Problems With the Continuous Doping TCAD Simulations of Decananometer CMOS Transistors

In this paper, we compare results from atomistic and continuous simulation of decananometer scale CMOS transistors. We study the behavior of important figures of merit, including threshold voltage, subthreshold slope, OFF-current, and ON-current. We provide physical explanation for the origin of the discrepancies between the averaged values obtained from the statistical simulations and the results from the continuous doping simulation. Based on our analysis, we clearly demonstrate that there are increasing errors in the doping distributions when device TCAD simulations are calibrated using continuous doping profiles. This questions the use of continuous doping profiles in the routine calibration and TCAD-based optimization of decananometer scale CMOS transistors.

I. Introduction

With the scaling of the CMOS transistors to decananometer dimensions atomic scale effects start to play increasingly important role in their behavior [1]. It is well documented in both numerical simulations [2] and experimental studies [3]–[5] that the discreteness of charge and granularity of matter introduces unavoidable, purely statistical variations in the transistor characteristics, which progressively increase with the reduction of the device dimensions [6], [7]. Among the major sources of statistical variability are: 1) the discreteness of dopants [2] and trapped charges at the interface and in the oxide [8]; 2) the granularity of the gate oxide [9] and the gate material [7]–[10]; and 3) the line edge roughness resulting from the molecular structure of the photoresist and the corpuscular nature of irradiation [11]. Increasing variability in bulk CMOS associated mainly with random dopant fluctuations has mediated the introduction of revolutionary new transistor architectures that tolerate low-channel doping, including FinFETs [12], [13] and fully depleted Silicon on Insulator transistors [14].