BIOLOGICAL RESERVOIR COMPUTING USING MODULAR HUMAN IPSC-DERIVED NEURONAL NETWORKS

As silicon-based computing approaches fundamental physical limits, Synthetic Biological Intelligence (SBI) offers an energy-efficient alternative by leveraging the intrinsic non-linear dynamics of biological systems. We investigated this potential using the Cortical Labs CL1 platform to implement Reservoir Computing (RC) in human iPSC-derived neuronal networks. In this paradigm, the biological network functions as a physical reservoir, projecting inputs into a high-dimensional state space for linear decoding. While previous studies in animal models suggest modularity enhances dynamic separability, its role in human biological neurocomputation remains underexplored. We hypothesized that engineered modular topology, mirroring the brain’s balance of segregation and integration, is critical for complex processing. To test this, we compared unstructured monolayers against modular networks confined by microfluidic PDMS devices across three tasks: spatial classification, temporal sequence recognition (Morse code), and spatio-temporal pattern recognition (MNIST). Results showed that while simple spatial mapping is likely ubiquitous, complex spatio-temporal computation may be topology-dependent. Modular hippocampal-cortical co-cultures achieved up to 88% accuracy on MNIST, significantly outperforming unstructured monolayers and organoids. Controls matched firing rate and remained no different to chance. These findings demonstrate that structural modularity acts as a functional regularizer, expanding the reservoir's computational capacity. This study establishes the CL1 as a viable platform for easily programmable SBI and suggests that modular architecture is important for high-order biological information processing.