MEMS Design Synthesis Based on Hybrid Evolutionary Computation


A hierarchical MEMS synthesis framework based on hybrid evolutionary computation is presented. The synthesis framework integrates a design component library with a MEMS simulation tool (SUGAR) and a hybrid evolutionary computation algorithm that includes two levels of optimization: a stochastic global search method using a multi-objective genetic algorithm (MOGA) and a gradient-based local optimization technique. MOGA is an evolutionary algorithm that can potentially find a global optimal design in a search space with multiple local optima and copes with the simultaneous optimization of a number of objectives. MOGA requires a parametric encoding – genotype – of the evolving phenotype that is efficient yet flexible enough to describe new and creative solutions. To capture the hierarchical nature typical of many MEMS designs, a tree-structured component-based genotype representation for MOGA is developed, which is based on an object-oriented extensible design component library. All elements in the library encapsulate instructions and restrictions for genetic operations and are associated with their connectivity. This genotype representation implicitly encodes much domain- 2 specific engineering knowledge and allows meaningful and flexible genetic operations during the MEMS synthesis process, which helps the evolutionary process to converge to good design solutions more efficiently. A gradient-based local optimization technique is integrated at the end of the evolutionary process to further optimize and fine-tune promising designs with fixed design geometries. An interactive hybrid computation (IHC) algorithm is developed to improve the performance of hybrid evolutionary computation in MEMS design synthesis. The IHC process provides the flexibility for designers to influence the hybrid evolutionary process through human identification of good design patterns and the local fine-tuning of these designs. The advantages of the IHC algorithm are demonstrated using a series of resonator test cases. The MEMS synthesis framework has been successfully applied to surface-micromachined resonators and accelerometers. These test cases demonstrate how engineering domain knowledge can be efficiently integrated into the MEMS synthesis process and provide better designs in a practical computational time. The synthesized designs also show that the framework can create new design concepts in a computational effective manner.