Accurate standard cell modeling is significant for circuit timing analysis and yield estimation. With voltage decreasing to near-threshold, cell delay distribution becomes asymmetrical and has a longer tail due to various process variation effects. In this brief, we propose a novel statistical model to fit the shape of the cell delay by using log-extended-skew-normal (LESN) distribution. It estimates the values of mean, standard deviation, and $3\sigma $ delay precisely by matching the kurtosis of the cell delay distribution. By changing the parameters of the LESN model, it can be used for up to the normal voltage region (1.1V) and down to the sub-threshold voltage region (0.4V). Considering the effect of load capacitances, the parameters of LESN distribution are further modeled so that the model can be used for different fanout constraints. Tested by INV, NAND2, and NOR2 with 28 nm technology, the average error of estimating the skewness is less than 6%, while the error for the kurtosis is less than 4%. Meanwhile, the average errors of $-3\sigma $ and $+3\sigma $ delay estimation are about 1.27% and 1.82% from 0.4V to 1.1V, respectively.
15 September 2021 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
The advent of autonomous power-limited systems poses a new challenge for early design space exploration. The existing architecture-level power evaluation tools lose accuracy due to ignoring features of circuit-level behaviors and influences of process, voltage, and temperature variations. Although power estimations based on SPICE or PrimeTime PX (PTPX) are accurate enough, they come at the cost of long simulation time and are available only in very late phases of design flow. In this article, a fast and accurate dynamic power evaluation method is proposed, which estimates activity factors at the circuit level. The impact of process variation at the gate level is considered through the proposed effective capacitance model. Activity factors are then estimated by the model and input vectors of the circuit. Input vectors are generated by architecture-level simulations in the form of streaming. For real-time and high-speed power evaluation, a data streaming framework is proposed for massive parallelism. The cross-layer estimation is verified based on the functional units of PULPino processor running SPEC CPU2006 benchmarks. Compared with the SPICE results using SMIC 28-nm PDK, our cycle-by-cycle dynamic power analysis shows an average error of 5.4%. Meanwhile, our approach realizes 65.2% faster than the traditional PTPX simulation and 48.8% faster compared with the state-of-art cross-level evaluation method.
As transistor size continues to scale down, process variation has become an essential factor determining semiconductor yield and economic return. The Liberty Variation Format (LVF) is the current industrial standard that expresses statistical timing behaviors based on single Gaussian model. However, it loses accuracy when the timing distribution is non-Gaussian due to growing process variations. This paper proposes a novel LVF2 distribution model that combines two weighted skewed-normal (SN) distributions, which better captures the multi-Gaussian timing distribution while maintaining backward compatibility with LVF. Experiments using TSMC 22nm standard cells show that, compared to LVF, LVF2 reduces binning error by 7.74X in delay and 9.56X in transition time, and reduces 3σ-yield error by 4.79X and 7.18X in delay and transition time, respectively. The error reduction for path delay is diminished due to Central Limit Theorem (CLT). But it is still 2X for a typical circuit path with 8 Fanout-of-4 (FO4) inverter delays.
In the advanced technology nodes, process parameter variations are increasingly resulting in unpredictable device behavior. The issue is even aggravated by low power requirements which stretches the transistor operation into near-threshold regime. Despite device simulation gives precise results, it is time-consuming for static timing analysis and dynamic timing analysis. In this paper, we propose VASTA, a statistical timing analysis tool based on the variation-aware standard cell library. The tool efficiently supports statistical static timing analysis (SSTA) and statistical dynamic timing analysis (SDTA). The standard cell library models delay under operating environment effects by using quadratic regressions and multivariate adaptive regression splines. VASTA works on industry formats (. $v$ and. $sdc$ ) and is designed to run in parallel during both SSTA and SDTA. The statistical cell library is built and verified using SMIC 40 nm and 28 nm PDK. The mean and standard deviation errors of cell delay models are 2.77% and 1.68% compared with SPICE simulation results under 10k Monte Carlo samples. The SSTA and SDTA are tested with ISCAS85, ISCAS89, and EPFL benchmark suites. The average mean and standard deviation errors of SSTA are 4.01% and 2.03%, which are similar to SDTA. Meanwhile, our SDTA is 17.7 times faster than traditional corner-based dynamic timing analysis which relies on generating. $vcd$ cell activity files.
With technology scaling progressing well into deep nanometer region, the number of technology corners surges from dozens to hundreds. Timing closure at Engineering Change Orders (ECO) stage is becoming increasingly challenging and time-consuming. Existing methodologies mainly focus on improving efficiency by predicting the full-corner timing metrics based on known corners. However, the escalating demand of known corners essential for complete timing deductions significantly prolongs the application of these methods. In this work, we propose a framework called FACT, to fast and accurately predict full-corner timing metrics for timing closure optimization circle. Our approach simplifies the process by necessitating timing analysis under only one known corner. Moreover, our framework seamlessly integrates with commercial EDA design flows, making it practical in industrial environments. Experimental results on open-source designs indicate superior stability of our method. Additionally, our approach achieves a significant runtime speed-up over previous ML-based timing ECO flows.