Date of Award

5-5-2015

Document Type

Dissertation

Publisher

Santa Clara : Santa Clara University, 2015.

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical Engineering

First Advisor

Shoba Krishnan

Abstract

The problem of power densities in system-on-chips (SoCs) and processors has become more exacerbated recently, resulting in high cooling costs and reliability issues. One of the largest components of power consumption is the low skew clock distribution network (CDN), driving large load capacitance. This can consume as much as 70% of the total dynamic power that is lost as heat, needing elaborate sensing and cooling mechanisms. To mitigate this, resonant clocking has been utilized in several applications over the past decade. An improved energy recovering reconfigurable generalized series resonance (GSR) solution with all the critical support circuitry is developed in this work. This LC resonant clock driver is shown to save about 50% driver power (>40% overall), on a 22nm process node and has 50% less skew than a non-resonant driver at 2GHz. It can operate down to 0.2GHz to support other energy savings techniques like dynamic voltage and frequency scaling (DVFS).

As an example, GSR can be configured for the simpler pulse series resonance (PSR) operation to enable further power saving for double data rate (DDR) applications, by using de-skewing latches instead of flip-flop banks. A PSR based subsystem for 40% savings in clocking power with 40% driver active area reduction xii is demonstrated. This new resonant driver generates tracking pulses at each transition of clock for dual edge operation across DVFS. PSR clocking is designed to drive explicit-pulsed latches with negative setup time. Simulations using 45nm IBM/PTM device and interconnect technology models, clocking 1024 flip-flops show the reductions, compared to non-resonant clocking. DVFS range from 2GHz/1.3V to 200MHz/0.5V is obtained. The PSR frequency is set >3× the clock rate, needing only 1/10th the inductance of prior-art LC resonance schemes. The skew reductions are achieved without needing to increase the interconnect widths owing to negative set-up times.

Applications in data circuits are shown as well with a 90nm example. Parallel resonant and split-driver non-resonant configurations as well are derived from GSR. Tradeoffs in timing performance versus power, based on theoretical analysis, are compared for the first time and verified. This enables synthesis of an optimal topology for a given application from the GSR.

Download

Share

COinS