Trends in contamination control in IC production tools
E.H.A. Granneman, ASM International, Bilthoven, The Netherlands, and DIMES, Technical University Delft, Delft, The Netherlands
Many modern IC production tools allow intratool storage and transport in very pure nitrogen gas. In such systems, the level of oxygen and water vapor is of the order of a few ppm or less. Although this approach results in a very clean ambient from a macroscopic (particle) point-of-view, it can lead to increased levels of microscopic (molecular) contamination that must be eliminated by special measures. Further, the emerging strategy and standards for 300-mm wafer fabs in terms of wafer carriers and intratool minienvironments are discussed. Most 300-mm fabs are expected to use some form of minienvironment.
Microcontamination control requirements become more stringent with each new IC generation for two reasons: the dimensions of typical IC features are decreasing so rapidly that particulates that were harmless in previous generations are now "killer" defects; and despite the decrease in feature size, overall IC dimensions increase with each new generation. To maintain acceptable production yields, the number of defects/unit area must be reduced.
According to the Semiconductor Industry Association (SIA) Roadmap, US industry experts believe particle density/unit area will decrease by a factor of five over the next decade. At the same time, the minimum particle size for inclusion in the density count will decrease by a factor of four. Table 1 considers only particles that are added to the wafer surface once the cassette of wafers is placed in the intratool ambient. Particles added during cleanroom transport and storage are not included in these numbers, but, obviously, contamination from the cleanroom must also be reduced. Solutions include using integrated minienvironments for both tool-to-tool transport and intratool ambient (Fig. 1).
Figure 1. Transfer of cassettes with wafers from the SMIF pod to the tool minienvironment. At left, the SMIF pod is opened in the cleanroom. After the elevator lowers the cassette, the robot arm transfers it from the SMIF station to the tool enclosure. The wafers are briefly exposed to cleanroom air. At right, the airtight integration of the SMIF pod and the tool minienvironment is shown. The wafers are not exposed to cleanroom air.
Economics of SMIF fabs
The SMIF container ("pod") for tool-to-tool transport was first introduced in the mid-80s, but is only now gaining acceptance (Table 2) . Minienvironments would lead to a substantial reduction in contamination and hence a significant yield improvement, as demonstrated in the next section. In addition, SMIF pods allow more relaxed cleanroom requirements , reducing both cleanroom investment costs and operating costs . A recent study from Meissner + Wurst  indicates, however, that the direct cost benefit is not really worth the investment (Table 3).
The initial investment for a SMIF/minienvironment fab is considerably higher than that for a conventional fab, due to the high costs of the SMIF-based cassette handling system and its integration with the processing equipment. The operational costs of such a fab are lower, but if one only considers the initial investments and the annual savings in running cost, the additional investments have a payback time on the order of 10 years!
Instead, the annual revenue resulting from higher yield drives the minienvironment trend: a projected turnover of $1000-$4000/m2 (Table 3). Lower cleanliness levels also reduce the time it takes operators to change gowns. The quality of the workplace improves when cleanroom requirements are less stringent. Futhermore, SMIF pods offer the option of electronic lot tracking of wafer lots, reducing the risk of misprocessing. The tool can also maintain a lot-based processing history for correlation with metrology results to prevent yield losses and aid rapid detection of process excursions .
Apart from tool-to-tool minienvironments, intratool implementations of minienvironments are also popular in IC fabs. Figure 1 shows how a cassette with wafers is transferred to the minienvironment of the wafer-processing tool. Two types of intratool minienvironment exist. The first type is the most advanced (and costly): here, the vacuum load lock is pumped down immediately after introduction of the cassette (Fig. 2). Such systems are used in almost all single-wafer processing applications. Depending on the surface cleanliness requirements, the vacuum level of the wafer transfer module and processing modules can vary from 10-6 Pa (10-8 torr) to 0.1 Pa (10-3 torr).
