Timely engineering fixes rely on high-speed communications standards, but data inconsistencies are getting in the way.
The tsunami of data produced during wafer fabrication cannot be effectively leveraged without standards. They determine how data is accessed from equipment, which users need data access and when, and how fast it can be delivered. On top of that, best practices in data governance and data quality are needed to effectively interpret collected data and transfer results.
When fab automation and process engineers have access to quality data and metadata that conforms to the latest communication standards, they can more effectively provide engineering fixes in the manner needed to best maximize yield and productivity. Use cases for equipment and sub-fab data are now intertwined with ML algorithms, which rely on standards for better communication of that data. But there are limits on data size, data governance, and data availability due to the realities of a wafer fab’s IT infrastructure, equipment age, and the standards that engineering teams use. The good news is that emerging standards plan to address some of these limits.
Behind this flood of data is a growing understanding of how to utilize it. Data analytics suppliers and OEMs can provide access to instrument data deeply embedded in log files. As a result, there is growing demand for fab equipment data, spurred on by the wider adoption of Smart Manufacturing 4.0 strategies and more sophisticated process control algorithms.
Engineering teams use this data for a range of purposes, including process control, fault detection classification (FDC), and predictive maintenance. But to effectively leverage all of that data, factory data systems and the data analytics platforms need to communicate equipment events and associated data in standard ways. Given the multiple equipment providers and tool types — etchers, CVD, CMP, etc. — those standards are needed to minimize the number of unique naming conventions, data structures, and communication protocols.
Today most of these equipment data standards are governed by SEMI committees.[1]. The common SEMI standards evolve as the need to manage more data and at faster rates evolves.
In a wafer fab, the automation of work in process (WIP), equipment, and process recipes involves many control points. Data from process equipment provides crucial information about lot/wafer activity, lot IDs, instrument measurements, and equipment health, along with other data types. Fab equipment communication standards support multiple engineering use cases, including feedback/feedforward process control, fault detection and classification (FDC), yield management, and root cause analysis. Having timely and accurate data, and associated metadata, enables timely data decisions. The lack of such data, in contrast, can lead to incorrect decisions that negatively affect manufacturing operations.
“Factories collect high-volume data from the manufacturing equipment using the EDA/Interface A standards (Freeze Version 2) [2,3,4] that are fed into an advanced analytic component,” said Albert Fuchigami, senior standards specialist at Peer Group. “Based on the analysis results, the factory adjusts the manufacturing equipment through the SECS/GEM control channel.[5,6] An example of this would be adjusting recipe parameters on the next processing runs.”
While statistical process control (SPC) charts remain useful for process and equipment engineers, multi-variant analysis helps identify subtle relationships between equipment parameters and yield.
“The tool reports out the data,” said Joe Fillion, director of product management at Onto Innovation. “If there’s a measurable change, the tool will be able to report that out to an FDC software package. It can take that information and then monitor it. You can then integrate FDC software with yield analysis software, which accelerates the proverbial search for a needle in a haystack. Your yield changes a little bit for these wafer lots. Your FDC analysis informs you that a specific equipment exhibits a small variation. It may seem very obscure, nobody would notice, but analysis software performing a multivariate analysis actually can understand and reveal correlations to find the needle.”
SEMI standards for fab equipment communication
Equipment data has multiple points of value. Getting it into control and analysis information systems requires engineering teams to implement that communication in a consistent and reliable manner. This is where standards come in.
SEMI has long been the home for fab equipment standards related to control, configuration and measurement data. For both 200mm and 300mm fabs, the suite of SECS/GEM standards enables equipment control and configuration. A couple decades ago, industry experts developed a suite of equipment data acquisition standards (EDA, also known as Interface A). The standards committees intentionally separated SECS/GEM from EDA to guarantee that the equipment command and control remains stable while the dynamic nature of fab engineering team data requirements can be met.
Fig. 1: Distinctions in purpose for SECS/GEM and EDA standard. Source: Peer Group
“The manufacturing execution system (MES) is concerned with controlling the manufacturing equipment and getting data from it, typically through the SECS/GEM channel,” said Peer Group’s Fuchigami. “This information can be shared with the factory data management system and the data analytic platforms (commonly referred to as equipment engineering). Analytics components often will get data directly from the equipment. There are a variety of different protocols to get data.”
