Flue gas desulfurization (FGD) wastewaters present a considerable analytical challenge for inductively coupled plasma–mass spectrometry (ICP-MS) because of high and variable concentrations of potentially interfering elements. For this reason, the United States (US) Environmental Protection Agency (EPA) commissioned the development of a new procedure, based on methods 200.8 and 1638 with requirements and quality control specific to the matrix. This paper discusses the development and performance of that procedure.

Sulfur emissions from coal combustion have been a focus of concern for some time because of their contribution to the formation of acid rain, accelerated soil acidification, and forest degradation. Air quality regulations established in the United States require SO

 scrubbing for most coal fired plants, with the resulting formation of flue gas desulfurization (FGD) wastewaters. Steam electric power plants are responsible for a significant amount of the toxic pollutants discharged to surface waters by point sources and FGD systems are the source of many of these pollutants. Based on the findings from the Environmental Protection Agency's (EPA) multiyear study of the steam electric power generating industry, a plan to revise the current effluent guidelines is in the works. The revised guidelines will apply to plants "primarily engaged in the generation of electricity for distribution and sale which results primarily from a process utilizing fossil-type fuel (coal, oil, or gas) or nuclear fuel in conjunction with a thermal cycle employing the steam water system as the thermodynamic medium." This includes most large-scale power plants in the United States. This decision is largely driven by the high level of toxic-weighted pollutant discharges from coal-fired power plants and the expectation that these discharges will increase significantly in the next few years as new air pollution controls are installed. The discharge of coal combustion wastewater from coal-fired power plants has caused a wide range of environmental effects to local aquatic life. Coal combustion wastewater also has been linked to human health concerns because of elevated pollutant concentrations in surface and ground water (1). Effluents from these plants, especially coal-fired plants, can contain several hundred to several thousand parts per million of the "matrix" elements: calcium, magnesium, manganese, sodium, boron, chloride, nitrate, and sulfate. 

Environmental matrices, which may contain low concentrations of toxic contaminants and high concentrations of interfering matrix elements, have historically presented challenges for sample analysis. Inductively coupled plasma–mass spectrometry (ICP-MS) was developed in the 1980s and has been used increasingly in the environmental field because of its multielemental isotopic capability, high sensitivity, and wide linear dynamic range. Provided that some care is taken during sample preparation and that appropriate calibration strategies are used to circumvent nonspectroscopic interferences, the technique is readily applicable to the analysis of a wide variety of environmental samples (natural waters, soils, rocks, sediments, vegetation, and so forth) (2). Two EPA ICP-MS methods are commonly used for regulatory compliance in environmental testing: Method 200.8 for water and wastewater and Method 6020A for solid waste. Both have recently been updated or are in the process of being updated. 

However, analysis of FGD wastewater matrix for low parts-per-billion levels of toxic trace metals (for example, As, Cd, Cr, Cu, Pb, Se, Tl, V, and Zn) by ICP-MS presents a significant challenge because of high levels of dissolved solids and potential interferences from molecular species. FGD wastewater can vary significantly from plant to plant depending on the type and capacity of the boiler and scrubber, the type of FGD process used, and the composition of the coal, limestone, and makeup water. As a result, FGD wastewaters represent one of the most challenging sets of samples for ICP-MS. That is, they are both very high in matrix elements (for example, calcium, magnesium, and chloride), known to cause interferences, and are highly variable. 

Three Approaches to Reducing Matrix Interferences

A new technique for the handling of high-matrix samples has been developed that inserts makeup gas between the spray chamber, where the sample is nebulized, and the torch, where plasma is generated. The end effect is an "aerosol dilution" that does not require the introduction of more water to the system and protects the mass spectrometer components from the high levels of matrix elements in the sample. This technique enables the ICP-MS system to handle 1% total dissolved solids (TDS) or higher and reduces oxides and oxide interferences to unprecedented levels. With less water to decompose, the plasma is hotter and, therefore, more robust. This increased robustness is a key to improving the ability of the ICP-MS system to run very high and variable matrix samples routinely and accurately. 

Using inert helium in the collision cell instead of reactive cell gases is accepted as the preferred mode for effective and reliable removal of multiple polyatomic interferences in variable or unknown samples (3). It is difficult to simultaneously remove all the interferences that occur in real-world applications using reactive gases. Using helium in the collision cell and a process called kinetic energy discrimination (KED) selectively attenuates all polyatomic interferences based on their size. KED exploits the fact that all polyatomic ions are larger than analyte ions of the same mass, so they collide with the cell gas more often as they pass through the cell, emerging with lower residual energy. These low-energy ions are excluded from the ion beam by a bias voltage at the cell exit, thus reducing interference from polyatomic ions. 

Discrete sampling allows the analysis of samples containing high levels of dissolved solids by analyzing a discrete sample volume rather than a continuous sample stream. A very small volume of sample is pulled into a sample loop and then rapidly injected into the nebulizer, followed by a rinse of the sample loop. This approach speeds the analysis and reduces the interferences and deposition of matrix material on the interface cones that can reduce system performance when testing for very low levels of trace metals. 

To address the difficult analytical challenge of measuring low levels of toxic trace metals in FGD wastewater containing very high levels of dissolved solids, the EPA commissioned the development of a simple, robust analytical method for the analysis of regulated metals in highly contaminated FGD wastewaters. This method was developed and validated at Test America Inc. using an Agilent (Santa Clara, California) 7700x ICP-MS system in helium or hydrogen collision mode and equipped with a high matrix introduction (HMI) aerosol dilution system and an integrated sample introduction system with discrete sampling (ISIS-DS) system to handle the high matrix levels in FGD wastewaters. Hydrogen mode was used only for Se analysis to attain sub-part-per-billion detection limits when necessary. The standard operating procedure (SOP) is available on the EPA's website at: 

http://water.epa.gov/scitech/wastetech/guide/upload/steam_draft_sop.pdf

Quality control for FGD wastewater analysis is typical of other EPA methods. Initial method validation includes the determination of method detection limits, linear ranges, and analysis of multiple, single-element interference check solutions to document the performance of the collision cell at removing specific polyatomic interferences under the conditions used. Daily quality control in a typical analytical sequence follows the workflow shown in Table I.

Table I: Workflow used in daily quality control as well as the validation of the EPA method SOP for analysis of toxic elements in flue gas desulfurization wastewater 

New to this method are the synthetic FGD matrix sample and fortified FGD matrix sample, which are analogous to the interference check standard-A (ICS-A) and ICS-AB solutions required by EPA Method 6020. However, the synthetic FGD samples are much higher in TDS than the ICS-A and AB solutions and contain those matrix elements that are commonly high in actual FGD samples. The composition of the synthetic FGD matrix sample is listed in Table II. 

Table II: Composition of the synthetic FGD matrix solution 

Initial demonstration of method performance requires that the project data quality objectives can be routinely met, including the determination of method detection limits (MDLs) and linear dynamic range, validation of interference management in synthetic matrix samples, and documentation of sufficient washout after analysis of high matrix samples. 

MDLs were determined via the preparation and analysis of seven synthetic FGD matrix samples spiked at three to five times the estimated MDL concentration of each of the target elements. Spiking the MDL replicates into the synthetic FGD matrix solution is a key requirement to ensure that reasonably achievable detection limits are generated. The MDL is calculated as Student's 

 for the 99th percentile (3.14) times the standard deviation of the MDL replicate results according to 40

 Part 136, Appendix B. The MDLs determined using this method are shown in Table III. 

