Variability-Aware VLSI Design Automation For Nanoscale Technologies
MetadataShow full item record
As technology scaling enters the nanometer regime, design of large scale ICs gets more challenging due to shrinking feature sizes and increasing design complexity. Aggressive scaling causes significant degradation in reliability, increased susceptibility to fabrication and environmental randomness and increased dynamic and leakage power dissipation. In this work, we investigate these scaling issues in large scale integrated systems. This dissertation proposes to develop variability-aware design methodologies by proposing design analysis, design-time optimization, post-silicon tunability and runtime-adaptivity based optimization techniques for handling variability. We discuss our research in the area of variability-aware analysis, specifically focusing on the problem of statistical timing analysis. The first technique presents the concept of error budgeting that achieves significant runtime speedups during statistical timing analysis. The second work presents a general framework for non-linear non-Gaussian statistical timing analysis considering correlations. Further, we present our work on design-time optimization schemes that are applicable during physical synthesis. Firstly, we present a buffer insertion technique that considers wire-length uncertainty and proposes algorithms to perform probabilistic buffer insertion. Secondly, we present a stochastic optimization framework based on Monte-Carlo technique considering fabrication variability. This optimization framework can be applied to problems that can be modeled as linear programs without without imposing any assumptions on the nature of the variability. Subsequently, we present our work on post-silicon tunability based design optimization. This work presents a design management framework that can be used to balance the effort spent on pre-silicon (through gate sizing) and post-silicon optimization (through tunable clock-tree buffers) while maximizing the yield gains. Lastly, we present our work on variability-aware runtime optimization techniques. We look at the problem of runtime supply voltage scaling for dynamic power optimization, and propose a framework to consider the impact of variability on the reliability of such designs. We propose a probabilistic design synthesis technique where reliability of the design is a primary optimization metric.