Figure 2. Vacuum cluster wafer-processing system. The chamber is evacuated after a cassette is placed in the input/output (I/O) chamber. The wafers are transported in vacuum one at a time from the I/O chamber to one or more processing modules. Once processed, the wafers are transferred back to the same I/O chamber. After backfilling this chamber to air, the cassette is returned to the cleanroom. The SMIF cassette robot in front of the I/O chambers is not shown.
The second type of intratool minienvironment places the cassette in atmospheric clean air or a nitrogen ambient. An example is the modular clustered vertical furnace (Fig. 3). Process temperatures range from 600-1200?C. During loading and unloading, the furnace temperature is lowered to 600-700?C. Recirculating the air or nitrogen ambient is usually necessary to reduce operation costs. Although this design is much cheaper, it is only used in systems where it is acceptable to expose the wafer surface either to air, or to small amounts of oxygen and water vapor present in nitrogen-purged systems. In the latter, a typical background oxygen concentration can reach 1-10 ppm, corresponding to an oxygen partial pressure of 0.1-1 Pa (10-3 -10-2 torr). The nitrogen purge is mandatory when oxidation of the wafer surface must be prevented, especially in the furnace section, where the wafers are loaded or unloaded at a high temperature.
The contamination levels in a properly designed tool minienvironment can be very low. Figure 4 presents the airborne particle count (just above the wafers) as a function of time for two cases. In the first one, a conventional furnace is placed in a high-quality Class 1 cleanroom, with clean air flowing past the cassette and wafer loading area, keeping it clean in the process. The large particle bursts (average count = 13.4 particles/ft3) arise from activities taking place in the direct vicinity of the wafer.
In the second scenario, the fab operates in a Class 1000 cleanroom. Cassettes are transported in SMIF pods and transferred to the (closed) minienvironment of a vertical furnace without exposure to the cleanroom ambient. The corresponding particle counts are substantially lower (average count = 0.1 particles/ft3), even though the cleanroom is of a lesser quality .
Particulate contamination is not the only contaminant present in cleanrooms and in tool minienvironments. Molecular contamination can also cause defects in ICs. A non-negligible concentration of volatile hydrocarbons exists in the air in a cleanroom [11, 12, 13]. If wafers are stored in open cleanrooms, a few monolayers of hydrocarbons are absorbed on their surface, producing a film that has a negative impact on several specific processes. For example, the film influences the growth rate of thin gate oxides and the quality of epitaxial films grown at low temperatures. If heated rapidly, hydrocarbon contaminants form silicon carbide (SiC), a compound almost impossible to remove. These effects are becoming more serious with new device generations.
Carrying the wafers inside a SMIF pod would only partially protect them from hydrocarbon contaminants. When not treated carefully, the plastic cassettes holding the wafers inside the SMIF pod de-gas. Storage in such a cassette in an otherwise hydrocarbon-free ambient can result in the deposition of a monolayer of organic material in 10-20 hr.
Figure 3. Vertical furnace system with atmospheric minienvironment. a, b) This particular system has three independent modules: c) one for cassette storage, one wafer-handling robot, and d) one processing (furnace) module. The system can be extended with one additional furnace module . The principle of the air (or nitrogen) recirculation system is indicated in c). The wafers are processed in quartz boats d) in batches of 100-150. Once loaded, the boat is moved upward. The process can start when the boat is up and the process tube closed. In LPCVD applications, the process tube is evacuated.
Hydrocarbon contamination can also occur in the protective nitrogen (clean-air) ambient of the minienvironment of the tool, sometimes from unexpected sources. Consider the following example : TEOS-type oxide films (based on the thermal decomposition of TEOS, Si(OC2H5)4 in an oxygen ambient at temperatures of ~700?C) were deposited in a vertical furnace similar to the one shown in Fig. 3. When the boat with 100 freshly deposited wafers was lowered into the nitrogen-purged minienvironment below, tens of thousands of very small particles (haze) appeared, particularly on the wafers at the bottom of the boat (Fig. 5).
