Neuromorphic Deep‐Ultraviolet Signal Processing Using a‐Ga 2 O 3

ABSTRACT Neuromorphic signal processing has been performed by lateral metal‐semiconductor‐metal photodetectors supported on wide‐bandgap thin‐film amorphous gallium oxide (a‐Ga 2 O 3 ). Paired pulse facilitation (PPF) observed in the photocurrents of the devices was attributed to charge trapping and de‐trapping in the a‐Ga 2 O 3 and depended systematically on the amplitude, duration, and interval of the pulsed ultraviolet‐C (UVC) input. Four‐bit UVC binary pulse sequences of optimized power were input to the device, and these produced separable electrical output currents (conductance states), measured after the pulse sequences. A simple resistor‐capacitor (RC) equivalent circuit model replicated the neuromorphic device responses, and this model was used with and incorporated into single‐ and multi‐node reservoir computers. Classification of electrocardiograms (ECGs) was performed in simulations using an RC model whose parameters were extracted from and validated against measured device responses. The single‐node reservoir with virtual nodes achieved limited accuracy in classification depending on the heartbeat type, while the composite multi‐node reservoir (consisting of eight RC sub‐reservoirs with different time constants) exhibited significantly improved classification capability, achieving an average accuracy of ∼95%. These results highlight the potential of the a‐Ga 2 O 3 detectors as artificial synapses for neuromorphic signal processing and temporal pattern recognition.