Amorphous metal-oxide semiconductor thin-film neuron, synapse, and neuromorphic system.

Artificial intelligence will spread exponentially and requires compact size and energy efficiency, for which one solution is neuromorphic systems. On the other hand, thin-film semiconductor technologies have spread widely and feature as giant-micro electronics. In this study, an amorphous metal-oxide semiconductor (AOS) thin-film neuron, synapse, and neuromorphic system are researched and developed. First, an AOS thin-film transistor (TFT) is fabricated and evaluated experimentally, and the experiment results are reproduced by a simulation model. A thin-film neuron composed of the AOS TFTs is designed, fabricated, and confirmed to work in a circuit simulation and experimentally, and this behavior is desirable for the thin-film neuron. Next, an AOS thin-film memristor and AOS thin-film Schottky diode are fabricated and evaluated experimentally, these characteristics are desirable for them, and the experiment results are reproduced by a simulation model. Thin-film synapses composed of the AOS thin-film memristors and Schottky diodes are designed, fabricated, and confirmed to work in a circuit simulation using simulation models of the AOS TFT and AOS thin-film Schottky diode, and the simulation results are reproduced by a fitting model. Lastly, a neuromorphic system composed of the thin-film neurons and synapses is designed, fabricated, and confirmed to work in a logic simulation. Training is executed using circuit-aware training based on the circuit behavior and training images of the MNIST, inference is performed, and the recognition accuracy achieves 94.1%. Finally, it is concluded that the AOS thin-film neuron, synapse, and neuromorphic system are feasible.