Accelerated time to market for future 300-mm fabs
11/01/1997
Accelerated time to market for future 300-mm fabs
Ed Ward, Ron Horwath, Hamilton Hayes, Semi, Mountain View, California
Bob Bracamonte, Ehrlich-Rominger, Los Altos, California
The semiconductor industry`s legendary ability to reduce cost/function by approximately 30% a year has been a key factor enabling its explosive growth. It is also taken as an article of faith that the industry will continue to grow rapidly if it can maintain its historical learning curve. The current transition to 300-mm wafers provides a clear opportunity to improve the methodology of building and running fabs.
The industry is now at a crossroads. The continually escalating cost of building new fabs has made achieving acceptable financial return significantly more difficult. Key areas of improvement such as yield, wafer size, and feature size that have led to phenomenal productivity gains in the past are no longer adequate to maintain historical trends. Jim Owens of Sematech has quantified the projected shortfall in necessary productivity improvements (Fig.1). Owens indicates productivity improvement could fall to approximately half of its traditional rate without new sources of productivity enhancements [1]. Semiconductor Equipment and Materials International (Semi), Sematech, and the Semiconductor Industry Association (SIA) have identified several areas that must be addressed to maintain the industry`s historical learning curve. Noteworthy among these is improvement in the time needed to bring a new IC factory on line [2]. The modeled economic benefit of accelerating fab start-up is substantial [3]. For example, a logic fab up and running in half the normal time would have the potential to reap an additional $20 billion in cash flow in its first three years of operation. The model demonstrates different but significant improvements for all product types.
Figure 2 on page 100, the cumulative cash flows for a typical and a "fast-track" fab, illustrates the economic benefits of accelerating fab construction and start-up for a 200-mm fab. (Even greater benefit is realized with a 300-mm fab.) The time for investment recovery for a fab producing a medium-priced microprocessor can be roughly cut in half, from 32 months for the typical scenario to 18 months. The model assumes 20,000, 200-mm wafer starts/month, and 225 die/wafer. The average selling price is assumed to decrease from $500 to $250 within 15 months of first wafer production [4]. Figure 3 shows the time lines assumed in the model.
 |
Figure 1. Expected shortfall in productivity relative to historical trends.
The model assumes a $2 billion capital investment for the typical fab and a $3 billion capital investment for the fast-track fab, with the differential accounting for possible premium costs of accelerating tool production and delivery. (These are extremely conservative estimates and there is nothing in the model indicating a fast-track fab will cost more than $2 billion.) Capital costs for tool purchase, fab design, and construction are front-loaded in quarter one for the fast track, and quarters one through three for the typical fab. The computations assume overall equipment effectiveness and yield performance are the same for the two fabs and do not include these values.
The methodology
At Semicon/West `97, we presented a procedure to achieve a time to first silicon of one year and designed production output within two years. The goals of the methodology, to be achieved within the existing industry framework, are to produce first silicon within one year after initiation of design, initial product shipments within 15 months, and designed production output at mature yield levels within 24 months. The start-up technology is to be the most aggressive possible, consistent with high-volume manufacturing. The factory is to be upgradable for a minimum of two generations of technology.
 |
Figure 2. Time-to-market acceleration cumulative cash flow for typical and fast-track fabs.
This approach recognizes that an advanced fabrication facility must be constructed to support technologies still under development and to include new equipment not necessarily field-proven. It is impossible to avoid changing requirements or to forecast all potential obstacles. The challenge, consequently, is to design and build a fab in an uncertain and changing environment, that comes up immediately and works as designed.
 |
Figure 3. Fab construction and production ramp time lines for current and fast-track fabs.
According to Gary Gettel of Sematech [5], current procedures take approximately two years to achieve first silicon and four years to achieve mature fab production (Fig. 3). Work schedules to meet current time lines require seven days a week, 24 hours a day, making acceleration by simply working harder impossible. The approach proposed here combines cross-functional teams with the division of the fab into four modules, each as completely decoupled as possible from all other modules. The four modules are: the physical facility, the equipment, the material transport system, and the manufacturing execution system (MES). Each module can be optimized and fully tested as an independent entity before incorporation into the overall fab system.
The physical facility module. The physical facility should house the equipment, provide basic generic services (such as house gases, waste disposal systems, and power), and provide a comfortable working environment for factory personnel. It should be decoupled from the processing environment provided by the equipment. This decoupling allows an advanced fab to be housed in a noncritical building, designed and constructed from standard, predesigned templates and materials. Predesigned templates enable a fab with specific requirements to be fully designed and construction initiated in less than four weeks.
