The challenges of macro integration for fully automated 300mm fabs
10/01/2000
Michael Chase, Douglas Scott, Jeff Nestel-Patt, PRI Automation, Billerica, MA
SPECIAL REPORT: State-of-the-art manufacturing
overview
With 300mm wafer processing, the industry should finally realize the dream of hands-off wafer processing — a factory run by a finely tuned "silicon machine." The route to this concept will require some rigorous and fundamental changes by those responsible for fab automation, however. For example, the industry must address the accuracy, access, and application of process tool data; suppliers must step up to the ownership of fab-wide integration of production resources; and everyone must adopt the principles of industrial engineering.
 The Semiconductor300 cleanroom at the Infineon site in Dresden, Germany. |
In the factory of the future, fab-wide integration will create a finely tuned "silicon machine" — an operating system for the wafer fab. Wafer processing tools will be web-enabled, operating within a secure network connecting manufacturers with customers and suppliers. If a tool needs attention, it will alert the on-call technician, who may be in the fab, at the supplier's site, or even hiking. Catching a plane to fix a problem will no longer be necessary. The secure network will allow fast response and diagnosis, and immediate escalation can bring the world's leading experts to address the problem remotely.
Instead of being largely autonomous, as they are today, wafer-processing tools in the fab of tomorrow will be part of a mission-critical set of integrated solutions. Full fab automation will eliminate manual errors, making manufacturing virtually foolproof. Prior to any run, checks will be automatically performed to verify that the right lots will be processed at the right tools, and fab-wide material handling systems will ensure that the lots arrive at the tool on time. Advanced process control (APC) will fine-tune control parameters before recipe download. During operation, real-time data about the tool, the process, the materials, and the work in progress (WIP) will be automatically gathered and uploaded to the factory management system, the successor to manufacturing execution system (MES) software.
Where is the industry today on the road to what will be the fully automated fab of future? How long before we have finely tuned silicon machines building silicon machines?
On the road to 300mm
Today, the semiconductor industry is moving slowly, but inexorably, to 300mm-wafer fabrication. The dual premise of more chips/wafer and economies of scale is widely accepted.
What is less certain, and cause for much of the current caution, is how to reduce risk and optimize the investment in 300mm. How can fabs improve the productivity of their manufacturing operations to ensure that their investment pays off as quickly as possible and keeps delivering returns?
 |
The semiconductor industry decided early on that standards would be developed to ensure a smooth transition to 300mm manufacturing. Material and information handling standards are prime enablers, providing a unified platform from which different tools and systems can interoperate. Standardization means that manufacturers are able to buy components from multiple suppliers without regard for proprietary restrictions, but simply adopting standardized equipment or software components into an automated fab will not automatically produce optimized production results. Standard interfaces do not by themselves lead to tightly integrated systems. The plug-and-play approach may work to a manufacturer's advantage when discussing price with suppliers, but they may be opening the door to huge integration costs to get all the systems to work together seamlessly.
 PRI Automation's AeroLoader (above) and the AeroTrak transport system (right). |
Nearly everyone expects full-fab automation to be the rule rather than the exception as 300mm manufacturing matures. For example, both IBM's [1] and Intel's [2] strategies for 300mm include fully automated in-bay material handling.
Full-fab automation includes two key elements: automated material handling systems (AMHS) to move WIP throughout the factory from one process chamber to another, and factory management software that converts the flow of data into information, transforming the fab into an intelligent manufacturing environment. The expectation is that the combination of these two systems will create a highly efficient manufacturing operation capable of producing a large number of chips in the least amount of time and at the lowest possible cost.
The figure on p. 54 illustrates an AMHS layout, showing the integration of the main aisle of the fab with two process bays using an overhead hoist transport (OHT) delivery system. By integrating the process bays and the main aisle with the OHT, manufacturers can reduce the amount of AMHS equipment. Today, most fabs have a wafer stocker located at the head of each process bay. In the integrated fab, the number of stockers is reduced because the OHT can service the throughput requirements of the process tools using only one stocker, as shown in the figure.
While the productivity gains from AHMS installations (like that shown in the figure on p. 54) are clear, there are many questions that must be answered before the entire fab turns into the finely tuned "silicon machine" outlined at the beginning of this article. The answers lie in three key areas: accessing and making better use of process tool data, addressing fab-wide integration of production resources, and the speed at which the industry adopts the principles of industrial engineering.
