Morphologically tunable mycelium chips for physical reservoir computing.

Area of Science:

• Bio-inspired computing
• Materials science
• Neuromorphic engineering

Background:

• Neuromorphic computing aims to mimic the brain's structure and function.
• Existing substrates often face challenges in cost, scalability, and environmental impact.
• Biofabrication offers a promising avenue for novel computational materials.

Purpose of the Study:

• To introduce a novel, biodegradable neuromorphic computing substrate using PEDOT:PSS-infused mycelium.
• To demonstrate the potential of this biofabricated material for physical reservoir computing.
• To establish a low-cost, scalable, and sustainable platform for analog machine learning hardware.

Main Methods:

• Utilized a "design-grow-compute" workflow integrating morphological modeling, controlled hyphal network growth, and vacuum-assisted polymer infusion.
• Engineered mycelium into electrically active components (resistors, capacitors, non-linear elements).
• Leveraged physical reservoir computing principles to transform time-varying inputs into high-dimensional state trajectories for machine learning tasks.

Main Results:

• Demonstrated successful machine learning tasks, including NARMA-10 sequence prediction, using the mycelium-based substrate.
• Showcased morphological complexity's influence on charge transport and memory capacity.
• Achieved high production yields (over 3 million chips per cycle) with significant cost reductions and biodegradability.

Conclusions:

• Mycelium serves as a functional medium for analog inference, advancing biologically derived machine learning hardware.
• The developed substrate offers unprecedented sustainability advantages and cost-effectiveness compared to existing technologies.
• This work establishes a biodegradable reservoir computing platform with potential for single-use or large-scale applications.