Issue



Reducing DI water use


12/01/1996







Reducing DI water use

Thomas S. Roche, Motorola MOS 12, Chandler, Arizona

Thomas W. Peterson, The University of Arizona, Dept. of Chemical and Environmental Engineering, Tucson, Arizona

Vast quantities of deionized water are required for semiconductor manufacturing. Cost savings and regional water limitations motivate reduction, recycling, and reclamation. Substantial reductions can be achieved by simply eliminating excessive water flow through existing baths. A fully optimized fab DI system can only be realized through a thorough understanding of rinsing mechanisms and hardware design. Optimized baths allow for thorough cleaning of wafer surfaces with reduced total water. Recycling and reclamation generally require substantial upgrades to drain systems.

Ultra pure deionized (UPDI) water is the most heavily used chemical in semiconductor manufacturing. Current consumption runs at up to 2000 gallons of UPDI water/wafer produced in modern 200-mm process lines. Even though it is the most inexpensive chemical used, a large capital investment is necessary to provide the increasingly pure high volumes of UPDI needed. Combined with the growing environmental responsiveness of the industry in general, these cost concerns have led to the current interest in reducing water consumption. In addition to work in various fabs such as Motorola MOS 12 in Chandler, AZ, work on wafer rinsing is being carried out at Sandia National Laboratories, and at the University of Arizona`s Center for Environmentally Benign Semiconductor Manufacturing (CEBSM) which also involves Stanford, U.C. Berkeley, MIT, and NSF and SRC funding.

The principal use of water in the fabrication process is to remove other chemicals from wafer surfaces. UPDI can thus have as much effect on the wafers as any other chemical. The concentration of impurities in water must be much lower than those of even the most sensitive chemicals such as hydrofluoric acid, because water is either dried on the surface in spin dryers or is displaced from the wafer surface with IPA. Either method can leave contaminants on the wafer surface and, in the extreme, water spots can be seen.

In addition to metallic impurities, bacterial contamination is a significant concern in ultrapure water systems. Any modern water system for semiconductor manufacturing is designed to keep water flowing at all times, due to the belief that bacteria, which are always present in water systems, can be kept to a minimum if the flow in the pipe can be kept turbulent. Turbulence inhibits the growth of bacterial colonies. If this practice is extended to all tubing in the water system, a continuous "low flow" of water will go down the drain from every bath in the fab. Turbulence also avoids the formation of particle reservoirs; particles that deposit in stagnant water can be swept into the rinse bath when water flow is suddenly resumed. Thus, the water consumption in any fab is only partly related to the throughput of the fab. It is also dependent on the number of DI baths and other water-consuming tools in the facility, and the flow rate of the "low-flow" water through these baths.

In spite of the stringent purity requirements for water in semiconductor manufacturing, modern UPDI plants are able to produce the required water at a purity level unmatched by any other industry. As the complexity of processing increases and with it the number of cleans, the amount of water/wafer out and its cost increases.

Efforts to reduce water consumption depend on reduction, recycling, and reclamation. Reduction methods require actual lowering of water usage in a given process. Recycle methods involve reprocessing of some DI water waste streams for subsequent reuse within that process. Reclamation approaches use post-rinse water in an environmentally friendly way, in cooling towers or landscape watering, for example. This article will review recent work in the area of water reduction and will also discuss some of the changes that have been carried out to lower water consumption in cleanroom tools.

Reduction

Water use reduction is the focus of much interest because the production of one gallon of DI typically costs more than 2 gallons of incoming water supply. Prof. C.R. Helms and coworkers at Stanford are developing a model of the water usage in a typical fab that shows how consumption is affected by the number of cleans performed, the number of tools in the fab, the throughput of the facility, and the low flow rate of water in each bath. These are all significant factors, and this type of analysis allows for fab baselining and focuses reduction efforts. The model will also calculate the concentration of chemicals present in the rinse water.

Water purity in process is typically measured in the bath with a resistivity monitor, since resistivity is inversely proportional to the ionic concentration of chemicals. At 25?C, water can exhibit a resistivity of 18.2 MW if no impurities are present. This number will never be attained in an open bath, due to the dissolution of CO2 into the water. In fact, the resistivity measured will depend on the flow rate of water through the bath since the concentration of dissolved CO2 depends on the amount of time that the surface is exposed to the ambient atmosphere.

Resistivity is an extremely sensitive method for detecting chemicals in water. The concentration of chemical at the monitor and the concentration on the wafer are not necessarily the same, however. As a specially designed resistivity cell has shown [1], the tank exit resistivity can show a slower change than that on the wafer surface itself.

