Technology-Enabled Services for Fast Problem Solving
James Moyne, Parris Hawkins, Jimmy Iskandar and Michael Armacost
Manufacturing automation, specialized expertise and close collaboration combine to improve product, process and equipment performance.
The migration toward more advanced semiconductor device nodes (< 28nm), with stringent requirements for acceptable device performance, yield and productivity levels, is forcing the nanomanufacturing community to think differently. Manufacturing automation technology suppliers and their device-producer users are recognizing the technological and economic advantages of broader, deeper collaboration throughout the life cycle of production tools.
The latest International Technology Roadmap for Semiconductors (ITRS) Factory Integration (FI) chapter echoes this trend as follows: The rapid increase in FI requirements (e.g., Big Data) and capabilities (e.g., prediction) in recent years has led to a change in the approach to implementing and maintaining FI capabilities. Development and maintenance of emerging capabilities such as PdM (predictive maintenance), VM (virtual metrology), waste management, and utilities management incorporation into fab objectives, requires intimate knowledge of the fab objectives, process, equipment and the capabilities themselves.
Applied Materials has recognized this need, and is responding with a new class of integrated, advanced customer service solutions. These solutions combine Applied’s innovative manufacturing automation software with specialized expertise and extensive equipment and process domain knowledge. Delivered as service solution options for customers covered by an Applied service agreement, they are designed specifically to help customers resolve their toughest production challenges—those with the greatest impact on device performance and yield— faster and more cost-effectively.
TOUGH CHALLENGE. ADVANCED SOLUTION.
The example below illustrates various aspects of Applied’s new technology-enabled services approach, in which technical experts using cutting-edge technologies—including predictive maintenance (PdM), virtual metrology (VM), run-to-run (R2R) control and equipment health monitoring (EHM)—work together to uncover problems and suggest solutions. The resulting benefits for contract customers include (1) integration of technologies, (2) incorporation of tool, process and automation technology domain knowledge, (3) delivery of collaborative support service, and (4) solution implementation as part of a continuous improvement process.
The Problem: Reduce Downtime from Epi Lamp Failure
In many fab lines epi process availability is critical. Overall availability, consistent availability and predictable availibility are all important factors. Therefore, it is important to minimize process mean-time-between-interrupts (MTBI) and mean-time-to-repair (MTTR), and also be able to predict with some lead time when a repair might be needed. Lamps are a common source of failure in epi and, as shown in figure 1, there are a number of mechanisms that cause them to fail. These failure modes are generally difficult to anticipate until a sharp drop in resistance occurs right before failure. This results in a significant amount of unexpected downtime and undesirable variability in downtime. After a failure and subsequent lamp kit replacement, “tuning” is required as part of the maintenance recovery process. Test wafers are processed and measured, and the process is adjusted based on the results.
The process is deemed production-ready when the test wafer measurements meet specified quality criteria. This tuning process can often be costly, both in terms of high MTTR and highly variable MTTR, not to mention the cost of test wafers and metrology. The process of tuning can be rather ad hoc with adjustments determined manually and often in a univariate (one-by-one) fashion.
PdM technology uses process- and equipment-state information to predict when a tool or a particular component in a tool may need maintenance. This prediction data is then used as information to improve maintenance capabilities, such as avoiding unscheduled downtime. PdM technology-enabled services can be used to predict lamp failure, thereby reducing unscheduled downtime and improving MTBI and MTTR.
Applied’s PdM solution, shown in figure 2, was developed over two years, with research into PdM techniques in nanomanufacturing and other industries. It includes an offline component for PdM model development and optimization, and an online component for model integration, execution and updating. The offline component uses historical data to develop maintenance event prediction models that are optimized to customer objectives. This is achieved with a comprehensive set of tools that provide capabilities for data merging (generally fault detection [FD], metrology and maintenance data), data quality strengthening, automatic and manual (graphical) data manipulation, key parameter identification, model building, model assessment, and model optimization to customer objectives.
Advanced Services specialists use these technology tools to consult with both Applied and customer tool- and process experts so that the solutions are optimized to a particular customer application. These specialists not only develop the optimized maintenance event prediction models, but also provide analysis utilizing the model prediction quality and customer objectives to inform the customer of the expected benefit of model application before the model is deployed.
Once the model is developed offline, it is brought online through integration into the existing fab automation infrastructure, minimizing integration costs. As shown at the bottom of figure 2, configurable execution strategies allow the prediction model to be “eased” into fab operations, first as an indication-only capability, and later as an actual control capability interfaced to the maintenance management system once the user has sufficient confidence in the prediction capability.
