Enhanced synaptic characteristics of H _x WO₃-based neuromorphic devices, achieved by current pulse control, for artificial neural networks

Abstract

Artificial synapses capable of mimicking the fundamental functionalities of biological synapses are critical to the building of efficient neuromorphic systems. We have developed a H x WO3-based artificial synapse that replicates such synaptic functionalities via an all-solid-state redox transistor mechanism. The subject synaptic-H x WO3 transistor, which operates by current pulse control, exhibits excellent synaptic properties including good linearity, low update variation and conductance modulation characteristics. We investigated the performance of the device under various operating conditions, and the impact of the characteristics of the device on artificial neural network computing. Although the subject synaptic-H x WO3 transistor showed an insufficient recognition accuracy of 66% for a handwritten digit recognition task with voltage pulse control, it achieved an excellent accuracy of 88% with current pulse control, which is approaching the 93% accuracy of an ideal synaptic device. This result suggests that the performance of any redox-transistor-type artificial synapse can be dramatically improved by current pulse control, which in turn paves the way for further exploration and the evolution of advanced neuromorphic systems, with the potential to revolutionize the artificial intelligence domain. It further marks a significant stride towards the realization of high-performance, low-power consumption computing devices.

Few- and single-molecule reservoir computing experimentally demonstrated with surface-enhanced Raman scattering and ion gating

“…Since it is not necessary to acquire the entire SERS signal for edge AI computing applications, only the signal intensities of specific wave number components used for information processing need to be detected with high accuracy, and the detected signal (reservoir state) can be digitally converted and transmitted to the existing computing infrastructure to perform learning/classification in the readout network. Furthermore, if the detected reservoir states are directly input as analog signals to the linear classifier consisting of an array of variable resistor elements such as memristors ( 68 , 69 ), and only the analysis results are transferred to the computing infrastructure, then further reductions in computational resources, power consumption, and communication costs can be expected. FM-RC is particularly suited to be combined with electrochemical processes ( 70 – 74 ) or nanoarchitectonic materials ( 75 – 79 ) so as to enhance complexity as a dynamical system.…”

“…The readout network and weights were stored and operated on a personal computer. It would be possible to physically implement a readout network by installing an array of programmable analog resistance-changing devices (artificial synaptic devices) such as memristors and redox transistors ( 68 , 69 ). For training the readout weights, we used the supervised data T ( k ), shown in fig.…”

A high-performance deep reservoir computer experimentally demonstrated with ion-gating reservoirs

“…The learning and the sum-of-product calculation performed in the readout network were done on a personal computer using current data obtained from the IGR. However, by utilizing artificial synaptic devices that reproduce weights by conductance, the sum-of-product calculation performed in the readout can also be calculated in a physical process, which is expected to further improve efficiency 49 – 58 .…”

Review of solid-state proton devices for neuromorphic information processing

“…In such structure, the gate bias is coupled to the semiconductor channel through the dielectric directly, which may be called three-terminal vertical geometry. [24][25][26][27][28][29][30][31] Also, oxide-based protonic/electronic hybrid transistors with in-plane-gate configuration were reported. [32][33][34][35][36][37][38] As shown in Fig.…”