New Fab Construction
Ruth DeJule -- Semiconductor International, 1/1/1998
This 480,000 ft2 fab, located in Buchon, Republic of Korea, has an open ballroom configuration. (Source: Amkor)
Currently, 127 new fabs are in various stages of planning and construction, with total expenditures exceeding $115 billion. New fabs cost more than $1 billion, and 300 mm fabs may be more than double that amount. While fabs are being built worldwide, the Pacific Rim is seeing the greatest growth, with Taiwan the hottest place, noted George Burns of Strategic Marketing Associates (Santa Cruz, Calif.).
A device fabrication facility is a multiuse building consisting of office space and a mechanical area located below the manufacturing floor for support equipment, distribution ductwork, piping and power systems, all built around a production area or cleanroom. Chemical and gas storage along with waste collection zones are located outside this area. Each area has its own particular functions that have to be addressed in the design and construction of the facility. The production area is the focus of this article, typically a supercleanroom.
Architecture
The facility architecture is essentially a compromise of efficiency for today's process technology and flexibility to rapidly and inexpensively adapt to meet new process technologies and flows. Fabs traditionally are built around a ballroom-style cleanroom where process bays are connected to a central aisle and wafers are transported via an interbay rail system. This simple architecture offers a cost-effective manufacturing facility with optimal flexibility, according to John Weekley, vice president of marketing at Amkor Wafer Fabrication Services (Santa Clara, Calif.). Anam Semiconductor of Buchon, Republic of Korea, for example, has recently completed a 67,000 ft2 open ballroom-style cleanroom with waffle floor infrastructure (Fig. 1).
Fig. 1. The cross section shows a waffle floor construction between the 100 x 60 m cleanroom and the basement for facility connections. (Source: Amkor)
The open architecture imposes no physical constraints, and the underlying waffle floor provides feedthrough of all utility support from the basement and subbasement. The equipment can then be placed virtually anywhere in the facility and still have access to electricity, water, gas and waste disposal.
Minienvironments are expected to increase above the present ~10% for fabs built since 1995, according to Burns. Such was the choice at the Class 10 Anam facility. The wafers, housed in a SMIF pod, are kept in a Class 0.1 environment. The facility can be adapted to changes in a multiprocess mix supply and upgraded with the addition of new process nodes.
An architecture that is attracting interest is the modular fab. The most common approach has two modules. One shell encompasses both units, but only one side is initially facilitized. Another concept consists of six modules and a central area for distribution and general functions (Fig. 2). The layout provides a simplified flow of goods; it is less complex and therefore requires less expensive automation systems, said Stuart McIntosh, chief operating officer at Philips Semiconductors (Eindhoven,the Netherlands). The modules can be individual process steps or configured as mini- fabs. Instead of commissioning an entire fab, individual modules can be built and started as required. Theoretically, production could begin as soon as the first module is completed and subsequent modules installed as needed.
Fig. 2. This modular style consists of six units and a central area to simplify the flow of goods. (Source: Philips Semiconductor)
According to Meissner+Wurst (Stuttgart, Germany), for volume production in the United States and Europe, the rectangular bay/chase cleanroom remains the preferred design, while in the Pacific Rim, the minienvironment appears to be making more headway. Interest in minienvironments for next-generation devices is growing in Europe.
The three-level, as opposed to two-level, fab with two subfabs is growing in popularity worldwide (Fig. 3), noted Stefan Duerr of Meissner+Wurst. The advantages include improvements in overall space management of utility and process systems; smaller fab footprint; and reduction in contamination and sound level in the cleanroom. However, this could mean longer building times by one to two months and additional costs of ~$230/m2 of covered floor space. Additional utility distribution is also required on the first level for chemicals, power and exhausts and higher chemical dispense pressures for high-viscosity acids such as phosphoric and sulfuric.
Fig. 3. A trend toward three-level fabs could mean reductions in particles and molecular contamination. (Source: Weissner+Wurst)
Fire control
Polypropylene is the material of choice for wet decks because of its resistance to acids. However, the fuel value of polypropylene or polyvinylchloride (PVC) is roughly equivalent to "several tanks of diesel fuel," said Don Wadkins, manager of cleanroom protocol and tool integration at Jacobs Engineering (Portland, Ore.). Despite this liability, polypropylene is used in virtually every fab. To address the potential fire hazard, local fire suppression systems are now available within the wet deck itself, not just in the room or the duct work. Because of the presence of acids, water-free carbon dioxide systems are used for fire suppression.
