Simulation Study of Dominant Statistical Variability Sources in 32-nm High- $\kappa$/Metal Gate CMOS

Drain bias effects on statistical variability and reliability and related subthreshold variability in 20-nm bulk planar MOSFETs

“…This skew is even more prominent, and its spread is larger, at high drain bias. The skew is caused by the threshold--voltage asymmetrical roll--off characteristics at the nominal design point [8]. As clearly shown in Figure 7, the roll--off slopes become large with increasing drain bias.…”

“…The GSS atomistic simulator GARAND is deployed for both the uniform and the atomistic simulations through this study [9]. Its drift--diffusion module features density gradient quantum corrections carefully calibrated against experimental data [8]. The 'uniform' or nominal n--MOSFET achieves a drive current of 1.35 mA/µm and leakage current of approximately 100 nA/µm.…”

“…The geometry, body bias and temperature impact on statistical variability for the 20--nm bulk planar technology was presented in [6]. To the best of our knowledge, there are few systematic studies of the impact of the drain--bias on the statistical variability and reliability of bulk MOSFETs [6][7] [8]. In this paper the impact of drain--bias on bulk MOSFET threshold--voltage and subthreshold variability is studied in detail, taking into account the individual impact of all relevant major statistical variability sources: random discrete dopants (RDD), line edge roughness (LER) and metal gate granularity (MGG), as well as the random interface trapped charges (ITC) responsible for the bias temperature instability degradation.…”

An Adaptively Segmented Forward Problem Based Non-Blind Deconvolution Technique for Analyzing SRAM Margin Variation Effects

“…The guard band (GB) design for static random access memory (SRAM) is expected to become a critical challenge because the increased time-dependent (TD) margin variations (MV)-caused failures cannot be predicated any more by only the convolution analyses [1,5]. This is because (1) TD-MV, (i.e., unknown MV after being shipped to the market), will become much larger than the non-TD-MV, (i.e., given MV based on the measurements), resulting in the TD-MV dominating over the overall MV.…”

“…This is because (1) TD-MV, (i.e., unknown MV after being shipped to the market), will become much larger than the non-TD-MV, (i.e., given MV based on the measurements), resulting in the TD-MV dominating over the overall MV. This leads to a rapid increase in pressure to figure out the unknown factors by solving the inverse problem [6,8], even though SRAM designers are unfamiliar with such a methodology and (2) the tail distribution of the convolution results of the TD-MV and non-TD-MV no longer shows Gaussian behavior but more complex mixtures of Gamma distributions [3][4][5][6][7], as shown in Figs. 1(a) and (b).…”