Design and analysis of VLSI clock and power distribution networks

Ruilin Wang, Purdue University

Abstract

The design and analysis of clock and power distribution networks are becoming more challenging with the continual VLSI processing technology scaling. The focus of this dissertation are the following four important aspects of the design and analysis of clock and power distribution networks: the design of the next-generation clocking scheme, the analysis and modeling of clock meshes and hybrid structures, the statistical timing analysis of clock meshes, and the stochastic analysis of power networks. First, a multi-GHz on-chip clock generation and distribution scheme based on traveling wave oscillators with wave reflection and regeneration (R 2TWO) is proposed. It takes advantage of the voltage wave traveling, reflection, regeneration and energy recycling properties of transmission lines based on the loop formed with an inverter and a transmission line. The simulation and measurement results show that it can generate and distribute 6GHz of clocks with low power, low skew and low jitter, and that it also scales well with process scaling. Second, a frequency-domain technique of analyzing clock meshes or hybrid structures is proposed. The circuit topology is reduced with the transfer matrices manipulations in the frequency domain. The linear equations are formed and solved in the frequency domain, and the voltage signal waveforms, as well as the arrival times, can be obtained. The experimental results show that the proposed technique is as accurate as HSPICE transient simulations, and is 2 orders of magnitude faster than HSPICE transient simulations. Third, a statistical timing analysis technique for clock meshes is proposed. The proposed technique is based on the observation that the first and second order derivatives of the clock mesh signal arrival times with respect to the parameters of clock meshes and the inputs signal arrival times can be easily derived analytically in the frequency domain. The problem of performing statistical timing analysis can then be converted into that of solving a set of linear equations. The second order timing models can thus be established and the means and the variances of the arrival times of the clock mesh can be obtained. The experimental results show that it is much faster than the Monte-Carlo-based statistical timing analysis, and the relative errors of the means and the variances are less than 3%. Last, a stochastic analysis technique of power networks is proposed. It is based on the facts that the correlation functions, convolutions and power spectra are closely related, and that samplings and convolutions can be easily applied in both the frequency domain and the time domain. Based on the decomposition of the currents and voltages signals into random and non-random parts, the correlation functions of node voltages are calculated from the power spectra obtained in the frequency domain. The experimental results show the accuracy and efficiency benefits of the proposed frequency-domain technique for stochastic analysis of power networks.

Degree

Ph.D.

Advisors

Koh, Purdue University.

Subject Area

Computer Engineering|Electrical engineering

COinS