Morphologically tunable mycelium chips for physical reservoir computing.

We introduce a neuromorphic computing substrate based on PEDOT:PSS-infused mycelium, a biofabricated, morphologically tunable material that can be engineered into electrically active components including resistors, capacitors, and non-linear elements. Leveraging the principles of physical reservoir computing, we demonstrate that hyphal networks grown under controlled environmental conditions can transform time-varying inputs into nonlinear, high-dimensional state trajectories, enabling machine learning tasks such as NARMA-10 sequence prediction. The chips are produced using a "design-grow-compute" workflow that integrates morphological modeling, parametric growth protocols, and vacuum-assisted polymer infusion. Morphological complexity is shown to influence charge transport and memory capacity, offering a new axis of control for designing analog computational architectures. Our prototype chips interface with a custom carrier board enabling analog signal conditioning and readout. Benchmarking revealed robust nonlinearity, temporal dynamics, and task-relevant encoding. Unlike memristor arrays, photonic, or living-cell-based reservoir systems, our non-living analog mycelium chip is low-cost, biodegradable, and scalable using existing mushroom farming infrastructure, with production yields exceeding 3 million chips per growth cycle. This proof-of-concept demonstration establishes a biodegradable reservoir computing platform and reports standard RC characterizations (linear memory capacity, fading-memory relaxation, and device-to-device variability) across independently fabricated chips. We present performance trade-offs justified by unprecedented sustainability advantages and orders-of-magnitude cost reductions. This advances a novel direction for biologically derived, single-use/ disposable or very large-scale machine learning hardware and introduces mycelium as a functional medium for analog inference.