Hello. I'm Neha Kaushal, and my co-author, Srivida, is with me today. We work for applied materials in tech consultancy and do data analytics on semiconductor process tools. In this study, we are going to talk about how we utilize JMP dashboard for optimizing tool maintenance in semiconductor processes.

In the semiconductor industry, because of ongoing customer demand for lower-cost devices, tool log data analysis is important for efficient tool usage, deploying tech-enabled services with JMP visualization allow us to become more efficient in responding to preventive maintenance event for process tools.

Minimizing the tool time during maintenance, which is also called PM, enhances tool performance, which means we have better PM services, which results in better yield, and higher customer satisfaction. Analyzing the two logs using JMP distribution, box plot, etc., help to focus on the problem areas during the maintenance events.

Our objective for this study are to identify the improvement opportunities in production down-time during different types of PMs. For this process, we have minor and major maintenance events. Considering the first-time PM success or failure, the maintenance events are categorized into four types. Minor with first-time PM success, major with first-time PM success, then minor with first-time PM failure, and major with first-time PM failure.

The first step in this methodology is to prioritize the tools based on the production downtime percentage by using JMP Graph Builder. Let me show you that. For example here, we have plotted tool downtime percentage with tool IDs on the X-axis. Here we can quickly check and identify that tool A has highest downtime percentage. This could be our top priority tool for improvement.

Then we compare the PM types based on PM duration that finds opportunities for process optimization. Then the distribution models are on PM duration to check if it fits normality, otherwise, study normality violation factors like outliers, skewness, etc. As data deviates from normality, we have applied a non-parametric approach to perform hypothesis study to check variation among four types of PM events. Then finally, we have created a quick and interactive dashboard for analyzing tool downtime with PM performance.

Now, I will hand over to my co-author, Srivida, and she will explain the results part.

We will do the methodology and result in this section. The PM duration data is explored in distribution platform and the summary statistics show mean is greater than median as well as skewness is high. The normal quantile plot also verifies dataset is right-skewed with significant outliers.

Accordingly, the P-Value of goodness of fit test on the fitted normal distribution was violated, while P-Value for the goodness of fit test on the fitted Cauchy distribution is passed. So our data is a Cauchy distribution.

Accordingly, we have analyzed the presence of outliers using the robust fit outlier detection method and found 11 outliers for root cause analysis. Here, the PM types are compared based on PM duration by non-parametric analysis to find opportunities for PM process optimization. Based on the P-value, all the three models shown here, the Wilcoxon method, median test, Van der Waeden test, conclude significant difference among the PM types.

Further, from non-parametric comparison for each pair using Wilcoxon method, we concluded that minor PM success and minor PM fail have significant difference based on the P-value. Why? Major PM success, Major PM fail is not significantly different based on the P-value.

Now, we will see the JMP demonstration to create a dashboard for quick reporting on the dependency of PM success and failure on PM types. Let me take you to the JMP dashboard part. We can create dashboard from the file, new dashboard option, and select the platform and take our hypothesis test, box plot, etc., to create a dashboard. This is the dashboard we have created from our analysis. Using this dashboard, we can explain the dependency of PM success and failure on PM types interactively by selecting one tool at a time.

From the study, we conclude, first-time PM success has high impact on PM type. We conclude first-time PM success has high impact on minor PMs, while first-time PM success has negligible impact on major things. The median PM duration of first-time PM success and fail are equivalent.

Next step is tool-wise analysis to identify the scope of improvement. We would like to acknowledge our stakeholders in the projects to help for the Oasis support. The Charles C. Chen, Chandrasekhar Roy, Vikas Jangra, and Sidda Reddy Kurakula.

Okay, Elodie, so this will be the end of our presentation. If you have any question, we can answer that.