Ironically, most of the local suppression systems are used in metal decks for the control of fires because of solvents, and not in the polypropylene decks where the fire risks of the deck material poses considerable danger, Wadkins said. The concerns about polypropylene, though, may be creating a trend toward decks that are primarily constructed with stainless steel but with polypropylene surfaces. At this point, those available are specially designed systems.
For areas such as electric rooms, sprinkler systems, which are pervasive in the fab, can present a problem because of the deluge of water during a fire. A new technology being developed is an atomized water mist system. The fine mist suppresses fires by absorbing heat with very small quantities of water. This technology could also be used to replace CO2 systems used in wet benches. However, this technology may not be accepted by all fire jurisdictions in all fire applications.
Other new methods for fire control include several hydrofluorocarbon (HFC) products as alternatives to CO2, which is an asphyxiant. Unlike CO2, HFCs do not consume or displace oxygen when suppressing a fire. Furthermore, these HFCs are considered superior substitutes for Halon, a perfluorocarbon (PFC), formerly widely used in computer rooms where water can cause damage. PFCs are no longer permitted because of their ozone-depleting effects.
Smoke detection
Smoke alarm systems for early fire detection are being installed in all new fabs and are being retrofitted into most existing fabs. When polypropylene burns, it produces a very hot particulate smoke that rises in the cleanroom. Smoke exhaust systems designed to remove this dense smoke minimize the spread of smoke with separate exhaust systems and smoke detection units that trigger the exhaust.
Photometry-based, high-sensitivity smoke detectors (HSSD) are used extensively in fabs better than Class 100 where detection of very low levels of smoke is essential. Once smoke is observed, extreme damage to the fab environment has already occurred. HSSD systems allow the fab owner to detect a smoke event long before an actual, visible fire has started and to take protective action to limit its spread. The high airflow in a fab tends to prevent conventional detectors from picking smoke concentrations. Typical ionization detectors used in office buildings would require many detectors positioned very closely together to be effective.
A typical HSSD system consists of a photosensitive detector connected to a vacuum pump that pulls air samples through a series of 1/2 in. diameter tubes with ~1/8 in. holes, spaced 1 in. apart. Generally, there is one HSSD system per bay, with tubing in the path of the return airflow stream, positioned under the flooring or in the side wall return. Because of its high sensitivity, this technology is not suitable for use in dirty environments, noted Phil Ricker, manager of electrical and instrumentation design at Jacobs. In the fab, however, smoke is a major contamination event, requiring work-stoppage. While catastrophic loss is insurable, smoke shutdowns may not be, even though they can critically impact production flow and productivity.
Energy efficiency
This year has seen greater attention to semiconductor manufacturing energy consumption and efficiency than previous years. SEMATECH, Electrical Power and Research Institute (EPRI) and the Environmental Protection Agency's Atmospheric Pollution Prevention Division are sponsoring studies and seminars in search of more energy-efficient solutions. It is currently estimated that as much as 800 kWhr of electrical energy is consumed for all manufacturing related to semiconductor devices from a single 200 mm wafer, which is enough energy to supply the typical household for two months, stated Charles VanLeeuwen of CAVLON Associates (Rio Rancho, N.M.). More than one-third is used by the fab with the balance going to the manufacture of the raw wafer, back-end processing, chemicals, materials, equipment and facilities.
Cleanroom heating, ventilation and air conditioning (HVAC) are the major energy consumers in the fab, accounting for 50% or more, while wafer processing tools account for 30-40%. There may be as many as 150-300 20,000 cfm air recirculation fan units in a typical large fab, each with 15 horsepower motors. One solution has been developed to drop the motor size by a factor of 3, to 5 horsepower. In addition, two technologies becoming more prevalent in new fabs may also prove to be energy savers minienvironments and full material handling automation. According to VanLeeuwen, minienvironments and au-tomation could substantially reduce the total amount of ultrapure air that has to be conditioned and circulated.
