OEE Focuses a Slow Economy
With years of double-digit annual growth, fabs and equipment suppliers alike are learning to make the most of decreasing demands in a ramp-down environment, including the use of overall equipment efficiency (OEE).
Ruth DeJule, Contributing Editor -- Semiconductor International, 4/1/2009
Our current economic condition has seen the semiconductor industry resort to downsizing as average selling prices (ASPs) decline in all sectors (Fig. 1), most notably in memory. With fabs running at diminished capacity, chip manufacturers and equipment suppliers must become more cost-competitive and productive while continuing on a path to the next-generation technology node. This article explores how an unfamiliar ramp-down environment is causing manufacturers to develop new strategies to run their fabs more efficiently (Fig. 2), employing overall equipment efficiency (OEE) at various levels. Interestingly, OEE, a subset of cost of ownership (CoO), contains no cost information.
|
| 1. IC unit selling prices have dropped significantly over the past several years, contributing to a sharp decline of chipmakers’ operating profit. (*Profit data included 16 of top 20 semiconductor manufacturers.) (Sources: WSTS, company data, Lam Research Corp.) |
|
| 2. In spite of decreased demand, fab managers are still taking a close look at operational effectiveness. (Source: Applied Materials) |
"The proper level of OEE is the one that yields the optimal value to the factory as a whole. Productivity must be measured against the business and market needs," said Vallabh Dhudshia, affiliate consultant for Wright Williams & Kelly (WWK, Pleasanton, Calif.).
The concept of OEE goes beyond a single number. It is made up of three major components:
-
Availability — non-productive time components, including idle time, non-scheduled maintenance, equipment up/downtime.
-
Performance — throughput relative to the theoretical limit.
-
Quality — yield and rework.
OEE can be determined per tool or tool groups. However, its usefulness is highest on tools that limit factory output, known as constraint or bottleneck tools. Dynamic OEE calculations are often used, for example, on constraint operations such as lithography cells, which are ideally run as efficiently as possible to bear out the high capital costs.
Though OEE is represented as a single number, fab managers typically spend as much time looking at each individual component in determining overall operation effectiveness. A suite of reports is typically available to the production manager on a bi-shift or daily basis that shows inventory queue times, cycle times or work in process (WIP) turns; in essence, multiple metrics show how well the bottleneck is being exploited.
Cutting costs
The decrease in demand has affected the way fabs operate. For many chipmakers, like Texas Instruments (TI, Dallas), one solution is idling an entire toolset or portion of the fab depending on the demand or loading situation. Although operational costs such as labor, electricity and water can be controlled, the importance of maximizing OEE on the machines that are not idle does not diminish. "One of the worst mistakes a factory can make in a down market like what we're experiencing is to assume that OEE doesn't matter," said Alex Casimiro, director of information technology for TI's Technology and Manufacturing Group. "OEE matters and matters probably more. Because now is the time to get healthy; to fix things that are more challenging to address when a factory is focused fully on producing output."
Another strategy taking hold at TI is to soften its position on immediate response to some factory disruptions. Clearly this does not entail disruptions that may delay delivery or in any way impede quality to the customer, but may include a function that can easily be assumed by another part of the fab. However, Casimiro cautions about relaxing too much; for when the demand increases, "we'll be back to making sure we have that 'hair on fire' attitude for supporting the factory."
OEE parameters, such as high availability, in themselves may not be the only economically advantageous element to look at, said Sematech's Olaf Rothe. While this approach provides a very useful tool to improve productivity on constraint toolsets, it still misses opportunities to reduce waste throughout the fab. "Even with a 94% efficiency, if a tool goes down suddenly and you don't look at the whole situation, you might have just missed a window of opportunity. If WIP accumulates at this station, the constraint toolset downstream may starve and not be able to cope later on with arriving WIP." Rothe suggests a better way to manage all the resources is to have as low a WIP into the line as possible so that wafers go through on a one-by-one basis. Reducing the overall wait time on the lot, while still maintaining a continuous process flow, will increase the throughput of the fab and the factory output. That might involve increased idle time of some toolsets, but the net result is shorter production times.