Table III: Method detection limits (MDL) and instrument detection limits (IDL) for the analysis of flue gas desulfurization wastewater samples 

Interferences from the high and variable levels of TDS in the matrix present the biggest obstacle to the analysis of FGD wastewater samples. Both spectral (polyatomic ions) and nonspectral (suppression and washout) interferences must be dealt with. To ensure the absence of interferences, high concentration, single-element ICSs were analyzed for the presence of polyatomic interferences and the ability to rinse out in subsequent blanks (Table IV). Elements likely to be present in high concentrations and those known to cause common polyatomic interferences were chosen, including Ca, S, and Cl. These single-element ICSs are used initially to avoid preparation of a synthetic FGD matrix sample containing trace-level contaminants, which can be a problem with high concentration multielement mixes. The use of single-element check solutions simplifies the identification of the source of any contaminants detected. 

Table IV: Validation of removal of interferences from individual check solutions and subsequent washout (rinse) 

Washout (memory effects) were determined via the repeated analysis of blank samples immediately following the analysis of high-concentration single-element ICS (Table IV) and ICS spiked with target elements (Table V). 

Table V: QC summary including interference check standard (ICS), matrix spike recoveries (spiked ICS), carryover check (washout), and continuing calibration verification (CCV) and blank verification (CCB) 

In addition to the initial demonstration of performance, which must be completed at least once per year or after significant instrument maintenance or changes in procedure, regular quality control checks must be performed each day or with each new batch of samples. Daily quality control checks include (Table I):  

Initial calibration verification (ICV) solution (universal in all EPA standard operating procedures), which is from an independent source, contains target element concentrations near the midpoint of the calibration curve, and must recover within ±10% of the true value for all analyte elements.

Interference check blank (ICB) of the interference check solution used to prepare the ICV, to measure carryover.

Method blank results must be less than the reporting limit (typically 3–5 × MDL) for each analyte element.

Synthetic FGD matrix solution analysis must result in target analyte concentrations less than their reporting limits. 

Laboratory-fortified synthetic FGD solution is synthetic FGD matrix solution spiked at 40 ppb for all target elements except Zn (400 ppb) and Al (4000 ppb). All target elements must be recovered within 70–130% of the spiked concentration.

A reporting limit (RL) verification sample prepared at approximately 2 × the MRL must be recovered within ±50% of the actual value.

Continuing calibration verification (CCV) and continuing calibration blank (CCB) samples after every 10 actual samples. CCVs, which are done at the midpoint concentration of the calibration curve, must be recovered within 85–115%. CCBs must be less than the reporting limit.

Laboratory-fortified blank is spiked at 40 ppb for all trace elements, except for aluminum (4000 ppb). It must be recovered within ±3 standard deviations of past results, but must not exceed 85–115%.

Matrix spike or matrix spike duplicate samples are prepared for each batch by spiking a typical sample out of each group of 10 samples. The relative percent difference must not exceed 20% for any analyte and the recoveries must be in the 70–130% range.

Internal standard recoveries must be within 60–125% relative to the calibration blank for all samples. 

Figure 1: Recovery of calibration check samples (CCVs) over 89 analyses. The horizontal rules indicate the acceptable recovery limits.  

Before beginning the analysis of actual samples, an SOP validation sequence composed of 89 analyses was completed using the workflow shown in Table I, including calibrations, initial and continuing QC checks, and real samples. Table V summarizes the daily QC results including the synthetic FGD matrix ICS and spiked ICS, as well as carryover checks, CCV recoveries, and CCB results. CCV recoveries after every 10 real samples are shown in Figure 1. All required quality control checks were within acceptance limits, and more importantly, remained within limits for the duration of the 7-h, 89-sample sequence. Internal standard recoveries, shown in Figure 2, indicate minimal matrix suppression within even the highest TDS samples and less than 10% downward drift from the beginning to the end of the sequence.  

Figure 2: Internal standard recoveries over an 89-sample validation sequence. The horizontal rules indicate the acceptable recovery limits. 

 is the director of Technical Services at Test America in Arvada, Colorado.

 is the metals manager at Test America Denver in Arvada, Colorado. 

Articles

Atomic Perspectives

Wastewater from coal-fired power plants is linked to a wide range of environmental and human health concerns. Measuring low levels of toxic trace metals in these wastewaters is complicated, however, by high levels of interfering matrix elements. A new EPA method using ICP-MS addresses this analytical challenge.

Analysis of Flue gas Desulfurization Wasterwaters by ICP-MS

Francis William Aston's first mass spectrographs used photographic film to detect ions passed through the instrument, and photoplates continue to be used for ion detection in spark-source and glow discharge ionization instruments today. However, electron multipliers and photomultiplier detectors are installed in many modern beam instruments that are used for organic and bio-organic analysis, providing gains in excess of 10

 Specialized detection systems are used in instruments when, for example, precision abundance measurements or position-sensitive detection is required. In this column, we review the Faraday cup detector for mass spectrometry.  

An earlier column discussed the operation of electron multipliers used as detectors in mass spectrometry (MS) (1). You may remember that the electron multiplier was invented by P.T. Farnsworth (2), who also invented analog television, and in his later years, a device that claimed to provide controlled fusion. Analog television has been replaced by digital transmission, and controlled fusion remains firmly entrenched in the future. However, the electron multiplier became an extraordinarily useful device and is widely used in mass spectrometers. Despite this, the electron multiplier detection process is subject to a mass-discrimination effect (3). Additionally, because the detector produces a signal for both fast-moving ions and neutral particles, it also produces detector "noise" unrelated to the mass-selected ions. 

In this column, as well as the next installment, we describe two detector systems used in MS — the Faraday cup and position-sensitive array detectors. The Faraday cup (FC or FEC, for Faraday electrometer cup) is very simple in concept. Coupled with modern electronics, the Faraday cup is singularly useful in producing high precision measurements in isotope ratio mass spectrometers. Array detectors (which can include arrays of miniaturized Faraday cups) are used not only as detectors for dispersive-beam instruments, but also as developmental tools for characterizing the position and cross section of a beam of ions as it traverses an instrument. 

Figure 1: Photograph of a cylindrical FC detector. The ions enter the cup through the aperture on the right. An electron suppression plate surrounds the ion entrance aperture and keeps secondary electrons emitted from the impacted surface within the confines of the device. Adapted from the public domain figure provided in the wikipedia entry on "Faraday cup." 

The design of a Faraday cup is remarkably simple; it is indeed a cup. The metal cup (Figure 1 is a photograph and Figure 2 is a schematic) is placed within a vacuum system to intercept a beam of charged particles (electrons or ions). The charge on each particle (approximately 1.6 × 10

C) is passed to the metal on neutralization of the impacting ion. The cup is an element in a circuit; the current flow through the circuit can be very accurately measured and is directly proportional to the number of ions that have been intercepted by the Faraday cup. A current of 1 nA in the circuit corresponds to the arrival of several billion singly charged ions per second at the Faraday cup. Let's do the calculation, remembering that 1 A corresponds to a current of 1 C/s:  

 Because the detection is based solely on the charge, FC-based detectors exhibit no mass discrimination, which is an advantage in high precision measurements. Additionally, ions of higher charge states produce a correspondingly larger signal. Errors in the current measurement are reduced with the addition of an electron suppressor plate to the cup, as shown in Figure 1. The suppressor plate reduces losses because of backscattering of the incident ions and also reduces the probability of escape for secondary electrons that may be released on ion impact. Commercial FC detectors may have a weak magnetic field to prevent secondary electrons from leaving the Faraday cup (4), and they may operate with a slight positive bias on the impacted surface to reduce secondary electron emission. As expected, the limit of detection for a Faraday cup depends on the sensitivity of the electrometer in the circuit that it is connected to. The current passes through a circuit resistor, and the generated difference in voltage is measured (

). Even relatively simple circuits and low cost amplifiers can provide a 10 mV signal for a picoampere of input current. Revisiting equation 1 above reveals that measuring microvolts corresponds to a few thousand ions. Therefore, the FC detector can be used for high sensitivity analyses. The ability to avoid a scanning mass analysis is also advantageous, as we will describe shortly. The noise associated with the electronics necessary for "amplification" of the weak signal (usually involving a high-ohm resistor) is compensated for by a measurement time that can last several hundred seconds. 

