Uncertainty Quantification of CMOS Active Filter Circuits: A Non-Intrusive Computational Approach Based on Generalized Polynomial Chaos

Mecit Emre Duman,Onder Suvak

doi:10.1109/access.2020.3031215

Mecit Emre Duman, Onder Suvak

Open Access

https://doi.org/10.1109/access.2020.3031215

Copy DOI

Abstract

Semiconductor fabrication technologies as applies to the nanometer-era paradigms of nowadays have rendered uncertainty quantification analyses through component-level parameters compulsory and indispensable. Frequency responses of CMOS active filters are invariably observed to be affected by probabilistically modelled parameter deviations, and in this article the focus is on the fast and accurate quantification of the uncertainties pervading CMOS active filters in terms of their magnitude frequency responses. Previous work dominantly has preference for the widely recognized non-intrusive Monte Carlo methods, which bring about a disproportionately high computational burden. Also discomfitures are observed to arise due to apparently inadequate ensemble volumes and a limited variety of distribution functions that are chosen to be utilized, along with seemingly insufficient means of resulting data visualization and the lack of accurate probabilistic quantification. Generalized Polynomial Chaos (gPC) based stochastic spectral techniques, which usually offer reduced computational complexity with respect to Monte Carlo, targeting CMOS active filters do not seem to have drawn much attention; the few related publications offer utility in a limited scope of electronic components. In this article, we carry out uncertainty quantification analyses in order to compute partial or approximate stochastic characterizations of the magnitude frequency responses of several multi-component CMOS active filter circuits with the gPC-based stochastic collocation technique. The pertaining inherent non-intrusive nature is exploited, and the stated issues associated with the previous work are addressed. We utilize a stokhos-based MATLAB/C++ toolbox of our own design, on whose software architecture, features, and facilities we provide profound details, and present performance comparisons with Monte Carlo along with intuitive and insightful comments, in an endeavor to suggest that such observations may prove to be beneficial to circuit designers.

Highlights

Fabrication of short channel transistor devices and operation through lower supply voltages became possible thanks to the advent of novel semiconductor manufacturing technologies
THE Generalized Polynomial Chaos (gPC)-BASED APPROACH IN THIS ARTICLE In this article, we demonstrate the utility of gPC-based stochastic collocation as a fast uncertainty quantification technique on the magnitude frequency responses of several CMOS active filter circuits constructed with integrated circuits and passive components
One must make a note of the fact that while omitted in this article in the interest of space, indispensable pieces of information such as internal structures of the DVCC, VDTA, and DDCC circuit structures suitable to be utilized in integrated circuits, dimensions of the embedded CMOS transistors, and nominal values of the utilized passive components are included for reference, and for purposes of reproducability, in the original publications alluded to, where the relevant filter structures are introduced

Summary

INTRODUCTION

Fabrication of short channel transistor devices and operation through lower supply voltages became possible thanks to the advent of novel semiconductor manufacturing technologies. GPC-based stochastic collocation technique is one that is highly preferred due to its non-intrusive nature, i.e., similar to Monte Carlo methods, stochastic collocation is able to utilize almost any circuit simulator as a black box In this technique, each of the sets of random values for the parameters, at which the system will be solved and which as a set is commonly termed as a node, is determined by stacking together the roots of those orthogonal polynomials associated with the probability distributions of the selected random parameters, upon which the analysis is founded, according to the tensor product [11] or sparse grid [11] rules.

BACKGROUND

ANALYZED CIRCUITS

SIMULATION SETUP AND DURATIONS

NUMERICAL RESULTS

CONCLUSION