Principle hessian direction based parameter reduction forinterconnect networks with process variation

As CMOS technology enters the nanometer regime, the increasing process variation is bringing manifest impact on circuit performance. To accurately take account of both global and local process variations, a large number of random variables (or parameters) have to be incorporated into circuit models. This measure in turn raises the complexity of the circuit models. The current paper proposes a Principle Hessian Direction (PHD) based parameter reduction approach for interconnect networks. The proposed approach relies on each parameter's impact on circuit performance to decide whether keeping or reducing the parameter. Compared with existing principle component analysis(PCA) method, this performance based property provides us a significantly smaller parameter set after reduction. The experimental results also support our conclusions. In interconnect cases, the proposed method reduces 70% of parameters. In some cases (the mesh example in the current paper), the new approach leads to an 85% reduction. We also tested ISCAS benchmarks. In all cases, an average of 53% of reductionis observed with less than 3% error in mean and less than 8% error in variation.