Figure 2: Schematic diagram of a simple FC detector. 

An early concern in the analysis of higher mass biomolecules was detection efficiency. Electron multipliers, for example, exhibit a threshold velocity for emission of secondary ions. Ions must impact the first emissive surface with sufficient velocity for the electron cascade to be initiated. Higher mass ions move more slowly than lower mass ions; remember that the acceleration potential of the source represents the potential energy drop that is transformed into kinetic energy of the ions. Various designs incorporating higher accelerating potentials near the detector were explored. The performance of the Faraday cup in these high-mass applications, using a matrix-assisted laser desorption–ionization (MALDI) source coupled to a time-of-flight (TOF) mass spectrometer, was also explored (5). The detector response was found to be about 50 ns, sufficient for high-mass applications, and ions with masses as high as 300,000 Da were observed. 

FC applications in MS are a small subset of the broader suite of applications in charged-particle detection. Larger Faraday cups for scaled up beam instruments, such as particle accelerators (6), have larger interception surfaces and may need separate cooling systems because the beam current is so high. Developments in new designs (7,8) are continually reported in the research literature, reflecting specialized applications, miniaturization, improved ion modeling simulations, and improved electronics and measurement processes (9). 

A single FC detector can be used in a scanning mass spectrometer, or in a TOF instrument (5). Of course, multiple detectors can be used simultaneously or configured into an array. The advantages of such systems have long been clear (10), even as technology to create them lags our desire to deploy them. Consider one singular advantage of a multiple-detector system. In a scanning beam mass spectrometer, the mass analyzer is scanned to mass-select the ions that are collected at the single-channel detector, which could be a single Faraday cup. The detected signal is linked in software to the scan function to create a time or intensity file that is eventually plotted as the mass spectrum. In a dispersive-beam instrument, no scanning of the mass analyzer is required, and the mass spectrum (or selected ions within the mass spectrum) is assembled from the combined signals of multiple collectors or an array detector. Aston's mass spectrographs were dispersive instruments, with the ion signals assuming the shape of a parabola on the photographic film. Dispersive instruments with multiple FC collectors are commonly used in modern isotope ratio measurements. The mass analyzer is not scanned, but instead disperses the ion beams from an ionization source to a suite of detectors, each located at the proper position to collect the ions of only one mass (a simple schematic is shown in Figure 3). The geometry of such a device is easily calculated, given the strengths of the mass analyzing field, the physical size of the instrument, the intercept angle of the FC detectors, and the resolving power needed. 

Figure 3: Schematic diagram of a dispersive-beam instrument that sends ions of different masses to separate FC detectors. In isotope ratio instruments, the mass separation between adjacent Faraday cups is usually 1 or 2 Da, and adjacent Faraday cups correspond to the various isotopes of a single atomic elemental ion. 

Wieser and Schwieters (11) have reviewed the development of multiple collector instruments for isotope ratio measurements, and the applications in geological and cosmological sample analysis are simply extraordinary. Such applications require the utmost care in sample collection, sample preparation, and measurement. Proper use of certified standards and calibration procedures is mandatory and attests to the accuracy and precision of results (12–14). As an example of the robustness of an FC detection system, a 30-year old mass spectrometer was upgraded (15) to better than original performance specifications using the original seven FC collectors (one fixed and six movable collectors). Because of improved electronics, the sample loading requirement was reduced. The instrument is used for uranium and plutonium isotope ratio analysis, and the two instruments in the facility analyze 8000 samples annually. The authors note that the ion stack at the front end of the instrument requires regular care and replacement, much more so than the FC detection system. 

Isotope ratio measurements, which are founded in authoritative metrology and the statistical underpinnings for accurate and precise measurement, have been described (16) and clearly linked through calibration to standards. This field of analysis (no pun intended) represents an exact science that should serve as a quality example for all of analytical MS. The importance of standard materials for the isotope-ratio community has been emphasized repeatedly. For example, Santamaria-Hernandez and Hearn (17) described inductively coupled plasma–mass spectrometry (ICP-MS) measurements using FC detectors, comparing sulfur isotope ratios in accepted and suggested standards. It is a testament to modern instrumentation that the primary contribution to the total uncertainty in the reported measurements for sulfur in samples, such as methionine, is the stated uncertainty in the standards themselves.  

A commercial manufacturer's application note for an instrument used for argon isotope ratio measurement describes strategies for high-precision measurements of argon in air samples using different available detection systems (18). The instrument described consisted of five separate Faraday cups and a compact discrete dynode electron multiplier used in ion counting mode. The electronics for the Faraday cups used either a 10

-Ω amplifier. The multiplier was used in an ion-counting mode, using peak jumping from one ion mass to another. For larger samples (that is, 4.2 × 10

-Ω amplifier in the FC system was used to produce a precision of 0.2%; for smaller samples, the 10

-Ω amplifier produced nearly equivalent results. The ion counting system can be used for still smaller samples, but the data must be treated to correct for dead time of the counting data and the inaccuracies associated with the jump. The FC data showed a linear and robust response over the dynamic range necessary for common argon isotope-ratio measurements. Many isotope-ratio instruments are equipped with both FC and electron multiplier detectors. The gain achieved with the latter (10

 or higher) facilitates measurements with very low signal levels. But the simplicity and stability of FC detectors provides excellent precision in the simultaneous measurements of multiple argon isotopes.  

How small can a Faraday cup be? The relevant metric is the size and shape (cross-section) of the ion beam that the cup is intended to collect, and the spacing between ion beams of different masses at the desired mass resolving power. Alternatively, an FC array can be used as a position-sensitive detector. Array and position-sensitive detectors will be described in greater detail in the next installment of this column. Microfabricated Faraday cups must also confront the issue of secondary electron emission from impact, and usually do so through geometry rather than the fitting of a discrete electron suppressor or use of an auxiliary weak magnetic field. Bowers and colleagues (19) describe a dense one-dimensional array of miniature Faraday cups in which 64 cups with widths of 15–45 µm are separated by a spacing of 5 µm. The cups can be up to eight times as deep as they are wide, minimizing cross talk in the separate detection channels. Similar microfabricated FC arrays have been described by others (20,21).  

Photoplate detectors used for MS require the use of scanners to retrieve the integrated and recorded signal. Similarly, array detectors, including FC arrays, require an electronic readout process. Scheidemann and colleagues (22) describe a system in which only a low overhead time is used for the read-and-reset cycle, and a 99.7% ion collection efficiency is achieved for a 64-cup array. A five-decade dynamic range is also demonstrated. The array is used as a position-sensitive detector in a dispersive compact mass spectrometer with a 200-Da mass span.  

The Faraday cup is but one of a host of different detectors used for MS (23), and FC arrays are only one type of array detector (24). The Faraday cup has been associated with MS since the first instruments were assembled and will continue to be used as a specialized detector. The robustness of an FC detector system is a consequence of its inherent simplicity. As discussed in this column, the electronics can be updated and the physical act of intercepting a charged particle beam needs no rejuvenation. As noted (and shown with a reproduced figure) in Grayson's recent article about J.B. Fenn (25), the early electrospray ionization device described by Dole (26) used an FC detector. It was the use of a retarding variable potential plate placed directly in front of the entrance to the Faraday cup that allowed the high-mass and highly charged ions to be detected, a simple and unambiguous measurement with profound implications. Simple, basic measurements are almost always at the forefront of exploration. The plasma spectrometer on the Voyager instruments (now in an interstellar place) is equipped with a FC detector. Although the gold-plated record encoded with information sent out with Voyager is well-known to the public, a recent article attests to the fact that the FC detector carries the inscribed names of those who worked on the plasma spectrometer (27). Extraterrestrials may not be able to decipher the names, but with a little study should be able to discern the purpose of the FC detector. Those of us closer to home may be interested in reading about The Faraday Cup Award (28). 

 has not inscribed his name on any hardware currently traversing interstellar space. Nevertheless, he has often been described as "spacey," given his oft-repeated prediction that preparative MS in near-Earth orbit (especially for the production of isotope-enriched or isotope-depleted materials) will be a major growth industry by 2100. When this prediction turns out to be true, he asks only for a special mention at the 150th annual ASMS meeting, and a toast with the Faraday Cup. This column is the sole responsibility of the author, who can be reached at 

 (2) P.T. Farnsworth, U.S. Patent No. 1,969,399, "Electron Multiplier," filed August 7, 1934. 

