Circuit-level modeling of FinFet sub-threshold slope and DIBL mismatch beyond 22nm

Four-injector variability modeling of FinFET predictive technology models

“…In [4] the sets of four injectors were generated to reflect exactly the same variability as the 1000 model cards used as a reference in a one-to-one association. This was useful to validate the accuracy of four-injector variability modeling but its application were limited as the statistical compact model were necessary.…”

“…This was useful to validate the accuracy of four-injector variability modeling but its application were limited as the statistical compact model were necessary. We show here how injectors can be randomly generated based on the correlations obtained in [4]. This allows us to apply mismatch to technologies with no variability information available, to scale the mismatch magnitudes based on the device areas to figures reported in the literature as well as generating any arbitrary number of Monte Carlo points.…”

“…It was shown in our previous work [4] that a four-injector approach, adding two supplementary power sources to model DIBL and sub-threshold slope mismatches as shown in Figure 1 enhanced the accuracy of Monte-Carlo analysis when simulating the key performance and stability metrics of an SRAM cell, as well as the yield of a whole memory.…”

“…Notably, the voltage source that models threshold voltage is directly linked to the coefficient AVt, usually reported for a Fig. 1: Injectors used to model (a) threshold voltage and drainsource gain (two-injector approach), (b) sub-threshold slope and (c) DIBL proposed in [4] and their respective effects in the l-V curves of the transistors.…”

“…In this work we apply the four-injector variability modeling to predictive technology nodes from 20nm down to 7nm using randomly generated injectors, opposed to the injectors generated to fit already known statistical compact models used in [4], and compare the results to those obtained using a twoinjector only approach, also randomly generated.…”