Hello everyone. I'm Neha Gupta, and I'm working in the Applied Materials, India Private Limited. A brief background before I start this presentation. It's like I am a Process Engineer, which works for this R&D project, research and development project in the applied material.

For this current study where the title goes with the review, the Change of the Process Parameter with the Hardware Layer Change is one of the study, what we have done. Before starting the presentation, I would like to thank JMP team who has given me a chance to talk about this presentation, this work.

With that, I'll start my presentation. As I mentioned that I am in Applied Materials India Private Limited, which is the semiconductor industry, and it has the expertise of atomic layer coating. What we tried to do is we use this atomic layer coating on the adjacent market field, which is pharmaceutical field.

What we tried to do is we took our expertise of atomic layer coating on the pharmaceutical particle and tried to improve their properties, the pharmaceutical vertical particles properties. With that, we have already a proof of concept. We have already shown on the small scale that if we will do the ALC coating on the particles, we can improve its flowability, its bioavailability, and so on.

Now the task is we have to do the scale up. For current, since before going more into the JMP detail, I would say this is what we have done is, we use the JMP as a tool to explain our technical area, our technical expertise, our technical finding, and JMP has supported us in that. As I mentioned that for this particle coating, we can improve the properties of the particle.

Our ultimate product, our ultimate goal is the coating. That coating quality should be good. If the coating quality is good, we can do much better thing. This coating quality is also impacted by the multiple parameters. In the JMP, we have a very good tool to show that which is called as a cause and effect diagram.

As a layman in person, if I want to show something that this is my final product and this is going to be impacted by the multiple parameters or multiple factors, how I will show that? This is the one thing. This is the caused-effect diagram where I have shown that my coating quality is impacted by my process, my agitation of the powder, because this is a high surface area.

Just conventional ALD is on the fill, not on the powder. In the powder, the surface area is much, much higher than the thin fill. In that case, the process parameter, the PADL impact, agitation of the powder, how you are putting the chemistry in for the reactions, how you are getting out the residuals, and then how this distance between these two are playing the role, and what is the temperature impact on whole of the chemistry.

These are a few parameter which is impacting the coating quality. Now, if I will say it is a para, nobody can understand. But if I will show this diagram, cause and effect diagram, this is a much better way to explain it, which we have used right now.

Coming back to my original point, that coating quality is impacted by the six parameters. Right now, we have focused on the headspace because process has already been set. We have already shown for this small scale as a proof of concept. Our process is already set. Now, the part is the hardware. Now, how the hardware will play the role?

Let's say you have a particle sitting on the base, and you have the gaze delivery injection system, or you have a chemistry which is going into it. If it is taking so much long to reach that, or it is getting the less time, that will impact on your process parameters, your pressure, your residual removal, everything will change with this distance. That distance is basically we are testing with our test stand.

We used to call this tool as a test stand where we can use the different architectures. This architecture is basically the lead design in the chamber where we will put or reduce the lead design headspace with the first lead and this which is the hollow one that can increase the headspace size. If more the headspace, higher the chemistry, less the residual removal may be, may not be. That we have to check.

For that, with that concern, we have started to see the lead design impact on the coating quality. With this, we have collected the data for this particular thing. While we have run this recipe, when we are trying to coat some particle, we have some parameter Parameters to give in, and that parameters we have recorded, and we have used for the analysis.

In that, the first thing is the pressure change. If the distance will be higher, maybe if there is more reaction can happen, if the sitting time is the longer the residential time for the chemistry will be higher, that pressure change will be higher.

That can be seen with the normality check, this pressure change between two leads, what we have taken. We have seen if there is with respect to cycle, if we are increasing the cycle, either it is increasing a lot or it is increasing the less. For that in the JMP, what we have used is the normality check and the box clots to identify the outlier.

This is the beauty of JMP. You cannot just only detect the outlier, you can get the root cause of this as well by fitting Y by X. This Y by X, you can fit the linear polynomial fit, what relation your outlier has with your own data set. With that, we can recognize that, or we can identify what is the root cause for the outlier.

The third step was [inaudible 00:06:14], where we have compared these two design data sets, what we have collected through. Then we have confirmed our hypothesis, what we have technically says that, that either it is right or how it will be go on. With that, I will come back to my results.

