Treffer: Parallel computing for the topology optimization method: Performance metrics and energy consumption analysis in multiphysics problems.

• The topology optimization method is parallelized in order to use modern multi-core CPUs and many-core GPUs by using the MATLAB Parallel Computing Toolbox. • An electro-thermo-mechanical multiphysics problem is considered herein to show the potential of parallel computing in complex optimization problems. • An energy consumption analysis is performed for first time in the context of parallel computing applied to topology optimization. • The shorter time required to solve the topology optimization of multiphysics problems using parallel computing allows obtaining results in a reasonable time, increasing the resolution of the topology, including several physics in the same problem, and saving on energy consumption. The topology optimization method (TOM) is a valuable tool to obtain conceptual designs in many scientific fields. However, small-scale problems have traditionally been considered due to the high computational resources this method demands. For example, hundreds of costly optimization iterations are needed, in which millions of design variables are used and where simulation of complex multiphysics phenomena could be required. To address this difficulty, the computing capacity can be increased, or efficient code implementations can be used, or a combination of both. Herein, the computing capacity and efficiency are increased simultaneously by programming parallel codes for running on the central processing unit (CPU) and on the graphics processing unit (GPU). A multiphysics problem is used as the optimization application. Specifically, electro-thermo-mechanical (ETM) microactuators are designed by TOM. To achieve this goal, three computer code versions are developed: one optimized sequential code, another using the parallelism offered by the CPU, and a third one using parallel computing on the GPU. Typical code performance metrics such as the execution time and their acceleration are measured. Additionally, an energy consumption analysis is performed for the first time in the context of parallel computing for topology optimization, which is an important topic from large-scale supercomputers to laptops that seek energy-aware methods. The results show that topologies are obtained up to 25 times faster with up to 93% less power consumption when parallel computing is used. This time reduction in TOM allows increasing the topology resolution, the inclusion of multiple physics, and significant energy savings. [ABSTRACT FROM AUTHOR]