High-Efficiency Multiobjective Synchronous Modeling and Solution of Analog ICs

An artful complex algorithm to achieve high-efficiency high-precision multiobjective synchronous modeling and optimal solving of an analog IC is proposed. The complete approach combines response surface methodology (RSM) with an improved support vector regression (SVR) machine, which shows obvious superiority in nonlinear modeling and global solution by reduced training datasets. Targeting an analog comparator circuit as the experimental object, 33 sets of optimal design parameters are solved first in the RSM preprocessing stage, and these high-quality sample data can be subsequently input to the SVR machine for model training. In the second SVR solving stage, a differential evolution (DE) algorithm is adopted to further auto-optimize the modeling factors of the SVR machine to improve the accuracy of the nonlinear model. Finally, an SVR model of the comparator circuit can be obtained to accurately describe the response relationships of performance metrics and the corresponding design parameters that are expected by designers. With the comparator design based on TSMC 65 nm/1.2 V standard CMOS technology, we compare the SVR model-based prediction value with the Cadence-based simulation result by design test-bench and observe error lower than 3%. This work proved that the proposed complex RSM-DE-SVR algorithm is significantly practical and effective to benefit the aided optimization design of analog ICs.

Automated summary

This paper proposes a complex algorithm for high-efficiency high-precision multiobjective synchronous modeling and optimal solving of analog ICs. The algorithm combines response surface methodology with an improved support vector regression machine, and has been demonstrated to be practical and effective in the experimental verification.