One access point for data is equipment log files which require custom software applications for real-time access. More prevalent equipment data comes by way of the EDA standards. These permit engineers to gain access to such equipment data at a specified frequency.
“The EDA standard is meant for monitoring and collecting data,” explained Briani Rubow, director of solutions engineering at PDF Solutions’ Cimetrix Connectivity Group and co-chair of the SEMI NA Information & Control Committee. “EDA was developed for the process engineers to change data collection and change how the interface is being used as needed for troubleshooting and diagnostics of the equipment. They wanted something independent of the GEM interface. No one wants to accidentally stop the equipment or mess with the recipe selection.”
An engineering change order is not needed for implementation changes on Interface A, but it is needed on the GEM interface.
“The other major feature of EDA is its architectural flexibility,” said Alan Weber, vice president of new product innovations at Cimetrix Connectivity Group at PDF Solutions. “You change data collection approaches on the fly without necessarily going through an engineering change notice. Also, you may have multiple independent clients of the data collection. Thus, independent applications change their data collection requirements and activity as needed.”
The more specific a standard, the more helpful it is for implementing cost-effective automation solutions, and for enabling efficient debug of yield and equipment issues.
“Although many of these protocols may be open source or industry standards, there often isn’t any standard description for how to use the protocols,” said Fuchigami. “This includes how the messages are formatted, which data does the sender include in the message, and what is the expected behavior when the receiver gets a certain message. This is one of the biggest values of SEMI standards suites, such as EDA/Interface A, GEM300, SMT-ELS, and RITdb. They define expected behavior. For example, after setting up a data collection report with triggers, the receiver knows they will get collected data in ‘this particular format at this interval after this particular event occurs on the equipment.'”
Fig. 2: Standards and associated protocols that support the information highway within a wafer factory. Source: Peer Group
The equipment description forms the basis for these equipment communication standards. SEMI standards E120 and E125 enable details of physical equipment and the data structures and events that can be controlled.
Fig. 3: SEMI equipment description standards, E120 and E125. Source: Cimetrix by PDF Solutions
Together these standards describe the architecture of a fab, too. A flow chart includes data structures, automation, process control, etc. Fundamentally, a common equipment model encompasses the data and metadata, which gets reported into the host computing system. Industry experts repeatedly cited the need for data in context, which metadata provides. For example, chamber pressure value and chamber name (metadata) complements chamber pressure value (data). In effect the suite of SEMI equipment data standards is a metadata standard [8].
Fig. 4: Suite of SEMI Interface A/EDA standards. Source: Cimetrix by PDF Solutions
In the spirit of continuous improvement, these equipment standards are updated on a regular basis. Notably, these revisions can take five years or longer. In SEMI standard parlance, a suite of standards is now described by their Freeze Version number, i.e., EDA Freeze Version 2. This assists in establishing the requirements on OEMs by the fabs, as well as technology roadmaps for fab automation teams. Nevertheless, several industry experts noted that an engineer or technician may have outdated assumptions on capabilities and data access due to a lack of familiarity with the most current standard.
Limits, gaps and inconsistencies
Standards specify formats, time events, and data structures. But the realities of a factory’s IT infrastructure, equipment vintage, and how engineers and technicians use standards result in limits, inconsistencies and gaps between what is possible and what is probable. Inaccurate, incomplete, and unavailable data hampers the interpretation of that data, and subsequently the actions taken on the resulting analysis.
The speed at which data can be delivered matters. Twenty five years ago, SECS/GEM communication reported out data at 1 hertz. Today it supports data from 10 to 20 hertz. The barriers to both achieving that data rate and transmitting data rapidly include real-time operating system limits, packet sizes, and computing network latencies and data rates.
“You do start running into complications with the equipment operating systems, such as Windows and Linux,” said Cimetrix’s Rubow. “Eventually, the limiting factor is the network and the packet sizes. You can only send data in a packet so fast over a network. All standards that move data around suffer those same limitations. One of the advantages of a new version of the EDA standard (EDA Freeze Version 3) is we’ve come up with a way to buffer data, rather than sending it one packet at a time. We can make the data collection a lot more efficient. This standard is not yet released.”
