Machine Learning + Decades of Experience
PDF Solutions通过将大数据基础架构和机器学习应用程序与数十年的制造和测试经验相结合,使企业能够从工业4.0获益,为客户提供量身定制的解决方案,以实现特定成果。
在大批量生产中得到验证
我们的Advanced Insights for Manufacturing(AIM)是一个可配置的知识型系统,该系统可从正在进行的计算和用户输入中学习,从而在大批量生产环境中快速做出明智的决策。 在过去的十年中,我们开发了一系列经过生产验证的AIM解决方案,这些解决方案利用我们的大数据和机器学习功能为制造、测试操作以及组装和封装领域的客户提供可观的投资回报率(ROI)。
AIM解决方案概述:
自适应特征诊断(ASD)
ASD是一个用于异常晶圆良率特征检测、分类和根本原因分析的自动化系统。 分析传入的新晶圆的空间特征,并将其分类,这些类别可使用基于用户输入的机器学习动态地进行细化,这种是我们称为协作学习。 对每种唯一的晶圆类别自动执行深入分析,并生成报告,以确认已知根本原因的再次出现或突出显示造成晶圆良率下降的新潜在来源。
投资回报率: 确定晶圆良率下降的根本原因,其确定速度比传统分析技术快5倍,并且可以在机器学习模型中获取专家知识以进行持续改进。
数据源: Wafersort / Binsort,PCM,LEH / WEH,计量学,缺陷,工具FDC
能力和效率提升(CEI)
CEI通过循序渐进的方法利用Exensio Analytics Platform内的设备性能跟踪(EPT)功能,通过匹配工具和腔室操作来优化OEE(总体设备效率)、晶圆厂产能和晶圆吞吐量。 从每个配方微步骤收集的数据可捕获工具之间任何的性能不匹配,然后通过对FDC数据的详细分析来消除这些不匹配。
投资回报率:瓶颈工具能力提高10%,效率/吞吐量提高20%以上,快速识别配方与设置以及设备硬件问题,这些问题会产生与生产能力模型的偏差。
数据源: 工具传感器FDC数据
Consumable Cost Reduction (CCR)
CCR解决方案利用Process Control模块(Exensio Analytics Platform的核心模块)收集的eBOM数据集,其中包括ERP,MES,EAM和FAC数据以及易损件、维护零件和材料以及化学和材料成分报告,以系统地减少材料消耗,优化使用并捕获材料成分偏移。 通过结构化的分析工作流可以识别性能不佳的设备、零件和供应商,从而指导用户完成优化过程。
投资回报率:减少材料消耗成本,减少良率和可靠性偏差,并优化零件和材料的使用
数据源: Consumable batch ID, Event data, Recipe ID, FDC data, PM information, Material Composition reports, MES data
早期失效检测(ELF)
ELF解决方案优化了产量和现场可靠性故障之间的质量成本权衡。经典的离群算法,例如“零件平均测试”(PAT),通常用于识别和筛选有早期寿命失效风险的零件。 Exensio的ELF超越了PAT,通过利用Exensio Analytics Platform的端到端数据库和基础架构提供了全面的die质量分级和风险分类解决方案。使用多种变量的机器学习方法分析从多个数据源生成的高级指标数,该方法可适应新信息的出现(例如8D报告,FA中发现的根本原因,其他RMA引入等)。
投资回报率:通过在晶圆上检测高风险芯片来防止质量和可靠性下降
数据源: Wafer Sort, Final Test, PCM, Burn-in, Returns, Defect, Metrology, LEH/WEH, FDC
Equipment Trouble Protection (ETP)
ETP is the next generation FDC solution for wafer fabs and assembly floors. Going beyond the standard approach of FDC data collection, feature selection and SPC alarm limits, ETP links FDC data with tool events and uses AI and ML to detect and classify abnormal equipment sensor traces into good vs. bad vs. unknown. The classification system adapts as users judge new signals and identify root causes, enabling fast issue detection and containment.
投资回报率:+ 1%DPW产量,+ 5%晶圆产量,+ 2%线产量,+ 2%工具利用率,节省工程FTE
数据源: 工具传感器FDC数据和设备活动
Intelligent Material Disposition (IMD)
A “Material Review Board” (MRB) is a common technique used to improve the quality of shipped product. The IMD solution dramatically reduces the manual work and frequency of human-induced variability in the MRB process. Automated workflows are implemented that capture the specific quality criteria of each customer’s business and product line to provide lot and wafer quality grading in minutes, rather than hours or days. Comprehensive analytics and full automation ensure uniform results and high quality decision making.
投资回报率:将lot处置的工程工作量减少了50%以上。 防止逸出,提高一致性,并确保晶圆配置的决策质量。
数据源: PCM/WAT, Wafersort/Binsort, Final Test
Smart Testing
Manufacturing complexity, advanced packaging technology, and high density chip designs conspire to drive up the cost of wafersort and final test. Exensio’s Smart Testing solution uses Machine Learning to find subtle signals in the massive data set associated with each product die and applies Artificial Intelligence to modulate test flows and achieve higher product quality at lower cost. The AI / ML approach identifies the highest quality die as candidates to skip expensive test insertions such as Burn-in, optimizing test cost while still meeting DPPM requirements. PDF can provide the machine learning algorithm, or you can provide your own. The system is designed for production operation, installed on your OSAT’s test floor “at the edge” for efficient, low latency operation, high uptime and minimal data loss.
投资回报率:将老化需求降低30-60%,根据测试的数量和成本,每年最多可节省数百万美元。
数据源: PCM/WAT, Wafersort, Final Test, (and Metrology, Defect, MES, and FDC data as available)
Yield Aware FDC
YA-FDC is a combination of technology and services that leverages Exensio’s “big data” platform to improve process variation, identify equipment conditions and sources of variability that influence functional and parametric yield, and set appropriate SPC limits with proprietary analysis and modeling techniques that identify critical parameters. Analysis is automated with reporting and dashboards to drive fast improvement yield, variation and excursions. AI / ML provides predictive models to for finer feedback and feed-forward control, predictive PM’s to optimize tool availability and Virtual Metrology for adaptive inline sampling.
投资回报率:良率提高8%,偏移降低40%,NPI斜坡学习速度提高7%
数据源: 工具传感器FDC数据,计量,缺陷,PCM / WAT,Wafersort,测试,组装