Facilities Technologies

Reducing water consumption in semiconductor fabs

Greg Klusewitz and Jim McVeigh, Fairchild Semiconductor

A series of studies has led to sink, water line, and other technology modifications, enabling a semiconductor fab to significantly reduce water and electricity consumption.

Semiconductor manufacturers use immense amounts of water. A large portion of that water is used to produce DI or ultrapure water. In the manufacturing process, water is employed to rinse and clean semiconductor wafers. Approximately 1500 gallons of city water are required to produce 1000 gallons of DI water. More than 2000 gallons of DI water can be used in the production of one 200-mm wafer, and large facilities can use 3 million gallons of DI water per day, according to "Energy and Water Efficiency for Semiconductor Manufacturers" (Pacific Northwest Pollution Prevention Resource Center, 2000). It costs approximately $12 to generate 1000 gallons of DI water.

Aside from the cost of the water itself, producing DI water involves ultraviolet lamps, filters, pumps, and recirculating systems, which require energy to operate. It is also costly to remove effluents via a water-treatment plant. Reducing DI-water consumption by 1000 gallons lowers power use by about 46 kWh. By reducing DI-water use by 10%, a facility using 3 million gallons of DI water daily can save 5 million kWh annually, or about $225,000 at an energy price of 4.5 cents per kWh.

This article describes the methodology and results of a study to reduce DI-water consumption conducted by a cross-functional team at Fairchild Semiconductor's 150-mm wafer fab in Mountaintop, PA. The team was composed of process technicians, process engineers, facility engineers, and equipment vendors. Reduction efforts initiated in February 2001 included rinse sink, process, and supply modifications. The goal of the study was to reduce DI-water consumption by 2 million gallons within one year. The annual savings from this effort was projected to approach 18 million gallons of DI water, or about $223,000.

Fab Organization

Fairchild Semiconductor, a large U.S. supplier of signal processing and power control devices to the automotive, telecommunications, industrial, and computer marketplace markets, purchased the Mountaintop wafer fabrication facility from Intersil in March 2001. The facility includes a 4-year-old 200-mm fab and an 11-year-old 150-mm fab with different toolsets and physical layouts.

The fabs operate on a four-shift, 12-hour compressed workweek. All production personnel are structured into self-managed work teams that are organized according to functional area. Each team consists of equipment repairers and production associates represented by Local 177 of the International Union of Electronic, Electrical, Salaried, Machine, and Furniture Workers, organized in the AFL-CIO. Equipment technicians are responsible for maintaining and improving the equipment used in the manufacturing process. Based on their manufacturing experience, technicians also make decisions about fab processes and product disposition. Although supervisors are not assigned to monitor daily fab activities, all production personnel report to the fab manufacturing leader. Engineering and facilities personnel are assigned to support groups.

The efforts of the cross-functional team formed to reduce DI-water use were concentrated in Fab 6 (the 150-mm fab) after it was determined that that facility clearly consumed more water than the 200-mm facility. Initial comparisons attempted to determine water consumption per square inch of silicon. However, because of the market downturn and reduced wafer output, that index proved to be ineffective. Consequently, the team investigated other methods to decrease DI-water use while maintaining product quality as well as environmental and safety practices.

Initial Efforts to Reduce DI-Water Consumption

The initial meetings of the cross-functional team focused on identifying DI-water use and on establishing reliable testing methods to assure that product quality would not be compromised by DI-water reductions. The team first concentrated on the Fab 6 wet etch sinks, which use traditional-style acid, spray/dump/rinse, and spin rinse/dry applications.

The team identified several immediate ways to decrease DI water consumption:

• A 1 L/min constant-flow sample line of DI water, used for obtaining bacteria samples, was shut off after an alternate method of obtaining samples was developed. Annual savings were projected to be $2000.
• Several leaking water lines to the sinks and difficult-to-find manifolds inside the sinks were repaired. In addition, the water line to an internal sink that leaked into the plenum, and eventually the drain, was repaired.
• Several production associates had been running DI-water dump rinsers without wafers in them to time wafer etch cycles, instead of using the established timers on the sinks. After being reviewed by the DI-water reduction team, that practice was eliminated. Assuming that 15 dumps were performed per 12-hour shift in a 4.5-gallon dump rinser, it was calculated that annual DI-water savings would amount to $600.
• The team discussed best-practice plans for sinks during shutdown periods, which had been implemented because of economic conditions.

The team identified another opportunity to reduce DI-water consumption with minimal or no impact on wafer quality. Dump rinsers, which are used to rinse the product after an etch operation, are configured to perform idle flow, a continuous DI-water overflow of the dump rinser that is used to prevent particles and bacteria growth. Team members suggested that reducing idle flow to more conservative levels, based on experience and similar successful efforts in the Fab 8 facility, would be sufficient to prevent bacteria growth.