The need for real-time advanced process control requires access to data at higher frequencies. Existing Interface A standards do not support real-time plasma arc detection, which requires 30 kHz sampling. It also requires specialized hardware and access to instrument generated log files.
“GEM300 is very limiting when it comes to critical tool sensor data, as it tops out at 3 hertz,” noted Boyd Finlay, director of solutions at Tignis. “This is fine for monitoring some tool sensors. However, a lot of the critical sensors require Nyquist measurement methods, which need a higher resolution of sensor data to avoid signal chopping. GEM300 was never created for ‘condition-based maintenance’ requirements. This is why we have seen Interface-A standards emerge over time, as well as a more intensive focus on using equipment log files. These contain higher-speed data captured by the OEM onboard tool computers, e.g., laser power intensity at 300 Hz. Obviously, you need to properly measure things based on instrument design principles, and not to assume what a tool publishes is what an equipment engineer needs.”
In addition to the delivery speed of equipment data, there are gaps in assigning data from different components in a process tool. Differences in OEM implementations present a small barrier to factory automation.
“There are many standards that cover the ‘how’ issue,” said Fuchigami. “How do we get data off the equipment? The answer is either through a well-defined standards suite, such as SECS/GEM, EDA/Interface A, RITdb, or SMT-ELS, or through custom behaviors built on top of industry or tailored protocols. Areas that can benefit from standardization are around the ‘what’ issue. That is, what equipment data is accessible to the factory, and whether we need to standardize how to approach this data. For example, we could develop a framework around the new SEMI E190 Equipment Data Publication standard so that equipment data from specific component types can be categorized consistently across different vendors. An example of this is the way SEMI E190.1 defines how data from an etch component could be grouped into gas, plasma, and optical emissions spectroscopy categories.”
Even if an engineering team faithfully implements standards (SEMI or internal), the interpretation of data can be hampered by differences in time-stamp accuracy and missing wafer IDs.
“The biggest challenge in comparing data at any level and at any time is always going to be time stamping,” said Cimetrix’s Rubow. “It’s important that data has accurate time stamps, and it’s surprisingly easy to have mistakes. For example, you might have one piece of software capturing data in the local time, and maybe another saving the data in a UTC format. Or maybe you have two computers that just aren’t well synchronized. If you don’t know what the timestamp means, you can’t really compare data.”
Meta data for context assists with rapid debug of tool generated yield issues. This context is especially useful when connecting data between wafer test and process equipment history.
“We still often face the situation that wafer IDs are not available when a wafer is processed in a chamber. Sometimes we have slot positions, which can be more or less used for tracing wafers, but sometimes there are only lot-level tracing data available,” shared Dieter Rathei, CEO of DR Yield. “None of these challenges are insurmountable. It’s mostly a question of whether additional investments are justified to get better data quality. In my experience, these investments always pay off in the long term.”
Conclusion
Effective, efficient and economical use of data is needed for fab automation. The problem is making that data consistent enough to leverage it for better and more timely analysis, and moving it to wherever it is needed at any particular moment. That requires continual updates in communication standards.
The challenge is meeting the demands for equipment configuration, maintenance and advanced process control. The latter two have grown in their need for more sophisticated analysis, which ML/AI can provide. Yet the implementation of data into these standards can hamper proper interpretation and subsequently the actions needed to resolve the engineering problems.
“Through collaborative industry efforts, standards development can be used to help your customer base ensure quality and reduce variability,” said Melissa Grupen-Shemansky, CTO and vice president of Technology Communities at SEMI. “Standards also accelerate and enhance innovation, such as those called for to enable future semiconductor factories by focusing the development process on truly differentiating performance features.”
Standards can be tedious to develop and implement. Yet they ultimately reduce engineering efforts devoted to managing yield, maintaining equipment and diagnosing seemingly inexplicable process issues.
References
Related Reading
Integrating Digital Twins In Semiconductor Operations
The industry must collaborate to develop a common understanding of digital twin technology across various hierarchical levels.
Aftermarket Sensors Boost Yield In Wafer Fabs
More data improves throughput, helps extend the life of equipment.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply