机器学习驱动的优化碳化化学机械抛光与表面粗性限制的优化
Hujingyue Wang1, Zhen Qin2, Lihong Zhang1
1School of Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, China.
Langmuir : the ACS journal of surfaces and colloids
|January 13, 2026
概括
一个可解释的机器学习框架通过平衡材料去除率 (MRR) 和表面粗度 (Ra) 来优化碳化 (SiC) 抛光. 这种方法可以确定高级半导体制造的最佳处理窗口.
科学领域:
- 材料科学 材料科学 材料科学
- 化学工程是化学工程的重要组成部分.
- 人工智能的人工智能
背景情况:
- 在碳化 (SiC) 上通过化学机械抛光 (CMP) 实现纳米尺度的表面表面要求高的材料去除率 (MRR) 和亚纳米表面粗度 (Ra).
- 对CMP过程的经验优化是具有挑战性的,因为它具有复杂的,结合的 tribochemical 效应.
- 现有的方法往往难以有效地确定SiC的最佳加工参数.
研究的目的:
- 开发一个可解释的机器学习 (ML) 框架来优化SiC CMP.
- 确定处理变量与CMP结果 (MRR和Ra) 之间的关键关系.
- 建立可操作的过程窗口,同时实现高MRR和低Ra.
主要方法:
- 一组设计的50个CMP实验使用氧化 (CeO2) /过氧化 (H2O2) 泥.
- 训练和对六个基线和两个增强的ML模型进行基准测试,确定具有多项式特征扩展 (5-MLP-PFE) 的多层感知子组合作为表现最好.
- 使用部分依赖,SHAP和RadViz分析来解释模型预测并了解变量重要性.
主要成果:
- 5-MLP-PFE模型实现了高预测准确性 (MRR的R2 ≈ 0.94,Ra的0.91).
- 机械参数 (压力,速度) 主要影响MRR,而泥化学 (H2O2,CeO2,pH) 则决定可实现的Ra极限.
- 实验验证证证实了ML推的设置,产生Ra ≤0.13nm,MRR高达230.8nm/h.
结论:
- 开发的ML框架为SiC CMP优化提供了一种可重现和可解释的方法.
- 确定了可操作的流程窗口,使得CMP设置的逆向设计能够实现所需的结果.
- 机器学习方法显著缩短了配方搜索时间,并且可以转移到其他半导体抛光工艺.
更多相关视频
08:48Selective Area Modification of Silicon Surface Wettability by Pulsed UV Laser Irradiation in Liquid Environment
Published on: November 9, 2015
8.6K
08:02Rendering SiO2/Si Surfaces Omniphobic by Carving Gas-Entrapping Microtextures Comprising Reentrant and Doubly Reentrant Cavities or Pillars
Published on: February 11, 2020
9.3K
相关概念视频
01:14Mechanical Efficiency of Real Machines
The mechanical efficiency of a machine is a fundamental concept that describes how effectively a machine can convert input work into output work. According to this concept, the efficiency of a machine is equal to the ratio of the output work to the input work. An ideal machine, meaning a machine that has no energy losses, has an efficiency of one. This implies that the input work and the output work are equal.
However, in reality, no machine can be truly ideal, and all of them experience some...
However, in reality, no machine can be truly ideal, and all of them experience some...
01:28Methods of Medium Optimization
Optimizing growth media enhances microbial proliferation and maximizes product yield. Statistical experimental design methodologies provide structured and reproducible approaches, offering progressively higher levels of robustness and efficiency.The One-Factor-at-a-Time (OFAT) MethodThe One-Factor-at-a-Time (OFAT) method involves adjusting a single variable while keeping all others constant. However, it cannot detect interactions between variables, often leading to suboptimal outcomes when...