Green and Clean

Assessing a system for CMP waste minimization and water recycling

Steven Browne and John Maze, Cirent Semiconductor; and Bob Heid, Lucid Treatment Systems

Installing an effluent segregation system may permit fabs to expand CMP operations without overtaxing existing water supply and waste treatment systems.

Although the semiconductor industry has gradually come to embrace chemical-mechanical planarization (CMP) as an enabling technology for the production of devices at or below 0.35 µm, one of CMP's big drawbacks has been its generally high cost of ownership in terms of water consumption and waste treatment. In fact, the CMP process can consume as much as 50% of a fab's ultrapure water (UPW). While manufacturers who are new to the technique and even those with one or two CMP polishers in their pilot production lines can get by without increasing their UPW capacity and adopting special waste treatment programs, they may be forced to do so as they add additional polishers.

When Cirent Semiconductor, a Lucent Technologies company, decided to expand CMP capacity at its fab in Orlando, FL, the existing UPW system was nearing full capacity and could not fully support the expansion. However, the cost of building a new UPW system was prohibitive. In addition, the fab is an ISO 14001­certified manufacturer, which means that water conservation, water reuse, and water recycling are important aspects of its overall operating plan. In light of these circumstances, the company decided to evaluate several types of water-recycling systems to help reduce the load on the UPW makeup system. Among the systems it evaluated was the Waste Interface System from Lucid Treatment Systems (Hollister, CA). This unit is designed to segregate process wastewater at the point of use, diverting the reusable water for recycling while concentrating the slurry waste going to the facility's central ultrafiltration treatment system. The case study discussed in this article describes the operation of this system and presents the results obtained after it was installed on the drain lines of oxide and tungsten CMP tools.

Growing Water Consumption

Many semiconductor fabs using CMP today were not designed for the water consumption and waste treatment demands inherent to the process. Manufacturers' specifications on water consumption by commercial CMP polishers range from 4.2 to as much as 12 gal/min. Taking the average of 8.1 gal/min, UPW consumption per polisher totals 11,664 gal/day, or more than 4.25 million gal annually. Thus, at an average cost of $0.016 per gallon of UPW and the subsequent average waste disposal/effluent charge of nearly the same amount, operating a single polisher can require an expenditure of $136,000 per year in water-related costs alone.

Additional costs may arise from expanding capacity. While a fab may be able to accommodate an initial installation of a limited number of polishers without surpassing the capacity of its UPW and waste treatment systems, there is often no way for the existing systems to keep up with the additional volume demands if expansion plans call for the installation of even more polishers. This problem can greatly diminish the returns a fab might expect to achieve from expanding the CMP process.

However, the nature of the process may hold a solution. The CMP process requires substantial periods of downtime, because the current state of the art requires that many time-consuming process steps be implemented to adjust for lot-to-lot variations in both wafers and process consumables. Although the amount of time CMP tools are used to polish wafers depends on shop flow, product design, product mix, polisher design, process configuration, and overall demand, at most fabs these tools are actually underutilized. However, to maintain the production readiness of each tool, it is imperative that the polishing pads be kept wet and that the process chambers be rinsed down. Timer cycles on wet cassette dump systems must all continue to operate even as the machine sits idle during inspection, test, and setup operations. Although it becomes part of the tool's waste stream, the water that flows uncontaminated through the tool at these times is nearly perfect UPW that is ideally suited for reclamation. Thus, with the appropriate systems in place, a fab may be able to:

• Reduce UPW consumption by reusing reclaimed CMP process water for nonprocess, non-wafer-contact applications such as acid scrubbing, CMP tub rinsing, UPW makeup operations, and process cooling water.

• Optimize the efficiency of its CMP waste treatment system by reducing the effluent's volume and increasing its slurry concentration.

System Design and Installation

The waste interface system evaluated here consists of a sensor for determining the slurry content of the waste stream as it enters the unit, a proprietary valve for segregating clean reusable water from slurry-laden effluent, two sumps for collecting the segregated effluent streams, and pumps for delivering the products to their destinations. The system is designed for installation directly adjacent to, or near, the output drain of most commercial CMP polishers. (The manufacturer also offers a combination of this system and a remote diverter, which does not include sumps and pumps.)

After minor adjustments had been made to accommodate the extended separation of the system from the fab's oxide polisher (a distance of 17 ft vertical and 10 ft horizontal), the system was connected via isolation valves to the waste drains of the CMP tool, as shown in Figure 1. Subsequently, this same system was reconnected to the waste drain of a similar tool set up for polishing tungsten. Both CMP tools are 200-mm-capable polishers used in production. Each has two separate drains: a 2-in. drain in the main polishing chamber, which accommodates the process waste and tub rinse, and a 4-in. drain, which accepts wastes from the buffing process, load and unload station, and machine automation system.

 

Figure 1: Schematic detailing the waste segregation system and its connection to the CMP polishers.

Because the buffing and polishing processes on the oxide polisher use the same commercially available colloidal silica slurry, a common feed line was employed to connect both drains to the waste segregation system. But because the buffing and polishing operations on the tungsten polisher use different slurries that must be segregated from each other, only the tool's main polishing chamber was connected to the drain.

Meters were installed in the escape lines from the two sumps to capture data on the total amount of effluent as well as on the amounts that were segregated as reusable water or waste. The meters were a necessary part of the evaluation because the CMP process tools consume UPW at different rates when processing and when idle. Sampling ports were also installed in the escape lines to enable the characterization of the different effluent streams. The escape lines were then replumbed into the existing drain system for normal treatment. Finally, a laptop PC with data collection software was connected to the COM port of the polishers' PLC control boxes to provide accurate, real-time data collection.

