An electro-photo-sensitive synaptic transistor for edge neuromorphic visual systems

The practical application of optoelectronic artificial synapses in neuromorphic visual systems is still hindered by their limited functionality, reliability and the challenge of mass production. Here, an electrophoto-sensitive synapse based on a highly reliable amorphous InGaZnO thin-film transistor is demonstrated. Not only does the synapse respond to electrical voltage spikes due to charge trapping/detrapping, but also the weight is modified directly by persistent photocurrent effects under UV-light stimulation. Representative forms of synaptic plasticity, including inhibitory and excitatory postsynaptic currents, frequency-dependent characteristics, short-term to long-term plasticity transitions, and summation effects, are successfully demonstrated. In particular, optoelectronic synergetic modulation leads to reconfigurable excitatory and inhibitory synaptic behaviors, which provides a promising way to achieve the homeostatic regulation of synaptic weights. Moreover, the analogue channel conductance with 100 states is used as the weight update rule to perform MNIST handwritten digit recognition, using system-level LeNet-5 convolutional neural network simulations. The network shows a high recognition accuracy of 95.99% and good tolerance to noisy input patterns. This study highlights the commercial potential of mature optoelectronic InGaZnO transistor technology in edge neuromorphic systems.