After careful investigation, we discovered the source was the process tube. Upon completion of each process, the elevator lowered the boat with processed wafers into the wafer (un)loading area (Fig. 3d). While the tube door was open to move the boat out of the furnace, large quantities of TEOS reaction by-products diffused out of the tube into the loading area. These partially cracked fragments recirculated with the nitrogen flow and condensed on the cold spots in the system. After another boat with fresh wafers is loaded into the furnace, the radiation from the furnace opening heats the top wafers in the new boat, in turn heating some of the cold spots. The heating causes some hydrocarbon molecules to desorb and condense on the wafers in the boat. In principle, such adsorbed hydrocarbons can desorb again if the surfaces are exposed to low temperatures (150-250?C) for a sufficiently long period. Desorption of hydrocarbon molecules generally occurs on wafers in the top section of the boat slowly heated by radiation from the tube. The wafers loaded into the bottom of the boat are transferred into the reactor almost immediately after loading. When heating of adsorbed hydrocarbons is rapid, the end result is not desorption, but decomposition and reaction with the underlying silicon to form silicon carbide. Such cracked hydrocarbons are hard to remove and can act as nucleation sites for enhanced growth, leading to particle-like defects in the film.
Figure 4. Airborne particle count vs. processing time in two vertical furnace configurations. Particle size > 0.1 ?m; a) conventional furnace in Class 1 cleanroom - average count = 13.4/ft3; b) furnace with closed minienvironment in Class 1000 cleanroom, cassette transport in SMIF pods - average count = 0.1/ft3.
Figure 5. Distribution of defects on wafers in various positions in a vertical wafer boat after deposition of a TEOS-based silicon-oxide film. The high sensitivity of the measurement tool allows observation of haze patterns (indicative of small particles and nucleation on molecular contaminants). The density of defects on wafers at the bottom of the boat is much higher than that on wafers at the top of the boat. The speckled pattern on the top wafers indicates the absence of haze.
Figure 6. The interface between the wafer container (with side-door access) and the processing tool. No separate (removable) cassette is present inside the container. The doors of the container and the tool are clamped together and moved downward inside the tool.
Figure 7. Interface between an Infab/Motorola wafer transport container and the front side of the wafer-processing tool ; a) parallel door opener -mechanism through which both doors are clamped together and moved downward inside the tool; b) load port (fixed position, 900 mm) or for microbuffer; cassette ready for (un)loading; c) speed load port for microbuffer - by adding cassette positions, the overall system throughput can be increased.
There are several solutions to this problem. Bringing the boat in slowly allows the hydrocarbons to desorb before they crack. Alternatively, loading wafers while the furnace is kept at a low temperature allows desorption to occur. The furnace temperature is not raised to the required wafer-processing temperature until desorption is completed. Neither of these approaches is attractive, as both severely reduce wafer throughput. A better solution is to prevent contaminants from reaching the wafer surface when the furnace is opened; providing a positive flow to the exhaust of the furnace (process vacuum pump is operated in a so-called "slow-pump" state) can completely eliminate the downward diffusion of TEOS by-products, and consequently defect generation.
This particular example demonstrates that a new class of problems, not present in conventional reactor systems placed directly in the cleanroom air flow, may arise in minienvironments -hydrocarbon contaminants, once allowed to diffuse into the nitrogen circulation system, are quite hard to remove.
300-mm production plants
As suggested by Tables 1 and 2, most (if not all) post-200-mm IC fabs will use minienvironments. Although not all standards are finalized, four wafer-transfer methods will probably emerge: cassettes with 13 or 25 wafers (including a test or a "dummy" wafer) and containers with the door opening to the bottom (as in the SMIF pods currently used) or to the side. Further, constraining each wafer in a cassette kinematically in a unique position via three pins prevents the wafers from sliding in the cassette slots during transport. Since the wafer position is accurately defined (to ?0.1 mm instead of ?1 mm), it is also easier for robotic wafer-handling systems to remove wafers from the cassette.
