CMOS-Compatible Artificial Optoelectronic Synapse for Neuromorphic Computing

A novel CMOS-compatible artificial optoelectronic synapse with combing a photodiode-body-biased MOSFET (PD-MOS) with a floating-gate Transistor (FGT) is proposed. The PD-FGT exhibits a broad spectral response (300–1100 nm) with a peak responsivity of 1.8 × 103 A/W at 600 nm, and a large memory window from -1.4 V to 4.1 V. Its opto-electronic performance makes a great promise of enabling emulation of biological synaptic characteristics (Excitatory Postsynaptic Current and Paired Pulse Facilitation) via optical and electrical pulse modulation. An optoelectronic artificial neural network based on this device can potentially obtain 92% accuracy in handwritten digit recognition. This optoelectronic dual-modulated synaptic device opens new avenues for artificial vision systems and future in-sensor computing and storage, laying a foundation for large-scale visual neuromorphic computing hardware systems.