The online component also includes a capability for model updating as necessary, and online PdM data analysis by the user or Advanced Services experts. The latter capability is illustrated in figure 3. Here a particular prediction event can be further investigated by highlighting an area of interest, utilizing embedded software analysis to determine the primary sensors contributing to the event and viewing the wafer-by-wafer data traces of the individual sensors.
Reducing MTTR with VM and R2R Control
As noted above, lamp replacement maintenance recovery can be time-consuming as there are usually multiple iterations of lamp parameter “tuning” that include running a number of test wafers with specific characterization recipes, analyzing metrology data, and making hardware and software adjustments. This process continues until the metrology data meets specified quality criteria. Four to ten iterations of this type are not uncommon, leading to MTTR on the order of two days or more.
Advanced Services experts apply R2R control and VM1 capabilities in this situation as shown in figure 4a. Specifically, they develop lamp-qualification tuning models using specialized software applied to historical test wafer process metrology and associated process FD output data. VM techniques are used to predict the test wafer metrology values, and multivariate R2R control techniques are used to determine what lamp tuning adjustments can be made simulatenously to bring the metrology values within acceptable limits, thereby completing the lamp-tuning process. When a lamp replacement event occurs, these models are re-centered using the data from the first lamp-tuning iteration, and then applied during the second iteration to provide tuning recommendations in a multivariate fashion. The result is that fewer tuning iterations are required to bring the chamber to a satisfactory matched state for release back into production, as illustrated with the conceptual diagram of figure 4b.
Reducing Unscheduled Downtime and MTTR with EHM
EHM is a technology that monitors tool parameters to assess tool health as a function of deviation from normal behavior. Advanced Services specialists leverage multivariate EHM throughout the production and maintenance cycle in concert with PdM, VM and R2R control as shown in figure 5.
During production it is used to monitor overall tool health as well as the health of specific target components. During the maintenance process EHM is used to provide multivariate “fingerprints” that help assess whether a particular maintenance procedure has been successful. In the case of lamp failure, EHM monitors the health of the tool and lamp during production, and provides information on the quality of the lamp-replacement procedure during maintenance.
Getting RESULTS: CUSTOMER PROBLEMS SOLVED
These technology-enabled service solutions from Applied Materials are adaptable to all our process tools and available as options to customers participating in Applied service agreements. They can help customers resolve difficult problems faster and save money by utilizing a number of technologies collectively to provide maximum benefit.
For example, figure 5 illustrates the use of these services throughout the maintenance cycle to reduce MTBI and MTTR. They also include continuous improvement mechanisms that provide feedback that can be factored into best practices, process- and equipment-operation design, and other automation capabilities such as maintenance and FD, delivering benefits that extend beyond the initial solution of a problem. The type and level of benefit depends, of course, on individual customer needs and the agreed roadmap for deployment of technology-enabled services. In the epi lamp-replacement example, application of PdM services resulted in the ability to predict lamp failure 5 days in advance with 90% accuracy. Further application of VM and R2R control to lamp MTTR would reduce the tuning time by an average of two iterations or more.
With a toolkit of advanced and proven analysis technologies customized to nanomanufacturing applications, including PdM, VM, EHM and chamber-matching options, Applied can address our contract customers’ most urgent issues quickly on multiple fronts, providing rapid and significant benefits. Further, we can incorporate these solutions into a continuous improvement process that will allow customers to utilize technology-enabled services to optimize equipment and process configurations and resolve difficult problems efficiently.www.itrs.net.
 J. Moyne, J. Iskandar, P. Hawkins, T. Walker, A. Furest and B. Pollard, D. Stark and G. Crispieri, “Deploying an Equipment Health Monitoring Dashboard and
Assessing Predictive Maintenance,” Proceedings of the 24th Annual Advanced Semiconductor anufacturing Conference (ASMC 2013), Saratoga Springs,
New York (May 2013).
 J. Moyne, “Method and apparatus for optimizing profit in predictive systems,” United States Patent Application (filed February 2014).
 A. Khan, J. Moyne and D. Tilbury, “Fab-wide Control Utilizing Virtual Metrology,” (invited) IEEE Trans. on Semiconductor Manufacturing-Special Issue on
Advanced Process Control, Nov. 2007, pp. 364-375.
 Proceedings of the 25th Annual Advanced Semiconductor Manufacturing Conference (ASMC 2014), Saratoga Springs, New York (May 2014). See also:
“Improving Yield with Fleet Chamber Matching,” NanoChip Fab Solutions, Vol. 8, No. 2, 2013.
 J. Zou, “Minimizing Pilot Runs with Non-Threaded Control Technology,” Applied E3 User’s Conference, Phoenix, Arizona (February 2014).