A new TCAD-based statistical methodology for the optimization and sensitivity analysis of semiconductor technologies

Abstract

A new TCAD-based statistical methodology for the optimization and sensitivity analysis of semiconductor technologies has been developed and demonstrated on a 0.18-/spl mu/m SOI CMOS process. Two new screening techniques applicable to deterministic systems (Lenth's test and normal probability plots) were introduced and compared with correlation analysis. The graphical nature of the new techniques provided easier analysis of the screening results by clearly displaying which process factors surpass predefined significance limits. A multiresponse steepest ascent analysis was developed to locate regions of improved process performance before beginning response surface experimentation. To perform the analysis, a composite function representing the response criteria was constructed using desirability functions and incorporated within the steepest ascent methodology. Locating the region of improved performance allowed smaller experimental designs to be used for the response study significantly improving model accuracy. The response models were used to optimize the SOI CMOS process and perform sensitivity analyzes on both the baseline and optimized processes. Optimization resulted in a 15% increase in I/sub dsat/ without violating any other criteria. The results of the sensitivity analyzes, which showed the greatest benefit from the increased model accuracy, indicated no conspicuous device performance degradation caused by anticipated manufacturing variations.