Two other options will also coexist. In the conventional approach, SMIF pods carry a separate cassette that holds the wafers (Fig. 1). In the second option, pioneered by Infab, in close collaboration with the Motorola 300-mm task force (Fig. 6) , the cassette and cassette holder are merged into one container that (kinematically) holds all wafers. A robotic wafer handler transfers the wafers directly from the cassette into the wafer-processing tool.
To maintain throughput, an additional wafer stocker inside the tool is needed to allow unloading of a complete cassette of wafers prior to processing. While a second wafer-handling robot transports these wafers from the stocker into the reactor to either fill up a batch or to feed a single-wafer-processing tool, the empty wafer container can be removed and replaced by a full one. Generally, two such interfaces between a wafer container and the processing tool allow for unloading containers, and then loading processed wafers back into a second container.
Since the doors of the container and the tool are only opened when there is tight connection between the container and the tool, this arrangement can be exceptionally clean. Once the airtight connection is made, both doors are clamped together and moved downward, away from the tool opening. The contaminated exteriors of both doors are pressed together with an "O"-ring seal, protecting the inside of the tool. Figure 7 shows the interface between the wafer container and the tool front surface.
Although the introduction of minienvironments has been slow during the last decade, the number of fabs based on the concept will gradually increase. Moreover, the majority of 300-mm wafer fabs will most likely be based on minienvironments. The primary reason for this increased acceptance is yield enhancement, not the reduction in cleanroom investment and operational costs.
In a number of applications, minienvironments can lead to a new class of problems related to the ineffective removal of molecular contaminants in stagnant or recirculated ambient gas. Without proper measures, such contaminants might find their way to the wafer surface.n
This paper was first presented at the 13th International Symposium on Contamination Control, ICCCS, September 16-20, 1996, The Hague, The Netherlands.
1. The National Technology Roadmap for Semiconductors, Semiconductor Industry Association, San Jose, CA, 1994.
2. A. Bonora, "There and Back Again," Solid State Technology, p. 141, May 1997.
3. Asyst product documentation, Fremont, CA, 1994.
4. A. Honold, Jenoptik Infab, private communication, 1996.
5. ASM Japan, ASM Korea, ASM Taiwan, private communication, 1996.
6. D.L. Tolliver, "Improve Control, Reduce Costs with Mini-environments," Channel, p. 5, May 1993.
7. H. Schneider, Meissner + Wurst, private communication, 1996.
8. "Bruce Sees SMIF Move for Process Control," Technology News, Solid State Technology, p. 62, June 1996.
9. ASM A400 batch cluster system, ASM Europe, Bilthoven, The Netherlands.
10. C.Y. Shu, "Three Years of Production Experience with Isolation Technology," TSMC report, Taiwan, 1992.
11. A.J. Muller, L.A. Psota-Kelty, J.D. Sinclair, Ext. Abstr., ECS fall meeting, Hollywood, Florida, p. 554, October 15-20, 1989.
12. J.E. Olsen, F. Shimura, J. Vac. Sci. Technol. Vol. A7, p. 3275, 1989.
13. W. Storm et al., "Measurement and Characterization of Organic Contamination on Silicon Wafers in Cleanrooms," CleanRooms International, p. S10, February 1996.
14. A. Hasper, ASM Europe, private communication, 1996.
E.H.A. GRANNEMAN received his PhD degree in atomic physics from the FOM Institute for Atomic and Molecular Physics in Amsterdam in 1976. From 1976-1984, he was group leader of the plasma physics group at the Institute, where he worked on particle accelerator development. In 1980-1981, he worked in the Magnetic Energy Division of Lawrence Livermore National Laboratories on plasma confinement experiments. Granneman joined ASM in 1984, and is presently chief technical officer at ASM International. He teaches in the Delft Institute for Microelectronics and Submicron Technology at the Technical University Delft. ASM International, Rembrandtlaan 7-9, 3723 BG Bilthoven, The Netherlands; ph 31/30-229-8411, fax 31/30-229-3823.