Abstract

The sensitivity, resolution and sample rate of optical particle monitors, counters and spectrometers are described. Important differences in conceptual design of each instrument class are detailed. Optical particle counters and spectrometers require uniform sample volume illumination and have been used in monitoring aerosol microcontamination for many years. Spectrometers have the highest resolution and the greatest number of size channels. Counters may have high intrinsic resolution but it is lost because of the fewer number of size channels provided.

Monitors are relatively new class of instrument which do not provide uniform sample volume illumination. They are becoming widely used in liquid monitoring. Monitors are simpler, less expensive devices and are characterized as having poor resolution but providing the highest sample flow rates and delivering the largest data base. When used on fluids with normal populations having an exponential size distribution, monitors show little size distribution distortion. When modal populations or deviations from exponential size distributions are encountered, counters or spectrometers are required. Filter penetration tests generally demand the highest resolution offered only by spectrometers.

Data are presented from field visits to 17 semiconductor plants having deionized water processing facilities where monitors and spectrometers were used simultaneously to characterize water quality. The data set provides an interesting comparison within the industry as well as an opportunity to compare the low resolution monitors with high resolution spectrometers under field conditions.

1.0 Introduction

The measurement of particles is of fundamental importance to a great number of disciplines. Size measurements span the range of the smallest macromolecular nuclei to the largest hydrometeor hailstones - 8 decades in diameter and 24 decades in mass. Interest in particles can be from the standpoint of natural particle phenomenon, particle products, or as particle problems. Particle phenomenon include natural atmospheric aerosol, clouds, fog, and precipitation. Particle products range from the obvious aerosol sprays, powdered foods, paint pigments, and ink toners to an almost infinite array of chemical products including the majority of pharmaceutical and cosmetic products. Particle problems relate to contamination of which microcontamination is receiving the greatest amount of recent attention and is out topic of immediate interest here.

It is of interest to note that most of these fields of particle study desire the measurement of particle size distributions. Oftentimes success hinges on the ability to measure and sometimes control, particle sizes extremely well. For example, in the particle product arena, a paint pigment performs most efficiently at one size only. Here measuring the total number of particles, or concentration of a sample, is of no importance at all, only the relative size distribution matters. Turning to particle phenomenon in the field of cloud microphysics, the size distribution at or near cloud base is extremely narrow due to growth competition of droplets for water vapor. Following the initial size spectrum through the ultimate development of precipitation requires high resolution and broad dynamic range capabilities oftentimes tracking the subtle development of multimodal size distribution characteristics with very accurate number density measurements. A common characteristic of most of these measurements is the need for high resolution size spectral data in differential form. Most often the supply of particles in inexhaustible and sizing accuracy and resolution are more often limited by problems with coincidence error than insufficient sample statistics.

The field of microcontamination is perhaps unique in that it does not share the above measurement needs. Microcontaminants are "rare events" with the primary goal of contamination control to indeed "exhaust the supply." Differential size distribution data are less import than the cumulative concentration totals at sizes above benchmark size thresholds. Coincidence error is replaced by problems of loss of statistical confidence from under populated samples. In short, we need to view the needs of microcontamination from a different perspective. The measurement problem here is simple not solved by the measurement of a few particles to exceedingly high tolerances if we have to accept nearly zero confidence that they are representative of the overall population we are seeking to quantify. Populations of microcontaminants are themselves fairly well behaved. In nearly all cases, the size distribution is exponential with the number density invariably increasing dramatically as size decreases. Increasing size sensitivity often becomes the all-important criterion in order to acquire improved measurement statistics. However, increasing size sensitivity ordinarily requires decreasing the sample volume. Hopefully, the greater the sensitivity the more quickly and confidently a situation can be characterized, in spite of the sample volume lost.

Manufacturers of instruments to measure particulate microcontaminants have generally tried to meet the needs of users by providing instruments that are capable of measuring sufficient numbers of small particles to characterize a clean media in a reasonable time (i.e., a few minutes). The ability to detect real short-term, out-of-spec fluctuations (episodic events) is often of nearly equal interest. For the most part, these needs are met with relatively low-resolution instruments. For real time measurements, the optical particle counter (OPC) is the overwhelming choice. For certain specialization applications, higher resolution particle size spectrometers are preferred. Both of these devices are fairly well understood with their performance and characteristics the subject of countless articles. However, in recent years a new type of instrument has emerged which differs significantly from OPCs or spectrometers. It is what we shall term a particle "monitor" simply for ease of differentiation. The device has high sensitivity, high sample rate, but quite poor resolution. It represents a trade-off in design criteria to enhance the special needs of ultra-low levels of microcontamination (rare events). It is becoming the preferred instrument choice for measurements in high-purity D.I. water. This paper provides an introduction to the performance and characteristics of this new instrument.

We begin with a review of aerosol measurement instruments including spectrometers and OPCs. We then discuss liquidborne particle size measurements introducing "in situ" measurement concepts. The monitor is then described and its theoretical and actual performance described. Finally, we present field measurements taken from 17 semiconductor plant D.I. water systems taken with monitors and spectrometers.

2.0 Underpinnings from Aerosol Measurements

Those of us who have out backgrounds in aerosol physics have often required instruments with reasonably high size resolution to support size distribution analysis work. A typical instrument may have 15-64 size channels and provide sufficient size resolution to reveal bimodal, trimodal or otherwise multimodal characteristics of aerosol populations. The differential rather than cumulative distribution is most revealing. Such detail provides clues as to particle origin (e.g. whether a condensate, agglomerate, or scavenged particle), particle growth characteristics, and integrated mass, optical depth and related particle ensemble properties. Such devices are designed optical particle size spectrometers and are characteristically low flow devices. A maximum flow rate of 0.01 cfm is typical and oftentimes must be further reduced to avoid coincidence errors when sampling polluted environments.