To reduce the amount of water used in rinsing, the process must be well understood. Rinsing chemicals off a solid object can be described theoretically, and the resulting model can be tested in a well-controlled experiment. In this case, the solid object is a wafer boat containing fifty 200-mm silicon wafers. Both the quartz boat and the silicon wafers can be considered as nonpermeable to the chemicals in use, so the removal of the chemical from the surface is not believed to depend on any specific interaction of the chemical with the surfaces.

Of course, the amount of chemical brought into the rinse bath is very important. The surface area of the wafers cannot be changed but the surface area of the boat needs to be minimized. The objective of the rinse process is to remove the chemical "completely." The exact definition of "completely" can only be decided by electrical or surface analysis tests on wafers.

In our work and in work previously reported, the diffusion of the chemical from the surfaces of the wafer and the boat was a fast process compared with the removal of the chemical from the bath. Thus, the design of the bath is extremely important to the water utilization of the tool.

Bath design

Substantial work has been done in the area of rinse bath modeling, including full-scale numerical simulations presented by the Sandia group [2], as well as a very good analytical model, based on principles of the continuous stirred tank reactor (CSTR) presented by Helms from Stanford and coworkers at SCP Global Technologies [3].

Click here to enlarge image

Figure 1. Schematic of a CSTR design that approximates the flow through an immersion rinse bath.

Since experiments have shown that immersion rinse baths in wet tools operate substantially like a CSTR, the design of the rinse bath is important to the effectiveness of the rinse. Due to the characteristics of a CSTR, water usage can be minimized in limited ways. One is the size of the bath itself. As long as the resistivity of the water is the primary means of tracking the rinse to completion, the rinse bath volume must be kept to a minimum so that a minimal amount of water is required to remove the chemicals from the bath.

One way to look at the removal of chemicals from wafers is to consider this removal mechanism as a two-step process. The first step would be the diffusion and transport of the chemicals from the wafer surface through a boundary layer in the vicinity of the wafer. The second step would be the dilution and removal of the chemicals in the rinse water itself. One could represent these processes, as a first approximation, using the two-tank CSTR model represented schematically in Fig. 1.

In Fig. 1, Tank 1 represents the "overall" rinse tank, and Tank 2 represents the diffusion of the chemicals from the wafer surface to the "bulk" of the rinse tank. Therefore, the "input flow" into Tank 2 is characterized by the chemical composition of Tank 1. The governing differential equations for this model are

Click here to enlarge image

Each tank is characterized by the volume of the tank (gallons) and, assuming the volumes are constant, the flow rate (gpm) into and out of each tank. From these two parameters, a characteristic residence time for each tank can be calculated (t1 and t2). In practice, the volume and flow rate of Tank 2 are only relevant in calculating the residence time of Tank 2, which represents the characteristic time scale for diffusion and transport of chemicals from the wafer surface.

These residence times are given by:

Click here to enlarge image

Figure 2. Concentration of chemical in UPDI over time, as a function of diffusion time in a CSTR, assuming a residence time of 30 sec; a) linear scale, b) semi-log scale. Compare the results to Fig. 4.

Figures 2a and b demonstrate expected behavior of c(t) in a CSTR including surface diffusion. For illustration, a CSTR residence time of 30 sec is assumed. Further, characteristic surface diffusion times of 1, 2, 5, 10, 25, and 50 sec are shown. The models are first presented as c(t) vs. t, and then as log c(t) vs. t, which produces a linear plot for the classical CSTR.

Click here to enlarge image

Figure 3. Schematic of a Plug-Flow Reactor (PFR) design.

Click here to enlarge image

Figure 4. Concentration of chemical in UPDI over time, as a function of diffusion time in a plug-flow reactor, assuming a residence time of 30 sec; a) linear scale, b) semi-log scale. Compare the results to Fig. 2.

Some tool manufacturers have argued for the efficacy of "plug-flow" configurations in their chemical rinse tanks [4]. The model described herein is shown schematically in Fig. 3. Clearly, if the resistivity in the rinse water is the most direct indication of wafer cleanliness, there are indeed configurations whereby substantially less water can be used in plug-flow rinse tanks than in well-mixed tanks (Fig. 4). The same surface diffusion process is assumed as in the calculations for Fig. 2, but the "second step" of the process involves "carrying off" the chemically laden water in a plug-flow configuration rather than in a well-mixed flow. A "residence time" of 30 sec is again assumed, and different characteristic diffusion times of 1, 2, 5, 10, 25, and 50 sec are chosen. In all cases, the "area under the curve" should be the same, as this quantity is proportional to the initial amount of chemical carried into the tank with the wafers. Keep in mind that, in all these plots, residence time is equivalent to "water consumed," since the flow rates are assumed to be constant.

