Carbon nanotube FET for spiking neural networks

Abstract

The increasing demand for Spiking Neural Networks (SNNs), characterized by high power consumption and the need for high speed, has rendered conventional silicon-based Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs) incompetent for spiking applications. Carbon Nanotube Field-Effect Transistors (CNTFETs) have emerged as a promising alternative to replace MOSFETs and achieve significant energy efficiency benefits without compromising the speed performance compared to silicon-based FETs. This research paper begins by addressing the limitations of MOSFETs in the context of SNNs, highlighting their inefficiency in addressing the specific demands of neuromorphic hardware. The investigates then proceeds to examinate the advantageous attributes of CNTFETs over MOSFETs. The examination delves into the device electrical properties, mechanism, power and speed performance, power-delay product (PDP), and bandwidth capabilities. Recent findings in power and delay characteristics of CNTFET circuit designs are also reviewed and compared to those of MOSFET-based in this article. The study demonstrates the practicality, potentials, and scaling challenges of CNTFETs as a viable solution for fulfilling the stringent requirements of spiking applications, showcasing their ability in providing the optimal speed and power performances. The insights gathered from this research hold great significance implications for the future development of SNN-specific hardware beyond complementary metal-oxide semiconductor (CMOS) electronics.