Measuring Moisture in Semiconductor Feed Gas
Why measure moisture levels?
Modern semiconductor fabrication involves an intricate series of steps to pattern circuits onto silicon wafers. To make the smallest patterns possible, many of these steps are carried out using gas-surface chemistry so that the reactions can be precisely controlled. Processing steps include sputtering and chemical vapor deposition to deposit metals or dopants, and plasma processing or reactive ion etching to remove material. When moisture is present, the rates of the chemical reactions can change, causing circuit traces to be too thin or too thick. In addition, water vapor accelerates corrosion damage to the pipes which supply the gases to the reactors. Corroded pipes release tiny particles which can also damage semiconductor circuits by causing short-circuits.
To control the chemistry and prevent corrosion, gas manufactures have supplied ever purer feed gases, with impurities reduced below one molecule in every billion delivered. However, water is present in the atmosphere at levels of about 1%, so that even a very small leak can reduce the purity of the gas significantly. Verifying the gas purity challenges existing analytical instruments. The most sensitive technique available is atmospheric pressure ionization mass spectroscopy (APIMS), but at $100,000 or more per instrument, it is also the most expensive. Electrochemical sensors perform poorly at very dry conditions and can not operate in hydrogen, oxygen, or corrosive gases. Frost point sensors have very slow time response below 1 part per million due to their requirement to build up a detectable layer of frost on the sensing element. What is required is a new analytical technique that combines high sensitivity with low cost.
How do we measure moisture?
Diode lasers can be used to detect moisture by detecting the laser power absorbed in the infrared wavelength region near 1400 nm. We use a miniature diode laser about the same size and power as the laser in a compact disk player, but designed to operate at a wavelength where water vapor absorbs strongly. Then we scan the laser wavelength 50,000 times each second across the water absorption feature. At this high scan rate, there is little change in laser power due to vibrations or electrical noise, so that very small changes (less than 0.001%) in the laser power can be detected. In turn this means that we can detect very small (less than 1 part per billion) amounts of water vapor.
The instrument is a 1.4 m long tube divided into two parts. One part houses the laser and detector. It is specially treated to remove all water vapor so that it will not contribute to the signal. The other side contains the semiconductor gas. The laser beam bounces back and forth in this sample region about 50 times, resulting in a path of about 50 meters through the semiconductor gas (long paths improve sensitivity). After bouncing around, the laser beam goes to a detector, which converts the laser power to an electronic signal. Because we modulate the laser wavelength 50,000 times per second, the laser power varies periodically when water vapor is present. Special low noise electronics capture this periodic signal (the electronics work like an AM radio), then a computer logs the data and converts it to a water concentration.
One of the hard things about measuring water vapor is that it "sticks" to just about everything. All the components of the measurement system are made of electropolished stainless steel or other high vacuum-compatible materials, to minimize the stickiness. Gas constantly flows through the measurement region to further minimize the effects of the surfaces. The cell can be heated to help unstick any moisture present after the cell has been exposed to the atmosphere for maintenance.
Calibration at NIST: Thermodynamic vs. Optical Calibration
The instrumentation was brought to NIST for calibration and delivery. NIST has built a new instrument which can produce calibrated levels of trace moisture, from relatively high levels (more than 10 parts per million) to trace levels (5 parts per billion). To produce these levels, air is blown through a long copper tube containing ice at a precisely defined temperature. By changing the temperature of the ice or the total air pressure, the water concentration can be controlled. The NIST moisture generator is calibrated from the thermodynamic properties of ice.
The usual method for calibrating diode laser sensors is to supply a known gas concentration and adjust a "calibration factor" to make the instrument reading correspond to that concentration. For this project, we employed a different approach: we calibrated the electronic gain factors and the optical properties of the laser, then we used a detailed numerical simulation to compute the electrical signal strength as a function of water concentration. This new method traces its calibration to the published optical properties of water vapor. There is no adjustable "calibration factor."
Next we connected our moisture measuring instrument to the NIST moisture generator. We made measurements at a series of moisture levels in the range from 5 parts per billion to 2,000 parts per billion. This lets us compare the thermodynamic standard to the optical standard over a wide range. The result is shown in the figure (From G. E. Scace, D. C. Hovde, J. T. Hodges, P. H. Huang, J. A. Silver, and J. R. Whetstone, "Performance of a High-Precision Low Frost-Point Humidity Generator," presented at the International Humidity Conference, London, 1998). The straight line corresponds to perfect agreement; the circles are the actual measurements. Both instruments are in very good agreement across the measurement range, with a slight deviation at low signal levels that appears to be due to outgassing within the diode laser instrument. Additional tests showed the high sensitivity of the laser instrument, which is as high as 65 parts per trillion at the optimum signal averaging time.
Additional information is available in "Wavelength–Modulation Laser Hygrometer for Ultrasensitive Detection of Water Vapor in Semiconductor Gases," D. C. Hovde, J. T. Hodges, G. E. Scace and J. A. Silver, Appl. Opt. 40, 829-839 (2001).. This research was sponsored by the Department of Commerce through the SBIR program.
Contact Information
e-mail info@swsciences.com