 (3) M.L. Alexandrov, L.N. Gall, N.V. Krasnov, L.R. Lokshin, and A.V. Chuprikov, 

 (4) M.G. Inghram and R.J. Hayden, "A Handbook on Mass Spectroscopy," in 

 (National Academy of Sciences, Washington, D.C., 1954), pp. 39-41. 

 (5) D.C. Imrie, J.M. Pentney, and J.S. Cottrell, 

 (6) L.M. Welsh, K.H. Berkner, S.N. Kaplan, and R.V. Pyle, 

 (8) J.D. Thomas, G.S. Hodges, D.G. Seely, N.A. Moroz, and T.J. Kvale, 

 (9) C.E. Sosolik, A.C. Lavery, E.B. Dahl, and B.H. Cooper, 

(12)  I.T. Platzner, K. Habfast, A.J. Walder, and A. Goetz, "Modern Isotope Ratio Mass Spectrometry," in 

vol. 145 (John Wiley, New York, New York, 1997). 

The Encyclopedia of Mass Spectrometry, Volume 5: Elemental and Isotope Ratio Mass Spectrometry 

, D. Beauchemin and D.E. Matthews, Eds. (Elsevier, New York, New York, 2010). 

(15)  J.V. Cordaro, S.R. Johnson, M.K. Holland, and V.D. Jones, "Extending the Useful Life of Older Mass Spectrometers," SRNS-STI-2010-00340, available at 

http://sti.srs.gov/fulltext/SRNS-STI-2010-00340.pdf

(16)  H. Kipphardt, P. De Bievre, and P.D.P. Taylor, 

(17)  R. Santamaria-Fernandez and R. Hearn, 

(18)  M. Krummen, D.G. Burgess, E. Wapelhorst, D. Hamilton, and J.B. Schwieters, Thermo Fisher Scientific, Application Note 30193. 

(19)  C.A. Bower, K.H. Gilchrist, M.R. Lueck, and B.R. Stoner, 

MS Forum

The Faraday cup has been associated with mass spectrometry since the first instruments were assembled and continues to be used today. Here's how it works.

Detecting Ions in Mass Spectrometers with the Faraday Cup

Operating temperatures are known to directly affect the performance and reliability of a range of modern microdevices, including light-emitting diodes (LEDs), microelectromechanical systems (MEMS), and high-power electronics. As a consequence, accurate temperature measurements have become imperative in the development of these technologies. Such measurements are complicated, however, by the complex multimaterial stacks typically used and by the fact that traditional probes (thermocouples) have a size and thermal mass on the order of the device being interrogated. In response to these difficulties, Raman thermometry is frequently implemented, because it is a non-contact, material-specific measurement largely benign to device operation that is capable of comparatively small spatial (~500 nm) and thermal (~1 °C) resolutions. Practically, these measurements can be made using a variety of spectral features including the position, linewidth, and intensity of the Raman signal associated with specific optical phonon modes. Each of these spectral characteristics offers particular advantages, depending on the type of device and its operational conditions. Here, the practical implementation of Raman thermometry using each of these spectral characteristics is reviewed to highlight the assumptions implicit with their use and to compare their effectiveness in measuring temperature. 

The performance and reliability of modern microsystems is increasingly defined not by the manner in which materials are capable of transporting charge but by the efficiency with which thermal energy can traverse the system. For example, high-performance electronics in applications ranging from data centers to electric vehicles (1,2), solid-state lighting (3,4), and even many microelectromechanical systems (MEMS) devices (5,6) are known to be adversely compromised by thermal effects. As a result, experimental techniques must be available that can accurately assess the thermal environment. This task remains difficult due to the small spatial scales involved (nanometers to micrometers) coupled with thermal gradients that are often large. Raman spectroscopy is particularly well suited to meet these challenges because it can perform noncontact thermal measurements with a resolution of ~1 °C in a manner that is unobtrusive to device operation. The technique also provides a spatial resolution that is on par with the structures themselves (~1  µm), coupled with the ability to sample in a material-specific fashion. For these reasons, Raman thermometry has been used for more than 20 years in a variety of material systems such as silicon (7), gallium arsenide (8), gallium nitride (9), and graphene (10) for applications ranging from laser facets (11) to radio-frequency electronics (12–15). Furthermore, because of its ability to measure temperature, Raman can be implemented not only to probe device self-heating but also to quantify the thermal conductivities of the materials making up the device (16–18). As a consequence, Raman thermometry is now used frequently in the thermal investigation of microsystems.  

The Raman response can be implemented in several different manners to deduce temperature. Each approach, in turn, is derived from a different physical mechanism and offers particular advantages and disadvantages. As such, there is not one proper way to perform thermometry using a Raman spectrum. Instead, the experimentalist must consider  

The physical mechanisms that give rise to the thermally dependent characteristic of the spectrum

The intrinsic ease with which this spectral characteristic can be acquired

The manners that these characteristics can be compromised by other effects that mask, or bias, the resulting temperature measurement. 

This review attempts to describe and summarize these considerations to provide the intellectual foothold necessary to optimize the usefulness of Raman thermometry.  

Figure 1: (a) A schematic of a typical Raman spectrum showing the anti-Stokes, Rayleigh, and Stokes signal. Raman thermometry may be performed by using any one of three different aspects of this spectrum. Specifically, either a shift in the peak position (Ï), a broadening of the linewidth (or full width at half maximum, FWHM, Î), or an alteration in the intensity ratio of the anti-Stokes-to-Stokes signal (I AS/I S) may be used, given their temperature dependence, as is shown schematically in (b) for the Stokes response. 

In Raman thermometry, temperature is deduced through the analysis of the inelastic energy transfer between the incident laser source (photons) and the quantized lattice vibrations (phonons). As the incident photons are invariant with the device temperature, all deductions are based on changes in the vibrational character of the analyte material. Thus, any aspect of the phonons that changes with temperature can be used to probe the thermal state of the system. Phenomenologically, these changes are reflected in the peak position, linewidth (full width at half maximum, or FWHM), and intensity of a Raman signal, all of which vary with temperature, as is shown in Figure 1. Practically, temperature is then measured in one of three manners: by quantifying the degree to which the peak position shifts (Δω), by assessing the broadening of the linewidth (ΔΓ), or by following the intensity ratio of the anti-Stokes signal to the Stokes signal (

). For ideal temperature measurements, each of these effects would be solely dependent on the thermal environment. In actuality, however, the physical mechanisms giving rise to these signals can be altered by a cause that is not thermally derived. These nonthermal causes can introduce errors in the deduction of temperature. To minimize errors, it is imperative that the underlying mechanisms linking the actual Raman signal to changes of the phonons in the material be understood so that the most appropriate signal characteristic may be leveraged. This choice is complicated, however, by the fact that each of these spectral features also varies in terms of its practical ease of use. Specifically, the peak position, linewidth, and ratio each differ in the readiness by which an accurate temperature measurement may be obtained, due to the different errors that are implicit in each. Consequently, the applicability of using any of these signals, each aptly termed 

, can be vastly different based on the nature of the device being analyzed and the experimental equipment used.  