After appropriate evaluations, idle flow was reduced by adding orifice restrictors measuring 1/16 in. in diameter to the appropriate flow line. To accomplish the job, a series of actions were performed. First, orifice restrictors were added to two Z-strip (chemical resist removal) DI-water holding tanks, which are used when work flow through the Z-strip is interrupted for one reason or another. With the restrictors in place, water flow was reduced from 1.5 gal/min to 0.4 gal/min on Z-strip 1 and from 8 gal/min to 0.5 gal/min on Z-strip 2. The total DI-water reduction was 8.6 gal/min, amounting to a projected annual savings of $54,100.

Next, all of the dump rinsers in Fab 6 were readjusted. Orifice restrictors were installed upstream of the idle-flow valve on the bottom fill of the dump rinsers (see Figure 1), and flow to the top sprinkler heads was reduced to a trickle to maintain reduced-flow conditions. Water fill-rate measurements were taken before and after the readjustment to determine the amount of water reduced. Idle flow on three Z-strip sinks, two metal-etch sinks, and two buffered-oxide-etch sinks was reduced from 31.6 gal/min to 8.6 gal/min, amounting to a total projected yearly savings of $145,000 in DI-water production costs.

Figure 1: Side view of the dump rinser after modification.

Another step involved correlating and replacing the resistivity probes on the spin dryers where needed. After an etch process, wafers are rinsed in a dump rinser and then rinsed again and dried in a spin rinser/dryer. Initially, the wafers receive a 2-minute DI-water rinse in the spin dryer, followed by a Q-rinse if necessary to reach the required resistivity level of 16 MW-cm. They are rinsed again with DI water until that resistivity value is exceeded, and then the DI drain water is measured for resistivity. Rinsing removes ionic contamination, and the resistivity measurement of DI drain water indicates when ionic contamination has been reduced to acceptable levels. However, correlation of the resistivity probes indicated that several probes were in need of replacement, indicating that the Q-rinse step was not functioning efficiently. After the faulty probes were replaced, DI-water consumption was lowered.

After meeting with the area production teams, the DI-water reduction team proposed eliminating one Z-strip sink. In order to prevent bacteria growth, the dump rinsers on the sink were constantly running on idle flow. By eliminating the sink, idle flow was reduced, amounting to a DI- water savings of 1.2 gal/min. After reviewing fab capacity models, manufacturing determined that the extra sink could be eliminated without affecting the work flow. The annual savings was estimated to be $7550.

Preserving Wafer Quality while Lowering Water Consumption

After focusing their attention on several sinks in the etch area, the team proposed that the number of dumps in the sink dump rinsers be minimized. The dump rinsers rinse wafers after a wet etch process. Wafers vertically positioned in boats are placed in the dump rinsers and the dump rinser controller is activated. Water in the dump rinser is drained out the bottom while fresh DI water is introduced from sprayers at the bottom and top of the tool. When water completely fills the dump rinser and covers the wafers, it is dumped again.

Reducing Z-Strip Dump Cycle. Concerned that a reduction in the number of dumps could lead to incomplete rinsing, the team proposed that the Z-strip and megasonic clean process dump cycles be investigated. Tests comparing the effects of running five versus three dump cycles in Z-strip were performed, and bare wafers were tested for particle counts. Wafers were first processed in a buffered-oxide-etch tool for 60 seconds to deposit more than 600 particles on each wafer. Five wafers were placed in a boat (two at both ends and three randomly in the middle) and processed through Z-strip with the standard five dumps, while another group of five wafers was processed with three dumps. The number of >0.3-µm particles remaining was determined, and cleaning efficiency in percent was calculated according to the formula:

The results were:

• 5-dump process: 23 particles, 96.9%.
• 3-dump process: 17 particles, 97.8%.

After the same wafers were coated with resist, baked, and processed again, particle-count data were:

• 5-dump process: 13 particles.
• 3-dump process: 23 particles.

Particle counts were also determined for three product lots that had been split at Z-strip prior to gate clean. The results were:

• 5-dump process (odd-numbered wafers): 33 particles.
• 3-dump process (even-numbered wafers): 21 particles.

Probe yields for the odd and even wafers were statistically the same.

Tests of spin-dryer resistivity performed on the resist-removal part of the Z-strip process only showed that the standard five-dump cycle and an exaggerated one-dump cycle required similar processing times to reach the required 16-MW-cm resistivity level. Wafers received an 8-minute hot DI-water rinse after sulfuric acid resist removal, which was sufficient to completely rinse the wafers. Samples of the dump rinser DI water were also taken from both a standard five-dump cycle and a worst-case one-dump cycle and then measured with ion chromatography equipment from Dionex (Sunnyvale, CA). Wafer anion levels in both cases were comparable (<1 ppm).