MES
Manufacturing execution systems (MES) feed data into OEE (Fig. 3). Volumes of data are collected to ensure equipment effectiveness, and volumes more ensure the manufacturing process is operating within certain limits. Tools are in place to catch anything outside those limits.
|
| 3. The flow chart illustrates how manufacturing execution systems (MES) feed infor mation to OEE, which is a subset of cost of ownership (COO).1 |
At TI, a single global instance over an MES performs ~13 million to 16 million transactions a day, just for tracking WIP movement and lots within a factory. From a test perspective, more than 50 million chips are tested each day, with the information stored in the databases. Management and engineering can then analyze the data and drive improvements.
Yield management
Software providers such as Synopsys (Mountain View, Calif.) provide fabs with manufacturing yield management solutions to control and enhance fab quality. Effective yield management maximizes the number of good units out of the total units processed through a particular operation. The connection of yield management to overall fab operations is mainly through various metrology operations including in-line defect, CD and film thickness metrology, as well as parametric and functional testing at the wafer and chip levels. All these operations represent a form of quality control. "But the key question is what is controlling the quality of these quality control operations," said Sagar Kekare, senior product marketing manager, manufacturing yield management, at Synopsys.
In metrology- and test-based screening, there is a risk either of having good die being called bad (an alpha risk), or of having a bad part escape the test and go into a customer's hands (a beta risk). Effective yield management reduces both these risks by capturing the total variance of the screening measurements and enables better understanding of the systematic and random components of total variance. This provides a statistical basis for the values set forth as specifications for the screening operations. Accuracy and yield-relevance of these limits is what dictates the quality of the QC operation they represent. Additionally, they provide guidance for optimizing inspection or test strategies to improve the throughput of the screening operations. The result is better feedback for process control in the most efficient and cost-effective manner.
At the current technology nodes, yield learning is gated by interactions between the design, the process variability and the test program. The logic IC industry has converged on design for testability (DFT), an approach that builds the circuitry of a given design in a manner that facilitates automated test pattern generation (ATPG). DFT also enables diagnostics of ATPG test results to identify where the chip logic has failed and under what circumstances.
One of the natural benefits of the diagnostics approach is capturing the very subtle design-process-test interactions that cause a chip failure. These interactions demand a strong statistical basis for specifications and limit setting for the outcome of the tests. The manner in which the test engineers set up a test program for the given set of specifications could impact up to 2% of yield. A thorough characterization effort during the yield ramp phase may flush out such interactions and provide a path to test program optimization. Therefore, unlike test plans with perhaps four different test conditions and 1000 test vectors each, three test conditions with 400 test vectors each may be seen as sufficient for equivalent coverage of all failure mechanisms.
In order to get to the level of confidence needed to recommend such test program optimization, one must be able to analyze the test results in light of fab and design information. Moreover, the data needs to be accumulated over a statistically valid sample size and thoroughly analyzed for multi-variable correlations. An integrated yield management architecture that analyzes design information with test and fab data will further enhance the control capabilities to a level that is critical for future technology nodes, Kekare added.
Suppliers respond
To address the changing economic environment, tool manufacturers are implementing programs to assist chipmakers with operating cost reduction, productivity improvement and lifetime of the equipment, to maximize the use of their installed base. At Lam Research Corp. (Fremont, Calif.), a number of productivity programs are being implemented to optimize cost per good wafer out. Spend reductions of 15–25% and up to 20% output improvements have been demonstrated, according to Abdi Hariri, Lam's group vice president, Customer Support Business Group.
