Ultrapure Gas Analysis

Characterizing an electrochemical oxygen sensor for process gas monitoring applications

Mohammad Razaq, Advanced Instruments

Ultra-high-purity (UHP) bulkgases undergo stringent quality checks and their distribution systems are always qualified prior to commissioning. However, during routine fab operations, an inadvertent error by a process tool operator or the failure of a delivery system component such as a valve or pressure regulator can cause gas contamination, which, if not detected in a timely manner, can result in a significant amount of scrapped wafers. The presence of even very low parts-per-billion levels of oxygen in such semiconductor process gases as nitrogen, argon, and helium can adversely affect both manufacturing yields and device quality. For example, oxygen can react with silicon during processing, forming undesirable oxides on the wafer surface. The result is uncontrolled gate-oxide growth that alters the characteristics of the devices. The effects of parts-per-billion to low-parts-per-million levels of oxygen contamination are generally detected at the annealing or epitaxial process step, since the affected wafer surfaces will appear hazy and dull. However, when contamination is more subtle, an uncontrolled thin (<50-Å) oxide growth may go unnoticed until the contaminated wafers reach a testing stage. Therefore, it is desirable to continuously monitor both the purity of the process gases and the integrity of their delivery systems, especially close to the point of use and at tool exits.

One innovative electrochemical sensor design developed during the last decade achieved an almost thousandfold increase in signal output by employing a large (~1 to 1-1/2-in.-diam), highly porous, metal-catalyzed gas diffusion electrode. The increase in signal output enables the sensor to exhibit a sensitivity of <1 ppb and a signal noise of <0.2 ppb. However, the large size of the porous electrode facilitates the loss of moisture from within the sensor, so that a large internal volume is required to accommodate sufficient electrolyte for a reasonable operating life.

With an internal volume of up to 250 cm³, more water and electrolyte must be added to the sensor after 3 to 6 months of use to maintain its operability. This periodic addition of water and electrolyte is not only disliked by operators, but it also takes the analyzer off-line for several hours. Furthermore, after an upset, these sensors recover very slowly. After only a few minutes of exposure to air, it may take more than 24 hours for such a sensor to recover to within 10 ppb of its stable value. This sluggishness is due to the slow consumption of the oxygen that gets dissolved in the electrolyte at the time of the exposure. (In an electrochemical sensor, oxygen is consumed via reactions at the cathode and the anode, or in certain designs, at an additional secondary electrode. The rate of the reaction is strictly limited by the rate of diffusion of oxygen in the electrolyte, which in the absence of convection is a slow process.)

The remainder of this article focus on a new sensor technology that was developed by Advanced Instruments (Pomona, CA) to address the shortcomings of conventional electrochemical sensors. The Pico-Ion oxygen sensor is described briefly, and research into its capabilities is discussed.

In addition, the design of the gas chamber above the cathode maximizes the oxygen reaction rate at the cathode and hence minimizes the amount of oxygen that escapes unreacted and gets dissolved in the electrolyte. These features combine to produce a sensor with a nominal sensitivity of 1 ppb, a noise level of <0.2 ppb, a 90% full-scale response rate of less than 40 seconds, and the ability to recover quickly from a process upset condition. The unique physical construction of the gas path to the sensing cathode also minimizes the temperature dependence of the sensor's signal even at very low oxygen levels, thus ensuring the long-term stability of the sensor even in environments in which temperature can vary significantly during a single day-and-night cycle.

To validate its sensitivity to oxygen, the sensor was incorporated into Advanced Instruments' Model GPR-16HP analyzer, which contains a fast-start bypass sampling system shown schematically in Figure 2. For the test runs, gas samples containing different concentrations of oxygen in increments of 1 ppb were obtained by mixing a 5-ppm-span oxygen gas with pure nitrogen. The temperature of the sensor was maintained by a PID controller set at 85°F, and sample gas flow was set at 1 std cu ft/hr. The data obtained are shown in Figure 3. The clear step changes seen in the figure demonstrate that the analyzer can detect changes in oxygen concentration down to 1 ppb. Beta-site testing of the analyzer at various manufacturers of UHP gases in the United States and Japan confirmed the results shown here.

Figure 3: Test results showing the sensor's response to incremental 1-ppb changes per hour in oxygen levels in a nitrogen gas.

Long-term testing of the sensor in both inert and reactive gases, such as hydrogen, showed that it exhibits excellent stability and sensitivity and that its output signal has a noise level of ±0.2 ppb, which results in a very high signal-to-noise ratio. It can recover to within 1 ppm of its stable value following a 5-second exposure to air in less than 15 minutes, as shown in Figure 4. This recovery period is almost 25 times faster than that of conventional oxygen sensors. No deterioration in sensitivity or recovery time was observed over a 1-year test period, during which the sensor was exposed to high levels of oxygen (8–9 ppm) at scheduled intervals.

Figure 4: Test results showing the sensor's ability to return to within 1 ppm of its stable value following a 5-second exposure to air.

 

Figure 5: Monitoring data showing the analyzer's stability during a 24-hour period when temperature varied ±10°F.

When the stability of the analyzer that contained the sensor was checked by periodic recalibration (every 2–4 weeks) using a certified 8.9-ppm-span gas, it remained within ±5% of the span over more than 6 months of operation. The stability of the analyzer with oxygen-free nitrogen (zero gas) was also checked by monitoring the signal for several months. The signal drift was less than ±1 ppb during day-and-night cycles where ambient temperature swings were typically ±10°F. Figure 5 illustrates the stability observed within one 24-hour interval.

Furthermore, because of the sensor's small internal volume, the amount of oxygen that gets dissolved in the electrolyte during a system upset is quickly consumed so that recovery from the upset occurs within minutes. In sum, this sensor technology offers a viable alternative to existing expensive and maintenance-intensive equipment for monitoring oxygen in process gases.