Ehrlich-Rominger, an architectural engineering firm, is investigating the feasibility of using predesigned templates for fab designs and believes that this idea is promising. Designs using templates have been developed for various fab configurations. Figure 4 shows an example of a fab layout concept based on a template approach with an exploded view of a cell concept. Further development of the concept is underway and will be presented at a future date.
The template concept uses standard, readily available construction materials to create standard cells that can be replicated as needed to satisfy capacity and technology requirements of specific fabs. Each cell provides basic house services on a grid system. A specific design is easily edited so services are only routed to the areas requiring them. A given fab design can include as much flexibility as desired and a fab site can also be easily laid out to accommodate additional standard cells for future capacity expansions. Decoupling the fab`s physical facility from the processing environment facilitates upgrades for new equipment and processes. The building can be upgraded to accommodate new, more stringent requirements as long as the basic fab organization remains appropriate
 |
Figure 4. Fab layout designed from templates with exploded view of unit cell concept.
Processing equipment. Processing equipment must create its own processing environment in special cases by locally upgrading the provided house services. If necessary, the equipment must also upgrade its effluents to a level that can be accepted by the house disposal systems.
Processing equipment must be plug and play. Equipment that has not reached a beta level of development at the start of a fast-track fab design cannot be considered as a candidate for the fab. The beta equipment must have demonstrated a minimum level of reliability (generally between 168 and 500 hr) under simulated fab conditions and be capable of achieving a Cp (process capability) of at least 1.3 on first generation generic processes. (A Cp will generally not be available for each fab-specific process. A Cp for each Sematech generic process, however, should be available and used to judge the capability of a piece of equipment.) The beta equipment software must be controlled, fully released, and capable of supporting all key wafer fab requirements. Wafer output and quality cannot be compromised by a missing software feature.
Initial equipment selection should consider the requirements for the next two generations of technology, as defined by the National Technology Roadmap for Semiconductors. Tools should have a clearly defined upgrade path for the next generation`s requirements. The equipment must perform flawlessly when measured against the requirements of the start-up technology, and users must be confident that it can be extended to the next generation of technology.
Equipment should be procured with standard options, and customization should be strictly avoided. With little field data available for standard configurations, imposing a custom requirement on the equipment vendor defocuses his efforts and produces alpha-level equipment. All equipment features must be able to exceed a predetermined mean time between failures on a 7 ? 24 basis under user conditions.
The material transport system. The material transport system safely stores and transports material without introducing particulate or molecular contamination. It must automatically dock with all fab stations through a standard mechanical interface using standard software protocols.
In order to execute an aggressive manufacturing ramp and maintain high yield, major yield-limiting problems like contamination must be designed out of the system. In order to minimize the effect of any contamination introduced between processing stations, processing equipment should receive wafers in a passivated state, complete a value-added process, and return the wafers to the handling system in a passivated state. A complete IC process, then, consists of the movement of passivated material sequentially from one processing station to the next with a minimum sensitivity to contamination.
MES. The MES should be restricted to recipe management, material scheduling, and inventory management. MES should reduce the overall complexity of factory construction and operations, assisting the modular methodology. Therefore, the system should not be used for any level of process control, which would couple the system to the equipment and significantly complicate the system. Sensor-based process control systems are required, but appropriate clustering should be employed to eliminate the need to tie equipment together at the fab level.
Cross-functional teams
This methodology depends on cross-functional teams with each member contributing specific expertise. The team includes the owner, designer, builder, key suppliers, and local government representatives. The aggressive goals set down here cannot be realized unless all team members are involved in the project from the beginning and are able to work in an environment of mutual trust and respect.
The owner. The owner is the team leader. He or she is responsible for creating a positive work environment and articulating clear project requirements. The owner must clearly understand the charter of the fab being built, its expected capacity, the technologies to be supported, and how much, if any, development is to be conducted in the fab. The owner is responsible for the fab output model and must determine the equipment set at the start of the project, in order to provide a minimum six-month lead time for the equipment vendor and to enable the designer to properly size facility requirements.
The designer. The designer is responsible for rapid turn designs based on standard templates. He must maintain a database of equipment requirements in a standardized format to instantly size facilities based on equipment selection. He must also provide guidance in selecting the most cost-effective design.