Inherent in the definition given (i.e., a spectrometer) is the ability of these instruments to provide the size "spectral" resolution necessary. The number of size channels required depends on the dynamic range of sized covered. Typically, an order of magnitude of particle size would dictate a minimum of 10 size channels. A number of workers have chosen to use multi-channel analyzers having up to 1,024 size channels. In general, this is at least an order of magnitude more size channels than can be supported by the "intrinsic" instrumental resolution. The "intrinsic" instrumental resolution is the underlying instrument's resolving power or, simply stated, "the smallest increment in particle size that can be distinguished." This is ordinarily a function of factors other than number of size channels, such as, signal-to-noise ratio, illumination uniformity, and response function characteristics. Providing excessive numbers of size channels is akin to a magnification of 10,000 by an optical microscope. Excessive resolution is termed "empty" resolution by microscopists. Even so, multichannel analyzers do provide a number of computational statistical tools that are of considerable value independent of their excessive "empty" resolution.

For aerosol microcontamination work, particle size spectrometers are much less common than OPCs. OPCs principally differ from spectrometers by providing only a few a size channels. Both devices require gas laser cavities to generate the watts of power needed for the most sensitive or high flow rate requirements. OPCs may or may not have high "intrinsic" resolution but, even if they do, they are generally incapable of revealing spectral size distribution properties because of the limited number of size channels. One reason that OPCs are generally preferred for microcontamination work is that the particle environments of quality clean rooms are "rare event" regimes totally unlike the particle rich ensembles of ambient aerosols. As such, the optical particle counters forgo high resolution in favor of large volumetric sampling rate - a 1 cfm flow being typical. The particle size distribution in a clean room would also be generally uninteresting to an aerosol physicist because cyclic HEPA filtration invariably generates a steeply exponential distribution with a slope of D-3 to D-4. Cumulative rather than differential size distributions are preferred. A typical clean room particle counter will have a maximum of eight size channels and more often only four to six.

Aerosol particle size spectrometers are largely used in microcontamination for calibration work, filter testing, and a variety of R&D activities. The HS-LAS manufactured by PMS is an example of such an instrument. This 32-size class instrument is a modern version of the 16-channel LAS-X introduced by PMS around 1980 (Knollenberg, 1989). With the optical system shown in Figure 1 its intrinsic resolution is quite high. At 0.1µm it is sufficient to reveal the underlying size distribution of the monodispersed PSL used to calibrate OPCs. This is shown in Figures 2a and 2b which reveal the size distribution of 0.100µm PSL before and after passing through an electrical aerosol classifier. With its lower sensitivity of 0.065µm , it provides an adequate reference instrument to assess the counting efficiency of OPCs - particularly near 0.1µm .

Figure 1. Optical and Flow System Diagrams for PMS HS-LAS and LAS-X Instruments. The optical system sis the same for both instruments, however, the laser cavity power is much higher in the HS-LAS and a superior detector is used.

Counting efficiency is properly limited to defining the characteristics of a particle sizing device near its threshold. It is that percentage of particles passing through an instrument's sensing volume that generated sufficient amplitude signals to be registered as a positive indication of a particle's passing. It should be clear that if a particle size spectrometer is capable of revealing the total population of a test particle, such as that of Figures 2a and 2b, it must have 100% counting efficiency as shown in Figure 3, although one is unable to determine this for certain because of the limited number of size channels available. This example illustrates the premise that, in order to test an instrument’s counting efficiency with reference to a standard test instrument, that standard instrument must itself be of greater resolution and sensitivity than the instrument being tested - a logical but not always followed premise.

Figure 2.Size Distributions of Nebulized 0.100µm PSL as Measured by HS-LAS.Before (A) and after (B) passing through a TSI electrostatic classifier. The electrostatic classifier removes background contamination materials as well as narrows the actual PSL size distribution.The HS-LAS's resolution is about a factor of three less than the standard deviation of the PSL.In either case, 100% counting efficiency is indicated by the closed distribution.

A 50/50 split for 0.300µm PSL is shown for an optical particle counter (a PMS Micro LPC-110) in Figure 6. The lower resolution of this device compared to the HS-LAS is apparent. In summary, one can state that the OPCs are generally low resolution devices with sufficiently high flow rates to provide adequate statistical confidence in measuring rarified particle populations, while useful particle size spectrometers typically have very high resolution but very limited flow rate.

The inherent resolution of a spectrometer and the high flow rate of the OPC are attributable to a number of factors- not the least of which are the viewing volume properties. Figure 7 illustrates comparative sample cross sections for these two instrument types. An OPC might typically use a rectangular jet with a 1 x 10mm cross section. A multimode laser cavity affords the greatest percentage use of the beam diameter with some intensity variations as shown in Figure 8. Spectrometers, on the other hand, can aerodynamically focus the aerosol to a circular cross section only 50µm diameter. Because of this small cross section, a Guassian TEMoo mode laser beam can be used affording maximum sensitivity yet with negligible variation in intensity. Of additional importance to a spectrometer's enhanced resolution is the small object field over which the collecting optical system covering a 1 x 10mm field are a prime factor in limiting the intrinsic resolution of OPCs.

Figure 3. Size Distribution of 0.073µm PSL as Measured by HS-LAS. Here the resolution of the HS-LAS is insufficient to provide a confirmation of 100% counting efficiency.

Figure 4. Counting Efficiency Curves for Three PMS OPCs. The counting efficiency of these instruments was determined using the HS-LAS as a comparative instrument standard.

Figure 5. Diagram Illustrating the Effects of Varying Resolution on Threshold Settings. Using a 50/50 split at a threshold removes errors caused by varying intrinsic instrumental resolution.

Figure 6. Size Distribution of 0.305µm PSL as Reported by PMS Micro LCP-110 Instrument. These histograms illustrate a near 50/50 split of a 0.3 µm PSL calibration standard. The difference in resolution in Figure 6 and Figure 2 is obviously sizable.