Note that the scales for both ordinate and abscissa are identical in Figs. 2a and 4a. This choice is intended to allow visual comparison of the resultant concentrations in each model, even though some of the results in Fig. 4a are off-scale. By comparing Figs. 2a and 2b to Figs. 4a and 4b, several observations can be made. For short diffusion times compared to the residence time of the tank, plug flow is more efficient than well-mixed flow. However, as the diffusion time approaches or exceeds the residence time of the tank, the water savings due to plug flow are minimized.

A significant number of experiments utilizing varying flow conditions have been conducted at MOS 12 to determine ways to minimize DI water usage. Figure 5 represents data collected for a selected set of experiments involving rinsing of sulfuric-peroxide (piranha) chemistry. The data provide compelling evidence that, for much of the time during the rinse processes, these processes behave as a classical CSTR. The residence time in the rinse tank, defined by the tank volume divided by the flow-rate, was 30 sec for these experiments, and the line representing a 30-sec residence time is shown.

Click here to enlarge image

Figure 5. Conductivity vs. time in a primary rinse tank following a sulfuric/peroxide (piranha) wet etch. Conductivity directly corresponds to the ionic concentration of chemicals in a rinse tank.

Reclamation and recycle

In addition to the other work being conducted at the CEBSM (mentioned previously), the Center Director, Prof. Farhang Shadman, is examining the important issue of DI water recycling. Current practice calls for DI rinse water to be used for a single pass only, even though the water in many ways is cleaner at that point than the "city water" used to make DI water in the first place. However, certain trace contaminants in the rinse water are potential sources of concern to semiconductor manufacturers. By developing methods to remove these specific contaminants in the rinse water, the water can be reprocessed and used again. Particularly challenging problems include the removal of metals and certain recalcitrant species such as trace organics and surfactants.

An interesting optimization problem arises when one compares reducing DI water usage in rinsing vs. recycling DI rinse water. As water reduction efforts increase, the total water throughput in rinse processes decreases, and therefore the concentration of species within the DI rinse water increases. Higher concentrations of contaminants in rinse waters may make recycling more difficult. Clearly, both potential water-saving methods must be considered simultaneously. The optimum solution would be to reduce water usage in each rinse tank only until contaminant concentrations in the resultant baths are still easily removed by reprocessing.

Lowering water consumption

As mentioned, work at some of the wet process station manufacturers has been published in cooperation with academic researchers. In addition, work is being carried out at Sandia National Laboratories at the SEMATECH-funded CFM Research Center to improve the rinsing capabilities of immersion tools, and work on both wafer rinsing and UPDI recycling is about to begin at the CEBSM.

Low-flow reduction. Since low flow is the common practice in semiconductor fabs, the rate of flow necessary to keep bacteria and particles from becoming an issue should be established. In our fab we have seen that about 50% of total UPDI water usage is due to nonprocess reasons, since we keep a low flow of water going through the baths in all DI water tools. The low flow is a balance between two considerations. The minimum flow rate is specified as that flow which establishes turbulent flow in a pipe or yields a Reynolds number (Re) >2100 for a pipe of a given diameter. Simultaneously, the flow rate achievable for a given pipe diameter depends on the maximum pressure drop sustainable across a given equivalent length (L), taking into account valves, elbows, etc. (Fig. 6). A nonadjustable bleed for low-flow control is preferable since adjustable valves can degrade consistency from bath to bath. Adjustable valves may be turned off when work is being done on the rinse bath with no guarantee that they will be turned on when the work is finished. Also, low-flow rates in every rinse bath in the facility should be measured as part of a regular preventative maintenance (PM). A single rinse bath flowing only 1 gpm above the required amount wastes about 0.5 million gallons of UPDI/year and at least 1 million gallons/year of raw water supplied to the facility.

Click here to enlarge image

Figure 6. Flow rates to ensure turbulence for given pipe diameters. The region of acceptable flows is defined as above the thick "diagonal" 2100 Reynolds number line and below the parametric lines for different pressure drops. Turbulent flow is required to prevent bacterial growth and eliminate particle settling sites.

