Scholarlink

<p>Atomic Layer Deposition (ALD) is critical for producing ultra-thin, precise coatings in nanotechnology, particularly in the semiconductor industry. Nonuniform gas flow in complex chamber geometries often causes coating inconsistencies, and a way of addressing this inconsistency is by optimization of the process parameter. This study introduces an AI-driven framework combining computational fluid dynamics (CFD), Self-Normalizing Neural Networks (SNNs), and Particle Swarm Optimization (PSO) to optimize porous baffle hole distributions, stabilizing gas flow and enhancing coating uniformity in horizontal ALD chambers. SNNs, which employ Scaled Exponential Linear Units (SELUs) for self-normalization, mitigate overfitting and improve generalization even with sparse CFD training data. Acting as surrogate models, SNNs enable rapid and accurate prediction of flow uniformity, significantly reducing computational costs compared to conventional machine learning models. Integrated into the PSO loop, this framework efficiently identifies optimal baffle designs for flow stability and uniform deposition. Experimental validation demonstrated an 85.9% improvement in flow uniformity and a 50% increase in Al<sub>2</sub>O<sub>3</sub> coating thickness uniformity compared to baseline chambers. These results confirm the efficacy of the proposed framework and its potential for broader application in manufacturing processes constrained by limited data, setting a benchmark for integrating AI-driven optimization in advanced manufacturing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing chamber systems for deposition processes in the semiconductor industry with deep learning framework: tackling small simulation datasets

  • 期刊论文

  • 发表日期：

    2025年2月27日

查看全文
  • Donggeun Park,
  • Hanbin Cho,
  • Changseob Kwon,
  • Youngyeon Ji,
  • Seunghwa Ryu

摘要

Atomic Layer Deposition (ALD) is critical for producing ultra-thin, precise coatings in nanotechnology, particularly in the semiconductor industry. Nonuniform gas flow in complex chamber geometries often causes coating inconsistencies, and a way of addressing this inconsistency is by optimization of the process parameter. This study introduces an AI-driven framework combining computational fluid dynamics (CFD), Self-Normalizing Neural Networks (SNNs), and Particle Swarm Optimization (PSO) to optimize porous baffle hole distributions, stabilizing gas flow and enhancing coating uniformity in horizontal ALD chambers. SNNs, which employ Scaled Exponential Linear Units (SELUs) for self-normalization, mitigate overfitting and improve generalization even with sparse CFD training data. Acting as surrogate models, SNNs enable rapid and accurate prediction of flow uniformity, significantly reducing computational costs compared to conventional machine learning models. Integrated into the PSO loop, this framework efficiently identifies optimal baffle designs for flow stability and uniform deposition. Experimental validation demonstrated an 85.9% improvement in flow uniformity and a 50% increase in Al2O3 coating thickness uniformity compared to baseline chambers. These results confirm the efficacy of the proposed framework and its potential for broader application in manufacturing processes constrained by limited data, setting a benchmark for integrating AI-driven optimization in advanced manufacturing.