In response, the following discussion examines each of these three techniques commonly used to perform Raman thermometry. First, the physical cause for each of the signals is described, to highlight the origin of its temperature dependence and also explain why other effects might bias the results. Subsequently, the uncertainties typical with each of these techniques are compared, to judge the methods' practical usefulness. Finally, a case study is presented in which the differing capabilities of the techniques are examined to highlight the considerations that must be made when performing Raman thermometry.  

In most micro-Raman systems used for thermometry, signal acquisition takes place by focusing laser light onto a sample using a high numerical aperture and then collecting the Raman-scattered light in the backscattering geometry. With this arrangement, light interacts most favorably with zone-center (Γ-point) phonons that give rise to the Raman signal. Therefore, phenomena that alter these zone-center optical phonons will alter the Raman spectrum. The following section outlines the physical origin of these phenomena for each of the three spectral features used to measure temperature. Understanding these phenomena, in turn, elucidates the mechanisms that give rise to the temperature dependence as well as the factors that could potentially obscure this capability.  

The original means by which Raman was commonly used to measure temperature, the anti-Stokes-to-Stokes ratio, arises because the intensity of a signal is proportional to the number of species present to be sampled. Specifically, at a given frequency, ω, the intensity of the Raman signal will scale with the number of phonons present to take part in the process (19). For example, in the anti-Stokes process, the number of phonons that may be absorbed at a given temperature can be calculated from the Bose-Einstein distribution function described by the equation 

where ħ is Planck's constant divided by 

 is the Boltzmann constant. Similarly, for the Stokes process, the total number of phonons will be this equilibrium distribution, 

, plus the emitted phonon leading to a cumulative total of 

 + 1. The ratio of the two signals is then proportional to  

 is an experimentally determined calibration constant. An example of such a trend found from a gallium nitride temperature calibration is shown in Figure 2a.

Figure 2: The Raman response and the associated calibration curve as a function of temperature for the three different spectral features used for thermometry, namely the (a) intensity ratio, (b) peak position, and (c) linewidth. 

Temperature measurements based on the ratio of these intensities are then dependent only on physical mechanisms that change the distribution of phonons from their equilibrium, Bose-Einstein, distribution. During the operation of most microdevices, changes in the equilibrium population of phonons are not common and thus the intensity ratio is implicitly robust from the perspective of competing physical mechanisms driving the signal. Under certain conditions, however, temperature measurements can be compromised as a result of the movement of certain phonon modes away from an equilibrium phonon distribution. For example, in gallium nitride transistors, nonequilibrium distributions of the longitudinal optical (LO) mode have a high likelihood of being present as a result of the so called 

 effect that results from preferential scattering of the electrons with this vibrational mode. These effects, in turn, have been shown to directly influence the temperature measurements inferred from the intensity ratio (20). 

The peak position of the Raman signal is derived from the energy of the zone-center optical phonons that are probed during the Raman experiment. Understanding the temperature dependence of this signal thus requires an understanding of how the frequency, and hence the energy, of these modes change. To this end, consider the fact that, to first order, phonons can be described as a spring-oscillator system. Thus, in much the same way that vibrations change depending on the stiffness of the spring, so too will the phonons change their energy, and hence their Raman peak position, based on alterations in the interatomic forces in the crystal lattice. As the lattice is heated or cooled, the equilibrium positions of the atoms are displaced, resulting in an overall volumetric expansion or contraction of the lattice and a change in interatomic forces as a result of the anharmonicity of the bonds (21). These changes in the interatomic forces modify the phonon vibrational frequencies that are reflected in the resulting variance of the Raman peak position. 

In addition to this volumetric contribution, phonon–phonon interactions themselves augment the frequency shift, because the mere presence of a phonon will also alter the equilibrium spacing of the atoms in the lattice and, by extension, the interatomic forces. As a result of this change in the interatomic forces, the frequency of oscillation of both this and other phonons will be modified, thus affecting the frequency of the Stokes and anti-Stokes scattering. Because the phonons responsible for this modification are governed by the Bose-Einstein distribution of thermal occupation, the resulting Raman shift varies as a result of this contribution in a manner synonymous with the intensity ratio described previously. Together, these effects induce a very convenient linear change as a function of temperature over the typical operating range of most microdevices (Figure 2b).  

It is of particular relevance that the volumetric contribution to the peak position is sensitive to any effect that changes the distances between the atoms. As such, the peak position is particularly sensitive to strain and frequently is used to this end (22–24). Because of this fact, temperature measurements based on this aspect of the signal can easily become significantly biased as a result of the effects of the thermoelastically induced strain that commonly evolves with device self heating. For example, in the case of silicon, a biaxial stress on the order of 150 MPa will lead to a bias error of 24 °C in the resulting temperature measurement (23). Therefore, examinations of the temperature using the peak position must consider the level of strain or stress in quantifying the applicability of the technique.  

The linewidth of a Raman spectrum evolves as a result of the finite lifetime of the zone-center phonons that are being investigated. Specifically, this signal originates as a consequence of the Heisenberg uncertainty principle, which stipulates that the energy of the phonon can be measured only to within a certain specificity if the mode being investigated is itself available for only a finite amount of time. This principle is described mathematically according to the following energy–time uncertainty relationship: 

 is the scattering time of the phonon. More simply, mechanisms that affect the lifetime of the zone-center phonons will, in turn, change the linewidth of the Raman line. The scattering time of this phonon mode is dependent on a variety of factors including microstructural defects, material boundaries, and, most importantly, other phonons. It is this dominant phonon–phonon scattering at typical operational temperatures that gives rise to the temperature dependence of the linewidth, because the number of phonons available for scattering is dependent on the temperature-dependent Bose-Einstein population distribution. As the temperature increases, so too does the number of phonons present, thereby increasing the likelihood of a scattering event. This increased likelihood reduces the phonon lifetime, thus increasing the linewidth and allowing the linewidth to be used as a probe of temperature. 

Incorporating this feature to measure temperature implicitly assumes that the only factor altering the lifetime of the zone-center phonons being probed is their interactions with other phonons. Such an assumption is most often a very safe one. Although other factors (such as grains and defects) do affect the scattering of the phonon, their contribution is typically temperature independent (25,26). In some situations, however, secondary factors beyond temperature have been shown to be capable of altering the interatomic potential to a degree sufficient to alter the nature of the phonon scattering and hence the resulting linewidth. This has, for example, been reported in gallium nitride transistors where piezoelectric effects in the polar direction of the crystal are believed to induce changes in the linewidth (27). In spite of this fact, normal operation of most microdevices does not induce these phenomena, allowing the linewidth to be governed chiefly by temperature dependent phonon–phonon interactions. 

Practical Considerations and Uncertainty 

In the preceding discussion, the linewidth and intensity ratio are shown to have physical origins that are less likely to be influenced by nonthermal effects in comparison to the peak position (Table I). Intuitively, these signals would seem to be optimal for temperature measurements. In reality, however, each technique varies in its ability to produce precise, accurate, and repeatable measurements. It is therefore necessary to quantify the pragmatic usefulness of implementing each technique. To this end, the typical methodology used to perform a Raman thermometry experiment is described. With this description, the uncertainty involved in using each technique is quantified and additional nonquantifiable factors also are considered.

Table I: Physical origin of temperature dependence and competing mechanisms in Raman thermometry 

Raman thermometry begins with a calibration of each signal to a series of known temperatures across the range of interest. This calibration is done by uniformly heating a sample and following the degree to which the peak shifts, the linewidth broadens, or the intensity ratio decreases as a result of a known temperature rise. These changes in the spectrum are quantified by fitting the spectrum using a known functional form — typically a Lorentzian or Gaussian form, or a convolution of both, termed a 

 function — thereby producing the calibration curves shown in Figure 2. After obtaining these calibrations, the thermometry measurements are performed by using these calibration curves to correlate changes in the spectrum induced by operation of the device to a temperature. The specificity and ease with which these changes can be directly correlated to the temperature define the usefulness of the different techniques.  

