We describe an optimization strategy for minimizing total power consumption using dual threshold voltage (Vth) technology. Significant power savings are possible by simultaneous assignment of Vth with gate sizing. We propose an efficient algorithm based on
Minimization of Dynamic and Static Power Through Joint
Assignment of Threshold Voltages and Sizing
Optimization
David Nguyen, Abhijit Davare, Michael Orshansky, David Chinnery, Brandon Thompson, and Kurt Keutzer
University of California at Berkeley
{nguyendt, davare, omisha, chinnery, thompsbj, keutzer}@eecs.berkeley.edu
ABSTRACT
We describe an optimization strategy for minimizing total power consumption using dual threshold voltage (Vth) technology. Significant power savings are possible by simultaneous assignment of Vth with gate sizing. We propose an efficient algorithm based on linear programming that jointly performs Vth assignment and gate sizing to minimize total power under delay constraints. First, linear programming assigns the optimal amounts of slack to gates based on power-delay sensitivity. Then, an optimal gate configuration, in terms of Vth and transistor sizes, is selected by an exhaustive local search. Benchmark results for the algorithm show 32% reduction in power consumption on average, compared to sizing only power minimization. There is up to a 57% reduction for some circuits. The flow can be extended to dual supply voltage libraries to yield further power savings.
current of a gate but increases its delay. However, with careful optimization, power consumption can be reduced without penalizing the performance by only increasing the delay of gates that have timing slack. Having an additional Vth on the chip requires an extra mask, and while costly, this practice is justified in the production of high-volume high-end parts. Transistor sizing is also a powerful tool for power minimization. Gates with timing slack can be downsized leading to reduced area, capacitance, and power.
To achieve greatest power savings, the optimization algorithm must be able to simultaneously assign Vth to transistors and size them. This, however, is a difficult optimization problem. The discrete nature of Vth assignment and sizing makes the problem combinatorial. In addition, accurate models of power and circuit delay in terms of optimization variables are non-linear. Thus, in its most general formulation, the optimization problem is an integer (or mixed-integer, if we allow for a continuous range of transistor sizes) non-linear programming problem. Neither of these problems can be solved efficiently.
Previous research has studied Vth assignment separately from transistor sizing [11][12]. A simultaneous Vth and sizing optimization introduced in [7] uses sensitivity-based size updating, which is bound to lead to sub-optimal results. Another algorithm relies on a binary search approach but is very computationally expensive [8]. Previous approaches also did not minimize total power, limiting themselves to stand-by power minimization using dual Vth [11][12] . The reality is that leakage power is now a large component of total power even in the active mode of a chip’s operation [6][9]. The work in [8] does acknowledge the need to minimize total power, but it suffers from other drawbacks, such as the lack of global circuit view during Vth assignment.
Our approach explicitly attempts to reduce total power by a simultaneous optimal assignment of Vth and transistor sizing. The basic idea of our approach is to convert the power minimization problem into the problem of optimal slack assignment to gates in the circuit. However, the slack assignment is such that the overall power savings, due to Vth change and/or resizing, are maximized. Formulating the problem in this way casts the combinatorial assignment problem to a linear slack assignment problem, which is then solved efficiently.
The result of our method is sizing and Vth assignment such that delay on a particular gate is increased, but the final circuit meets timing constraints while minimizing power. By assigning slack to a gate, the slack of gates in the transitive fan-in and transitive fan-
Categories and Subject Descriptors
B.6.3 [Design Aids]
General Terms
Algorithms, Performance, Design
Keywords
Dual Threshold, Dual Supply Voltage, Sizing, Simultaneous
1. INTRODUCTION
In deep submicron technologies, minimization of leakage power becomes the dominant concern as we try to combat the increase in the overall circuit power consumption. At the 90nm technology node, leakage power may make up 42% of total power [4]. The primary reason for this increase in leakage power is the reduction of threshold voltage (Vth) of devices, and an exponential increase in leakage current that it causes.
A potent way to reduce leakage power consumption is to raise the Vth of some gates. A higher Vth reduces the subthreshold leakage
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ISLPED’03, August 25-27, 2003, Seoul, Korea.
Copyright 2003 ACM 1-58113-682-X/03/0008…$5.00.
热门文档