Extending Fab Life: Measure, Monitor and Control

In legacy fabs, dynamic, remote monitoring of gauges can lead to significant improvements in productivity and cost savings.

Dick Deininger and Rebecca Taylor, Taylor-Deininger Partners Inc., Austin, Texas, www.td-partners.com -- Semiconductor International, 7/1/2008

Two-thirds of semiconductor products made today are manufactured in older fabs (150, 200 and 300 mm)¹; they are focused on making products that do not demand leading-edge technology to be profitable. These fabs face significant, increasing pressure to cut costs, conserve resources and continuously increase productivity to be competitive in world markets.

When these fabs were built, thousands of static-reading, analog gauges were installed throughout the facility. These gauges monitor critical parameters, but are typically located under the floor, up in the ceiling and in other remote places. They cannot be easily accessed and read. For the most part, they are not dynamically monitored. At the time these gauges were installed, the cost to connect them to a network was too high.

Wireless monitoring

The advent of low-cost, low-power wireless technology has driven the cost of remotely monitoring these gauges much lower than implementing invasive and expensive hard-wired sensors and gauges. Using external clamp-on capability, which does not require infrastructure changes, remote monitoring and dynamic alarming of parameter changes becomes affordable. This scenario defines a classic paradigm shift opportunity.

We developed a modeling capability to demonstrate the value of implementing remote monitoring in a number of applications. The model analyzes fab wafer losses, die yield losses, gas usage, electrical usage, water usage and air handling, and then estimates the value of implementing measurement, monitoring and alarming in the highest-value areas. It can show how losses and usage of key materials can be reduced, thus improving productivity. The model shows that remote dynamic monitoring helps identify problems before they can adversely affect the product. This detailed modeling capability will be described in a subsequent paper.

This article describes specific situations with demonstrated savings in ultraclean facilities.

Shrinking margins

It’s not your imagination...it really is tougher out there these days. DRAM, flash and other IC prices have fallen roughly 50% this past year, and while volumes have gone way up, the margin for errors and waste have gone way down. Net margins of 3% are not atypical. In this environment, there really is no room for a scrapped wafer. The three-year trend in both IC and DRAM pricing, as tracked by VLSI Research (San Jose), is depicted in Figure 1.²

1. Three year IC/DRAM pricing trends. (Source: VLSI Research)

This situation puts extreme pressure on fab operations. Couple that with rapidly changing product demands requiring lower costs and you have a recipe for money-losing operations if they are not tightly controlled.

Monitor, manage, control

Without control, it is impossible to manage a fab to the tight financial tolerances required in low-margin situations. Legacy fabs need methods to further cut chilled/process cooling water usage, reduce electric usage and unneeded exhaust, cut expensive gas usage, avoid costly repairs and yield losses, and more rapidly find problems in the fab or any process-based manufacturing operation.

Following are four real-world case studies, undertaken within eight months at four different companies, which enabled fab managers to find hidden variations in water pressure/temperature, gas flow and equipment status that were driving undefined yield variation.

Remote gas bottle monitoring

Micrel (San Jose) installed a wireless gauge reader (WGR) system to automate the monitoring of manual gauges at the company’s 150 mm, 30,000 wpm fab. The projected annual savings generated by this effort is estimated to be $215K, with an investment payback period of less than seven months.

Micrel’s facility, similar to other fabs built in the 1980s, has hundreds of manual gauges that measure critical parameters on gas cylinders, pressure valves and other numerous facilities equipment. Key gauges are read manually twice a day, requiring valuable time from skilled technicians. Gauge excursions may not be detected for up to several hours, which can lead to unplanned downtime and scrap.

Micrel installed WGRs (Fig. 2) on ~100 of their gas systems; this helped them automate the gauge reading process, and led to $40,000 in reduced downtime, $95,000/year savings in labor cost, and 10% lower consumables use (i.e., process gases), roughly $80,000.² The system’s software allowed engineers to define and send early warning alarms to their computer, pager or cell phone in the event of excursions. Cost per point solution was $1250.

2. The clip-on wireless gauge reader includes a wireless access point, server and web interface. (Source: Cypress Systems)

This compares to a traditional transducer installation cost of $4000 per point, which includes an estimated downtime cost of $1000, the $300 transducer/sensor, wiring/installation/design cost of $1500, fire code conformance of $1000, and I/O panel termination $200.

Gas usage debugging

The Stanford Nanofabrication Facility (SNF, Palo Alto, Calif.) is a high-mix facility with a wide range of customers. In any given month, nearly 250 researchers from Stanford, other academic institutions and industrial firms make use of this facility to build prototype devices and structures. The installed base of nearly 100 instruments are used in the fabrication of advanced microelectronic devices, integrated optics structures, MEMS/NEMS devices, biological and biomedical devices, and structures based on nanotubes. Because this facility was originally opened in 1985, there is a large installed base of analog gauges of various types with very few electronically monitored points.

Because nitrogen usage is the single largest non-salary expense in operating this facility, the facility’s management wanted to better monitor this key system.

SNF installed 21 WGRs on a variety of pressure gauges throughout the facility. Having this data led to rapid cost reductions. After installing the WGRs in the facility, engineers noted certain periodic drops in the nitrogen pressure delivered to the facility that had not been previously observed. Given the timing of these pressure drops, they quickly determined that one of their spin rinse dryers was entering a nitrogen purge cycle much more frequently and for a longer duration than needed. Once identified, they altered the purge cycle to both save nitrogen and reduce the pressure fluctuations in the nitrogen system.