Uncertainty in the predicted values of temperature stems from a variety of factors that influence the calculation of the position, intensity, and shape of the Raman peak. These sources of error arise from any of several sources, including changes in the spectrometer or laser during testing; variation in the sampled volume as a result of drift in the microscope stage; spatial nonuniformities in the device response; alterations in the materials of the device itself; and differences in the fitted spectra as a result of  the inherent difficulty in fitting a continuous function to a discrete pixelated data set. Quantification of the total error must take into account each of these effects through a vector summing of the individual contributions (28). Mathematically, this is accomplished through analysis of the calibration equations predicting the temperature from a change in the analyzed spectral component. With respect to the peak position–based measurement, the curve shown in Figure 2b is first solved for temperature using the equation 

 is the slope of the line in Figure 2b and 

 are the deduced operational and ambient temperatures, respectively, acquired from the shift from the reference peak position ω

. Individual uncertainties may then be linked to the final uncertainty via a vector summing procedure through analysis of equation 4 as follows: 

 are the uncertainty metrics, typically some level of confidence interval of the temperature, calibration constants, and peak position at the tested and reference conditions, respectively. In the above relation, it is assumed that the reference temperature, 

, remains constant throughout testing and that additional sources of error not explicitly mentioned could be present as well. Analogous relations exist for the calibration curve of the intensity ratio via equation 2 and the empirical second-order polynomial relationship describing the linewidth broadening seen in Figure 2c. Uncertainties introduced by other external sources, such as any laser heating effects, positional drift, and spectral drift of the spectrometer, also must be considered in equation 5 to obtain an accurate prediction of the total measurement uncertainty. Laser heating of the sample is a source of error that is becoming increasingly important as devices become smaller. The thermal conductivity of nanostructures, for example, can be significantly reduced relative to that of the bulk. In this case, heating induced by the laser could significantly alter the resulting temperature measurements. This effect must, therefore, be considered (29). Ultimately, any Raman thermometry experiment must be crafted carefully to mitigate these extraneous effects whenever possible.

Figure 3: Uncertainty in Raman temperature measurements varies depending on the methodology implemented. Shown here are uncertainties stemming from temperature measurements of a gallium nitride high electron mobility transistor. Quantitatively, these values depend on a host of experimental factors and can, therefore, vary significantly from magnitudes shown here. Qualitatively, however, the general trend of peak position exhibiting the smallest uncertainty and the intensity ratio having the largest is almost always observed. 

Using these relationships, the magnitudes of uncertainty associated with a typical temperature measurement of a  gallium nitride high electron mobility transistor are shown in Figure 3. Although the quantitative values are highly dependent on the methodology used, comparisons of the uncertainty between the differing spectral features provide insight into the usability of each. For example, the intensity ratio displays significantly more uncertainty as a result of an extensive nature that requires an absolute, quantitative, measurement. Therefore, any change in the laser, device surface, collection optics, or detectors will introduce uncertainty into the measurement. The extensive nature of the method is also sensitive to secondary effects such as photoluminescence and fluorescence that can become convoluted with the Raman spectrum and make it difficult to distinctly quantify the intensity of the signals. Additionally, the ratio method requires the use of an exponentially based calibration that itself provides difficulty. In fact, it has been our experience that the total uncertainty in the intensity measurement stems from a comparable degree of scatter between the measurement and calibration. In total, these factors highly complicate the measurement and make the implementation of this method difficult. 

In contrast to the extensive nature of the intensity ratio method, the linewidth and peak position are instead intensive in nature. Rather than measuring an absolute quantity, both of these methods measure a relative change from a reference condition. As such, they require a lesser degree of experimental control because even changes in the spectrometer can be mitigated through periodic measurements of the reference condition (such as the unpowered device) or the use of a spectrally static reference source such as a spectral lamp (30). The linewidth shows a greater level of scatter relative to the peak position as a result of its nonlinear calibration and the greater complexity needed to fit the width, or shape, of a function in comparison to its maximum. In addition, the linewidth, like the intensity ratio, is sensitive to other non-Raman-specific features in the spectrum, whereas the peak position is more robust to these effects. Finally, it is worth noting that whereas the peak position has a calibration that is consistent across multiple spectrometers and laboratories, neither the linewidth or intensity ratio can be extended to systems other than the system used to obtain the calibration. For the linewidth, this limitation arises from spectrometer-specific broadening, whereas the intensity ratio is implicitly variable because of its extensive nature. As a result of these complications and its lower uncertainty, the peak position usually is the most straightforward portion of the spectrum to use in practice.  

This fact, unfortunately, is a complication. Peak position, the spectral feature that is most easily implemented, is also particularly sensitive to nonthermal biasing of the signal. It is thus necessary to weigh the relative contribution of the nonthermal biasing effect with the additional uncertainty that accompanies the use of either the linewidth or intensity ratio. To make this comparison, the linear response of the peak position to both temperature (Figure 2b) and stress (23) can be related to quantify the stress-induced bias. For example, the rate of change of the peak position to temperature in gallium nitride is –0.014 cm

/°C. Similarly, the peak position also shows a linear dependence to a biaxial stress at a rate of –2.91 cm

/GPa (24). Using the ratio of these rates, the bias error induced by stress is quantified to be 0.2 °C/MPa for gallium nitride. Similarly, in silicon, this bias rate is calculated to be 0.17 °C/MPa.

Figure 4: Stress (biaxial) induced bias in the measurement of temperature using the peak position method for gallium nitride. At levels of stress greater than 25 MPa, the level of uncertainty induced by the stress is greater than that of the other spectral features. 

Using these rates, we find that equivalent errors resulting from stress become greater than the uncertainty in both the linewidth and intensity ratio at a relatively moderate level of 25 MPa, as shown in Figure 4. If expected stresses are greater than this level, the peak position should be used only with caution. 

Case Study: Thermometry of a Gallium Nitride High Electron Mobility Transistor 

Thermometry of gallium nitride high electron mobility transistors provides insight into the effects described thus far. These devices, utilized in applications ranging from wireless base stations to radar systems, have both their performance and reliability limited by temperature. It is, therefore, necessary to obtain accurate temperature measurements to monitor these effects. To this end, Raman measurements frequently have been used to examine these types of devices (9,34,35). Here, we compare the efficacy of using each of the spectral components to measure temperature in these devices and compare the obtained temperatures to a finite element simulation of such a system. (Full details of this study are found in references 24 and 27).  

Figure 5: (a) A schematic of a gallium nitride high electron mobility transistor similar to the one measured using (b) infrared (IR) thermal imaging and (c) Raman spectroscopy. Because of high levels of thermoelastic stress that develop during transistor operation, the peak position measurement significantly underpredicts the measurements acquired using the linewidth and intensity ratio, which corroborate with thermal simulations of this transistor. When stress is very small, all methods converge and the more easily implemented peak position can be used. 

Figure 5c shows the temperature profiles obtained using the peak position, linewidth, and intensity ratio. For the peak position and the linewidth measurements, a pinch-off reference condition was used in which the device is biased, but the charge-carrying channel is closed off through gating. This step is necessary to minimize inverse piezoelectric effects that induce changes in the peak position as well as the linewidth. Using these reference conditions, temperatures can be compared between the methods in a manner independent of piezoelectric influences. In this manner, it is found that measurements of the device temperature acquired using the linewidth and intensity ratio match quite well with computational predictions of the temperature. The peak position, meanwhile, significantly underpredicts the temperature because of the fact that thermoelastic stresses of greater than 200 MPa develop in the material. As shown in Figure 4, this underprediction leads to a bias well above the uncertainty of either the intensity ratio or the linewidth, thus rendering the peak position a comparatively inaccurate measurement technique. If stress were not this high in the material, the peak position could have been leveraged, thereby allowing for a quicker and easier measurement as a result of the lower practical uncertainty of this approach. Consequently, before choosing a methodology, it is necessary to consider not only the thermal effects but also those other factors that could alter the Raman spectra. 