Figure 7. Diagram of Sample Air Flows Intersecting a Laser Beam. The sample cross sections of these two examples differ by more than 100X in area.

Figure 8. Laser Intensity Distributions for Air Flows in Figure 7.

3.0 Liquid Instruments

Measurements of particles in liquids differ substantially form those of aerosols. Until most recently, all liquid optical particle counting devices were of a "volumetric" type. A volumetric instrument is one in which all of the sample flow passes through the illuminated viewing volume. Such devices sample 100% of flow, and all particles at sized slightly above threshold should be counted. The "volumetric" terminology in liquid instruments is necessary to distinguish them form a second class of instruments that has no aerosol equivalent in microcontamination - "in situ" instruments. All microcontamination aerosol instruments are fundamentally "volumetric." Liquid "in situ" aerosol devices are currently offered is that, thus far, manufacturers have been successful in plumbing aerosol and, if necessary, even focusing aerosol fluid dynamically into well-controlled particle beams which pass through the laser beam in constrained paths. The construction of jetted sample flows, as shown in Figure 7, has never been satisfactorily devised for use in liquids.

Traditionally, liquid sampling cells for particle size measurements have been constructed using a sandwich of glass and metal or a transparent capillary (see Figures 9 and 10). A unique property of such cell designs is that the flow path us always much smaller than the laser beam diameter illuminating the sample flow. In this manner, all particles are illuminated and viewed, and the ratio of laser beam diameter to sample cell width largely determines the potential intrinsic resolution of such a sensor. Spectrometers or OPCs can be devised by meeting the appropriate requirements of size resolution and number of a PMS laser diode IMOLV spectrometer using PSL microspheres. This sensor has a 0.2µm lower size threshold and a 20 ml/min flow rate. Laser diodes are particularly important to liquid measurements because they offer higher power than conveniently sized gas lasers, and laser cavities are not yet viable in liquid media.

Figure 9. Sample Cell for a Liquid Volumetric Instrument. This liquid flow cell has four surfaces contributing background stray light and noise to limit sensitivity (Surfaces A, B and two window inside surfaces.)

Figure 10. Capillary Type Liquid Volumetric Flow Cell. This flow cell uses closer refraction index matching to reduce surface scattering at fluid/interface boundaries.

Figure 11. Histogram of 0.56µm PSL as measured by PMS 0.2µm Volumetric Sensor. The resolution of this instrument is highest at smaller sizes where the response curve is steepest.

The need for "in situ" devices for liquid measurements arises because of a fundamental limitation in the use of volumetric flow cells. Since the entire sample cell cross section must be illuminated, the interface between the cell wall and the liquid is also illuminated. The liquid/cell wall interfaces are a source of stray light largely due to surface defects. The noise generated from this stray light limits ultimate sensitivity.1 One can match refractive indices between certain liquid and cell materials to limit the scattering form defects, but the match is never perfect and surface contaminants are also stray light sources. Furthermore, matching cell wall and fluid refractive indices limits cell design to a narrow range of fluids.

In a properly designed "in situ" instrument, the laser beam passes through entrance and exit windows as shown in Figure 12. The collecting optical system is typically oriented at 90 degrees to avoid directly viewing the illuminated interfaces at the entrance and exit windows. This greatly reduces the collection of light scattered by defects and the sensitivity approaches the fundamental limits imposes by noise from the liquid media scatter alone. Since the particles are not constrained mechanically to intersect the laser beam in any special manner, the laser beam dimensions become a "free" design parameter and the beam diameter is ordinarily made extremely small to maximize sensitivity. Generating the smallest beams requires the use of Gaussian intensity distributions.

Figure 12. Liquid In Site Instrument Flow Cell Diagram. The wetted surfaces in a properly designed in situ instrument are out of the field-of-view of the collecting optics.

3.1 Properties of Gaussian Beams

A property of the uniphase TEMoo mode of a laser is that is has a Gaussian intensity distribution. For all practical purposes, any uniphase wave front will have a Gaussian intensity fall off at its edges (even multimode laser beams as shown in Figure 8). This essential feature of the lowest order TEMoo mode is required to produce the highest intensity, least divergence and best collimation (see Siegman, 1986, or Verdeyen, 1981, for details of laser and Gaussian beam properties). Any other pattern that might appear more desirable such as that produced by truncating the beam with an aperture, even though of constant phase initially, will have phase reversals in any focused image and even the unfocused beam after traversing a sufficient distance due to diffraction. There are no diffraction rings in the far-field pattern of the wave has an approximately Gaussian slope.

A Gaussian field (see Figure 13) is defined by a relative amplitude (A) (Download PDF for equation)

Figure 13. Gaussian Intensity Distribution of a TEMoo Mode Laser Beam. The definition of a beam size for a laser beam has to be defined at an intensity level. The standard definition is that diameter where the field strength falls to 0.368 of center mazimum and the intensity falls off to 0.135 of maximum. This is the laser beam's so called "spot size".

3.2 Liquid "In Situ" Spectrometers

To obtain a uniform region of illumination using a TEMoo Mode Laser, one is restricted to working with regions near beam center. Such a requirement exists if one is to devise a spectrometer around an "in situ" technique just as has been shown for volumetric aerosol instruments. Since particles are not restricted in where they may pass through the laser beam, they are just as likely to pass through regions of weak intensity as regions of high intensity. At PMS our approach to solving this problem has been to map particle trajectories using high resolution imaging systems.2 The instruments are called High Sensitivity Liquid In Situ (HSLIS) spectrometers.

The PMS HSLIS spectrometers operate on the principal that light scattered by a liquidborne particle resident in a laser beam is directly proportional to its size. Particles of the same size transiting the laser beam produce the same amplitude pulses and pulses of finite width (the exact width depending upon the flow rate being used). The pulses of radiant energy are sensed by photodiode detectors, amplified, and their maximum amplitude (peak detected) stored with a conventional pulse height analyzer.

