Realistic worst-case modeling by performance level principal component analysis

A new algorithm to determine the number and value of realistic worst-case models for the performance of module library components is presented in this paper. The proposed algorithm employs principal components analysis (PCA) at the performance level to identify the main independent sources of variance for the performance of a set of library modules. Response surfaces methodology (RSM) and propagation of variance (POV) based algorithms are used to efficiently compute the performance level covariance matrix and nonlinear maximum likelihood optimization to trace back worst case models at the SPICE level. The effectiveness of the proposed methodology has been demonstrated by determining a realistic set of worst case models for a 0.25 /spl mu/m CMOS standard cell library.

[1]  Kurt Antreich,et al.  Improved Methods for Worst-Case Analysis and Optimization Incorporating Operating Tolerances , 1993, 30th ACM/IEEE Design Automation Conference.

[2]  Derek J. Pike,et al.  Empirical Model‐building and Response Surfaces. , 1988 .

[3]  Kurt Keutzer,et al.  Getting to the bottom of deep submicron , 1998, ICCAD '98.

[4]  M. A. Styblinski,et al.  Yield and variability optimization of integrated circuits , 1995 .

[5]  Zhihong Liu,et al.  Realistic worst-case SPICE file extraction using BSIM3 , 1995, Proceedings of the IEEE 1995 Custom Integrated Circuits Conference.

[6]  Carlo Guardiani,et al.  An Assigned Probability Technique to Derive Realistic Worst-Case Timing Models of Digital Standard cells , 1995, 32nd Design Automation Conference.

[7]  Sung-Mo Kang,et al.  An integrated approach to realistic worst-case design optimization of MOS analog circuits , 1992, [1992] Proceedings 29th ACM/IEEE Design Automation Conference.

[8]  Andrea Neviani,et al.  Impact of unrealistic worst case modeling on the performance of VLSI circuits in deep sub-micron CMOS technologies , 1998, IWSM. 1998 3rd International Workshop on Statistical Metrology (Cat. No.98EX113).