In one case study, along with declining ASPs of memory devices, a fab with a large installed base of etch systems sought to reduce operating costs. Additionally, they wanted to increase productivity while going to the next technology node, due to additional etch steps. Lam's strategy was to reduce the fab's operational spending by optimizing application-specific part designs, both consumables and non-consumables, extending the lifetimes of their parts and improving maintenance practices. The strategy further targeted reducing overhead time associated with tool operation and optimized process recipes for higher productivity.
The impact (Fig. 4) included a >25% reduction in spending while maintaining constant output using the existing installed base of systems.
|
| 4. Operational spending reductions were achieved by optimizing application-specific part design, part lifetime and maintenance practices for etch systems at a major customer site. (Source: Lam Research Corp.) |
Applied Materials (Santa Clara, Calif.) helps its customers develop and implement strategies to maximize output and shut down unused equipment in order to shrink "operating footprint" and reduce labor, maintenance, energy and facility costs. The company uses OEE as a key measure of customers' productivity and follows a rigorous discovery process to identify improvement opportunities to boost OEE of bottleneck tools and increase tool predictability.
In addition to providing one-time improvements that are implemented as a result of the discovery process, automation solutions enable continuous real-time monitoring of OEE performance. MES is used to track throughput and make comparisons against maximum speeds. A platform called E3 monitors the input variables to the tools, including pressure, temperature and gas flow, and compares them against "golden signatures." Deviations from normal operation are highlighted, enabling predictive maintenance at a planned time and reducing unscheduled breakdowns.
"It's not just a snapshot in time. You want to see the variation week to week, month to month. And that is only captured by specialized software that can pull the data from different data sources," said Norm Armour, vice president of Applied Global Services Fab Consulting at Applied Materials. The focus, therefore, goes beyond a given number, to the variability and trending of OEE and its components. "If you can shrink the variability of OEE, you have even flow of product, much more output and smaller loss of productivity, " he said.
Armour emphasized that WIP management is critical for achieving high OEE. The key is to have a buffer WIP upstream of the constraint tool, which runs at maximum OEE. This tool must not run dry or become WIP starved. Advanced performance software products such as products for real-time dispatch provide capabilities that allow the fab manager to ensure inventory is always in the right place to run high OEE at the bottlenecks. This includes FOUP delivery to critical bottlenecks, and optimizing delivery times by adding zero footprint stocker capabilities where FOUPs can be stored at the critical bottleneck tools.
Greater predictability
Most process equipment is utilized in the 60–70% range with bottleneck tools ~85%. However, there are competing factors such as cycle times. Running at >85% in terms of total utilization, commented WWK's president David Jimenez, can build inventory and ultimately hinder cycle times. As queue time builds, there is a "knee" where cycle times through the factory start to accelerate upward; a characteristic curve for every factory. However, with greater predictability in terms of equipment availability, the knee can be pushed down, enabling a fab to operate at a higher loading without significantly increasing cycle times.
Though not a new concept, predictability is nonetheless powerful, moving equipment availability from an essentially random issue to a systematic one. Unlike other industries, high-tech companies typically do not run a piece of equipment long enough to eliminate randomness, Jimenez noted. That is perhaps why predictive maintenance and remote diagnostic capabilities that companies have on the equipment side need to be implemented on the factory side.
He suggested the solution is in 3-D operational simulations, giving management the ability to test a variety of realistic scenarios without tying up equipment and determining start times that match factory capacity. To predict an event and plan around it adds tremendous control in facility operations.
The bottom line is the concept of OEE does not slow down when business slows down. Now is the time to improve the numbers so when the orders pour in, the efficiency of the fab will be better than ever.
| Reference |
| 1. V.H. Dhudshia, Hi-Tech Equipment Reliability: A Practical Guide for Engineers and Managers, iUniverse, 2008, p. 59. |
32 nm Marked by Litho, Transistor Changes
03/13/2010Gaining Control over STI Processes
03/13/2010Who Will Gamble at 0.13µm?
03/13/2010
VIEW ALL PODCASTS
