In situ monitoring for semiconductor fab chemical supplies
|After a series of chemical supply issues at Philips Semiconductor (now NXP) in Fishkill, NY, the fab identified several clear deficiencies, particularly a lack of chemical monitoring to detect future failures. A review of monitoring techniques found that the refractive index measurement is the most cost-effective and reliable way to monitor the variety of chemistries encountered in the fab. This paper reviews several applications of the refractive index technique, for both process monitoring and fault detection.|
It's safe to say that chemicals delivered to semiconductor fabs are among the most carefully tested, impurity-free products of their kind. Chemical suppliers who deliver anything but top quality product simply don't succeed in the semiconductor industry. Though mistakes do happen, they are quite rare.
So, why did Sematech data show that chemical supply is one of the top fab reliability concerns for more than 10 consecutive years ? Because very little monitoring occurs after chemicals reach the fab's loading dock. Along the complex in-fab supply chain are ample opportunities for human errors and mechanical failures, leading to misprocessing and contamination—a fab's custom blend might not be mixed to the correct specification, one slurry or photoresist might be swapped with another by mistake, or failure to purge a supply line might cause cross-contamination.
Even when all procedures are followed to the letter, the composition delivered by the supplier may not be received at the process tool. Chemical baths pick up contaminants and reaction products from wafers passing through them. Between delivery and process, highly reactive chemistries break down, water evaporates, and chemicals become less effective as they get "used up."
Fabs depend on established procedures to minimize product risk. Without accurate monitoring, however, they can't actually know the composition of a given chemical stream. Without monitoring, product wafers become the de facto chemical monitors. Errors, when they occur, may not be detected until the etch rate drops below specification, the particle counter spikes, or a lithography or CMP step fails to deliver the desired results. By that time, wafers have already been lost and large-scale equipment contamination may have occurred.
Without accurate monitoring of chemical solutions, fabs must add fresh chemicals or replace the bath entirely at regular intervals to maintain performance. While this approach can prevent bath degradation, unnecessary replacement of a good bath increases chemical consumption and disposal costs.
Balancing cost, speed, accuracy
Monitoring chemical composition along the intra-fab supply chain is difficult. The ideal method would provide fast, accurate measurements, obtained without immersing probes into the chemical, and made instantly available through the fab's information network. Keeping the cost per measurement down—by minimizing consumables, chemical waste, and equipment costs—allows fabs to limit process risk by sampling frequently.
Few technologies can meet all of these requirements. Thus, fabs must constantly balance cost, accuracy, and other factors.
Titration—the gold standard for measurement accuracy—is slow and expensive. It requires consumable reagents, creates a new waste stream of sampled chemicals and their reaction products, and introduces possible contamination through the probe used to collect samples.
Conductivity, pH, and IR spectroscopy methods can be useful in some situations, but are not universally applicable across the many compositions encountered in a semiconductor fab. Conductivity measurements require ionic solutions. Neither conductivity nor pH is precise enough, as several different compositions might produce the same measured value. Both techniques also require chemical contact between sampling probes and the sampled chemistry, potentially introducing contaminants. IR spectroscopy is an optical, non-contaminating method that can identify individual components in a solution. But it cannot easily determine their relative concentrations, and it is difficult to calibrate.
In place of these techniques, in situ refractive index measurements are emerging as a cost-effective tool for fast, accurate, real-time measurement of chemical composition. The refractive index can be measured at the interface between the fluid and a chemically inert window. Thus, the light does not need to pass through the fluid. This method can be used with opaque fluids, and is not affected by bubbles and other flow irregularities. Overlapping refractive index values are rare. Even very similar mixtures generally have unique index values.
Converting refractive index to weight percent is straightforward for common chemicals. For proprietary solutions, it may not be possible to measure the concentrations of individual components. Instead, the refractometer can determine the concentration of the solution as a whole based on its optical fingerprint. Deviations from this value indicate a deviation from the ideal bath composition.
Research at the Fishkill fab found that refractive index measurements are not reliable for surfactants. Levels of these compounds, used in parts per million quantities to control surface energy, are best monitored by dynamic surface tension measurements.
Generally speaking, chemical monitoring applications can be divided between process monitoring and fault detection. In process monitoring applications, the desired chemistry has been established and must be maintained over time. In fault detection applications, the system must verify that the correct chemical is being introduced to the process.