Study Results

The overall results of the study are shown in Table I. During the six-week evaluation period, the polishers produced an average of 7342 gal of effluent per day. However, the volume fluctuated as a function of the tools' duty cycles. For both polishers, the effluent rate varied from 4.76 gal/min during wet idle to 5.88 gal/min during processing.

 

o Percentage Value (gal) Total effluent 0 337,724 Average daily effluent 0 7,342 Maximum daily effluent 0 7,928 Minimum daily effluent 0 6,891 Total waste 0 117,712 Average daily waste 35 2,559 Maximum daily waste 67 5,636 Minimum daily waste 3 190 Total reusable water 0 220,012 Average daily reusable water 65 4,783 Maximum daily reusable water 97 6,701 Minimum daily reusable water 33 2,293

Table I: Overall results of the 6-week evaluation of the waste segregation unit installed on two different CMP tools. The study period covered a total of 4344 process cycles.

When the waste segregation system was connected to the oxide tool, the PLC data indicated that 70% of the effluent was diverted to the clean, reusable-water sump. In actuality, however, the escape-line meters indicated that only 65% of the total effluent volume was diverted to that sump. When the system was connected to the tungsten tool, the comparable percentages were 76% and 70%, respectively. These discrepancies were probably due to process cycle time issues. The day-to-day data showing the division of the waste stream are presented in Figure 2.

 

Figure 2: Day-to-day data from the escape-line meters, showing the division between reusable water and slurry-laden waste.

Water Quality. Samples were taken from both the reusable-water and escape-line ports of the waste segregation system at random times over the course of the evaluation. Since the escape ports were pump-driven rather than gravity drains, the sampling sequence was dictated by the polisher duty cycle. A total of six samples were taken from each port. These samples were tested for solids, pH, total organics, conductivity, and trace-metal contamination. Metal levels were determined using an inductively coupled plasma atomic emission spectrometer (ICP-AES) with a periodic-table scan of 32 trace-metal contaminants. The limit of detection for the ICP-AES is 0.01 ppb for all 32 trace-metals.

One phenomenon revealed by the sampling was that there was an extended period during pad conditioning when the particle and conductivity levels caused the sensor to divert the effluent to the waste drain even though the turbidity of the waste stream was relatively low. The total organic contaminant level went from 40 ppb to 2.4 ppm during that period.

As can be seen in Table II, the quality of the effluent from the reusable-water line was sufficiently high to permit its reclamation and reuse in nonprocess applications. The only contaminant of any consequence was silicon, which is the main constituent in the slurry. When these results are compared to those for the slurry-laden waste, the efficacy of the test system in segregating the effluent and thereby concentrating the slurry waste is very apparent. Furthermore, because the segregated slurry-laden waste stream is lower in volume and far more concentrated than unsegregated effluent, the efficiency of a fab's CMP waste treatment system would be expected to improve. During this study, for example, the waste treatment system would only have had to accommodate 2559 gal of slurry waste with an average of 0.31% solids instead of having to process 7342 gal of waste with 0.027% solids.

 

Composition Reusable Water Waste pH 6.15 9.3 % solids 0.000 0.312 Total trace metals (ppb)   Potassium   Silicon   Sodium   Calcium   All other metals 3.738 0.000 3.447 0.145 0.074 0.072 1401.094 29.969 1370.341 0.075 0.050 0.659 Total organics (ppb) 34.2 n/a Conductivity (m‡) 1.2 n/a

Table II: Average contaminant levels in the reusable water and the slurry-laden waste. Samples were taken from the respective escape-line ports.

Conclusion

CMP consumes an average of 8.1 gal of UPW per minute per polisher, sending the same amount of effluent to the fab's waste treatment facility. Most polishers are fully operational less than an optimal amount of time, however, which means that a tremendous quantity of clean UPW is needlessly sent for treatment. This situation creates an unnecessary burden on the environment, the fab's UPW makeup and waste treatment systems, and the company's cash flow. The waste segregation system discussed here has been designed to facilitate the reuse of the clean CMP wastewater in noncritical applications.

When this system was evaluated on oxide and tungsten polishing tools, results indicated that it met the following criteria:

• It accurately and reliably segregates slurry-laden waste from clean, reusable water in polisher effluent in both oxide and metal CMP applications.

• It successfully reduces the volume of effluent and concentrates the slurry in the stream, thereby potentially increasing the efficiency of a fab's waste treatment system.

• It enables the reclamation of reusable water whose quality is acceptable for non-wafer-contact applications.

Such systems, therefore, should enable fabs to expand their CMP operations without having to expand their UPW and waste treatment capacity.

Steven Browne is a member of the technical staff at Cirent Semiconductor in Orlando, FL, a Lucent Technologies company. He is responsible for the design and operation of the site's ultrapure water, industrial wastewater treatment, and recycled water systems. He also oversees regulatory issues surrounding industrial wastewater effluent discharge and the site's UPW well-water permits. (Browne can be reached at 407/371-6443 or smb@lucent.com.)

John Maze is a member of the technical staff at Cirent Semiconductor, where he is responsible for CMP equipment reliability and performance, as well as tool modifications. In this capacity he seeks to reduce costs, enhance yields, and improve the quality of tool consumables. Maze has more than 10 years of experience in the semiconductor industry, a portion of which he spent holding a staff position in the Lucent Bell Labs development group. (Maze can be reached at 407/371-6145 or jmaze@lucent.com.)

Bob Heid is the director of business development at Lucid Treatment Systems in Hollister, CA. He has nearly 20 years of experience in semiconductor-related materials and equipment sales and business management. Previously he was the managing director of Solution Technology and held positions at Olin, OCG, and FSI. (Heid can be reached at 407/622-2085 or bheidhsd@earthlink.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit