By Pete Singer
The importance of data gathered and analysed in the subfab – the place where vacuum pumps, abatements systems and other supporting equipment operates – is growing. Increasingly, manufacturers are finding that these systems have a direct impact on yield, safety, cost-of-ownership and ultimately capacity and cycle time.
“The subfab is getting recognized evermore as a contributor to the overall fab effectiveness, particularly when the fab is looking to get last fractions of a percentage of performance efficiencies,” notes Alan Ifould, Global Market Sector Manager at Edwards.
There’s also keen interest in tying this data with process data from the fab, the MES (manufacturing execution software) system and ultimately the ERP (enterprise resource planning) system as part of today’s efforts to understand and control the entire data ecosystem.
Subfab data systems provide a volume of data related not only to vacuum and abatement equipment, but also upstream, to the foreline, gate valve and chamber. Of special interest is the monitoring of vacuum faults, which can negatively impact quality, cost and safety. “A vacuum fault is anything that results in a loss of a degradation in vacuum,” said Ifould.
Ideally, faults – and the overall quality of the vacuum system — are proactively managed. Potential faults are detected days or even weeks before they occur and addressed during regularly scheduled tool maintenance, for example. “We’re finding that our ability to detect vacuum faults in the wider vacuum system comes very much to the fore,” Ifould said.
Data seen at the pump or abatement can help determine the size and location of vacuum system leaks. Algorithms based around vacuum science and thermodynamics can lead engineers to problematic leaks that, over time, can have a significant impact on yield.
Often, the first reaction to a loss in chamber pressure is to blame the vaccum pump, Ifould said. Vacuum pumps can be swapped out in about 4 hours, but if the process tool goes down while in operation, it could be in excess of 48 hours to get everything back up and running. Even then, it might be something other than the pump that caused the initial problem, such as a leak in a gate valve or in the foreline. It’s essential to accurately diagnose the problem(s) at the onset, but that can be a challenge: “You only need a small leak in a gate valve, and you immediately have problems with maintaining the base pressure in the chamber. The pump may become overloaded because of the additional gas load caused by leaks,” he said.
Edwards has developed a verity of new data collection and analysis strategies aimed at improving such decision making. The SMA (Site Management Application) is latest addition to data analytics portfolio, focused on subfab. As shown in Figure, SMA is designed to provide insight into maintenance activities, equipment performance and fault resolution. It is implemented in parallel with the company’s VTPS (Vacuum Technique Production System), which drives standard work and behaviors based on LEAN principles and best known methods.
Edwards is also working on what it calls “sensorization” where, for example, the use of vibration analytics can detect anomalies otherwise missed by traditional monitoring techniques.
Ifould said the SMA and sensorization helps improve the stability of fab operations by bringing veracity to the data. “It’s one thing to have a volume of data, but the data itself is of little value unless it’s of good quality,” he said. “When we’re looking at equipment operations and the way you have operators involved, being able to bring discipline to the behaviors of those operators to the task that they perform brings discipline to the data and improves the veracity of the data,” he said.
He said Edwards has been using this approach to “great effect” over the last year. “We can help our customers see where some of their maintenance practices need to be improved to eliminate some of the sources of error that cause some of those vacuum faults,” he said.
More recently, Edwards is looking to move beyond a simple predictive maintenance model (PdM) to a model that include quality (PdMQ). The model includesnot only the condition of the subfab equipment, but of the quality of the vacuum it provides, and therefore the process it supports. “We’re not just considering the condition of the subfab equipment and being able to predict when that may fail, but considering the quality of the vacuum that system actually provides.”
Harnessing data from all parts of the fab ecosystem is essential, Ifould notes, but has its challenges, especially when it comes to IP. “In an ideal world, we would like to receive contextualized data which allows us to relate what’s happening in the vacuum pump into the process itself. That becomes challenging because of the IP sensitivity,” he said.