Harness Background Lighting for Separability Enhancement in Optical Sensing Reservoir for Temporal Signal Processing Utilizing Amorphous IGZO Neuromorphic Transistors

Abstract

By drawing inspiration from the human visual system, utilizing optoelectronic devices for processing temporal and unstructured data within an insensor reservoir (RC) has gained increasing attention in the field of advanced machine vision systems. Despite advancements in the integration of optical sensors with RC architecture, the challenge of enhancing output state separability for better performance remains. In this study, we present a transistor based on amorphous indium−gallium− zinc oxide (IGZO, film stoichiometry In:Ga:Zn:O = 1:1:0.8:3.4) with a ZrO x dielectric layer. By leveraging the migratory properties of oxygen vacancies in the ZrO x dielectric through electrical stimulation, the proposed IGZO transistor demonstrated potentiation/depression cycles with low variation and achieved 90% accuracy in the pattern recognition simulation, indicating its potential as a readout layer in RC systems. Furthermore, by leveraging the inherent light-responsive properties of IGZO and the short-term memory behavior induced by optical stimuli, we can effectively extract distinct reservoir states from the 4-bit optical temporal signals. Notably, the incorporation of additional background light (890 and 635 nm) amplifies the distinctiveness of these extracted reservoir states. The current difference between the maximum and minimum states is 0.76 nA without background light, while it is extended to 1.4 nA with 890 nm background light and 2.27 nA with 635 nm background light. This enhancement makes a substantial contribution to the in-sensor RC architecture. The effectiveness of our IGZO transistor in processing optical temporal signals within the in-sensor RC architecture highlights its potential for advancing neuromorphic applications and state-of-the-art machine vision systems.