In the.1µm HSLIS, a 5mW He-Ne TEMoo mode laser beam is focused to the sample region using astigmatic condensing optics as shown in Figure 14. Condensing optics generate an elliptical beam cross section of approximately 100 x 20µm; the waist (minimum beam dimension) in each axis is co-located at the center of the sample volume. The laser is rotated so that the polarization gives maximum signal generation along the orthogonal axis, telecentric with the paired primary objectives.

The optical system for the 0.05µm HSLIS spectrometer is 20X system shown in Figure 15. The laser used here is a 30mW polarized 780nm TEMoo mode solid state device. The laser output is collimated and again re-focused forming the desired beam cross section at the sample volume. The beam cross section at the sample volume is elliptical with a 10µm minimum waist in the ellipse minor axis. The beam waist is positioned at the center of the sample volume.

Wide angle collecting optics are required to produce high sensitivities and obtain a monotonic calibration curve as shown in Figure 16. The amount of scattered light collected is a function of the collecting optics numerical aperture (≈(N.A.)2). The light collecting objectives used in these spectrometers must have large N.A.'s. At small solid angles the response becomes ambiguous - particles at larger sizes. A shift in wavelength only proportionately shifts the peaks and valleys of the oscillating response function as shown in Figure 16 for wavelengths of 488, 633 and 780nm. Clearly, the larger collecting angles are essential to generate monotonic response characteristics regardless of the choice of wavelength.

Figure 15. Optical System for PMS 0.05µm HSLIS Spectrometer. This optical system uses a cast housing to maintain alignment.

The objectives of either HSLIS instrument collect light scattered from 50-130 degrees. Each objective senses the same particle and has identical solid angles of collection and thus receives identical amounts of signal. Each objective relays its energy to independent photo-detectors and pulse processing electronics. Both objectives have focusing elements which produce a 10:1 (0.1µm HSLIS) or 20:1 (0.05µm HSLIS) magnification at their image planes.

The "sizing" objective shown in Figures 14 or 15 focuses the beam onto a strip photodiode detector. The photodiode detector used with this optical train has a 10mm x 1mm active area. If the instrument has 10X (20X) magnification, an area of 1mm x 0.1mm (0.5mm x 0.05mm) is viewed inside the sample cell.

On the opposing side of the sample cell the "acceptance" objective produces the same magnification at its image plane. The photodiode used here is a linear photodiode array. The photodiode array has 20 active elements spaced on 0.5mm centers. (Less than 20 are actually used in order to make the effective array length slightly less than that of the sizing strip detector.) Since the sizing optical train using the 10:1 or 20:1 magnification is viewed over a full 10mm detector length, all events viewed by the array are simultaneously seen by the 10mm strip photodiode.

Figure 16a Figure 16b Figure 16c

Figure 16. Scattering Cross Sections for PSL in Water for Various Collecting Angles, at 488 (A), 633 (B), and 780 (C) nm. These data indicate the need for large collecting solid angles to produce a monotonic calibration and maximize sensitivity.

In addition to the photodiode array width being determined by the number of active elements and their spacing, a slit aperture is placed in front of the photodiode array to reduce its effective viewing height to approximately 0.25mm. This, in combination with 10X (20X) magnification, results in an effective sample volume field height of only 0.025mm (0.0125mm).

The above optical systems, when combined with an elliptical laser beam profile rotated at 45 degrees, allows one to define a region of suitably uniform intensity to make reproducible measurements independent of the Gaussian beam profile shown in Figure 13. The overall situation is depicted in Figure 17.

Figure 17A.

Figure 17B.

Figure 17. Ray Tracings of Particle Images in Viewing Volumes of HSLIS Spectrometers. In (A) particle positions in the depth-of-field shift image position and image size. In (B), particle position along beam axes generates image localized to individual array elements.

The operation of the detector geometry will now be explained. When a particle is in the preferred region, as defined in Figure 17, it is simultaneously viewed with full amplitude by each detector; however, when the particle is outside of this region, the energy received by the apertured array detector is substantially reduced because of the vertical shift in image position and the increase in image size with depth-of-field position. Pulse comparison circuitry is used to simultaneously compare the amplitudes (all summed elements) from the array is equal to or larger than that of the strip photodetector, the particle is adjudged to be in the preferred region of the uniform intensity and the pulse height measurement is accepted. When the amplitude produced by the array detector is less than that observed using the strip photodetector, the measurement is rejected. In addition, signals from the individual elements on the array photodetector are individually compared via parallel processing to a common threshold. The simultaneous presence of a signal determines particle validity. In this manner, high resolution measurements can be made and a region of uniform illumination is achieved "in situ". Because signals must be simultaneously observed by two independent detectors, random noise can also be more easily rejected.

In the HSLIS spectrometers, the overall flow cross section in the sample region is approximately 10mm2. The sample volume definition scheme in combination with the flow profile generates a sample volume rate that is approximately 1 mL/10 minutes for the 0.05µm sensor with total liquid flow rates of 300mL/minute and 100mL/minute, respectively.

With the above sample volume definition scheme, one can utilize as small a portion of the Gaussian beam as desired. The trade-off is a reduction in sample volume in favor of improved size resolution. The HSLIS spectrometers provide the greatest size resolution at the smallest sizes because this is where the calibration response curve is steepest affording the greatest resolution opportunities (see Figure 16). In face, the 0.05µm HSLIS instrument has nearly its entire operating range in the region of Rayleigh scattering where the scattering response is approximately proportional to the particle diameter to the sixth power. The amplitude spread for a monodispersed particle can be quite high and still provide good performance. For example, a ±50% uncertainty in intensity generates less than 0.01µm error at 0.1µm. A good fraction of the laser beam's power can thus be utilized. Figure 18 illustrates the resolution of a 0.1µm HSLIS spectrometer for 0.157µm and 0.203µm PSL.

