Perspective on oxide-based three-terminal artificial synapses in physical neural networks

Abstract

The physical implementation of artificial neural networks, also known as “neuromorphic engineering” as advocated by Carver Mead in the late 1980s, has become urgent because of the increasing demand on massive and unstructured data processing. complementary metal-oxide-semiconductor-based hardware suffers from high power consumption due to the von Neumann bottleneck; therefore, alternative hardware architectures and devices meeting the energy efficiency requirements are being extensively investigated for neuromorphic computing. Among the emerging neuromorphic electronics, oxide-based three-terminal artificial synapses merit the features of scalability and compatibility with the silicon technology as well as the concurrent signal transmitting-and-learning. In this Perspective, we survey four types of three-terminal artificial synapses classified by their operation mechanisms, including the oxide electrolyte-gated transistor, ion-doped oxide electrolyte-gated transistor, ferroelectric-gated transistor, and charge trapping-gated transistor. The synaptic functions mimicked by these devices are analyzed based on the tunability of the channel conductance correlated with the charge relocation and polarization in gate dielectrics. Finally, the opportunities and challenges of implementing oxide-based three-terminal artificial synapses in physical neural networks are delineated for future prospects.

Binarized neural network of diode array with high concordance to vector–matrix multiplication

“…In the diode array, p + , n + , and gate electrodes were connected to the input lines (ILs), output lines (OLs), and weight lines (WLs), respectively. WLs and OLs were parallel to each other and perpendicular to ILs 37 – 39 . The multiplication of V IN and the conductance (i.e.…”

“…5 b. The uniformity of the component diodes in the array can minimize the BNN computational errors 39 – 41 42 . Thus, a massive array architecture comprising gated p + -n-p-n + diodes can be realized with a high ratio of the bistable current magnitude to uniformity.…”

Super-additive interaction of homo- and heterosynaptic plasticity in a hot electron transfer optosynapse for visual sensing memory and logic operations

“…Optoelectronic synaptic devices based on transistors are particularly advantageous compared to two-terminal memristors, as they offer high device stability, synergistic integration of different stimuli, and long-term retention characteristics. [8][9][10][11][12] Amorphous oxide semiconductors (AOSs) have been explored for the emulation of visual perception, as they exhibit inherent persistent photoconductivity (PPC) due to the energy needed for the recombination of ionized oxygen vacancies and photoexcited electrons. [13][14][15][16] However, it requires sufficient photonic energy (43 eV) or lighting time for the ionization of oxygen vacancies located near the conduction band of AOSs to exhibit the PPC effect in the devices, 17,18 leading to high-energy consumption.…”