Based on a production schedule of 5000 wafers/wk (330 lots), the three-dump cycle requires four fewer dumps for every two lots processed through Z-strip than the standard five-dump cycle. Accordingly, based on a manufacturing schedule of six Z-strips, the number of dumps/wk is reduced by 3960 dumps. With a dump totaling 4.5 gallons, that results in a DI-water savings of 17,820 gal/wk, or 1.8 gal/min. The projected savings is $11,300 per year.

Reducing the Number of Dumps for Aluminum Etch Processing. An evaluation was also performed to determine the feasibility of reducing the number of dumps from eight to five for the wet aluminum etch process. Dump rinser water samples were procured and analyzed, spin-dryer resistivity tests were conducted, and split lots were run. All results were positive, indicating that decreasing the number of dumps would not affect wafer quality. Based on a production schedule of 5000 wafers/wk (330 lots), three fewer dumps per lot, or 990 fewer dumps/wk, can be processed through wet aluminum etch by reducing the number of dumps from eight to five. With a dump totaling 4.5 gallons, the overall DI-water savings is 4445 gal/wk, or 0.44 gal/min, translating into a savings of $2800 per year.

Improving Non-DI-Water Applications

While the DI-water studies were being conducted, the facilities members of the team concentrated on reducing non-DI-water use throughout the plant. They took several key measures:

• They outsourced the production of compressed air. Since the internal air compressors were water-cooled, shifting to an external supplier resulted in reducing yearly water consumption by 13 million gallons.
• They shifted to the use of recovered water to run the water-cooled packaged air conditioners. Furthermore, the air-conditioner systems were put on time-of-day control and on-demand control, leading to an overall reduction in water consumption of 1.5 million gal/yr.
• They placed several house fume scrubbers on reverse-osmosis reject water, further reducing water consumption by 3.6 million gal/yr. Plans were also drafted to convert additional fume scrubbers in the future.
• They began to use recovered water as a supply source for the production of DI water.

Conclusion

This article describes a series of measures undertaken to lower water consumption on the fab floor. The total projected savings from the project of $223,550, based on reducing consumption in areas with measurable water flows, are broken down in Table I. Other measures, such as repairs to internal sink leaks, are not accounted for. Figures 2 and 3 illustrate the changes in Fab 6 water and electricity use in 2001. Since the measures were implemented, wafer quality has not been adversely affected.

Water-Reduction Measure Savings (gal/min) Savings (gal/yr) Approximate Yearly Savings ($) Shut off unused DI-water sampling line 0.3 157,248 2,000 Utilize process timers instead of dump rinser cycles 0.1 49,960 600 Add flow restrictors to Z-strip holding tanks 8.6 4,500,000 54,100 Reduce idle flow on sinks in etch 23 12,000,000 145,000 Reduce dump rinses from five to three for Z-strip 1.8 1,000,000 11,300 Reduce dumps from eight to five for aluminum etch 0.44 230,630 2,800 Eliminate one Z-strip sink 1.2 629,000 7,550 Total yearly savings 35.44 18,566,838 223,350

Table I: Fab 6 DI-water reduction results.

Figure 2: Fab 6 water savings in 2001 over 2000.

Figure 3: Fab 6 electricity savings in 2001 over 2000.

Future efforts will include the installation of additional orifice restrictors in Fab 6 to reduce idle flow. Moreover, two new sinks, with production capacity equal to or better than that of the four existing sinks, will be constructed in the etch department.

Acknowledgments

This article is based on a paper presented at the 13th annual IEEE/SEMI Advanced Semiconductor Manufacturing Conference in Boston, April 30–May 2, 2002.

Greg Klusewitz is the Fab 6 process engineering manager at Fairchild Semiconductor (Mountaintop, PA). His areas of responsibility include epitaxial growth, all fabrication processes, and parametric test probe. He received a BS in electrical engineering from Pennsylvania State University in University Park. (Klusewitz can be reached at 570/474-3296 or greg.klusewitz@fairchildsemi.com.)

Jim McVeigh is the Fab 6 etch engineer at Fairchild Semiconductor. He is responsible for wet and dry etching of thin films (oxide, poly, metal, and borophosphosilicate), and for wet/dry resist removal. He received a BS in chemical engineering from the University of Pennsylvania in Philadelphia and an MS in chemical engineering from Drexel University in Philadelphia. (McVeigh can be reached at 570/474-6761, ext. 4836, or jim.mcveigh@fairchildsemi.com.)