Edwards sees promise in sub-fab data analysis

By Dave Lammers

Integrating data from various sensors in semiconductor fabs is a key focus in the industry now, and the sub-fab is an increasingly important part of the equation. As process steps become ever more sophisticated and expensive, keeping pumps and other sub-fab equipment running optimally involves integrating multiple pieces of data into useful information.

Paul Rawlings, president of the SEMI Service Division at Edwards, Ltd., said while the semiconductor industry has a history of analyzing tool data and relating it to end processes, more progress is needed in considering the entire fab as an ecosystem.

“The sub-fab equipment also has a bearing, not just in terms of how efficiently we are using data to manage the sub-fab efficiently, but also in terms of improving overall fab performance and yields,” Rawlings said.

Edwards is developing “a structured approach to the kinds of data that we have, and how we use it,” Rawlings said, including a database of best known methods (BKMs) which includes the optimum configurations for pumps, abatements and other systems.

Edwards currently is launching EdCentra, a fault detection and classification (FDC) software platform aimed at the sub-fab. It provides a comprehensive solution to vacuum security, to take one example, by combining equipment monitoring, data acquisition, and analysis of operational data.

EdCentra, Edwards Sub-Fab FDC platform, provides process-critical vacuum and abatement equipment information, complementing Fab-based platforms and supporting industry efforts to create integrated Fab data-management systems.

EdCentra, Edwards Sub-Fab FDC platform, provides process-critical vacuum and abatement equipment information, complementing Fab-based platforms and supporting industry efforts to create integrated Fab data-management systems.

Besides monitoring the performance of the equipment, the EdCentra sub-fab information management system has built-in predictive capabilities. And it complements another Edwards tools, which records service activities.

When a pump is taken back to an Edwards service center, it is stripped down and serviced. “Then we update all of that on to our central database on the lifetime and the performance of our equipment. We have an ecosystem there,” he said.

The company’s overall goal, Rawlings said, is to connect what have been “fairly separate systems,” maximizing up-times in part by comparing the performance of different tools in the same area.

By extending the data analysis ecosystem, Edwards can increase up-times and refine service scheduling. “The data is there. When we connect these systems, that’s where we get all of the benefits from the data. It’s no longer about taking data at individual points along the life cycle. It’s about connecting the data across the journey of the equipment and then looking for optimization, adjustments, and maybe upgrades on the equipment,” he said.

Engineers and data scientists throughout the semiconductor industry are developing more fab-level techniques, using multi-variate analysis of data coming off the tools, as well as other inputs.

For Edwards, that involves looking at data coming from pressure gauges, temperature sensors, the power spectrum, and, increasingly, the very useful information derived from vibration monitoring.

“What we are finding is that by looking at the combinations of data, that’s when things become really interesting. Rather than just relying on one or two points of data on the equipment, we’re starting to build up a library of different parameters. Then we are looking to see how we combine those to give us the most accurate predictions of tool lifetime,” he said.

Edwards is working closely with two fabs to develop data-sharing protocols that allow for optimum monitoring of the sub-fab equipment. “We are looking at it both from their point of view, and ours, discussing different ways of processing the data and the analysis of the performance of the equipment.”

Because of the huge amounts of data involved, Rawlings said “we only transmit data, if you like, that moves. It’s not worth sending data at 10+ Hertz when that particular parameter is not moving. It’s only worth sending that data when a change occurs.”

The work with customers is done within strict limits to ensure data security. “Clearly, you would never want to share any data that could give any indication of what customers do in their fabs and we go to great lengths to safeguard this” he said.

One encouraging sign is that, industry wide, cloud security is becoming more effective, reducing concerns about moving data to the cloud. “The way that end-user data is segregated, the built-in security, is resulting in a little bit more openness in terms of using this data,” Rawlings said.

That said, it is still a difficult area to work with customers. “I won’t mention names, but there are a few folks where we are engaged, so we can move to the next level of validation. I think it’s going to be an area of development over the next few years, as we really focus on the right parameters to measure to predict lifetimes.”

A lot of sub-fab equipment is affected by the processes, more than the basic mechanics of the equipment. It is fairly rare to have a bearing failure on a pump, for example. What is more likely is that the pump will have some process-related issues, such as corrosion or condensate build-up from materials in the fab.

“We spend a lot of time studying the process materials, temperature settings, and those kinds of things to extend equipment lifetimes, but there’s obviously always a limit to what can be achieved. What we are now doing is looking at how the different types of sensors that we have, both already on the equipment, and other ones that we’re developing, can give us the best combination, the best ways of predicting lifetimes. That’s an area that we’ll be working hard on over the next few years for our customers,” Rawlings said.


Easily post a comment below using your Linkedin, Twitter, Google or Facebook account. Comments won't automatically be posted to your social media accounts unless you select to share.