A Systematic Workflow for Intelligent Process Capability and ROI Simulation (2026-US-30MP-2931)

Welcome to 2024 JMP Discovery Summit, Exploring Data, Inspiring Innovation. Hi, my name is Pin. I'm the Senior SE from Global Technical Enablement team, based in Singapore. Together with me is Dr. Charles Chen, who is from Applied Materials. He is also a Master Black Belt and JMP program leader.

Today, we two together to present a topic, talk about intelligent process capability. Around this topic, we are going to present two big things. First thing is talk about a systematic workflow. The second part is talk about the ROI strategy. Let's first talk about the systematic workflow.

JMP streamline our analytic workflow from very left: the data access, to data analysis, use all kinds of analytics method to the right side: our insight sharing. Everything can be done in one single software. In between, the big part is JMP analytic capability. Today, we only focus about quality and the process.

Along our way for the process improvement, stability, capability, and lean are the three major focus at the different stage of our process. We focus about process capability. In order for us to improve capability, first thing we need to quantify it.

We can use some parameters like CP, CPK, and non-conformance. In short, we call it NC percentage to measure to quantify it. This is the curve that shows the CPK versus NC percentage, which you can see with the CPK increases, the NC percentage actually decreases a lot. Vice versa, for the yield performance will increase a lot.

From the chart, you see at the very first stage, when our CPK increases, our Sigma level increases from 1 Sigma to 2 Sigma, the NC percentage will drop from over 30% to below 5%. That's a great reduction of the NC percentage. NC percentage reduction actually saves the money we scrap at the very early stage.

This stage, what we focus on is the stability of the process and the capability of the process. We try to improve our CPK is equivalent to reduce NC percentage, save a lot of the scrap money at this stage. Then later on, you will see the curve start to shift.

When we reach a stage when our CPK is further improved, however, the NC percentage seems to come to a plateau. It's saturated. At this stage, we are more focus about the lean process. The CPK improvement is no longer our top priority. Our top priority is NC percentage reduction. Any type of NC percentage and any type of waste improves the overall efficiency.

With this picture in mind, with this mindset, let's begin with our process capability. We quantify it using a set of parameters, let's say short-term CP, CPK, long-term PP, PPK. However, for today's presentation, we do not separate them, we sometimes use them interchangeably. In most of the time, we do the process capability analysis for the normal process.

Normality need to fulfill, however, not always the case. When this time happens, we need to study why the process is not normal, what are the reasons, what are the causes causing the process by right should be normal, however, in actual, it's not normal. This is our Focus A.

Sometimes, our process is incapable. That could be due to the mean shift, missing target, central tendency of the mean. The second questions could be because of the Sigma, because of spread too big. These are the two things which can impact our CPK performance, which can be perfectly visualized in our JMP goal plot.

Let's take a look at them one by one. First, focus about the normality. JMP has a different type of normality test. When it does not meet the normality requirement, it could be many different other type of distributions. Let's just take a look a few examples. The first one is a Bimodal distribution. From visual display, you can very easily see it is Bimodal.

JMP will assess, compare different type of distribution and choose the best fit. In this case, it's a normal 2 mixture. On top of that, gave us overall Sigma capability and non-conformance calculation is based on this mixture of two normal distribution.

Dr. Charles based on this has his own additional few criteria to judge this bimodal distribution. Why he's doing this is simply because he wants to drive the root cause analysis and eventually for problem-solving. Even before he accepts the PPK calculation and NC calculation. He can choose the representative data coming from peak Number 1, coming from peak Number 2.

From there, he can carry on to do the root cause analysis by compare the input factors performance, which could be most influential for the two different peak. Eventually, he found the problem and drive the problem fixing.

This is for bimodal distribution where Dr. Charles has his additional requirement to drive. Then let's take a look at another one. Uniform distribution is quite let's say flat for this distribution across the range.

JMP will assess the distribution and fit a Johnson, from there carry on for the CPK and the NC percentage. While Dr. Charles will additionally look at the skewness look at the kurtosis. From there, he derives his own customized formula to come up the PPK and the NC percentage calculation.

So, the PPK will be relatively low as compare to the formula JMP give, and also the defect rate will be much higher if you compare to the JMP value versus the value come from Dr. Charles, since there are at least three times difference. There are actually no right, no wrong, just depends on which formula that you are using to come up the mean, come up the Sigma.

So, the moral of the story is based on your alpha or beta risk, see yourself where is your position in the whole business chain? Are you the customer? Are you the supplier? Which position are you in?

