Compact Modeling of Variation in FinFET SRAM Cells

Abstract

POWER CONSUMPTION OF EMBEDDED SRAM is a significant concern in state-of-the-art processors. Lowering the supply voltage (V DD ) effectively saves power. However, V DD scaling for conventional SRAM is limited by the random variation of the device threshold voltage (V TH ), because of random dopant fluctuation (RDF) effects, which are expected to increase with technology scaling. On the other hand, high doping is not required for multigate devices such as FinFETs. 1 In such devices, V TH variation is smaller and V DD can be scaled to lower voltages. 2 These devices will likely be used in future CMOS technologies. In the absence of RDF, variation in V TH will originate mainly from the lithography-defined gate length (L) and fin thickness (T fin ). It is crucial that these variation sources in multigate devices are modeled.Numerical simulations based on finite-element methods or TCAD (technology CAD) tools are useful for technology evaluation and design exploration of FinFET-based SRAM cells (see, for example, the work by Guo et al. 3 ). Mixed-mode TCAD simulations can be combined with Monte Carlo simulations to predict the impact of device variation on circuit performance. Many simulations are needed, however, and such a task is time-consuming. An efficient compact model (or Spice model) such as BSIM-CMG is more suitable. 4 The model employs physical expressions to capture the effect of device parameters such as L and T fin on the electrical characteristics of multigate devices. Through proper parameter extraction, the effects of these variation sources can be captured.In this article, we present a procedure to model variation in FinFET SRAM cells using BSIM-CMG, which we used in a study for the design and optimization of a six-transistor (6T) FinFET SRAM cell. We review design considerations of FinFET SRAM cells and the advantages of multigate devices, and give a brief overview of BSIM-CMG. We explain the process of separating global and local variation components and of calibrating variation to data. SRAM design considerationsFour design metrics quantify the read/write stability and performance of an SRAM cell (see Figure 1a):A read static noise margin characterizes the read stability of the SRAM cell. The RSNM is defined as the side length of the maximum square that can fit inside the butterfly curve (see Figure 1b). We form a butterfly curve by plotting the voltage transfer characteristics of the two inverters in an SRAM cell when both the bit line (BL) and word line (WL) are biased at V DD . If the two squares inside the butterfly curve do not have equal side lengths, we define RSNM as the side length of the smaller square. Compact Variability Modeling for Nanometer CMOS TechnologyEditor's note: FinFET technology is a possible solution to achieve a better power/performance trade-off for SRAM cells. This article provides a comprehensive analysis of the variations in FinFET devices, their impact on SRAM stability, and a statistical design procedure for FinFET SRAM cells.

A novel stability and process sensitivity driven model for optimal sized FinFET based SRAM

“…We are optimizing IDG-FinFET 10T SRAM (PPN10T) cell in terms of device width (W) and length (L), fin-thickness (T Fin ), oxide thickness (T ox ) and leakage current (I lkg ) for SRAM stability. Lu et al discussed in [12], device-length is technology dependent parameter and deviation in it causes shift in performance and stability. We take experimental variation data for 32 nm FinFET technology using Synopsys Sentaurus Device simulator by taking 6r ¼ 10% variation from their nominal values.…”

“…We use short channel based BSIM-IMG model in which second-order effects such as high field mobility, short-channel effects, quantum-mechanical effects, and leakage currents are considered [30]. Variations in L; T Fin ; H Fin and T OX parameters influence device V TH and play a vital role to model stability of FinFET based 10T SRAM [12].…”

“…In this work only device-length and width are taken under consideration and surface orientation is taken as constant. To analyze stability due to variation, researchers mostly used statistical tools such as Gaussian distribution and boxed whisker plot [12]. To quantify stability of 10T SRAM cell along with varying device parameters, we have assumed 6r ¼ 10% variation [31] in device parameters from the nominal value, where r is the standard deviation.…”

“…The impact of varying device-channel-length (L) is represented with the help of box-whisker plot drawn using 5000 random values considering process variation as discussed in [12]. Five value summary of box whisker plot presents sensitivity of variation through Inter Quartile Range (IQR), median, mean, first quartile, third quartile, lower and upper whisker points (from 2% to 98% data).…”

“…Compact modeling of process variation in terms of stability is modeled by Darsen et al and they proposed a behavioral model based on BSIM-CMG model [12]. In this model, device dimensions (W/L) are optimized through number of fin, therefore, this considers only LER effect.…”