Overview
- The peer-reviewed PNAS study published March 12, 2026 shows cultured rat cortical neurons learned to produce periodic and chaotic time-series in a reservoir-computing loop.
- Researchers used microfluidic guides and high-density electrode arrays to shape modular circuits that avoided synchronized firing and kept rich dynamics.
- The system learned sine, triangle, and square waves and reproduced the Lorenz attractor, with one culture switching among 4 to 30 second target frequencies.
- Performance fell after training ended and feedback latency of about 330 milliseconds limited fast or sharp targets.
- The team plans to cut delay, improve post-training stability, refine the FORCE readout method, and test uses in drug screening, disease models, and brain-machine interfaces.