If we elected to provide only a few size channels of data, the above "in situ" instrument would be classified as a liquid OPC. To be an OPC, the sample would still be necessarily restricted to regions of reasonably uniform illumination and thus most of the sophistication of the spectrometer would still necessarily have to be retained. However, there is another class of instrument that can be defined that is much simpler and is currently only used for liquid measurements. It is what we call a "monitor" instrument. The properties of monitors are rather unique and, because of their very recent introduction, they have not been well documented or characterized. Our remaining work here will be to provide a detailed characterization of monitors.

Chart

Chart

Figure 18. Histograms of 0.157 and 0.203µm PSL as Measured by a 0.100m HSLIS Spectrometer. The HSLIS can provide equivalent resolution to the HS-LAS by further reducing the width of the laser beam for sizing acceptance.

3.3 Liquid Monitor Instruments

A monitor instrument is an "in situ" device which, unlike a spectrometer, provides no means of restricting, or otherwise sorting out, where particles pass through the laser beam. The light source may be a solid state or gas laser. The optical collecting system may be simply ½ of that required by the spectrometers of Figures 13 and 14, although coincident detection by a detector pair can greatly improve random noise rejection. The length of beam viewed is a function of detector size and imaging system magnification while the full width of the beam is used. Parallel processing again increases sample volume without S/N penalties.

Since no attempt is made to determine where a particle passes through the laser beam, the same size particle has the opportunity to generate a range of scattering amplitudes ranging form a maximum when traversing the beam center to minimum values at the beam extremities. Needless to say, monitors are not high resolution devices! They do, however, fulfill a special niche when monitoring ultra-clean fluids in spite of their inherently poor intrinsic resolution. As we will see, they can provide relatively accurate cumulative size distribution information when restricted to regimes characterized by steeply exponential size distributions and when properly calibrated. They can also provide greater sensitivity with higher flow rates at lower cost and complexity. In short, they provide the greatest statistical data base for the least cost.

For most instruments, calibration is itself an ordinarily straightforward task. In the case of a spectrometer, proper size calibration is self-evident in the data presentation. Even in an OPC, size splits at threshold boundaries provide opportunities to employ a standard half count size calibration method as was shown in Figure 6. However, in a monitor it is difficult to define a proper size threshold by data inspection. The problem results from the lack of uniformity in intensity.

Consider the Gaussian intensity distribution in Figure 13 and the size between 0.050.2µm using the response function scattering cross sections in Figure 16c. We have intentionally re-plotted the Gaussian intensity distribution logarithmically in Figure 19 to illustrate the wide range of scattering cross sections.

Figure 19. Relative Amplitude of PSL Particles as Function of Transit Path Through Beam in a Monitor. This figure illustrates the variation of scattering signal amplitudes produced by PSL particles passing through a Gaussian intensity distributions at varying beam positions. The amplitudes have been normalized to unity at 0.2µm. Thresholds at 0.05, 0.10, and 0.20 µm reflect settings at 0.135 of peak amplitude.

The full range of amplitudes generated by each particle size outline Gaussian amplitude probabilities. To begin with, we will only consider a single size channel device. As with other instruments, we need to define a minimum threshold corresponding to the smallest particle size (0.05µm)3. Since it will need to be substantially lower than the maximum amplitude if we are to utilize a reasonable fraction of the beam, we will assume that it has been set to 13.5% of the maximum pulses amplitude produced by a 0.05µm particle. This particle size generates signals above the 0.05µm threshold for an effective portion of the beam width (A-A') corresponding to the laser beam spot size diameter as we had previously defined it (where I = 0.135 Io, i.e., 13.5%). If we increase the particle size to 0.1µm a new set of amplitudes is generated. This set of amplitudes includes much larger peaks, and the fraction of the peaks which exceed the 0.05µm threshold is much greater. Hence, the useful beam width is extended from A- A' to B-B'; a 75% increase. At 2.0µm, the effective beam width (C-C') is again wider by an additional 50%. The net result is that the effective viewing volume, which is proportional to the beam width utilized, and thus sample volumes are increasing functions of particle size, differing by a factor of 2.25 for particles with diameters of 0.05µm versus 0.2µm. The reason for the sample volume disparity for only a factor of 4 in size is the large range in signal strength these two sizes produce (>3,000 times).

The setting of the threshold with respect to particle size is somewhat arbitrary. One could select a threshold at 50% of the maximum amplitude instead of 13.5% and , in such case, one would obtain a slightly modified set of results with sample volumes differing by a factor of 3⅔ between sizes of 0.05µm and 0.20µm. However, the differences are not extreme of the smallest sizes use a reasonable beam fraction - especially considering the factor of 3,000 in signal range.

The complete curves for size versus sample cross section beam width representing all sizes between 0.05µm to 0.2µm for the two threshold situations discussed above are given in Figure 20. Sizes as small as 0.035µm and 0.045µm produce threshold amplitudes and represent the minimum detectable particle sizes in each case. The portion of the curves around 0.05µm resemble the counting efficiency curves of Figure 4.0 near threshold; however, there is no asymptotic leveling (e.g., at 100% counting efficiency) as size increases - it simply continues to increase. What then is the proper sample volume to assign to a monitor? Furthermore, considering its low resolution, what use can be made of more than a single size channel?

Figure 20. Effective Beam Width for Various PSL Sizes in a Monitor. Thresholds set at 13.5 and 50% of peak signal produced by 0.05µm PSL.

