Physical reservoir computing (RC) provides a low-power framework for complex spatiotemporal tasks, but its accuracy is fundamentally limited by the state richness and separability of physical nodes. While existing approaches often rely on increasing the number or types of devices to enhance reservoir dimensionality, we propose a materials-centric strategy to intrinsically improve node nonlinearity and discrimination. Here, we introduce a transient molecular templating method to engineer the semiconductor-device interface, using a volatile molecular template (o-pdnqr) to guide the assembly of PDVT-10 and achieve a highly ordered channel. This approach reduces trap density by 48%, extends carrier lifetime, and enhances photoresponsivity-collectively yielding synaptic transistors with prolonged relaxation, high paired-pulse facilitation, and state-rich nonlinear dynamics. Implemented as a single-type physical node for RC, our device achieves 99.8% accuracy on static gesture recognition (Sign Language MNIST) and 96.2% on dynamic gestures. This work demonstrates that precision materials engineering through transient templating can address intrinsic bottlenecks in neuromorphic hardware, providing a scalable pathway toward high-accuracy and efficient in-sensor computing.
Ge et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: