COMPANY CONFIDENTIAL. DO NOT DISCLOSE.
Nikon Precision is precise. They have the data to prove it.
After switching from Excel to JMP – and putting the R&D engineering organization through a rigorous SAS Education training program – the company is finding they’re better able to troubleshoot, save time and be more predictive with their data.
____________________________________________________________________________
Challenge: With the explosion of new custom-fit applications for wafer technology, customer specifications are becoming increasingly difficult to meet. Nikon Precision’s Applications Team is tasked to work with internal and external stakeholders to reduce defects and resolve wafer production problems.
Solution: After years of using Excel, Nikon Precision switched to JMP, a tool that could both boost engineer’s data capabilities and reduce the time they needed to get results. To provide engineers a better foundation in statistical thinking, the company procured training support via SAS Education.
Results: Says application engineer David Godfrey, “I can do in an hour what would have previously taken us all night.”
____________________________________________________________________________
The incredible processing power of the more than 100 billion semiconductors in use on any given day means billions of computations per second and an unimaginably vast quantity of data. But as semiconductor technologies continue to advance, so will the pace of innovation; today the sector is one of the most rapidly evolving areas within the tech industry. Wafer manufacturing is becoming exponentially complex as new models build on the capabilities of previous designs, and market opportunities mean manufacturers are racing to develop new products fast enough to meet market requirements. Photolithography is a big part of the development process, and no company has been more influential than Nikon Precision.
With decades of experience as a manufacturer of optical lenses and precision equipment, Nikon Precision developed the world’s first production-worthy step-and-repeat photolithography tool in 1980. Photolithography is a patterning process in which a photosensitive polymer is selectively exposed to light through a mask, leaving a latent image in the polymer that can then be selectively dissolved to provide patterned access to an underlying substrate. In the past four decades, over 60 models of Nikon Semiconductor Lithography systems have printed half of all the integrated circuits ever manufactured.
Accurate to the nanometer
“In a few words, there's a very complicated system that allows you to print something on the semiconductor wafer with the precision of a few nanometers,” explains Mikhail Ladanov, an application engineer at Nikon Precision. The company’s advanced tooling systems involve a complex network of sensors that collect data in real time as wafers proceed through the printing process. This data enables Nikon’s analytics group to screen for defects and monitor product quality and tool performance. If any alarm bells are sounded, the Applications Team – of which Ladanov and his colleague David Godfrey are a part – is called in to troubleshoot.
“Working in R&D, we don’t always know what we’re looking for – that’s the biggest challenge,” explains Godfrey. “There are so many times where we know there’s a problem but we don’t know what’s causing it.” In precision manufacturing, tool performance can have an outsize impact on a company’s bottom line. When tools are down or under-performing, yield can drop precipitously. Reacting to arisings in the tooling landscape takes time; that’s why companies like Nikon Precision are investing heavily in proactive approaches to managing and analyzing the data generated in their plants.
Data visualization helps narrow the gap between data and understanding by enabling the team to more easily spot issues among a sea of numbers in columns and rows. “We look at the data visually because it’s much easier to clean, prepare, plot in the right way and run tests. We can more easily find anomalies this way,” Ladanov says.
For many years, Nikon Precision’s engineers relied on Excel to mine and explore data. But with sensor technology – and quality expectations – becoming more and more advanced, data sets were growing in size. Today, Godfrey notes, a single data table can run in the millions of rows. That’s why it was time they adopted a new tool with more advanced capabilities. Godfrey raised the question with the company’s leadership and immediately set about searching for a new tool. It wasn’t long before he settled on JMP, an industry standard software that Godfrey had first encountered in a prior role with Motorola.
“Excel works, but it’s very hard to handle with large data sets,” Godfrey explains. “JMP on the other hand is wonderful for data mining; it’s not something I think I could do in Excel without losing my mind.”
Increasing capability… and saving time
Though the primary motivation for switching from Excel to JMP was the need to expand engineers’ analytical capabilities, Godfrey says the team soon discovered the tool was also saving time: “The analysis capability is amazing. But the speed difference is amazing too.”
To illustrate, Ladanov offers an example. “I was in a meeting with my counterpart in Japan. He made a request and the expectation was that this request be completed by tomorrow. But I was able to complete the request before he had even finished talking,” he recalls. “That’s the difference between [our previous standard] how fast JMP can go.”
