Nanoimprint‐Enhanced Organic Heterojunction Phototransistors for Bipolar Optical Encoding in Reservoir Computing

ABSTRACT

Optoelectronic reservoirs require transient states with sufficient electrical contrast and diverse temporal dynamics. However, weak optical‐to‐electrical transduction reduces the resolvability of reservoir states, whereas unipolar photoresponses restrict how successive optical inputs modulate the residual device state. Here, we report a nanoimprint‐enhanced PDVT‐10/F ₁₆ ‐CuPc heterojunction organic phototransistor that combines amplified photocurrent output with wavelength‐dependent bipolar dynamics. Periodic nanostructures introduced into the PDVT‐10 layer enhance broadband optical absorption and photocurrent generation, enabling a minimum detectable irradiance of 0.65 µW cm ⁻² at 430 nm. Meanwhile, the heterojunction generates wavelength‐dependent photocurrent transients with opposite polarities and distinct relaxation behaviors, enabling bipolar optical encoding in which binary 0 and 1 inputs are represented by optical pulses at two wavelengths that drive the photocurrent state in opposite directions. Consequently, successive inputs can selectively reinforce, offset, or reverse the residual photocurrent state, yielding more distinguishable and history‐dependent reservoir trajectories. This bipolar encoding improves reservoir‐state distinguishability by expanding the accessible photocurrent range and reducing state overlap under fluctuations, increasing nine‐class human action‐recognition accuracy from 77.04% to 83.3%. Combining two bipolar reservoir groups further increases recognition accuracy to 90.0%. This work establishes a compact device paradigm that unifies weak‐light sensitivity with temporal dynamics, advancing hardware‐efficient processing of complex temporal information.