At very first, as I mentioned that we have two designs. In the first design, it is the higher headspace, which is being known as the design, which we said that as a design one. The another one is the design two, which you can see here where the data is more like uniform or normal. You will say that there is a very less differentiate.

But here, if you will see in the design one and if you will corresponding box plot, if you will see there is a lot of outlier, you can see that there is also. Now, again, coming back to the point, while we have done the outlier test, which is a very powerful tool in the JMP, where we can change the q-factor and change the tail quantile to check or the compact your spec range, that what you are going to say is your outlier. Either it is a marginal outlier or it is a process outlier or it is an extreme outlier.

In our case, we are looking for the extreme outlier. This extreme outlier, we could figure out in the design one only, not in the design two, which has the less headspace. That less headspace doesn't show any outlier. That discussion we will take later on.

But before that, I would say with the extreme outlier, we have just tried to take the root cause. We tried to fit with the Y by X linear fit, polynomial fit, different degrees in the polynomial fit. But what we could figure out is nothing is working in these two particular reasons, neither the linear ray fit or the polynomial fit. Then we came across that there is one more point called as the nonparametric cluster density analysis, which is the significance.

Although it is a nonparametric, so the confidence interval will be 95% compared the parametric. That is fine. But right now, that shows for the current study is that there are two different models because there are two different clusters, as you can see here. These two clusters are different from each other, and this cluster is basically going to tell you that there are two mechanism is going on.

If I will connect this with my technical theory, that will be like initially the process supposed to be increased linearly or with a separate slope, but then it has been moved. Overall, the process has been drifted. This is what our technical concern. As a technical person, if you will think that in the chamber, if you have the higher head space, that means if your distance is too much from the powder to your removal of the chemistry or the injection of your chemistry, if it too much high, the residential time for the chemistry and the residuals will be very high, which can impact not in the initial cycle, but on the later on cycle, it can impact a lot.

It can drift towards a complete process because that is the built-up process. One layer will build, then second layer will build, and the third layer will build. If this is the layer-wise structure, your first is having a lot of condensation, then the upper will be drifted a lot. This is the hypothesis what we have taken. With that, when we have plotted with the help of the control chart, which is again one of the amazing tools where we have seen the moving range with respect to the cycle.

If you will see here in the moving range, initial cycles, if you will see the initial cycle is at the different level, and suddenly there is a sudden change, actually, sudden drop I would say, sudden drop in the moving range, which again confirms the drift in the process.

Basically, the hypothesis and this data overall suggested that process started normally. Process started normally with the first few cycle. But since because there is a lot of headspace, residential time is higher, so that is why this process has been drifted. That is why it is giving a lot of issues in the coating quality. When we compared this with the ANOVA, this is our experimental data is also saying the same.

This is the statistical proof for the same hypothesis what mechanism is going on. Although coating is getting worst in case of design one, but what is the mechanism? To prove this mechanism, that process has been drifted is what we can see as well with that moving range chart.

At last, what we did is we compared this two leads with the one-way ANOVA and just try to see, statistically, are they same or the different? In the statistical point of view, while we have done the one-way analysis, these two are very far away from each other. That analysis said that our hypothesis that these two variants are same failed. That is why I would say these two leads are not similar.

Overall, the conclusion of this whole of the study, JMP, we have used as a tool to show that we have a hypothesis which says the coating quality in design one is poor because of the drifting of the process. That can be confirmed, and that is confirmed by the moving range chart. There are a lot of outliers, which is, again, suggested the multiple model fitting that we have shown. Then at last, these two leads are not same.

At last, if I would say, if I will conclude this whole of the presentation, how this is helpful for me as a Process Engineer, if I have done some work, I have collected my data, I have analyzed my data, I know that how it is working, but how I will communicate this to my stakeholders or my customers. In that sense, because I have to show, or I should have some supportive data.

I have this data and with that, I am able to convince that the hypothesis of the drifting process because of the higher headspace, and we are able to make decision with that the new tool when it will come up, the headspace should be reduced and will go with the design.

This JMP tool helped me to show or proof my hypothesis and to make some decision which is the correct as a design. This is all about my tool. At last, I would like to thank my team members, my instructor, Charles, and my mentor, Sukhti. At last, thanks JMP team for giving me a chance once again. Thank you.