Temperature measurements can be made using spectral features such as the position, linewidth, and intensity of the Raman signal associated with specific optical phonon modes. Each of these spectral characteristics offers particular advantages, depending on the type of device and operational considerations.

Raman Thermometry of Microdevices: Comparing Methods to Minimize Error

Microwave plasma–atomic emission spectrometer

The model 4100 microwave plasma–atomic emission spectrometer from Agilent is designed to run on air. According to the company, the device uses a nitrogen-based plasma that runs on air and does not require external cylinder connections or gases. 

Teledyne's application note on the analysis of trace elements in tungsten carbide using the company's Prodigy DC Arc spectrometer is now available at  

http://www.teledyneleemanlabs.com/resources/applications/dc_arc/1078-Tungsten_Carbide.pdf

. The publication discusses the ability of the instrument to analyze tungsten carbide samples in solid form. 

The RaySphere spectroscopic system from Ocean Optics is designed for irradiance measurements in the 380–1700 nm range. According to the company, the system includes two thermoelectrically cooled detectors. Applications include absolute irradiance measurements of solar simulators and other radiant sources. 

Glow discharge spectrometry application information

LECO's application notes for glow discharge-atomic emission spectrometry describe the analysis of titanium alloy, tin-based lead-free solder, and boron in low-alloy steel. The notes feature the company's GDS500A instrument, which uses the glow discharge method to uniformly sputter material from the sample's surface. 

Thermo Fisher Scientific's precalibrated FT-NIR analyzers are designed for feed and ingredient analysis in agriculture, and flour and milling analysis in the food processing industry. The systems consist of the Antaris II FT-NIR analyzer with Integrating Sphere Module with a 5- or 12-cm sample cup spinner accessory, RESULT and TQ Analyst software packages, and either the INGOT Level 4 Flour and Milling or Feed and Ingredient calibration software. 

Amptek's XRF Kit is designed to help users begin performing elemental analysis via X-ray fluorescence. According to the company, the kit includes the company's X-123 complete spectrometer with an SDD or Si-PIN detector; a mini-X USB controlled X-ray tube; XRF-FP QA software; a mounting plate; and a test  sample. 

Moxtek's XPIN detector includes a silicon PIN diode, a multilayer collimator, and a thin DuraBeryllium window. According to the company, the detector's preamp provides a low-noise signal output to an analog or digital pulse shaping amplifier. The detector reportedly can be used for benchtop and portable applications for which an integrated temperature controller is not required. 

Optical components from CVI Melles Griot are designed for ultrafast spectroscopy. The company's mirrors, beam splitters, polarizers, and prisms for Ti:sapphire or fiber lasers can be used in time-domain measurements, spectroscopy, imaging, sensing, and biomedical applications. 

Compact UV–vis spectrophotometers from Shimadzu are designed to reduce stray light during routine analysis and research applications. According to the company, the model UV-2600 spectrometer has a measurement wavelength range to 1400 nm, and the model UV-2700 achieves stray light of 0.00005% at 220 nm. 

Andor's iDus InGaAs detector array series provides a platform for spectroscopy applications up to 2.2 µm. According to the company, the TE-cooled, in-vacuum sensors reach cooling temperatures of -90 °C. 

The ShakIR ball mill from PIKE Technologies is designed to grind and mix IR-transparent diluents for diffuse reflectance measurements. According to the company, the accessory features electronic control and a protective shield for safety during use. 

The EZRaman-N analyzer from Enwave Optronics is designed for fast carbon nanotube characterization. According to the company, the 100–3300 cm

  spectral range covers full carbon nanotube bands of interest from the radial breathing mode to the G'-band. 

Inorganic certified reference materials from SPEX CertiPrep are designed for use with US EPA and ASTM methods. The products can be used for compliance monitoring of drinking water samples and for the analysis of ground and surface water. Common methods include US EPA SW-846, method 1310, US EPA 200.7, and US EPA 200.8. According to the company, each standard comes with a comprehensive, detailed certificate of analysis, and MSDS. 

Thermo Fisher Scientific's imaging ATR accessory is designed to integrate with the company's FT-IR microscopes. According to the company, the accessory provides enhanced spatial resolution and high-speed data acquisition capabilities for FT-IR chemical imaging applications such as composite structures, paper products, pharmaceutical dosage forms, biological tissures, and polymer films and coatings. 

Departments

Products

This column, which is the second of a two-part series, looks in more detail at a periodic review.Here, we will discuss how to conduct the review, the use of checklists, and reporting the outcome. 

In part I of this column (1), we discussed what a periodic review is, how to define the scope of a system scheduled for a review, and the main elements in the planning and execution of a review. In the second part of this discussion on periodic reviews, I want to focus on the execution phase of a periodic review and discuss in more detail the following activities:  

writing and approving the periodic review report and action plan  

implementing corrective and preventative actions. 

Who should be involved with a periodic review, apart of course, from the person who is conducting the review (as we discussed in part I of this column)? Ok, here is a list of potential people. Some of these roles may be combined depending on the size of the system being reviewed and especially if the laboratory is responsible for the whole system including the IT support, if the system is relatively small, or if there are only a few users. Other persons may be involved, depending on the roles and responsibilities for the system:  

Laboratory manager: Although the laboratory manager may not be directly responsible for the system itself, the system operates in the laboratory for which the manager is responsible and accountable. At a minimum, the laboratory manager should be present at the opening and closing meetings to introduce him or herself and listen to the review conclusions and discuss any findings.

Process owner: The individual in the laboratory who is responsible for the system. For a standalone spectrometry system, this may be the spectroscopist who runs the system. For larger systems, say the high performance liquid chromatography–mass spectrometry (HPLC–MS) networked system that was discussed in the first part of the column (1), it may be that the process owner is the laboratory manager.

System owner: The system owner is the person responsible for supporting the system from an IT perspective. If the system is standalone (where IT support comes from within the laboratory), the role of system owner can be taken by the process owner. If the the system is networked, the system owner may be an individual in the IT department. In the latter situation, the system owner could be from an external company if IT operations have been outsourced. 

Power user or laboratory system administrator and users: Depending on the size of the system, these roles could be taken by a single person or by many individuals. However, the reviewer needs to have access to somebody within the laboratory who knows the system in-depth from a technical perspective (the power user or system administrator) and may also require an experienced user to discuss procedures and use of the system.

Computer validation: Because the periodic review will examine the last validation of the system, there needs to be a person available to answer the reviewer's questions on this topic. Depending on how the work was performed, this could be the system administrator from the laboratory or a member of the computer validation group within the organization.

Quality assurance (QA): A member of QA who is responsible for the laboratory may also be involved with the periodic review. Depending on the depth of this person's knowledge, his or her involvement may range from interested observer to full participant.

Here is where the timetable for the periodic review comes in. Instead of having everybody hanging around waiting to be called, a timetable for the review should state what topics will be covered and roughly when they will take place. This allows individuals to carry out their normal jobs and turn up at an appointed hour. There may be some variation in the timetable depending on what the reviewer finds, so prepare to be flexible with the timing. 

Types of Computerized Systems to Be Reviewed

Typically, there are two types of computerized systems that will be covered by a periodic review: hybrid systems (electronic records plus paper printouts with handwritten signatures) and homogeneous systems (electronic records with electronic signatures where the only printouts will be summary sheets or at the request of an inspector or reviewer). Let's look at the advantages and disadvantages of each, starting with a completely paper-based system as shown in Table I. Paper is a medium that we are all familiar with; it is tangible, we can handle it and pick it up. Also, the audit trail is apparent because the paper is written or printed in a linear fashion. However, the poor reviewer has a disadvantage as he or she must wade through piles of paper to understand what was done, by whom, and when.  

