Abstract: Silicon anisotropic etching simulation is time-consuming due to its complex chemical process. In order to accelerate the process, the GPU-based silicon anisotropic etching parallel simulation was discussed, and an improved parallel method was proposed to improve the direct parallelization of serial algorithms method with low efficiency. In this method, the parallel load of the direct parallelization was increased, the memory management cost was reduced, and the acceleration performance was improved. The experimental results show that the method can achieve a simulation with higher speedup.

Key words: anisotropic etching, Cellular Automaton (CA), simulation, Graphics Processing Unit (GPU), parallel computation

摘要： 硅各向异性腐蚀过程复杂，采用元胞自动机模拟硅各向异性腐蚀非常耗时。为了加速腐蚀模拟过程，研究了基于图形处理器(GPU)进行硅的各向异性腐蚀模拟。针对串行算法直接并行化方法存在加速效率低等问题，提出了一个改进的并行模拟方法。该方法增加了并行部分的负载，减少了内存管理的开销，从而提高了加速性能。实验证明该方法能够获得较理想的加速比。

关键词: 各向异性腐蚀, 元胞自动机, 模拟, 图形处理器, 并行计算