Optoelectronic Synaptic Memristors for Bio-Inspired Computing

Nowadays, technology is rapidly evolving. Artificial intelligence(AI) is one of the fascinating technologies in the modern world. AI has made significant technological progress in face classification, speech recognition, strategic game, and decision-making [ Yao, et al., Nat. Commun., 8, 1, pp. 1-8, 2017; Jarrahi, Bus. Horiz., 61, 4, pp. 577-586,2018]. The processing of AI computing heavily depends on big data availability and requires a large amount of computation. Bio-inspired neuromorphic computing is expected to develop a more proficient computing architecture that mimics biological neural networks. Neurons and synapses are the two fundamental elements of neuromorphic architecture, where synapse plays an important role in learning and memory(C. Mead, Proc. IEEE, 78(10), 1629–1636, Oct. 1990). Memristors can become a potential candidate to behave as synapse due to their chemical compositions and electrical properties. In the last few years, photonic memristors have attracted increasing attention for AI systems. Such memristive devices have excellent capabilities to reduce the von Neumann bottleneck issue [Indiveri et.al., Proc. IEEE, 103, 8, pp. 1379-1397, 2015]. Additionally, integrated system consists of arrayed optical memristors to accept as building blocks of biospired vision system.<br/> <br/>This talk will present that the photonic oxide memristor has high potential for synaptic application. It will report the conduction mechanism of the memristor device and working principle of the synapse. The synaptic features including potentiation/depression, paired-pulse facilitation(PPF) and spike time dependent plasticity(STDP) will be briefly introduced. In our recent work, Zn₂SnO₄(ZTO) based memristor devices are fabricated. The SL device shows over 80% optical transparency for the entire visible region. Significant improvements in bipolar resistive switching properties of the device with low SET voltage and long DC endurance cycles are observed in the 200°C, N₂ annealed device. The linearity of such memristive synapse is improved for 350 training epochs with a total number of 175000 pulses. The STDP learning rule for the annealed device is demonstrated through the electric field. The optical sensing capabilities of this device including photonic potentiation, photonic PPF, learning experience behavior, and multilevel memory feature by the repetition of optical pulse are demonstrated under the blue light illumination. On the other hand, the linearity and on/off ratio of the optoelectronic synapse are further improved by employing ZTO/MgO DL device. Such device has improved reliability with stable endurance and synaptic characteristics. The nonlinearities of potentiation and depression of the device are 1.96 and 0.33, respectively. The device shows 300 training epochs with 300000 pulse numbers. The synaptic features of the ZTO-based memristors make it to be suitable for optoelectronic synaptic application