With regard to sample volume, a reasonable first approach would be to define the sample volume at mid-range in scattering signal instead of mid-range in size. Defining the sample volume for a 0.05µm particles would result in underestimates of particle concentrations, while use of 0.2µm particles leads to overestimates. Using a sample volume at larger sizes partially offsets the loss of sample volume at smaller sizes. This is shown in Figure 21. Indeed, the sample volume crosses at values of 40% and 60% at the 0.05µm lower size limit for the two threshold cases in a quite similar manner to the 50% counting efficiency values characteristic of OPCs at threshold.

The fact that over-sampling at larger sizes exists is not much of a problem when one considers that the number of larger particles decreases rapidly with increasing size. Consider a cumulative size distribution which has a slope of D-3. The total number of particles greater than 0.2µm is 1/8 that >0.1µm and 1/64 of that >0.05µm. The integration over all sizes generated by a monitor, with the viewing volume defined for 0.1µm, can be quite simple and is found to be ≈ 95% of the actual total. These results are shown in Figure 22. For a slope of D-4, the monitor counts 85% of the total. It oversamples slightly if the slope is D-2.

Figure 21. Variation of Sample Volume with Particle Size for a Monitor. The oversampled and undersampled regions tend to average out in populations with exponential size distributions.

Figure 22. Response of Monitor to Size Distributions of Varying Slopes. The monitor can still provide meaningful size distribution information if cumulative distributions are well-behaved.

Ideally, one would like to count 100% of the correct value, and if one has an ideal Gaussian beam, refinements can be made knowing approximate distribution shapes. However, several non-idealities are usually present. First we have only considered the beam's intensity distribution in two dimensions. If any other than an infinitesimal beam length is used, the beam's convergence and divergence will exaggerate its Gaussian non-uniformity. Also, larger secondary beams are present in the illuminated field coincidence with or adjacent to the primary beam - their effects becoming noticeable at sizes larger than 0.2µm. These non-ideal effects typically increase counts by 10-20% over theoretical values, and it invariably all shows up in the first size channel on multi-channel monitors. The general tendency of monitors is this to over count and not under count populations.

Addressing the issue of additional monitor size channels, one's first inference is that they may not provide much additional useful information, and a question arises as to how to set additional thresholds? Having additional thresholds is, however, the only way to know if particles of larger size are indeed present. Consider a three-channel device with additional size thresholds at 0.1µm and 0.2µm. Setting all thresholds in proper proportion to the nominal size response (peak amplitude proportional to threshold) is the best first choice as is shown in Figure 19. This also results in each size threshold being set the same with respect to peak amplitudes of particles of threshold size. One can easily determine the possible binning resulting from selected thresholds graphically using Figure 19 or by direct computation. These are shown in Table 1 for the hypothetical 3 channel instrument. In general, a monitor's binning will show ≈ 50% of a PSL test particle concentration in the bin bounded on its lower limit by that threshold size.

During the last year, PMS has introduced three new monitor instruments whose characteristics are summarized in Table 2. The M-50 is a 0.05µm threshold unit designed for D.I. water use only. The M-65 has chemical compatibility with a higher flow rate but a sensitivity reduced to 0.065µm. The M-100 is a 0.1µm threshold device with up to 5ml/minute sample flow rate and chemical compatibility. The M-65 and M-100 were initially He-Ne based but, as with all liquid instruments at PMS, these have transitioned to solid state light sources. All monitors are available with 1, 2, or 4 size channels.

Table 1.

Table 2.

These monitor differ from each other in terms of actual versus theoretical performance characteristics but are extremely repeatable from instrument to instrument within a given type over widely varying concentrations of monodispersed or ambient populations. In certain cases they can be used to test filters if monodispersed challenge is used. In general, however, a spectrometer is the strongly advised instrument choice for this critical level of activity. We generally recommend monitors to find problems and spectrometers to study them. The performance of three M-50 monitors compared to a 0.1µm HSLIS is shown in Figure 23. These cumulative distributions are shown to match extremely well in the overlap range between 0.1µm and 0.2µm. There are slightly higher counts in the M-50s at 0.1µm as compared to the spectrometer as predicted. These results are form PMS's D.I. water system and are similar to field results measured at customer semiconductor plant sites discussed below.

4.0 Field Performance of Spectrometers and Monitors in Deionized Water

The observation of sparse populations of very small particles in deionized water (D.I.) systems requires detection at the maximum possible sensitivity available. During 1990, PMS was privileged to observe the particle populations in effluent from 17 semiconductor plants (D.I.) water facilities. A brief description of the observing system utilized together with a short discussion of the practical application of the theory presented concerning sizing accuracy and sample volume determination for the spectrometers and monitors is presented. The results of the observations form the 17 locations are summarized and presented.

Figure 23. Comparative Cumulative Size Distributions for M-50 Monitors and a 0.1µm HSLIS in D.I. Water. Data taken from PMS D.I. water system.

4.1 Observing Systems Configuration

At most installations, four sensors were used to observe the particle population present in the deionized water. Two of these instruments were 0.1µm HSLIS spectrometers. M-50 and M-65 monitors were also used. The instruments were always connected in series to ensure that each instrument sampled the same steam of water. The flow rate was necessarily set to 100 ml/minute - the nominal flow rate of the monitors.4 On occasion, the connecting lines between instruments were found to be a source of particles to the next sensor in the series. If the line did not clean up, it was changed and allowed to clean up and the counts stabilized. Intermediate line contamination was assessed by having a pair of identical sensors placed first and last. When the system was initially turned on, it was always necessary to flush the connecting lines until a stable level of counts was achieved. Depending on the source condition, a few minutes up to a few hours of operation was required to reach a stable background. Data obtained from each sensor were recorded on a PMS Facility Monitoring System.

HSLIS Spectrometer- The spectrometer used had first threshold sensitivities of 0.10µm. The HSLIS spectrometer provided the needed sizing resolution as well as quantitative information on the concentration of particles in each size class over the size range of the instrument.

HSLIS-M50- The M-50 used had a first threshold sensitivity of 0.05µm. Most of the measurements used a single channel M-50, but a few has a 4-channel unit.