Develop that kind of formula which is the best suit your position to best present to precisely come out the quality, which is later on transfer to the cost control and also give you a very good output prediction. Eventually, for operation, we want to deliver just the good enough high-quality products and just the amount customer needed.

This is the whole moral of the story behind this uniform distribution. Just now, we share only two of that. Dr. Charles assess the visual display by looking at the normality, by derive his own set of formula come up the mean calculations, standard deviation calculation, provided a flexible guidance for such intelligent process capability.

Because eventually capability is just a parameter which is assess how capable is the process. Eventually, is we need to see through the CPK value, see through the NC percentage value to find the problem, to fix them, to provide the best quality. That is the end goal. Okay, let's now move on to the mean shift and Sigma too wide.

Both we have a parameter P and PP to assess how severe is the mean shift, how severe is the standard deviation, how bad is the performance. Dr. Charles apply their own stringent criteria. Not only using this ready-to-use formula, he even used an equivalence test to assess that versus a practical threshold.

We know a hypothesis difference doesn't guarantee the practical also difference, meaning something for the engineer, for the business. The reason behind the thing is both for PP and the K are just a point estimate, single point.

It's not an interval estimate. We know using interval estimate can give us much reliable direction of improve decision-making. That is the gist of us to make use of statistical discovery software like JMP to analyze the data, give you a much more reliable decision-making. That is the data-driven decision-making.

But of course, behind the scenes, you also need to have a good software driven by statistical discovery, and also the knowledge the people should use, should know how to use, how to combine stats, how to combine the software together to analyze the data, to give you the much more reliable direction guidance. That is the value.

To combine part A, normality, part B, which is to assess the mean shift, the central tendency, and also part C, which talk about the standard deviation, talk about the spread too wide, Dr. Charles develop his own systematic and intelligent workflow. Each part of this one, he goes very in-depth.

Not only this one, he even automates the entire work package using JMP Workflow Builder function. This is no-coding automation, greatly save the engineer effort, high efficiency. Not only this, the knowledge which is acquired from one person can be share with the entire organization, level up the entire company of all scientists and engineer competency.

I think that is the extreme value. To summarize this part, first talk about the process capability. Dr. Charles decompose into three big parts A, B, and C. Each part, we need to quantify, go very in-depth about that. Once we have the thorough understanding, we can develop a systematic workflow and automate the workflow. This is a continuous process sharing.

The end goal is we are trying to minimize the human dependency. Sometimes, some people, some engineer, they are lack of understanding about this one, but with a very mature setup system workflow to guide them, they are unlikely to make such severe mistakes. Now, let's move on to the ROI strategy.

How can we make further directions for improvement based on all what we learned? When we talk about the process capability, actually, what we are seeing is overall total process capability, because the true process cannot be measured purely, this one integrates of the measurement system. What we are talking about here actually is the overall total process capability.

By integrating our measurement system actually provide a holistic view for that. You can use the JMP potential graph. That's a graph under our Process Screening platform to visualize the plot of percentage of measurement Sigma square versus our process capability, CP or PP value.

Using this one, we have different guidance which area. This is a desired space. Your process, your measurement system is perfectly fine. Here, you can do the measurement improve measurement. On some place here, you may focus more about your process improvement. Using this one give you a very good guidance which direction you should go.

Let's take a look at one of the example coming using the sample data semiconductor capability. I focus one parameter called IVP5. From the calculation, the CPK is lower than 1, it's not good, we should further improve. Then later, we carry on by looking at the control chart.

From the control chart, first thing is that you see the mean actually is away from your target, a little bit higher, and also the set of calculate control limit is a little bit far from the both upper and lower spec limits.

The space between this upper spec limit to the upper control limit for both it's same for the lower side, it's a very narrow, that means the response time for the engineer is very minimum. What we want is a very wide range from the control limit to the spec limit, and the control limit should be within the spec limit.

This is a very good process. This is what other information we can derive by using a simple control charts. On top of that, JMP using the goal plot to visualize the standardized mean on the X. We can see this parameter IVP5 mean actually is somewhere on the right side versus the given target 0, which means on target.

This is the same as what we see over here. When you do the process improvement, you can first center your process if this is easier for you to do so. Then secondly is you can see the standard deviation according to a standardized 1, 0 are far away from that. This, it gives you a direction of improvement.

I may reduce the overall total Sigma or total variance, whatever you call it. JMP also gives you an additional tool which called Interactive Capability Plot to let you do simulations to answer those what-if questions. What if I simply want to reduce my Sigma from the original about 1.2, reduce 50% to current 0.6?

