An open software platform to bolster IC manufacturing yield

文章：Yuji Minegishi

As the pandemic exacerbated what was already arguably a formidable supply chain problem, the semiconductor industry continued—as it has for years—to focus intensely on discovering ways to increase efficiency, precision, and yield. The pandemic and rapid consumer buying enthusiasm simply functioned as an accelerator to hammer home what the industry already knew: swift advances are required that facilitate increased productivity.

Among the most promising approaches for achieving higher yield within the semiconductor industry is greater reliance on software and collaboration throughout the industry to cohesively integrate fragmented and siloed processes that inhibit progress.

While software for equipment monitoring, problem troubleshooting, data analysis, and report generation is in place throughout the industry, the software solutions are typically either home-grown or based on an assortment of solutions from multiple vendors. As the resources available to any given manufacturer are limited, it can take months to install new features and enhancements. Even after repeated requests, the new software capabilities often never actually appear. With solutions from multiple vendors in place, the onus is on the IC manufacturer to contact each individual vendor, update all of the software at play on the production line, and then hope that everything is compatible and works seamlessly together.

根据最近一项调查的参与者的说法（Figure 1), professionals in the semiconductor manufacturing chain want tools that are easier to use, more intelligent, able to support all their equipment, can automate the process of reporting and alerting, and are verifiably secure. Yet when manufacturers attempt to integrate equipment monitoring performance on their own, the results can be lackluster.

Figure 1A民意调查conducted by Gigaphoton provides insights into the current state of the semiconductor manufacturing industry.

Prescriptive analytics

最大的罪犯包括分析工具，这些工具具有基本的描述性和诊断功能，但没有有效地解决预测性和规范性维护。预测维护使用统计和建模技术来根据历史数据来确定将来可能发生的事情，并尝试确定机器何时需要服务。它不做的是确定应采取什么行动。它只是通知用户需要维护。在行业4.0的背景下，预测性维护方法正在慢慢集成到制造业中。

相比之下，规范性维护虽然不那么普遍且发达良好，但分析的档位更高。它不仅可以预测设备中的故障事件，还可以推荐动作。如果预测维护提供信息以确定是否执行或推迟资产维护，则规定维护提供了一套选项和结果。将来，规范性方法甚至可以安排维护并分配最佳的技术人员或团队以进行维修或更换。

由人工智能（AI）工具提供支持的规范性分析甚至可以通过在部署补救措施时提供详细说明来帮助维护团队。借助典型的晶圆厂内的大量设备和传感器，两种方法都可以改善每个制造过程，因为它们可以最大程度地减少某些重要组成部分可能会失败或降级，然后才能及时采取行动。

Figure 2从描述性到规定的分析工具提供了四个可能的见解级别。来源：Gigaphoton

共同的分母

What’s interesting about this scenario of predictive versus prescriptive maintenance is that while every semiconductor company’s processes are different and proprietary, they have many common problems. To find a solution, it would make sense to have a single software platform that would allow all equipment manufacturers to collaborate. That is, it would give chip manufacturers the ability to leverage the experience of others who have solved similar problems.

Of course, such a platform must ensure that participants’ intellectual property (IP) would be inherently secure, and similar efforts in other industries show that this goal can be achieved. With this assurance in place, the platform would allow cross-vendor equipment monitoring and analysis to be performed while enabling each vendor to solve their specific problems. The ability for customization would be essential to satisfy the unique needs of each participant. The platform also should be agnostic to various programming languages and accommodate all or at least the most prevalent languages. Another benefit of the platform would be more effective use of AI integration within critical processes.

这种协作方法最好在一个单个开放平台上完成，该平台消除了依靠来自多个（通常不兼容的）产品的需求。可能会减少停机时间和服务调用，因为制造商可以快速，轻松地自行确定哪种工具引起了问题，并且可以确定某些问题可能是跨职能的。该方法将显着提高效率，并产生这可以提高生产率。

简而言之，软件工具的进步对于半导体行业加速推广市场已经至关重要。但是，这些工具的大量在内部开发和由第三方供应商提供，也已成为该行业试图实现的目标的障碍。正如本文所表明的那样，利用所有这些工具的有效方法是在竞争对手之间共享有关它们的信息，而不揭示“秘密酱”，这是每个公司成功的核心。

A single software platform

半导体行业目前的监测和分析工具的分裂阻碍了其应对供应链中断产生的挑战的能力，这些挑战几乎肯定会在未来再次出现。新兴解决方案FABSCAPE, a new open platform imagined by the laser manufacturer Gigaphoton, provides actionable insights that were never identified before.

Figure 3Fabscape的渲染是一种新的开放平台概念，旨在将流线的数据监视，管理和分析整合到整个生产线上。

这种开放的协作平台解决方案并不新鲜，甚至国防部（DOD）越来越多地利用它们来鼓励与私营部门进行更大的融合。尽管国防部倡议面临着早期的障碍，因为公司担心自己的IP会受到损害，但保证会缓慢地，使他们更适合与竞争对手以及与政府合作。他们成功的关键是协作，信任和互惠互利，并且没有理由为什么半导体行业无法使用类似的东西。

In summary, to increase productivity on an industrywide scale, semiconductor manufacturing would be well served by a platform where all analytics, fault detection, optimization, and communication occur in a single place and with the depth of insight that only vendor-specific intelligence could provide.

This article was originally published onEDN.

Yuji Minegishi is the general manager at Gigaphoton Inc., a manufacturer of light sources for semiconductor lithography processes. He leads the engineering and development efforts for Fabscape.