论文部分内容阅读
450mm晶圆刻蚀机开发中大量应用确定性仿真来模拟腔室内部物理、化学环境,并通过仿真结果指导装备结构的详细设计。为控制仿真试验的采样规模以缩短开发周期,本文详细介绍一种新型的基于采样密度和非线性度的序贯设计方法。此方法通过蒙特卡洛方法,在设计空间中获得采样密度信息,进而对低采样密度区域增加采样点。另外,通过对每个采样的领域进行发掘,以获得采样的梯度和非线性度信息,进而对高度非线性的区域增加采样点。以450mm刻蚀机约束环设计模型和Goldstein-Price模型为背景,采用拉丁超立方和新型序贯设计方法同时采样,以代理模型精度和特征捕捉能力两个角度来对比采样结果的优劣,结果证明,在达到同样精度的前提下,新型序贯设计方法能有效减小采样规模,符合刻蚀装备设计的需要。
The development of 450mm wafer etching machine uses a large number of deterministic simulations to simulate the physical and chemical environment inside the chamber and guides the detailed design of the equipment structure through the simulation results. In order to control the sampling scale of the simulation experiment to shorten the development cycle, this paper introduces in detail a new sequential design method based on sampling density and nonlinearity. This method uses Monte Carlo method to obtain the sampling density information in the design space, and then adds the sampling points to the low sampling density area. In addition, by sampling the area for each sample to obtain the gradient and non-linearity of sampling information, and thus to a highly non-linear region to increase the sampling point. Taking 450mm etching machine design model and Goldstein-Price model as background, we use Latin hypercube and new sequential design method to sample simultaneously, and compare the advantages and disadvantages of sampling results with the accuracy of agent model and the ability of capturing feature. Prove that, to achieve the same accuracy premise, the new sequential design method can effectively reduce the sampling size, in line with the design of etching equipment needs.