The manufacture of bare 300 mm wafers could become the most significant energy consumer if ways are not found to dramatically improve current polysilicon to finished wafer material efficiency, VanLeeuwen noted. For 200 mm wafers, the current overall conversion efficiency is believed to be <20% and ~6% for 300 mm wafers using today's technology. It is anticipated, however, that changes such as going from 100 kg to 300 kg crystal-puller charges could significantly increase the manufacturing efficiency of 300 mm wafers if ways can be found to maintain crystal structure with the large ingots.
Construction of 300 mm fabs
The past year has seen more activity worldwide in the 300 mm arena than anticipated. The advantages are clear enhanced productivity. First, 300 mm construction is expected to begin in the 1999-2000 time frame, and judging from accumulated information from the device community, new 200 mm fabs construction is expected to cease after 2000-01, according to George Lee, director of the 300 mm initiative at SEMI. Figure 4 displays the timeline of new building schedules and pilot lines of 27 device manufacturers worldwide. These figures are consistent with those of I300I and SELETE. While there is currently no change in stated plans, present economic conditions in the Asia/Pacific Rim could possibly cause delays.
Fig. 4. The transistion from pilot lines to full-scale production can be seen over the next five years.
A cost-effective alternative to new fab construction is conversion of current fabs to 300 mm. This is primarily a facilities upgrade of utilities and waste water. At this time, 300 mm fabs will most likely use more water than 200 mm as much water in a year as a city of 60,000 people. Numbers ranging from 1.5X to 2.5Xgreater water usage than 200 mm have been bantered around depending on dry vs. wet etch and the amount of recycling intended. Estimating the utility needs of the 300 mm facility has yet to be determined because much of the wet equipment that will dictate the amount of water consumption is just coming into place, noted Frank Robinson, president of I300I.
Faced with higher costs of building new 300 mm fabs, chip manufacturers are making greater efforts to maximize productivity by minimizing fab space. To support these efforts, Lam's (Fremont, Calif.) approach has been to scale up the transformer coupled plasma (TCP) source and chamber size of its metal, oxide and poly etch technologies, while maintaining the 200 mm footprint by optimizing the space typically required for components. This approach eases overall fab space requirements, lowers the risk of transition to new technologies and achieves enhanced capital productivity, according to Sanjay Tandon, product marketing manager for 300 mm at Lam.
Software
For the past 10 years, fab modeling software has provided a tool for the design of the most advanced fabs. Today, a billion-dollar fab investment makes 3-D simulation software a necessity. Advanced Micro Device's Fab 25, for example, used AutoSimulationsAutoSched software to analyze alternative material handling systems. The company was able to study interbay and intrabay handling in terms of the amount of time a tool spent waiting for material in the simulation. The Fab 25 model became the benchmark for other decisions such as tool set, layout, traffic intensity, stocker and track sizing, plus location, staffing analysis, training and ramp up. Even before construction was completed, the building team could observe factory operation in 3-D animation.
Motorola (Austin, Texas) is currently re-evaluating some of its fab design concepts to provide better operational performance to the end-user and to be responsive to management's need for better cycle time and optimized cost of ownership, said Phil Naughton, part of Motorola's new construction team. This optimization includes performing analyses of conventional cleanroom designs with large ballrooms using fan towers and fan deck designs vs. modular designs using fan filter units and/or minienvironments. Also being evaluated are the most effective techniques for 300 mm fab designs that require simulation to determine average cycle times, for example (Fig. 5).
Fig. 5. Motorola's 300 mm program used AutoSched simulation to determine average cycle times for direct transport to tool vs. through stockers. (Source: AutoSimulation)
A challenge in new fab design and construction are finding ways to shorten ramp-up times. One solution is software, such as Triant's ModelWare/RT, which aids in new equipment ramp up by providing "visibility" into the machine. The equipment's learning process is accelerated by the software, thus shortening the time it takes for a fab to bring its equipment up to production-ready speed. Advanced modeling technology and proprietary algorithms are used to solve complex process control problems. Data, such as temperature, pressure, RF power and residual gas levels, are collected from good runs and automatically turned into a comparison model. Every variable is correlated with each other to determine any deviations in the overall health of the equipment, at which point an equipment fault is indicated and location identified.