The builder. The builder is responsible for using the most cost-effective construction techniques for the area, maintaining a stable of subcontractors, and completely understanding construction activities in the area. The builder must procure all required building materials in a timely manner, and so may need to inventory certain long-lead-time items.
The supplier. The supplier is responsible for timely delivery of production-worthy equipment and trouble-free installation and start-up. On-time delivery should be within two days of the original order date and within two hours of the forecast provided two weeks before delivery. Equipment should be facilitized within three days, with initial process data within one week. It should meet generic specs within two weeks and meet fab-specific specs within three weeks. The supplier must provide cost-effective seven- days-a-week/24-hr-a-day support, whether using on-site technicians or remote diagnostics capabilities through a modem.
Local government. Local government must be sympathetic to the project and be willing to expedite all required permits. Building a fab on the proposed schedule is possible only if local government is involved in the project. Conceptual agreement between the owner and local government is an essential parameter in site selection.
The separation of the fab into four distinct, decoupled modules, each of which can be fully tested before being incorporated into a specific fab design, is critical to achieving the rapid design, construction, and start-up schedule depicted in Fig. 3. The pretested module concept eliminates all critical unknowns, while maintaining the flexibility necessary to produce a fab meeting any specific capacity and technology requirement. Using standard components in the physical facility design template allows the builder to achieve the required rapid construction schedule.
The rate at which a fab can be ramped to full production is largely determined by the technical problems encountered and logistical issues associated with recruiting and training a work force. Decoupling the processing environment from the physical facility, controlling the processing environment in each tool, and ensuring that each tool has a demonstrated process capability of 1.3, reduces the technical problems substantially. A closed material transport system controlled by a factory-level recipe and material scheduling system substantially reduces the risk of errors by a new, inexperienced work force, and allows the rapid ramp of fab production with minimum risk.
Summary
The rapidly escalating costs associated with fab construction and the imminent transition to 300-mm wafers have brought the IC industry to a crossroads. Significant productivity enhancements must be identified if the industry is to maintain its historical growth and achieve an adequate financial return. This article outlines a method that compresses the historical fab design, construction, and start-up time line. Models suggest that compression may result in as much as a $20 billion improvement in revenues over the first three years of a fab`s operation. Such potentially substantial time-to-market improvements should encourage the industry to investigate the possibilities of the modular approach presented here.n
Acknowledgment
The authors would like to thank Jim Linden of DPR Construction, Bill Rowe of Rowe Consulting, George Lee of Semi, and Dan Fleming of SVGL.
References
1. J. Owens, Sematech Chart, 1995.
2. "Factory Integration," National Technology Roadmap for Semiconductors, p. 146, 1994.
3. Cash flow model available from Business Transformation Associates, One River Point Plaza, #607, Jeffersonville, IN 47130.
4. Microprocessor pricing data from Microdesign Resources` Micro Processor Report, October 28, 1996.
5. G. Gettel, "Future Factory Level Issues and Needs," 2nd Annual Semi 300-mm Symposium, Santa Clara, CA, June 19, 1997.
Ed Ward received his PhD degree in electrical engineering from Purdue University in 1970. He has had extensive experience in both the IC and capital equipment industries. Ward has held executive positions at Advanced Micro Devices and Silicon Valley Group, where he was responsible for design and construction in capital equipment and IC fabrication areas. Semi, 805 East Middlefield Rd., Mountain View, CA 94043; ph 415/964-5111, e-mail [email protected], www.semi.org.
Ron Horwath received his degrees in metallurgical technology and applied physics from New York State University and Adelphi University in Garden City, NY, respectively. He has held senior management positions at IBM Corp. and at Sematech in the IC strategic planning organizations. Horwath is presently director of industry relations at Semi in Mountain View, CA.
Hamilton Hayes operates a consulting business, Business Transformation Associates, focused on business and manufacturing process improvement and economic analysis. He is active in the semiconductor industry in the areas of equipment and performance analysis and standardization, software systems development, factory performance modeling, and Semi standards development.
Bob Bracamonte is a principal with the architectural firm of Ehrlich-Rominger. His 25-year architectural career includes a lead design role on major cleanroom commissions for clients like Micron Technology, Taiwan Semiconductor Manufacturing Co., Advanced Micro Devices, Delco Electronics, Digital, IBM, Raytheon, and International Rectifier. Most of these projects were completed using design/build accelerated project delivery.