Rinse water return. Modern wet-processing tools can be designed to return water to the DI plant when it is above a certain resistivity. A separate waste drain and valve, activated based on the water resistivity, can divert the low-flow water and even some of the water used for rinsing wafers back to the UPDI water return or to some other use in the DI plant.

Rinse process modification. Rinse processes can also be designed to use less water. In addition to the bath design described above, small changes to the rinse process can have major impacts. For example, when a dump rinser is in the act of dumping the water, the incoming high flow can be turned off since it is going directly to drain. In a single dump rinser flowing about 20 gpm with 5 dumps, the savings could amount to 30-40 gallons of water during one process. The high-flow portion after the rinse bath is refilled may in fact have little effect on chemical removal since the water filling the rinse bath will be drained again in the next dump. Recent work [5] has shown a 5? water reduction in a dump rinse bath.

Rinsing past resistivity. When a rinse cycle has attained the desired resistivity level, the high flow of water can be turned off, regardless of whether the wafers can move into the subsequent process. About 25% of high-flow process water in our automated tools is used after the rinse is complete and the wafers are waiting to be moved by a robot.

Hot water reclaim. Many facilities heat UPDI water for rinsing of chemicals like phosphoric acid and sulfuric acid that are difficult to remove. Our DI water heaters operate at a constant temperature only when there is a flow of a few gpm through the heater. Instead of sending the flow through the heater out through the rinse bath, we have diverted much of the flow to the return loop to the DI plant, saving a few gpm from each of the water heaters in use and still keeping the rinse baths warm and able to heat to operating temperatures almost instantaneously.

Conclusion

Working fabs have published relatively little on efforts to cut the consumption of water. Some process tool manufacturers have devoted significant effort to the problem in recent years, but their efforts have paid off only for the most advanced tools available. At the same time, the cost of producing ultrapure DI has increased; the water facility is a significant portion of the cost of a new fab. More efforts in water savings will pay dividends in the near term as the easy problems are attacked, but there are limits to the amount of water that can be saved. When water is recognized as a significant chemical in the fabrication process, the efforts to reduce the amount used may bear real fruit. n

Acknowledgment

We would like to thank the members of the DI Reduction team at MOS 12, and Eric Hansen from SCP Global Technologies for their efforts to reduce water consumption in our facility and for some of the ideas mentioned here. We would also like to thank the management of this facility for supporting the work we are doing.

References

1. P.G. Lindquist, R.N. Walters, J.O. Thorngard, J.J. Rosato, "Determination of Rinsing Parameters Using a Wafer Gap Conductivity Cell in Wet Cleaning Tools," Proceedings of the Materials Research Society, San Francisco, CA, 1995.

2. S.N. Kempka, et al., "Evaluating the Efficiency of Overflow Wet Rinsing," Micro, pp. 41-48, May 1995.

3. J.J. Rosato, et al., "Studies of Rinse Efficiencies in Wet Cleaning Tools," presented at 3rd Int`l. Symp. on Cleaning Technol., in Semiconductor Device Mfg., Electrochemical Soc. Proceedings, Vol. 94-7, p. 140, 1994.

4. C. McConnell, H. Thomas, S. Verhaverbeke, S. Bay , J.W. Parker, "Water Consumption and Rinse Considerations in Semiconductor Wet Processing," presented at 4th Int`l. Symp. on Cleaning Technol., in Semiconductor Device Mfg., Electrochemical Soc. Proceedings, Vol. 95-20, 1995.

5. W. Chen, C.R. Helms, R. Parker, presented at the SRC Technology Transfer Course, Stanford Univ., Palo Alto, CA, January 1996.

THOMAS S. ROCHE received his BS degree in chemistry from Manhattan College, Bronx, New York, and his PhD degree in inorganic chemistry at SUNY at Buffalo. He has worked for Olin Chemicals in their research center in the area of chemicals for the electronics industry, and is currently a principal staff engineer with Motorola in the Cleans Process Engineering group at the MOS 12 facility. Motorola, MOS 12, 1300 N. Alma School Road, Chandler, AZ 85224; ph 602/814-3475.

THOMAS W. PETERSON received his BS degree at Tufts University, MS degree at The University of Arizona, and his PhD degree at California Institute of Technology, all in the area of chemical engineering. Peterson has been on the UA faculty since 1977, and has been head of the Chemical and Environmental Engineering department since 1990. He spent a sabbatical in 1996 at Motorola working on pollution prevention activities at Mesa, and with the Cleans Engineering group at MOS 12 in Chandler.