Table I: Advantages and disadvantages of paper, hybrid, and electronic auditing systems 

Moving to a hybrid computerized system, we have the same advantages and disadvantages of the paper plus the complication and correlation of electronic records. It is the correlation of the paper with the electronic records that provides the biggest challenge in an inspection or periodic review of heterogeneous or hybrid systems. The failure by Food and Drug Adminstration (FDA) inspectors to cross check the paper printouts with the corresponding electronic records in a chromatography data system allowed Able Laboratories (Cranbury, New Jersey) to conduct fraud on a large scale for a long period of time (2). If an electronic (homogeneous) system is audited, the reviewer needs to ensure first that the computerized system itself is validated and has remained so from the last full validation to the time of the periodic review. However, the reviewer needs an experienced user when checking the way the system operates and how the electronic records are produced to implement the requests from the reviewer through the system. Often, the system used for part of the review may use the validation or training instance and the reviewer needs to ensure that the operational and other instances are equivalent. However, the great advantage when reviewing an electronic system is the speed of the audit; a lot of ground can be covered relatively quickly because of how the system supports the business process. What also needs to be addressed are the technical controls within the system to ensure data integrity, for example, checking the audit trail to make sure that results have not been manipulated to pass. As an electronic system can cross organizational departments, the review scope may be wider than normal and if networked will also include the IT department responsible for supporting the system. 

Are Computerized Systems Designed to Help Periodic Reviews?

No, is the simple answer to a simple question. Let us explore the reason why we want systems to help us with periodic reviews. 

(Turn on moan-mode). Much of the emphasis in the design of the majority of computerized systems used in the analytical laboratory has focused, unsurprisingly, on providing new functions for users and incorporating regulatory compliance, such as audit trails when requested by regulated users. However, as we are moving toward electronic systems with minimal paper output, we really do not have the functions in software to allow effective and efficient second-person review of the data on a day-to-day basis and also for periodic reviews on a 1–3 year frequency. It is the need for second-person review that we need to concentrate on because the periodic review will use the same functions. Both the second-person check and the periodic reviewer need to assure themselves that the data quality and integrity are good and have not been compromised. This is also a concern for regulatory agencies, especially the FDA, which recently launched an initiative on ensuring data integrity (3).  

The United States good manufacturing practices

US GMP) regulations require, in section § 211.194(a), that laboratory records include complete data derived from all tests necessary to assure compliance with established specifications and standards, including examinations and assays, as follows:

     (4) A complete record of all data secured in the course of each test, including all graphs, charts, and spectra from laboratory instrumentation, properly identified to show the specific … drug product, and lot tested. 

     (8) The initials or signature of a second person showing that the original records have been reviewed for accuracy, completeness, and compliance with established standards (4). 

These two clauses are important because the regulation requires that 

 data be collected. This includes the checks carried out on the spectrometer before sample analysis commences, as well as any rerun samples. The checks performed by the reviewer (second person) must include complete data, including audit trails of hybrid and electronic computerized systems. 

Referring to Table I, the complete data should be present in the laboratory notebook or uniquely numbered laboratory sheets and instrument printouts. Built into the paper will be the cellulose audit trail — that is, where mistakes that have been identified and corrected will have the original entry struck through with replacement data entered by the side, the initials of analyst who made the change, and the date of the change. The correction may also include a reason for change as per the applicable good X practice (GxP) regulation that the laboratory works under. 

When we consider hybrid systems, our data integrity problems start because the audit trail becomes jumbled and the linkage between electronic records and the corresponding paper printouts is not always obvious. Often, the audit trail is not very good as there is just a single audit log with little or no search function. By its nature, an audit trail will normally contain a multitude of entries: user access, failed attempts for access by users, additions, modifications, and deletions to data, user account management and access control, and so forth. Many of these entries are routine events, but unless there is a simple and easy means of separating different types, the audit trail can be useless.   

Similarly, when fully computerized systems are used, the linkage between records and reports is usually easier and more straightforward. However, the key requirement is an effective audit trail, which may not be the case. Typically, systems are not designed to work electronically: The business process may be fine but the Achilles heel of the majority of systems is the audit trail as it is the key to ensuring data integrity. What is required are data filters that can be configured by users to identify when changes to data have been entered into the audit trail that will allow effective second-person review and, by implication, periodic review. In addition, there is the new European Union (EU) GMP requirement (5) for review of audit trails and few if any systems have the ability to demonstrate that this has been done. So, we are left with systems that are typically not designed to facilitate reviews of data, audits, or inspections. Welcome to the periodic review world of hybrid and electronic computerized systems! (Turn moan-mode off.) 

So having moaned about the systems you have to review, let's begin the process. This starts with the opening meeting, which is typically a short session where the reviewer meets the key people who will be involved with the review. It allows introductions to be made and for the reviewer to explain what the aims of the periodic review are and set expectations from the individuals who will be contributing from the laboratory side. It also provides an opportunity to fine-tune the schedule to take into account any last-minute availability of staff and to move items around. The point here is that the reviewer should be flexible and, so long as the main topics are covered, the order of review should not matter.   

One item that is important is at the end of the opening meeting and before the start of the review, the reviewer should ask for a short presentation of the system by the process owner or validation manager to outline the main functions and how the system was validated. This allows the reviewer to understand the context of the material he or she has read before the review starts and also to pick up and understand terminology.  

I want to consider how you actually carry out a periodic review. In my view, you have two basic options:  

Static: Sit in a room and read and evaluate system documentation or logs against procedures and polices. At this point, the more intelligent readers among you may be wondering if watching paint dry is a more intellectually stimulating option. 

Dynamic: Look at the system and see how the software and instrument operates, examine the IT support, and have discussions with users and IT staff. You will be amazed at what you can find out by simply being observant, which we will discuss later in this column. 

In reality, the periodic review is a mixture of the dynamic and static options. Walking around the laboratory and seeing the system in its operational environment allows a reviewer to put the documentation into context and see the reality rather than what is portrayed in policies and procedures. So, the reviewer should not just sit in an office and read — walk around and see the system in operation in the laboratory. Therefore, the review schedule should contain time for a tour of the laboratory and system and, if applicable, the IT department. 

You will need to obtain information from the people that you will interview and have discussions with during the audit. The best way to do this is to ask open-ended questions (for example, when? who? where? how? why? and what?) to get people talking. Then, of course, comes the question that makes you ask if all periodic reviewers are from the state of Missouri. Because the answer to the open questions listed above is usually followed by the infamous "show me" to verify that what has been stated actually occurs and there is documented evidence available to demonstrate this. To verify the information provided, the reviewer should use closed questions that the responder can respond to with either yes or no answers. Another technique that allows a reviewer to understand information is to ask questions such as, "What happens if … ?" 

Some of the dos and don'ts of conducting periodic reviews are shown in Table II, and many of these are derived from how to manage people correctly during the review. Part of the role of a reviewer could be to persuade users and IT staff to change their way of working if noncompliant practices are observed; you can't do this if you have put their backs up when you conduct the review. 

Table II: The dos and don’ts of conducting a periodic review 

Continued discussion of the periodic review process, including how to conduct a review, the use of checklists, and reporting the outcome

Spectroscopy-11-01-2011

Analysis of Flue gas Desulfurization Wasterwaters by ICP-MS

Detecting Ions in Mass Spectrometers with the Faraday Cup

Raman Thermometry of Microdevices: Comparing Methods to Minimize Error

Products

Periodic Reviews of Computerized Systems, Part II

Vol 26 No 11 Spectroscopy November 2011 Regular Issue PDF