With that achievement, I can reach my PPK is about 1.67. It's a very good process capability. If we integrate the measurement system, what happens? Now, I'm just assuming the measurement system the standard deviation is 2 for all the parameters.

With that, I rerun the process screening JMP will give me additional two columns which talk about the measurement Sigma and the measurement portion. In this case, the measurement contribute 288% of the overall. Recall what we learned from MSA, the gauge should contribute less than 30% of the overall variance.

By this case, the measurement system actually contributes nearly 300 times. I believe now you change your focus. Instead of focus overall total Sigma variation, you can pinpoint the major things that you should focus that is to reduce the Sigma coming from measurement system itself. Give you a much more direct guidance rather than the overall pinpoint to the top one priority you should focus.

That is the value we can get from the potential graph. Using this one, JMP already tells you this IVP5 actually is far away from what usually we see. In the range, you should focus on the measurement system Sigma reduction rather than the overall. Give you a straight to the point direction guidance. With this, I believe you can see the whole picture of the elephant rather than just see a partial of the elephant trying to guess where you should go.

This is the value for us to make use of software to give you a much more efficient, highly achievable guidance or direction. When it comes to the how-to part, it's generally very easy. To the preference setup, just tick the Process Potential Graph, and go to the Column Property to set up your process screening as one of the property.

When you run the process screening, this measurement system will come as one additional column, giving you additional information. With such, you will have much holistic view by looking at the overall system rather than just focus one single part. Dr. Charles get inspired from this potential graph. He makes it even better by using the contour plot.

Contour plot actually display three variable in a two dimension. On the X, he uses measurement system Sigma. On the Y, he can come up the true process Sigma by using the formula here. The line actually is the third dimension using for PP or CP.

With such, you can clearly see the direction of higher CP or lower CP, where is the direction. Here, both are low. This is the good region we should maintain. Well, on this side, obvious measurement system needs to be further improved.

This is we need to focus about the process. While in between is both the process itself and measurement system need to be improved, reduce the Sigma. I believe this is a quite good visual way to display the three dimensions and provide guidance for continuous improvement direction. Furthermore, Dr. Charles quantified our improvement effort with the direction using a set of data to build this model.

First of all, we must have a set of spec conditions. For every process, we have the spec limits so that we have the CP, CPK calculations. Then we have the end goal. We know at the end what I want to achieve for the measurement system, what is the precision-to-tolerance ratio I want to get? What are the CP I want to get?

For this one, higher and better. For this one, of course, the lower, the better. We just address the Sigma issue, not talk about the mean shift issue at this moment. With the end goal, you can derive the end-goal process performance for both Gage R&R Sigma and the overall Sigma. Of course, you can derive the true process Sigma. We have our current status quo, which is our starting point, the baseline performance.

What are the baseline CP, CPK? What is the baseline P/T ratio? What is the baseline Gage R&R Sigma, and so on and so forth? From the current status quo to the end goal, what you want to achieve, in between there is a gap. The gap is what you need to achieve. This is the delta to achieve for both Gage R&R Sigma and also for the true process improvement Sigma.

By using this delta coming from the measurement system, delta coming from process-to-process improvement, Dr. Charles provides engineering guidance using this table. Let's say this delta is higher than 2, the focus will be the measurement system to improve first. If the delta is very low, our focus will be the true process improvement.

This provides us with a very good action priority. If we have limited resources, limited engineering effort, which direction should we focus first to get the result faster and quicker? Visualize the improvement effort by using this chart. On the X is our baseline Gage R&R Sigma. On the Y is the true process Sigma, of course, coming from the formula. This is the color coming from the visual using a log scale because sometimes the delta can be really huge.

Not only this one, provide a local data filter using the end goal, what are the CP you want to achieve, what is the P/T you want to achieve, then you know which direction should you go first, whether the gauge or the true process improvement. Of course, this is just at the very early stage, with more data, more people start to use, the model itself can be much more reliable.

But at least, this is the prototype give us a what-if scenario. If we have limited resources from the current status quo, we want to achieve this higher, which one should you prioritize? I think this one values much for our decision-makers, for our business owners. To summarize today's paper, we focus process capability.

This is a big topic, but we talk about two major things. One thing is how to set up a systematic workflow by decomposing and going very in-depth so that we can set up a systematic workflow and automate it to share with everyone, to level up the overall organization's competency. Secondly, is to address the ROI strategy. With the limited resources, which direction should you strictly go at a higher priority to achieve your end goal faster and quicker?

This one can help us to visualize and prioritize. Together, we can inspire each other to innovate to grow further. Thank you, everyone.