Much of the expected gains in productivity resulting from the "silicon machine" will come from optimizing all of the fab resources into a coordinated and highly intelligent manufacturing environment. The transition to 300mm is an important inflection point whereby semiconductor manufacturers and equipment suppliers can form greater partnerships in addressing the most critical areas on the shop floor to ensure greater productivity. One area that is receiving attention is advanced planning and scheduling software. Ensuring that the right material is in the right place at the right time is the work of robust scheduling software. Realizing the productivity gains required in 300mm fabs will come from optimizing all of the shop floor production resources (see "The challenge of scheduling a wafer fab," page 54). New, powerful scheduling solutions, capable of scheduling and quickly rescheduling all shop floor resources to meet daily production goals, are finally coming to market.
Addressing inaccurate tool data
Over the last 30 years, reduced feature size, larger wafers, and improved yields have helped Moore's Law fulfill its prediction of doubling the number of transistors on a chip every 18 months. According to the 1999 International Technology Roadmap for Semiconductors (ITRS), though, most of the known technologies will reach their limits within the next 10 to 15 years, if they haven't already. Throughput is already replacing yield as the most significant discriminator of product cost in the fab.
As the industry matures, success will increasingly be defined by production efficiency.
Because 300mm fabs will become fully automated over time, it is imperative that we begin by addressing and resolving problems not addressed in today's 200mm fabs. Specifically, factory automation is severely complicated today by inaccurate information coming out of the tools. For example, using their tool performance tracking platform, engineers at International Sematech undertook a detailed evaluation of semiconductor equipment communications standards (SECS) messages at 14 fabs [4]. They found that 5-20% of messages transferred from tools to host systems were inaccurate. In three leading-edge fabs, the error rate was an astonishing 30-40% [5].
The Sematech study showed that there are many improvements needed in SECS implementations by equipment suppliers [4]. The study found problems with
- minimum list of event identification messages — the ones most often missing include transport time events and recipe step starts-stops,
- attaching reports to messages — some SECS implementations provide wafer start messages, but do not provide any information about the wafer number, the chamber, or the recipe,
- standard message numbers — suppliers use different numbering conventions in their SECS implementations, even for different tools from the same supplier (e.g., one tool uses message No. 84714 to designate a cassette start while another uses No. 33554447), and
- message inconsistencies — the lot start message may be issued when the cassette is loaded, when the wafer is taken from the cassette, or when the wafer is placed in the chamber.
 AMHS layout for 300mm wafer fabs where main aisle is integrated with two process bays. |
From the Sematech study, errors in SECS messages also included [4]
- data in conflict with the known equipment process flow — equipment emits messages that violate the known logical sequence of events (e.g., if each wafer is supposed to receive 1) a metal etch, 2) a strip, and 3) a cooling recipe, messages are emitted out of order: 1, 3, 2 or 2, 1, 3, etc.),
- data dropped in transmission — for example, a lot start message is received, but the lot complete message is dropped (this particular problem was found to be significant and widespread),
- data with incorrect time stamps — time stamps from equipment messages deviate from the fab "standard clock" by up to 10 minutes, and this becomes critical when a fab wants to use automatically collected data to improve manufacturing processes,
- conflicts in time stamps — for example, the event time for the lot start message contains a time different from the transmission time (some deviations of over an hour were found), and
- data critical overload — some tools can provide more than 5000 alarms, and some sites have enabled all the SECS messages on a tool to determine what messages are available, only to have the sheer volume of messages bring the tool down (work needs to be done not only in determining which messages are missing, but which messages are unnecessary).
In addition to these communication errors, experts have stated that 50% of tool downtime problems are caused by software [6].
Our analysis of these facts is that there is a major disconnect in the industry between expectations for full automation in 300mm fabs and the quality of today's tool software.
The expectation is that standardization enables semiconductor manufacturers to buy and integrate components from different suppliers, but the reality is that this often means that they must then develop in-house integration expertise to create the integrated system from the various components. This approach can expose manufacturers to tremendous integration risks, delaying implementation and increasing development costs on already overworked development resources.
Automation's integration ownership
In an ideal world, one would like to have solutions that are pre-integrated (integrated by suppliers before installation at the customer site) or integratable (capable of being integrated with minimal time and effort). Each component of an integrated solution can be a self-sufficient point solution or be integrated into a larger suite of products. This enables a semiconductor manufacturer to buy the components needed today, and then add further components in the future as needed, giving manufacturers greater responsiveness and flexibility. Fully integrated solutions provide rapid implementation, reduced complexity, adaptability to changing environments (agility), cost-effective integration, ease of use, and reduced cost of ownership. When correctly implemented, these solutions allow one to mix proven legacy systems with new or expanded solutions, and to mix and match material handling automation (MHA) and information handling automation (IHA) products from different suppliers.
