Solar Fab Process Control Software Saves Time and Money
Looking to take its Discover Enterprise solution from semiconductor fabs into the growing photovoltaics market, Rudolph Technologies found that they would instead have to start anew with a significantly different solution to serve solar's distinct requirements.
Aaron Hand, Executive Editor, Electronic Media -- Semiconductor International, 1/12/2009
|
Rudolph Technologies Inc. (Flanders, N.J.), which specializes in process characterization solutions for the semiconductor manufacturing industry, has made its way into photovoltaics fabs with Discover Solar, a fab management software tool designed to help PV manufacturers increase cell efficiency and reduce costs.
Reengineered for the PV industry from the company’s Discover Enterprise, the software provides comprehensive analysis of process performance information to improve energy conversion efficiency of solar cells. It also enables process engineers to monitor the health of a complete production line and quickly identify tool and subcomponent problems or incoming material issues.
Discover Enterprise, for semiconductor manufacturing, had its genesis with Inspex, which sold its yield management assets to August Technology several years ago. August in turn merged with Rudolph in early 2006. Inspex was first to market with yield management systems for the semiconductor industry, according to Mike Plisinski, vice president and general manager of the company’s Data Analysis and Review Business Unit, and Rudolph was looking to make the same sort of entrance into the PV market with those capabilities. They got a surprise, however, when they found that the differences between semiconductor and PV production were too great for such a move, and Rudolph had to instead go back to the drawing board and rewrite the underlying algorithms, Plisinski said.
“It would’ve been nice for us to be able to keep the same platform, but what we see now is a tremendous increase in speed, and the ability to pinpoint their issues based on the new algorithms vs. slugging through old algorithms that required and expected a lot more data from the manufacturing process,” he said.
That manufacturing data, in actuality, does not exist in the solar world the way it does in the semiconductor world. PV manufacturers are dealing with entirely different economic and throughput models — producing as many as 500,000 solar cells a day at a cost of perhaps $5 each, compared with a big semiconductor fab producing some 100,000 wafers per month with a considerably higher value placed on each wafer. Solar cell manufacturers are trying to move those wafers through the line quickly, and the cost of inline metrology is not a worthwhile value proposition.
Therefore, solar cell manufacturers have essentially been running blind, Plisinski said, and Discover Solar gives them some eyes. The software tracks data all the way from the incoming raw materials through final test, making use of path analysis to pinpoint tool or material issues. Manufacturing engineers can use the capabilities to reduce their cost and their scrap, manage their equipment, and pinpoint equipment failures quickly. Process engineers can use the software to optimize and tweak the process in order to maximize cell efficiencies, for example.
Rudolph has been working on the software for about a year, including initial discussions, and has had installations in the field beginning last quarter, according to Plisinski. During that time, customers have been able to make significant cost- and time-saving discoveries that would’ve previously taken them much longer to discover, if at all.
In one case, Discover Solar was measuring some 14 output variables at a solar cell manufacturer, and found an increased number of wafers with a very low shunt resistance value. Discover was able to drill down and pinpoint the problem to a specific print line, laser and table. A daily report would’ve caught the problem, but 7000 wafers would’ve been lost in the meantime. Discover’s alarm monitoring, however, caught the problem 18 hours earlier, saving 4000-6000 wafers. “What would’ve taken normally hours or even days to identify and fix, Discover was able to pinpoint very, very quickly for this customer,” Plisinski said, noting that this saved the company $44,000 to $66,000 in scrapped cells.
In another example, Discover identified a drop in cell efficiencies, which the customer wanted to better understand. “So they initiated a path analysis, where Discover looks at all those bad samples and tries to determine any kind of first- or second-order effects,” Plisinski said. “Is there any combination of tools that are contributing to the problem, or is it one particular tool? And we can also go down to chamber or even tube or zone, so internally to the tool, if we’re getting the data, we can even pinpoint within the tool what’s going on.”
With just a couple clicks on the program, the cell manufacturer was able to see that they had a batch that was underperforming, indicating a raw material problem. Just to make certain that there wasn’t more to it, the customer was then able to separate out the bad batch and see how the remaining batches performed. They then easily found that a particular diffusion tool was actually underperforming the rest of the line as well. “So that was a hidden problem that they would’ve never been able to understand or identify without the capabilities that Discover provided them,” Plisinski said.
Because there is so little inline metrology going on in a PV fab, path analysis is a key aspect to Rudolph’s latest engineering of its software. “In the semi world, we’ve always had something called commonality analysis, but it uses inline metrology in order to help identify the quality of different steps,” Plisinski noted. “We also look at the MES information as well, but in this case, that analysis would’ve taken two or three hours for a couple million of wafers, and now we have that down in the seconds range, less than a minute.”
There are a couple key reasons why the solar world is not yet making use of metrology as much as the semiconductor world is, Plisinski said. For one, because solar wafers are relatively inexpensive, the cost of the metrology does not make economic sense. “The other is throughput,” he said. “A lot of the metrology systems are going to be orders of magnitude slower than this line, so they’d have to do some kind of sampling strategy.” Any kind of economical sampling strategy would likely not be statistically meaningful. “So most of the customers that I talk to are not interested in any short-term solution there, but they are seeing the need, as they drive performance and they drive their technologies forward, they do see that metrology will play a bigger and bigger role.”
These factors made path analysis an important part of the changes that Rudolph had to make to its original Discover product to make it fit into solar cell manufacturing. Also, just the sheer volume of data meant the software developers had to change the way all the information was stored, according to Len Labua, software applications manager. “We hadn’t planned on pulling millions of data points at a time, where in semiconductor, it may be hundreds or thousands,” he said.
Another important difference is was Rudolph calls health monitoring. “That was an area where, in semi, they’re very cell-oriented. So they’re looking at lithography and they’re making sure that step’s okay, then they go to CMP and etch — as long as each step is okay, that’s good enough for them,” Plisinski explained. “But in the case of PV, where it’s so fast, they’re looking at the overall line. And we had to build in the methodology for them to be able to tell is their line performing okay or not with one quick shot.”
“One big thing here, with semi, they’re all concerned with using SPC, which tracks one variable at a time, which is fine, and we do that also,” Labua added. “But we also developed a method that allows customers to look at really all the variables at the same time.”
Although Rudolph has put a lot of work into redesigning Discover for the solar world, some of those developments could potentially be applied back to semiconductor manufacturing. Path analysis, for example, will be directly applicable to semiconductor manufacturing, Labua said. That will be particularly true as big manufacturers develop gigafabs, Plisinski added. “In fact, that’s a direct correlation that we’ve taken advantage of already. Because of our redesign of the databases, we were able to prove to one of our potential Discover Enterprise customers that are looking to install a gigafab. And they’re saying that none of the vendors in the semi world can support gigafab volumes of data. And we were able to show that we actually can do that.”
