Circuit Optimization Using Efficient Parallel Pattern Search

Circuit optimization is extremely important in order to design today's high performance integrated circuits. As systems become more and more complex, traditional optimization techniques are no longer viable due to the complex and simulation intensive nature of the optimization problem. Two examples of such problems include clock mesh skew reduction and optimization of large analog systems, for example Phase locked loops. Mesh-based clock distribution has been employed in many high-performance microprocessor designs due to its favorable properties such as low clock skew and robustness. However, such clock distributions can become quite complex and may consist of hundreds of nonlinear drivers strongly coupled via a large passive network. While the simulation of clock meshes is already very time consuming, tuning such networks under tight performance constraints is an even daunting task. Same is the case with the phase locked loop. Being composed of multiple individual analog blocks, it is an extremely challenging task to optimize the entire system considering all block level trade-offs. In this work, we address these two challenging optimization problems i.e.; clock mesh skew optimization and PLL locking time reduction. The expensive objective function evaluations and difficulty in getting explicit sensitivity information make these problems intractable to standard optimization methods. We propose to explore the recently developed asynchronous parallel pattern search (APPS) method for efficient driver size tuning. While being a search-based method, APPS not only provides the desirable derivative-free optimization capability, but also is amenable to parallelization and possesses appealing theoretically rigorous convergence properties. In this work it is shown how such a method can lead to powerful parallel optimization of these complex problems with significant runtime and quality advantages over the traditional sequential quadratic programming (SQP) method. It is also shown how design-specific properties and speeding-up techniques can be exploited to make the optimization even more efficient while maintaining the convergence of APPS in a practical sense. In addition, the optimization technique is further enhanced by introducing the feature to handle non-linear constraints through the use of penalty functions. The enhanced method is used for optimizing phase locked loops at the system level.