Journal Home Online First Current Issue Archive For Authors Journal Information 中文版

Frontiers of Information Technology & Electronic Engineering >> 2016, Volume 17, Issue 2 doi: 10.1631/FITEE.1500168

An efficient bi-objective optimization framework for statistical chip-level yield analysis under parameter variations

1. Technology Innovation Center, Jiangsu Academy of Safety Science and Technology, Nanjing 210042, China
2. Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks, Nanjing University of Posts and Telecommunications, Nanjing 210013, China
3. School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China

Received: 2015-05-25 Accepted: 2016-02-02 Available online: 2016-02-26

Next Previous

Abstract

With shrinking technology, the increase in variability of process, voltage, and temperature (PVT) parameters significantly impacts the yield analysis and optimization for chip designs. Previous yield estimation algorithms have been limited to predicting either timing or power yield. However, neglecting the correlation between power and delay will result in significant yield loss. Most of these approaches also suffer from high computational complexity and long runtime. We suggest a novel bi-objective optimization framework based on Chebyshev affine arithmetic (CAA) and the adaptive weighted sum (AWS) method.Both power and timing yield are set as objective functions in this framework. The two objectives are optimized simultaneously to maintain the correlation between them. The proposed method first predicts the guaranteed probability bounds for leakage and delay distributions under the assumption of arbitrary correlations. Then a power-delay bi-objective optimization model is formulated by computation of cumulative distribution function (CDF) bounds. Finally, the AWS method is applied for power-delay optimization to generate a well-distributed set of Pareto-optimal solutions. Experimental results on ISCAS benchmark circuits show that the proposed bi-objective framework is capable of providing sufficient trade-off information between power and timing yield.

Related Research