I-Corps: LabMate: Accelerated Empirical Process Optimization

今天，微米和纳米制造工艺的配方开发和优化通常基于耗时且昂贵的实验试验和错误。有些流程可能需要长达一年的时间才能创建和充分优化，严重限制了技术发展。像JMP这样的统计软件工具试图利用经典的实验设计(DoE)技术来降低配方创建和优化的高昂成本，但它们忽略了可能从对过程物理的理解中获得的信息，并且通常需要进行大量实验才能进行精确拟合。其他过程工具，包括Synopsys和Coventor，缺乏许多预测能力。缩短开发周期提供了一个明显的机会来节省时间和金钱，并使新的纳米技术成为可能。这个i-Corps团队发明了一种使用基于物理的模型和集成贝叶斯统计的方法，与经典的能源部相比，大大加快和降低了微纳米制造过程的经验优化成本。拟议的技术“Labmate”在建立在可靠的理论基础上的模型和实验之间采用迭代反馈。此外，Labmate使用户的知识和经验能够定量地融入到模型中，进一步减少优化时间，使其非常适合商业应用。在干法蚀刻配方开发中的初步应用表明，与能源部要求的实验次数相比，实验次数可以减少两到三倍。这意味着每年在蚀刻配方创建和优化方面节省数十万美元。重要的是，虽然最初的目标市场位于半导体领域，但所提出的技术可以很容易地转化为任何具有大量未知参数和有限实验数据的物理过程。该团队的目标是为LAM、英特尔、东京电子、Global Foundries、台积电制造公司和应用材料等公司优化干法蚀刻工艺的配方。最初的市场调查告诉团队，目前用于蚀刻配方创建和优化的少数模拟工具缺乏我们发明的灵活性和能力。该团队将采用基于订阅的模式，并提供客户支持，以将提议的软件商业化。

Today, recipe development and optimization for micro- and nanofabrication processes are typically based on time consuming and expensive experimental trial and error. Some processes may take up to a year to be created and fully optimized severely limiting technology development. Statistical software tools like JMP attempt to capitalize on classical design of experiment (DoE) techniques to mitigate this high cost of recipe creation and optimization, but they neglect information that might be gained from an understanding of process physics and often require large numbers of experiments for precise fits. Other process tools, including Synopsys and Coventor, lack many predictive capabilities. Shortening the development cycle offers a clear opportunity to save time and money and enables new nanoscale technologies. This I-Corps team has invented a methodology for using physics based models and integrated Bayesian statistics to dramatically speed up and reduce the cost for the empirical optimization of micro- and nanofabrication processes compared to classical DoE. The proposed technology "LabMate" employs an iterative feedback between a model constructed on a robust theoretical foundation and experiments. In addition, LabMate enables the knowledge and experience of the user to be incorporated quantitatively into the model to further decrease the time to optimization and making it well-suited for commercial application. Preliminary application of the proposed method to the development of dry etching recipes shows that the number of experiments can be reduced by a factor of two to three compared to the number of experiments required by DoE. This translates to hundreds of thousands of dollars in annual savings for etch recipe creation and optimization. Importantly, although the initial target market lies within the semiconductor space, the proposed technology can easily translate to any physical process with a large number of unknown parameters and limited experimental data. This team is targeting recipe optimization for dry etching processes at companies including LAM, Intel, Tokyo Electron, Global Foundries, Taiwan Semiconductor Manufacturing Company, and Applied Materials. An initial market survey taught the team that the few simulation tools that exist today for etch recipe creation and optimization lack the flexibility and capacity of our invention. The team will employ a subscription based model with customer support to commercialize the proposed software.