A Compact SPICE Model for Carbon-Nanotube Field-Effect Transistors Including Nonidealities and Its Application—Part II: Full Device Model and Circuit Performance Benchmarking

TLDR

The study introduces a full circuit‑compatible compact model for single‑walled CNFETs, extending the work of Part 1. The model incorporates channel elastic scattering, source/drain resistance, Schottky‑barrier resistance, parasitic gate capacitances, supports multiple nanotubes per device, and is evaluated via HSPICE simulations against standard CMOS digital library cells. When benchmarked in HSPICE, the CNFET model yields a universal, circuit‑compatible implementation that outperforms 32‑nm silicon MOSFETs in intrinsic gate delay (6× for nFET, 14× for pFET), but interconnect capacitance reduces the speed advantage by a factor of five to eight; in realistic circuit simulations, CNFET logic with one to ten nanotubes per device runs two to ten times faster, consumes two to seven times less energy per cycle, and achieves a 15–20× lower energy‑delay product compared to CMOS.

Abstract

This paper presents a complete circuit-compatible compact model for single-walled carbon-nanotube field-effect transistors (CNFETs) as an extension to Part 1 of this two-part paper. For the first time, a universal circuit-compatible CNFET model including the practical device nonidealities is implemented with HSPICE. In addition to the nonidealities included in the companion paper, this paper includes the elastic scattering in the channel region, the resistive source/drain (S/D), the Schottky-barrier resistance, and the parasitic gate capacitances. More than one nanotube per device can be modeled. Compared to silicon technology, the CNFETs show much better device performance based on the intrinsic gate-delay metric (six times for nFET and 14 times for pFET) than the MOSFET device at the 32-nm node, even with device nonidealities. This large speed improvement is significantly degraded (by a factor of five to eight) by interconnect capacitance in a real circuit environment. We performed circuit-performance comparison with all the standard digital library cells between CMOS random logic and CNFET random logic with HSPICE simulation. Compared to CMOS circuits, the CNFET circuits with one to ten carbon nanotubes per device is about two to ten times faster, the energy consumption per cycle is about seven to two times lower, and the energy-delay product is about 15-20 times lower, considering the realistic layout pattern and the interconnect wiring capacitance.