Programmable DNA-Based Molecular Reservoir Biocomputing Network Circuits with Emerging Biomemristors for Solving Complex Nonlinear Problems.

Biomolecular reservoir computing, despite its potential for nontraditional information processing, encounters difficulties in realizing intricate nonlinear dynamics within biochemical systems. This research proposes a biomolecular reservoir computing framework utilizing DNA-based molecular neural networks to implement reconstructed echo state networks (RESNs) and reconstructed delay-feedback reservoir (RDFR) computing for addressing the aforementioned complex nonlinear challenge. Three key innovations underpin this work: (i) a new chemical reaction networks (CRNs)-based reservoir computing structure utilizing idealized molecular interactions with adaptive parameter optimization through gradient descent algorithms, validating short-term memory capabilities; (ii) methodical topological analysis clarifying the operational mechanisms of various biomolecular reservoir computing topologies─including RESNs and RDFR using DNA chemical reaction networks; (iii) DNA strand displacement-driven implementations of RESNs and RDFRs, allowing for the resolution of complex second-order problems and nonlinear autoregressive moving average systems, respectively. This work demonstrates feasibility and efficacy in solving intricate nonlinear systems but also establishes a programmable molecular computing paradigm, providing theoretical foundations and potential implementation architectures for biomolecular information processing in unconventional computing.