Defect engineering in parylene memristors with ZnTe nanoparticles for enhanced switching reproducibility,...

To date, no single memristive material has been identified as optimal for neuromorphic computing, so systematic refinement of well-established devices remains a valuable route to practical application. Parylene (poly(p-xylylene)) memristors are a representative example: more than a dozen studies of pristine and nanocomposite parylene devices published over the last decade have cumulatively lowered the switching voltages, energy and variability. Here, we propose new parylene-based memristors with incorporated ZnTe nanoparticles, whose favorable electronic properties and interfacial sites for Cu accumulation promote more deterministic filament formation and add an optical degree of control over the switching. The resulting Cu/parylene-ZnTe/ITO memristors combine in a single device the advantages previously demonstrated separately for other nanocomposite parylene memristors: the lowest switching voltages reported for this family (0.6 V and -0.36 V), low device-to-device variations of 6% and 15% for the set and reset voltages, 14 resistive states, an endurance of 10⁴ cycles, a switching time of <10 ns at an energy of 6 pJ, reversible photosensitive switching, and the first batch-to-batch comparison for parylene memristors. The switching mechanism is studied in detail using transmission electron microscopy, I-V slope analysis, impedance and optical spectroscopy, temperature dependencies and photoresponse measurements, and is attributed to percolation chains of ZnTe nanoparticles bridged by Cu clusters with thin tunneling gaps. Finally, simulations of a memristive neural network and an in-sensor reservoir computing system, both based on the measured device characteristics with their nonidealities, indicate that the devices are promising for in-memory and in-sensor computing.