The gate length size impact on the performance of carbon-nanotube-based 1T1R configuration for neuromorphic computing

Abstract

Benefiting from its excellent scalability, the carbon nanotube (CNT) has emerged as a promising material for advanced transistors to drive the resistive random-access memory (RRAM) in a one-transistor-one-resistor (1T1R) configuration. However, very few reports have investigated how the size of the CNT field-effect transistor (CNTFET) affects the performance of 1T1R. In this work, the gate length (L_G) size impact on the electrical behavior of 1T1R has been investigated. It is found that the current of CNTFET modulated by different L_G can significantly affect the switching behavior of the integrated RRAM. The 1T1R device with smaller L_G exhibited superior switching reliability but inferior ratio performance. This trend may be attributed to the reduced conductive filaments (CF) in the RRAM cell originating from the low SET current induced by the smaller-channel CNTFET. As the L_G decreased, the relative contribution of parasitic resistance to total device resistance increased. Hence, the gate control of CNTFET is weakened, leading to a degradation of the driving current. The amount of oxygen vacancies (V_OS) caused by the electric field in the integrated RRAM is thus reduced. Therefore, the CF constriction becomes narrower, resulting in less distinguishable conductance states associated with a smaller switching window. The small amount of migrated V_OS results in fewer damage sites and improved switching reliability. Furthermore, the long-term plasticity of the 1T1R configuration has been verified by pulsed testing. This work paves the road for further optimization of 1T1R devices based on CNTFET and their applications in large-scale neuromorphic computing arrays.