Top Gate Overlap Carbon Nanotube Transistor Electronic Synapses Arrays for High-Performance Image Recognition

Abstract

Carbon nanotube field-effect transistor (CNTFET) electronic synapses is of great potential for brain-like neuromorphic computing thanks their low power consumption. However, the modulation of diverse biological synaptic plasticity of CNTFET remains a significant challenge due to small dynamic range, abrupt conductance modulation and limited hardware structure. In this work, we developed a top gate overlap structure carbon nanotube field effect transistor (TGO-CNTFET) with a large dynamic range, and successfully mimicked synaptic functions, including excitatory and inhibitory synaptic behaviors (EPSC/IPSC), paired-pulse facilitation (PPF/PPD), and spike-timing-dependent plasticity (STDP). We further investigated two groups of transistors, comparing the synaptic performance of as-fabricated and the air-annealed device arrays. The ideal dynamic range of STDP and low power consumption per spike (1.27 pJ) were achieved with annealed TGO-CNTFET. Ultimately, the TGO-CNTFET and the air-annealed TGO-CNTFET synaptic transistor achieved a high accuracy rate of 90.8 % and 93.2 % in an image recognition task on the CIFAR-100 database using ResNet 50, respectively. This work introduces an architectural strategy for developing neuromorphic computing systems that incorporate functional oxides as dielectric layers.