HSLIS-M65- The M-65 used had a first threshold sensitivity of 0.065µm and was always a 4-channel unit.

4.2 Sizing Accuracy and Sample Volume Determination

Verifying the sizing accuracy and sample volumes of the instruments used in the field was typically accomplished by passing a minodispersed distribution of particles of known modal size and known concentration through the sensor array. For size calibration it was only necessary to compare the proper split of PSL standards into their respective size channels. When using a single channel monitor instrument, pulse amplitude verification also was helpful.

Various ways can be employed to determine the concentration of particles in a liquid media. If the distribution of PSL is monodispersed and the modal size is known, gravimetric analysis can be used to determine the concentration. As of this writing, the only current independent source of the "so called" count standard particles at submicron size is the Japanese Synthetic Rubber Company (JSR) Clintex particles. The smallest available particles are 0.168µm in diameter. In their specifications, JSR indicates a countable lower size limit. For example, 0.506µm particles of a specified concentration have a countable lower limit of 0.4µm. When evaluating the counting accuracy of an instrument which has size resolution outside the specified JSR Clintex limits, the total concentration observed must take account of that fraction outside the countable lower limit.

The sample volume of the PMS HSLIS spectrometers was verified by passing a monodispersed distribution of Clintex count standard particles through the sensors. With a known flow rate, modal particle size and concentration, the sample volume or percentage of total flow sampled was calculated. The sample volume of the HSLIS spectrometers was checked at nearly all of the sites prior to observations of ambient conditions in D.I. water by using two sizes of Clintex count standard particles (0.168µm and 0.506µm in diameter) having concentrations of 105 particles per ml. These measurements indicated that the variation of sample volume across the size range of the instrument to be less than five percent.

The sample volume for the monitor instruments was empirically established using the HSLIS spectrometers as a "referee." As was previously shown, when large particles are observed by a monitor type instrument, a portion of the particles will be missed and the illuminated sample volume is which a minimum signal produced is inflated. A monodispersed distribution of PSL particles 0.117µm in diameter, instead of 0.100µm in diameter, was selected because of desire for the referee spectrometers to see the entire distribution. The 0.117µm particle was a compromise mid-range size for the two different monitor instruments. With the concentration established using the spectrometer as a referee, the sample volume of the monitors was established. These sample volume relationships are shown in Figure 24. The empirically determined monitor sample volumes were seldom significantly different form theoretical values. Large differences would suggest other problems. These calibration characterizations are summarized in Table 3.

Figure 24. Sample Volumes of M-50 and M-65 Monitors Compared to a 0.1µm *HSLIS Referee Instrument. The volumes have been normalized using a 0.117µm PSL size.*

4.3 Discussion of Observations

Data were collected from 17 D.I water facilities. A number of reports have described the results of measurements taken within a single facility or company (for example, see Yang and Tolliver, 1989); however, this is believed to be the first data set taken with the same instrumentation as many different sites. In general, the observed concentrations could be approximated with the power law expression:

N = ADB

Where N = number of particles per milliliter, D = particle diameter in microns, and A = constant.

The concentration of particles per milliliter was determined, for eleven cases, for the 0.05µm, 0.065µm, five cases, the <-65 was not available. In one case, the M-50 was unavailable. A best fit to these data, using a power law expression, was calculated. The coefficient of correlation of the best fit equation and 95% confidence intervals for each threshold were also calculated. These data are shown in Table 4.

The original observations from the 17 data sets are also shown graphically in Figure 25. The data suggest that clean D.I. water systems have ten or fewer particles per milliliter larger than the 0.05µm in diameter, about one per milliliter larger than 0.10µm and nearly always less than 0.05 per milliliter larger than 0.5µm.

Figure 25. Comparison of Populations of Particles from 17 Different D.I. Water Plants in U.S. Semiconductor Plants. The data represent a minimum of 12 hours accumulation at each site. The wide range of results is not easy to explain by causative factors known to PMS but represents an accurate relative data base.

As expected, the particle concentration observed in the various D.I. water systems showed large increases with decreasing particle size. The slope of the power law distribution ranges from about -2 to -3.5 (see Table 4). A typical power law distribution representative of a clean system is N = .0006 D-3 which yields about five particles per milliliter 0.05µm in diameter and larger and only five particles per liter 0.5µm in diameter and larger - a ratio of 1,000.

Table 4. Statistical Summary of D-I Water Plant Measurements at 17 U.S. Semiconductor Plants

5.0 Instrument Selection

It is well understood that when the number of particles observed increases, the variance decreases. Therefore, for an acceptable level of variation (sufficiently small confidence interval), large sample volumes (high sampling rates) are desired. Clearly then, instruments with limited sample volume must integrate over much longer periods of time to reduce the variation of the observation. One result, however, of the long time constants is the probability that short-term episodic events will be missed.

Evaluation of the appropriateness of the type of particle counting instrument to be used for a specific application involves: (1) A determination if, within the desired sample time, a sufficient sample volume is obtained to yield an acceptable level of statistical variation and still see episodic events. If, for example, a clean D.I. water system has typical distribution of particles such as N=.0006 D-3 and we observe this hypothetical system with an instrument which has a sampling rate of 0.30 ml/min, we would expect to see a particle rate = .00018 D-3 particles per minute. We would expect to see about 1.5 particles per minute (5 particles/ml) 0.05µm in diameter and larger and about 1.3 particles per hour (75 particles/liter) 0.2µm in diameter and larger. In this example, the distribution is representative of a very clean system, and the instrument would yield an acceptable level of variation at the 0.05µm diameter threshold and for the very rare event, such as a particle 0.2µm in diameter and larger, a much greater level of variation over the same sampling time. If, on the other hand we had a distribution N = 0.02 D-3.3 and the same 0.30 ml/min sample volume, we would expect a rate = 0.006 D-3.3 particles per minute or about 120 particles per minute (400 particles/ml) 0.05µm in diameter and larger and 1.2 particles per minute (4 particles/ml) 0.2µm in diameter and larger. The level of variation, over the same sample time, for both the 0.05µm and 0.2µm particle diameter thresholds would be quite different than the previous example. These examples illustrate the two ends of the spectrum of out field observations.