With a well-structured, "integratable" automation architecture in place, additional benefits may then be gained by using a single supplier, or collaborating suppliers, for several components (i.e., buying them already pre-integrated):
- lower-cost solution — a large portion of traditional integration will now be a product solution, instead of a custom solution provided by a systems integrator or the customer,
- reduced risk — multiple CIM solutions can be developed, tested, upgraded, pre-integrated, released, and installed as a single set of products, reducing the risk of delays and cost overruns, and
- easier to take advantage of product upgrades — there will be less testing and resources required by the customer, products will be "in sync" regarding platform versions and third-party packages, and product enhancements will be validated across the complete integrated product suite.
Long-term profitability will not come from possible short-term capital savings on equipment but from creating a highly integrated set of components that make up the production resources found in 300mm fabs.
Gains from equipment performance and reliability have improved greatly over the past three decades, creating unmatched productivity improvements. Additional gains can still come from incremental tool improvements, but overall equipment effectiveness (OEE) is still at unacceptable levels because of hardware, software, or maintenance glitches. As these problems are resolved, there should be a corresponding improvement in overall factory effectiveness.
Significant factory operational gains will come not from the tool itself, but from the relationship of the tool to other objects: other process tools, metrology tools, processes, schedulers, APC and AEC (advanced equipment control) controllers, materials, people, and material handling systems. Optimizing individual tools alone may not lead to improvements in the overall productivity of the fab.
No machine is isolated, and the ultimate objective is a highly efficient integrated system, not brilliant individual tools.
Adopting industrial engineering
In addition to the industry's historical focus on equipment effectiveness and on physics, chemistry, and process technology, a new focus is now coming to the fore, namely industrial engineering, which embraces the goal of overall fab effectiveness. The prime example of this discipline has come from Rob Leachman and his team at the University of California at Berkeley; they have defined the various levels of material handling and information handling automation (see table). This shows increasing levels of both MHA and IHA systems typically found in fabs today.
 |
Increased automation permits a significantly reduced number of operators and supervisors, usually coupled with a slightly higher number of engineers and technicians. A relatively large reduction in the number of operators is observed when both MHA and IHA levels are increased. At mid-level production volumes (20,000-25,000 wafer starts/month [WSPM]), simultaneous improvements made in these two categories of automation produce a 10-20% reduction in the number of operators. At high-level production volumes (45,000-50,000 WSPM), moving from medium-high to high automation levels generates an even more significant reduction in the number of operators — greater than 35%.
When improvements to these two types of automation are considered independently, the reductions in direct labor are typically smaller than when both types of automation are varied together. This suggests a synergistic effect of MHA and IHA on fab staffing requirements [7].
Further, transitioning from no automation to full automation reduces staffing costs by about $41/wafer at the 25,000 WSPM volume level and by about $37/wafer at the 45,000 WSPM level [7]. This works out to a savings of $12.3 million/year for a medium-size fab and $22.1 million/year for a large fab. It is likely that savings from the modest staff reductions would be more than offset by the costs of equipment and systems, but automation may generate positive impacts on OEE, yields, development time, ramp time, and cycle time. These benefits are likely to be substantially greater than those accruing from staff reductions [7].
Conclusion
It seems clear that answers lie in adopting the principles of industrial engineering (see "The need for industrial engineering," p. 60) and using fab-wide automation as key enablers to meet the business goals established for 300mm fabs. The rules of semiconductor manufacturing are changing, and the companies that will remain competitive will be those that adopt the new rules of 300mm manufacturing and apply them throughout their manufacturing operations.
 Material handling automation (MHA) |
Realistically, manufacturers have a full-time job making faster, more complex chips at the lowest possible cost — in other words, running their core business. In the 300mm realm, where increased opportunities for chip yield and economies of scale abound, macro integration and optimization of the entire production-manufacturing environment is more important than ever. Manufacturers should look to partnerships with companies that provide expertise in automation systems and software and industrial engineering principles. This will be the key to avoiding the delays and overrun costs that can be incurred dealing with different operating system versions, multiple vendors, and different third-party products. Solving the integration problem right from the start will put manufacturers on the road to quicker implementation, faster ROI, and better yields.
Michael Chase holds a BS in electrical engineering from the University of Lowell and an MBA from New Hampshire College. He has more than 11 years of experience in the semiconductor equipment industry. Chase is VP of marketing, Factory Systems Division, PRI Automation, 805 Middlesex Turnpike, Billerica, MA 01821; ph 978/679-4270, fax 978/663-9755, e-mail [email protected].
Douglas Scott is a pioneer in the development of semiconductor factory management systems and was recently chosen to receive the 2000 Semi Award for North America for his contributions to factory integration. He founded the first MES company, Promis Systems, and has been delivering manufacturing systems to the semiconductor industry for 22 years. Scott is VP, strategic marketing at PRI Automation; e-mail [email protected].
Jeff Nestel-Patt holds a BA in English literature from the University of Wisconsin and advanced studies in high technology marketing. He has more than 15 years experience in marketing communications in high-technology companies. Nestel-Patt is director, corporate communications at PRI Automation; e-mail [email protected].