Refractive index for fault detection
In lithography, for example, the fab might use several different photoresists, depending on the etch resistance and optical properties needed for a particular mask layer. Whether the error is in the distribution system—such as a stuck valve—or in human container handling, dispensing the wrong photoresist leads to very costly production problems and wafer damage. The fab must often purge the wrong photoresist and clean the tool before more wafers can be processed. Photoresists are extremely expensive specialty chemicals, and lithography cells are among the most capital-intensive areas in the fab. Photoresist dispensing errors impose substantial costs—wasted chemicals, lost production time, and wafer scrap. Fortunately, a broad range refractive index sensor can differentiate among 16 different types of photoresist, most of them DUVs (Fig. 1) confirming that the requested resist is the one actually being dispensed.
|Figure 1. 16 different resists (most of them DUVs), each with a different refractive index. The difference marginal between every resist is larger than the measurement accuracy, RI0.0002, of the hardware used.|
Similarly, we found that each of 26 different CMP slurries, both commercial and custom-blended, had a different refractive index(Fig. 2). Typically, the chemical and abrasive components of CMP slurries are diluted and mixed in the fab, near the process tool, due to the risk of agglomeration and chemical degradation. K-Patents' refractometer provides a real-time method for monitoring CMP slurries in the POU blending and distribution loop.
|Figure 2. 26 different CMP slurries, each with a different refractive index.|
Refractive index for process monitoring
In other applications, fabs need to know how the content of a chemical bath changes over time. In MEMS production, for example, potassium hydroxide (KOH) etching chemically removes silicon from the wafer surface. The etch rate of silicon in a KOH bath depends on the bath temperature and the KOH concentration. As multiple wafers are etched in the same bath, dissolved silicate is released, forming a tertiary solution. An in-line refractometer provides a real-time indication of the concentration of KOH or other etchant solutions. K-Patents provides a method (patent pending) for compensating for the influence of the dissolved silicate in the refractometer output reading.
|Figure 3. Change in SC1 dilution with time.|
This technique can also monitor custom dilutions, except that in this case, the desired refractive index range must be determined experimentally. Figure 3 is monitoring data for a custom blend of ammonium hydroxide, hydrogen peroxide, and de-ionized water from the Philips Fishkill Fab. This is a common blend used by most semiconductor manufacturers, but the ratio is different for each fab and some use many different dilutions. In Fig. 3, taken from an older blending system, the blend is gradually getting stronger. After lab analysis confirmed the trend, it was determined that the day tank of this material slowly weakened, and that the addition of the fresh blends gradually strengthened the contents. The day tank has been removed and the chemical is now blended on demand. Process yield has improved as a result.
When a chemical bath degrades over time, the normal procedure is to simply replace it after a set time interval. For example, the effectiveness of an EKC resist removal solution depends on water content. Water evaporates from spray solvent tools over time, dropping from 18% down to 13%. Above 18% water, the solution is too aggressive; below 13%, it is no longer effective.
Refractive in-dex provides a measurement of the water content. The fab can simply monitor EKC's water content as it depletes,or can maintain it at a steady establishedlevel. With an accurate water contentmonitor, such as provided by the refractive index measurement, the system can auto-matically spike the bath with water after each wafer run. This approach maintains the water level at the desired set point and helps to extend the life of the EKC bath. In our tests, automated water spiking decreased consumption of the chemical by 20% per run, and doubled the bath life.
Note that the EKC resist removal process takes place at 65° to 75°C; the monitoring hardware must be able to tolerate temperatures in excess of that value.
|Figure 4. Flow chart of water spiking system for post-etch resist removal with EKC.|
Fabs considering the monitoring option have commented that, if they have three resist removal tools in a row, they will lose synchronization of the bath change for the three baths. Fabs currently change the chemistry for all tools at the same time; if they monitor the water content, they will most likely have to change the baths at different times, which would mean more work. The key is to maintain the water content at the same level for all three; then they can change all three at the same time—much later than they do now. The water spiking of the EKC bath can be arranged as in Fig. 4. EKC is supplied by Dupont; a typical fab uses 10,000 to 30,000 gallons (between 37,000 and 113,000 liters) of this chemical per year.
Though fab facilities such as water and air supplies have been monitored for a long time, routine monitoring of chemical supplies is rare. Yet chemical monitoring to reduce human and mechanical error and improve process yield costs less than the labor needed to correct a single major incident.
The Fishkill fab installed its first three refractive index devices in December 2004, and they were all qualified and released for full function in January 2005. The instruments have been networked, and the actual data is logged and charted in an SPC type application. Any drifts or trends can be quickly identified and addressed before they cause an incident in the manufacturing operation.
1. Private communication with Thomas Moseman, NXP Fishkill, based on SEMATECHreport, "2007 Facilities Systems ReliabilitySurvey," March, 2007.
Marcus Kavaljer received his MSc at Åbo Akademi U., and is an applicationengineer at K-Patents OY, P.O. Box 77,FI-01511, or Elannontie 5, FI-01510, Vantaa, Finland; +358 207 2915 70; firstname.lastname@example.org
Marja Kivenheimo received her MSc. at Henley U. of Reading, and is a marketing manager at K-Patents OY.
Ville Voipio received his MSc. and PhD. at Helsinki U. of Technology, and is a managing director of Janesko OY, P.O. Box 77, FI-01511, or Elannontie 5, FI-01510, Vantaa, Finland; +358 207 2915 70; email@example.com
Thomas J. Moseman received his BS at the Rochester Institute of Technology, and is a facilities engineer at NXP Semiconductors Fishkill, 9 Kendell Drive, Wappingers Falls, NY 12590 USA; 845-902-1396; tom.moseman@NXP.com