In another case, the team was facing a very pressing issue. But the analysis was taking hours as the customer’s process involved more than thirty steps in Excel. “They were spending two hours per test just typing in the results!” Ladanov recalls. Once he got hold of the data, however, he was able to plot the information in the way they wanted in no time. “It took me longer to copy the files than to actually do the analysis! The first time we did this, the customer couldn’t believe it. It was taking them hours to produce the same results.”
Adds Godfrey: “I now regularly get requests from Nikon Corporation asking me to do the data analysis because I can get results faster. I can do in an hour what would have taken them all night.”
Dynamic wafer maps in JMP help engineers spot quality patterns and variance
With each wafer printed in Nikon’s facility, a grid of sensors collects data across the wafer’s surface, generating what the team calls an exposure result file in X and O format. When data from a batch of wafers is loaded into JMP, Godfrey, Ladanov and their colleagues can see a dynamic map, color-coded for quality metrics. “We can look at all the wafers at one time here – so we can try to spot patterns across all 24 wafers,” explains Godfrey. “The colors on the map are what is called GC residual. We have two alignment systems within our scanner [an interferometer and encoder] that tell us the placement of the shot and position of the stage.”
The team uses these maps to spot batch variation. Even with tools operating at a high degree of precision, quality can be reduced over time due to variance in uncontrollable environmental factors like fluctuations in temperature, humidity and air pressure. Excess water might become trapped in the system’s encoder, causing variation in the delta between the two measurement systems. By spotting variation in printed wafers, engineers can recalibrate or adjust the scanning systems accordingly.
“We’re looking to see if it is repeatable every time. Does the coloration change throughout the batch?” Godfrey says. “With this visual in JMP I am able to see abrupt changes throughout the process…. And then we go look at the wafer and see the defects changing across lots and spot what the issue is.”
As the team expands their statistical capabilities, Godfrey explains, this work will become less reactive and more predictive. “We want to be able to predict errors based on a machine’s measurements. And if we can predict what is on the wafer, we can correct what is on the wafer. And our control can become much more precise.”
SAS Education helps engineering staff optimize their investment in JMP
With most of Nikon Precision’s engineering staff less versed in statistical analysis than in their technical subject matter, Godfrey and Ladanov knew that additional training was needed to help expand the company’s collective statistical knowledge if they were to move toward predictive approaches. “We were trying to beef up our ability in the statistics area,” Godfrey explains, noting that while in-house, on-the-job training may have adequately prepared engineers to use JMP for visualization, most were still far from tapping into the software’s full potential. The company’s leadership had, after all, invested in JMP; now it was time to provide their engineers with the training they needed to get the most out of that investment.
Godfrey and Ladanov went straight to the source, reaching out to SAS Education for an extensive menu of tailored training options. SAS Education’s JMP Training program uses a combination of comprehensive course notes, lectures, software demonstrations, question-and-answer sessions and hands-on workshops in a variety of training formats, all of which can be tailored to unique requests. By bringing JMP experts on-site for training, Nikon Precision was able to provide engineers the opportunity to train in a familiar, confidential environment where they could practice with their own data without worrying about IP security. SAS Education provided them with a tailored curriculum – as well as trainers with deep experience using JMP in industry – to do just that.
“We actually sent SAS Education a list of the items that we would like to understand better in JMP and they built a custom course around that request,” says Godfrey. “They took our list and said, ‘we already have a class that’s very close to this, but we can add some content here and take out what you don’t want there. And they were able to fit our timeframe, which was very nice.”
Nikon Precision opened these courses to all employees across the board. "It's new hires up to me, or even the guy who once mentored me who's still here,” Godfrey jokes. “Now that I've gotten better at JMP, I almost never use Excel anymore.”
“JMP has brought so many things that we weren’t able to do before,” Ladanov adds. “Plots in JMP are so much better and you have more control over how you present the data.” It’s all about optimizing the work and making use of every single scrap of data available to make Nikon Precision’s products better. “By constantly asking and learning what's new, and knowing what's out there, we can use it on our data,” he continues. JMP helps them execute on that aspiration, and should they ever have to make do without, well, Godfrey laughs, “that would be a misery!”
