Engineering a human stem cell-derived neural network platform for biocomputing

We introduce a Bio-adaptive Processing Unit (BPU) based on a two-reservoir microtunnel Brain-on-Chip with electrophysiological readout. We fabricated and validated three independent devices, demonstrating that human stem cell-derived ( Ngn2 + hiPSCs) cortical neurons extend axons over 1200 µ m, forming robust, long-range connections. Relative to reservoirs, tunnel electrodes exhibited substantially higher spike amplitudes and firing rates (median fold-changes of ×10.1 and ×24.4, respectively). Notably, deferred seeding biased ∼85–90% of propagation events from Reservoir A to B, enabling directed axonal conduction within the living substrate, with a propagation velocity distribution characterized by a median of 0.75 m/s (IQR = 0.46–1.00 m/s; n = 9973 events). These results confirm reproducible neuron-electrode interfacing and reliable axonal routing, establishing the BPU as an MEA-interfaced platform for engineering topologically constrained human neuronal networks and quantifying directed axonal signal propagation. Although this manuscript does not demonstrate computation (e.g., task training, learning, or long-term plasticity), we report enabling routing and readout primitives intended to support future biocomputing experiments using living neural substrates that leverage intrinsic neuronal dynamics (e.g., spike-based signaling, plasticity, and energy efficiency).