Under certain conditions, it may well not be possible to achieve the desired level of sensitivity and an acceptable level of sensitivity and an acceptable level of variation at a larger size with the same instrument. As a practical matter, as the detection sensitivity increases (to smaller particles), the sample volume decreases. One reason that the M-50 is not provided with an upper size threshold above 0.2µm particle diameter thresholds would be quite different than the previous example. These examples illustrate the two ends of the spectrum of our field observations.

Under certain conditions, it may well not be possible to achieve the desired level of sensitivity and an acceptable level of variation at a larger size with the same instrument. As a practical matter, as the detection sensitivity increases (to see smaller particles), the sample volume decreases. One reason that the M-50 is not provided with an upper size threshold above 0.2µm is that it simply has too low of a sample rate to quantify populations at larger sizes. It is 100X lower than the 0.2µm volumetric instrument discussed in Section 3.

As in all cases, when a certain phenomenon has been observed, one must carefully consider interpretation of the collected data. Quite often, one is tempted to assert an absolute value when discussing particle concentrations in D.I. water, which have been determined using an instrument that samples only a very small portion of the total flow. Clearly, the question of variability of the measurements is of extreme importance, and under such circumstances it is not appropriate to consider the observed values as absolute but rather to consider them valid within a certain range. For example, in the results form the 17 demonstrations, a mean concentration and the 95% confidence interval was calculated for various size thresholds. This is equivalent to being 95% confident that the particle concentration was within a calculated range. If the range is large, the measurement may not be of much value to the user. On the other hand, if the measurements may be of great value to the user for characterizing the process of system.

When considering the data collected with any instrument which samples a small portion of a system's output, it is more conservative to accept the observations as relative values and look for changes in the observations. Such changes can provide very useful information for characterizing the performance of a system. This approach also requires that the utilized instrumentation has sufficient response capabilities to be able to see episodic events which occur within a time period of concern. By establishing good sampling techniques and operating the instrumentation over an extended period of time, good baseline observations can be attained. Once solid baseline data are collected, it is much easier to detect anomalies form the established baseline. If the observing instrumentation continues to perform, these anomalies represent a change in the behavior of the observed system. The next important step in the management of any system is to track down the cause for the change in the system, make corrections, and see if the baseline is again achieved.

6.0 Conclusions

This work has attempted to compare the performance of different types of instruments used to measure microntaminants. We began with the more familiar aerosol instruments but, after providing this overview, placed greater emphasis on liquid instruments. Using out understanding of aerosol instruments, we built a framework from which to interpret the greater complexities of liquid instruments. Achieving maximum sensitivity in liquid media requires the use of "in situ" instruments. In situ instruments with high resolution potential demand more exacting optical and detector systems. For many applications a choice can be made to a simpler measurement having low resolution. This class of instruments we have defined as a monitor. Despite the inferior resolution, monitors provide a statistical advantage generating maximum sensitivity and highest sample rate with excellent reliability and repeatability. These simpler devices can still provide adequate resolution and cumulative distributions which typify clan process fluids. Their effective use in the characterization of D.I. water facilities during demonstration tests at 17 semiconductor plants confirms their application as a useful tool. In the future, it would not be surprising to see the use of monitors for clean room aerosol measurement at sub-tech micron sizes.

By Dr. Robert G. Knollenberg and Dr. Donald L. Veal Particle Measuring Systems, Inc.

Biographies

Dr. Robert G. Knollenberg has pioneered the development of a variety of particle-sizing techniques with applications to a host of scientific disciplines. He is best known for the design of optical-array-imaging, particle-size spectrometers for use on aircraft, "in situ" light scattering and extinction particle-size spectrometer for use in liquid contamination measurements, and is the inventor of laser resonant cavity techniques for sub-micron aerosol measurements. He has more than 60 publications relating to particle sizing and holds numerous patents on instrumentation related to particle-size measurement. He was the principal investigator on the Pioneer Venus Particle-Size Spectrometer Experiment, the only particle-sizing device having flown to another planet. After holding a faculty position at the University of Chicago, he founded Particle Measuring Systems, Inc., in 1972 and is currently Chairman of the Board of Research and Development Director.

Dr. Donald L. Veal joined Particle Measuring Systems, Inc., in 1987 after spending nearly thirty years as a professor and administrator in public higher education. He retired in the spring 1987 as President of the University of Wyoming. Dr. Veal's research interest was the development of instrumentation and measurements systems required to observe and understand cloud microphysical and dynamical processes in the atmosphere. He has more than 50 publications and papers related to atmospheric science and instrumentation.

References

Knollenberg, R.G., 1989: "The Measurement of Latex Particle Sizes Using Scattering Ratios in the Rayleigh Scattering Size Range," J.Aerosol Science, Volume 20, No.3, pp.331-345.

Hiraoka, M.; Zaitsu, Y.; Hoshikawa, H.; and Manabe, T. 1990: "A New Instrument for Measuring Particles in High Purity Water," presented at the Ultrapure Water Conference, Philadelphia, PA, April 1990.

Yang, M., and Tolliver, D., 1998:"Ultrapure Water Particle Monitoring for Advanced Semiconductor Manufacturing," J. Environmental Science, July/August 1989, pp.35-42.

Siegman, A.E.: "Lasers" (Mill Valley, California: University Science Books, 1986).

Verdeyen, J.T.: "Laser Electronics" (Englewood Cliffs, New Jersey: Prentice-Hall, 1981).