The challenge of scheduling a wafer fab
Scheduling a wafer fab with the intent of optimizing equipment utilization and cycle time is an extremely complex task. The sheer number of process steps, lots, wafers, valid products, active products, types of processing equipment, specific processing equipment capabilities, lot types, lot priorities and other related factors create an extremely difficult scheduling challenge. For example, the following numbers are not uncommon in the semiconductor wafer fabrication industry:
- a WIP level above 100,000 wafers,
- 1000 steps in a complete fab process,
- 5000 distinct and defined process flows,
- 2000 valid parts that can be built on those flows,
- 800 different products active in the line at any given moment,
- 500 different pieces of processing equipment, each needing a schedule of qualification and preventative maintenance,
- five or more types of lots requiring different processing protocols, and
- five or more lot priorities, each with separate target cycle times.
With modern factory management systems (FMS), the sequence of processing specifications and the actual processing done will be process flow-specific and, in ways that are sometimes important for scheduling considerations, lot-specific.
Adding to the challenge is the quietly chaotic nature of fab operations. This comes in several forms. The first is the steady rate of process and procedure changes that modify the way material is routinely processed in the fab. The second is the extent to which temporary unavailability of processing equipment, metrology equipment, or transportation equipment affects the explicit processing actions possible at a given moment. Finally, the path of the lot becomes determined not just by equipment availability but also by the results achieved so far. Lots exhibiting certain characteristics may require reworking or other variant processing, or may be held for additional evaluation and will therefore be temporarily unavailable for further processing.
Scheduling payoffs
Quantifying the benefit of optimized scheduling to a fab is not easy, but the system will help to improve equipment utilization and/or reduce cycle time, improve on-time delivery, decrease tool setup costs, and reduce transportation costs. For a facility that is operating sub-optimally, the implementation of an optimizing scheduling system could provide a huge payback. In a well-run fab, where serious attention is already being paid to all key factors, the advantages will be less dramatic. However, it is not necessary to posit a large increase in equipment utilization to justify the effort required, even if no other benefits accrue. Even a 1% increase in utilization produces $500,000/month for any moderate-volume fab that is running at capacity. This provides solid justification for all reasonable efforts to schedule and manage the WIP and workflow as precisely as possible.
The need for industrial engineering
The 1999 ITRS predicts that the technology barriers facing the industr are challenging suppliers and inspiring researchers to break them down. These technology breakthroughs (historically the mainstay of manufacturing success and the key contributors to Moore's Law) may include a device geometry shrink, a new tool or unit process, a new lithography technology, chemical mechanical polishing (CMP), copper, or 300mm. Surmounting these technology barriers has enabled the industry to achieve productivity improvements unmatched by any industry ever.
In addition to these technology challenges, the factory integration and economic challenges facing chipmakers are equally daunting, but often receive less attention. Consider two examples:
- multimillion-dollar machines often produce good product less than half the time, and
- the time taken to ramp a new plant, technology, product, process, or wafer processing tool is often twice as long as necessary and is the major competitive discriminator between leading and average performers.
Integration challenges transcend all bays in the fab, all departments, and all skill sets. Factory integration is often a patchwork quilt built on the quicksand of variant architectures, as well as different models of the plant, its resources, its unit processes, and its flows.
Jonah, the guru professor of Industrial Engineering in Eli Goldratt's novel "The Goal," taught us that a system of local optimums is not an optimum system at all; it's a very inefficient system [3]. Materials and processes have to be successfully choreographed among hundreds of tools to improve overall factory efficiency and profitability.
The importance of factory integration and a robust, industry-standard, fab-wide systems architecture to simplify the integration process is vastly underrated. Integration challenges can be mitigated by ensuring that the underlying technologies are compatible and that all components have published interfaces that conform to industry standards. In an ideal world, one would also like to have solutions that are pre-integrated, and a coordinated support strategy.
Industrial engineering (IE) takes a holistic view. Instead of viewing tools as autonomous, as is common today, industrial engineers view them as part of a complex web of highly interdependent activities. IE and manufacturing science improve the efficiency with which materials and information flow through the plant, and the efficiency with which resources are used. Although material transport and information management are often called "non-value-adding" activities, without them, no value is added. Factory management systems plan, schedule, and synchronize all the resources required to add value; for example, process tools, metrology tools, processes, maintenance activities, reticles, people, and material handling systems.
For years industry roadmaps have provided extraordinarily detailed requirements for the physics, chemistry and process technology associated with semiconductor manufacturing, 15 years into the future. It is interesting to speculate whether the worst excesses of our boom-and-bust cycles could be avoided if we attacked the business and economic challenges as rigorously as we attack the technology challenges.