reducing the training overhead.
• A major challenge is physical implementation and readout, including mapping real-world data into chemical inputs, operating in non-equilibrium regimes, and extracting outputs via analytical techniques (e.g. spectroscopy, chromatography) without disrupting the reaction.
• Open questions remain around scalability, reproducibility, latency, cascading of CRNs, fuel supply, and integration with hardware, meaning that practical CRN computing systems have yet to demonstrate real-world performance

Watch the video below to learn more.