CMP slurry filter clog

At an R&D foundry, a CMP slurry filter was clogging at unpredictable intervals, requiring 1-2 wafers to be scrapped before the clog was identified and the filter changed. This event occurred every 7-14 days. Because this occurred late in the process, scrapped wafer cost was estimated at $1650.

The foundry considered doing nothing, changing the filters more frequently, or monitoring the pressure across the pump to determine when the filter was nearing its operating limit.

Doing nothing scraps 1-2 wafers at market value of $1650 each every 7-14 days. The financial loss per month is 1-4 wafers or $1650-$6600. This is clearly not a desirable approach.

Changing the filters twice as often might reduce the problem, but the window of failure still exists. This solution requires at least 26 additional filters/year at a cost of $4160/year. This drives unnecessary tool downtime that significantly impacts the throughput of the CMP area. Most importantly, the filter swap is unrelated to an actual filter clog event, so there is still the possibility of a wafer scrap event.

The most effective approach involveed monitoring the filter pressure, tracking and trending it, and alarming operators when the filter nears its operating limit. This maximizes tool production availability and eliminates associated wafer scrap. Implementing this approach required one WGR per slurry filter at a cost of $1250 per point.

Using either the cost of a wafer or the lost revenue associated with a scrapped wafer as the basis for a financial model, the return on investment (ROI) for installing the monitoring capability and changing the filter accordingly is less than one month.

As a test, a WGR was installed on a CMP tool in the foundry. After the WGR was in use for one month, filters were found to clog 3 hours to 15 days after installation. Based in part on this information, the foundry derived a strategy for dealing with the clogging filters.

Figure 3 covers a 12-hour time horizon tracking pressure across the CMP filter. The pressure climbs over a three-hour period and spikes, indicating a pressure buildup indicative of a filter clog. The filter tends to recover in between jobs, as shown. The peaks and troughs over the next nine hours track filter clogging with usage. In the last hour, an unrecoverable clog was forming; the pressure peak was not followed by a deep trough. Using such trend plots, the equipment team determined filter replacement times.

3. CMP filter pressure trend plot

Cooling water pressure monitoring

A product development lab was experiencing a severe, unexplained drop in pressure of the process cooling water. It appeared to occur at random intervals. Without resolving the problem, these failures ultimately could result in tool downtime, pump repair costs ($2500 -$6000) and potential wafer losses.

The fab equipment engineering team decided to use 15 WGRs in their facility with a one-minute sample rate. Upon tracing the pressure profile, they discovered that it was not random but periodic. The problem was traced to a gate valve opening too fast in a tool.

In another instance, after the WGRs were installed, an alarm sounded as the pressure was falling. A leak was discovered in a piece of support equipment. No vacuum pumps were lost.

In another fab situation at a separate company, the main cooling water supply pipe suffered a rupture caused by old age at the fitting, which made the entire fab’s source of process cooling water disappear. It happened on a Friday night, and only one technician was on duty. The fab suffered downtime of one week, and five pumps had to be replaced at a cost of $75K. An inexpensive tool, such as a remote water pressure monitor, would have alerted the technician, perhaps in time to avert the burnout of all the pumps.

Improving tool productivity

An improvement has been developed in the way an end point is detected in plasma CVD/PECVD-based tools. RF sensors (Fig. 4) can reduces the time and material spent cleaning the chamber. This unit has been demonstrated in a number of customer sites on both Novellus’ (San Jose) and Applied Materials’ (Santa Clara, Calif.) plasma deposition tools. End points can be tracked more accurately, and gas usage and cycle times can be substantially reduced.

4. RF transducer (Source: Forth-Rite)

The impact has been significant. A 10%-50% reduction in chamber clean times (20 min on a 25-wafer batch) over conventional methods was achieved. It enables overall C₂F₆ to be reduced by 40%-50%. This technology helps optimize chamber cleans (Fig. 5).

5. RF endpoint chamber clean elapsed time.

This technology has been repeated in numerous tools and chambers with consistent results. OEMs are now making this capability an option that can be ordered directly.

The financial analysis summary of a customer scenario:

• Spot market pricing of clean gases is:
  • $28/lb C₂F₆
  • $18/lb C₃F₈
  • $36/lb C₄F₈
• Only 40% of C₃F₈ is needed to etch a film area, as compared with C₂F₆. Further, C₃F₈ is “easier” for abatement systems to “digest” and, thus, is less expensive to abate. These two factors translate into a 74% reduction in cleaning costs.
• Actual customer data indicates a savings of $2400 per 200 mm tool per month, when their C1 chamber clean was converted from C₂F₆ to C₃F₈ using the Forth-Rite control mechanism.

According to Forth-Rite data, including the cost of the gas handling conversion, ROI was <5 weeks per tool.

Conclusions

The impact of dynamically reading many of these hidden gauges, previously thought too difficult, has uncovered significant cost, resource savings and yield improvement opportunities.You must measure and monitor before you can control. Customers get the best results by using WGR first for diagnosis and troubleshooting, then expanding into more automated control methods.

Acknowledgements

We wish to thank Ted Berg of Stanford’s Nanofabrication Facility; Ron Farry, Guy Gandenberger and Tom Wenger of Micrel; Phil McGowan, Mike Moore and Ray Romero of SVTC; Pat Ireland, Dave Low and Carl McMahon of Novellus; Harry Sim of Cypress Systems; and Terry Turner of Forth-Rite Technologies.

References

1. Taylor-Deininger Partners Inc., “Hidden Gold in Your Manufacturing Facility,” FEO Magazine, February 2008.
2. Micrel Gas Cylinder Case Study, January 2008.