(Download this paper for all tables and figures) (170.3 KB)

Introduction

Particle monitoring of ultra pure water systems provides the ability to qualify new systems, monitor ongoing operation, and detect anomalies such as filter failure and bacteria growth in real-time. Particle Measuring Systems' HSLIS M-50 particle counter was specifically designed to monitor ultra-pure water systems and has been the standard for measuring DI water for a number of years. However, as ultra-pure water systems have improved the concentration of particles has decreased to a level that is very difficult to measure. In fact, state of the art DI water systems are so clean that the intrinsic noise of the M-50 particle counter, the zero count level, may be larger than the true concentration of particles. In response to these very clean water systems Particle Measuring Systems has developed a new instrument, the Ultra DI® 50 particle counter. The new instrument has 15 times the sample volume that provides statistically significant data faster and allows it to effectively monitor these very clean water systems. If the cleanliness of the water system is cruicial and accurate monitoring is essential, the best course of action is to replace the M-50 particle counter with the Ultra DI 50 particle counter. However, by understanding the source of the noise and the statistics of particle counting, the usefulness of the HSLIS M-50 particle counter can often be extended.

Zero counts and the HSLIS M-50 Particle Counter

Detecting problems such as bacteria growth is usually no problem because the number of particles is quite large. The difficulty occurs when monitoring systems that are very clean. The intrinsic noise of an M-50 particle counter is about the same magnitude of the particle counts on these systems, therefore it is impossible to distinguish between real particles and noise counts. This noise is commonly referred to as cosmic rays, but is in fact high energy particles that can come from a number of sources including, true cosmic rays, background radiation from the surrounding environment (this is very high in Colorado), and even from the glass covering the detector. When these particles strike the detector they penetrate through the silicon and generate a pulse of free electrons. These electrons are indistinguishable from electrons generated by the light scattered from particles, and so the cosmic rays are counted as if they were particles.

Cosmic ray noise is only a problem on systems that have very small signals from particles and small noise levels from other sources like light noise from molecular scattering, and detector and amplifier noise. Thus, systems that look at very small particles, like the M-50 particle counter, are sensitive to this noise source.

During manufacturing, the zero count level is checked by running the M-50 particle counter for 8 hours. The specification is that the maximum concentration must be less than 4 counts/ml over this time period. This measurement technique includes both real particles and cosmic rays. Therefore, with cleaner water systems the zero count level will be lower. The sample volume of an M-50 particle counter is .25 ml/min, resulting in 4 counts/ml x .25 ml/min = 1 cosmic ray per minute. A typical system runs about 2 counts/ml.

Statistical variation

The zero count specification is an average over an eight hour period, as mentioned above. However, both the noise and particle counts arrive at random intervals, and these intervals follow a Poisson distribution. If the sample interval is too short the statistical variation will make the data appear as if there are spikes in the particle concentration. Most of these spikes are simply from the random nature of the data and longer sample intervals would eliminate the spikes. Typically the shortest sample interval that should be used if the particle concentration is less than 4/ml is 20 minutes, and one hour is recommended. Table 1.

Actual number of occurrences vs. predicted

The problem with statistical variation is shown in Figure 1 <(Download this paper for all tables and figures) (170.3 KB) where a day's worth of data from an M-50 particle counter using 4 minute sample intervals is shown. The customer was concerned because there are several data points at 6 counts or higher even though the mean for the day was only 3.2 counts. Based on the mean and using Poisson statistics, one can calculate how many times in a day one would expect see each number of counts. The results are shown in Table 1 and plotted in Figure 2. For example, using the Poisson distribution, one would expect 6 counts 20 times over the day, but only 17 actually happened. The results are plotted in Figure 2, and as can be seen the agreement between theory and measurement is very good. The occasional high counts seen in Figure 1 are simply the result of the statistical variation, and reacting to these high counts is inappropriate.

Figure 2 (Download this paper for all tables and figures) (170.3 KB)Frequency of predicted and actual number of particles in 4 minute sample intervals over one day.

Other considerations

Often the data from an M-50 particle counter is compared to Scanning Electron Microscope (SEM) data. The SEM test often takes days or weeks and gives an average value over that time. In contrast, short sample intervals of a few minutes are often used on the M-50 particle counter. The M-50 particle counter will show some spikes of higher particle concentration, much higher than the long-term average of the SEM. However, if the particle data from the M-50 is averaged over the same time period, the particle counter and SEM should agree reasonably well.

Bubbles can be another confusing factor when comparing particle monitors with SEM data. On rare occasions microbubbles can form in the DI line and the particle monitor will count those as particles. The SEM, on the other hand, will not count bubbles as particles. Again, the data from the particle monitor and the SEM is different, and must be analyzed and interpreted accordingly.

One way to diagnose bubbles is to correlate the particle concentration with temperature. As the temperature of the water rises more bubbles will tend to form and the particle concentration will increase. However, if the temperature increases and no increase in particles is seen, then elevated particle concentrations are less likely to be related to bubbles. Another way to distinguish particles is to increase the pressure in the DI line. Bubble formation should decrease with increased pressure, so if the particle concentration decreases with pressure, bubbles are likely to be present.

The current HSLIS M50 particle counter is a reliable instrument and is still the standard for measuring DI water systems in the semiconductor industry. As is the norm in the semiconductor industry improvements in technology have forced improvements in Particle Measuring Systems' products, and we have responded with the Ultra DI 50 particle counter.

The new Ultra DI 50 particle counter was designed to address the zero count limitations of the current M-50 by increasing the sample volume by a factor of 15. Assuming the cosmic ray rate stays constant, then 1 cosmic ray /minute divided by the sample volume of 3.75 ml/minute gives a zero count level of less than .3 particles/ml. This should be adequate to monitor most DI water systems.

(Download this paper for all tables and figures) (170.3 KB)

Ultra DI® is a registered trademark of Particle Measuring Systems, Inc.

Contact us if you need more information or have questions.

Printed: 1999