Absolute characterization of laser-induced breakdown spectroscopy detection systems

Transcription

1 Spectrochimica Acta Part B 63 (2008) Views and criticism Absolute characterization of laser-induced breakdown spectroscopy detection systems M.T. Taschuk a,, Y. Godwal a, Y.Y. Tsui a, R. Fedosejevs a, M. Tripathi b, B. Kearton c a Department of Electrical and Computer Engineering, University of Alberta, Edmonton, Alberta, Canada b Andor Technology, S. Windsor, USA c Ocean Optics, Dunedin, USA Received 16 December 2006; accepted 24 January 2008 Available online 8 February 2008 Abstract A quantitative comparison of the performance of four different laser-induced breakdown spectroscopy detection systems is presented. The systems studied are an intensified photodiode array coupled with a Czerny Turner spectrometer, an intensified CCD coupled with a Czerny Turner spectrometer, an intensified CCD coupled to an Echelle spectrometer, and a prototype multichannel compact CCD spectrometer system. A simple theory of LIBS detection systems is introduced, and used to define noise-equivalent spectral radiance and noise-equivalent integrated spectral radiance for spectral detectors. A detailed characterization of cathode noise sources in the intensified systems is presented Elsevier B.V. All rights reserved. Keywords: Laser-induced breakdown spectroscopy; LIBS; LIBS detectors; Responsivity; Noise-equivalent signals 1. Introduction Since the beginning of laser-induced breakdown spectroscopy (LIBS), a key role in determining the overall performance of the technique has been the responsivity and characteristics of the equipment used. Characterizations of both the performance of LIBS using different systems [1 4], as well as studies of the performance and characterization of intensified detectors [5 7] have been performed. However, little or no attempt has been made to characterize the combined performance of the spectrometers and optical detectors used in LIBS in absolute terms. The results reported in LIBS literature are typically a combination of the response of the LIBS phenomena and the characteristics of the detection system used. As a result, it is difficult to make quantitative comparisons between different experimental groups or with theoretical predictions. This paper was presented at the 4th International Conference on Laser Induced Plasma Spectroscopy (LIBS 2006) held in Montreal, Canada, 5 8 September Corresponding author. address: mtaschuk@ece.ualberta.ca (M.T. Taschuk). LIBS is an optical technique and, as such, it would be very useful if LIBS data could be reported in radiometric units. In this paper, we will be using the nomenclature of O'Shea [8]. The radiometric approach can be adapted to the reporting of spectral measurements by correcting for the reciprocal linear dispersion of the detectors used, reporting units in terms of spectral radiance [W sr 1 cm 2 nm 1 ] or spectral intensity [W sr 1 nm 1 ]. However LIBS data is strongly dependent on the starting time of the observation and the total integration time. Thus, it may be more useful to report the total energy observed, rather than the power, while noting the gate delay and width of the integration. Spectrally integrated line intensities and equivalent noise power may then be reported in units of time-integrated radiance [J sr 1 cm 2 ] or timeintegrated intensity [J sr 1 ]. To present LIBS data in these units, it is necessary to have a good understanding of the optical properties of the experimental system. Every component of a LIBS system possesses properties that vary with optical wavelength. However, once an absolute characterization has been made, it is possible to report data that reflects the behaviour of the LIBS plasma itself, rather than a combination of the LIBS plasma and experimental equipment. With data representing the behaviour of the LIBS /$ - see front matter 2008 Elsevier B.V. All rights reserved. doi: /j.sab

2 526 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) plasma, it will be possible to make detailed, absolute comparisons between different experimental groups and with theory. The fundamental understanding of the LIBS process will improve, and the limits of the technique may be better understood. The aim of this paper is to show how a radiometric approach can be used in LIBS experiments and in comparisons of different instrumentation. The treatment is somewhat tutorial, since the concepts and the terminology discussed are well known. However, the application of this approach to LIBS experiments is, in our opinion, worthy of being illustrated and possibly followed by other researchers. In this paper the absolute performance of four LIBS detection systems is studied. A simple theory of LIBS detection systems is introduced, and used to define noise-equivalent integrated spectral radiance (NEISR) and noiseequivalent spectral radiance (NESR) for spectral detectors. Experimental results for absolute responsivity, NEISR and NESR are given, and a detailed characterization of cathode noise sources in the intensified systems is presented. 2. LIBS detection system theory While in many cases it would be possible to reliably distinguish between the components of a LIBS system, in experimental work or practical applications it is the performance of the combined system which is important. Further, in some cases such as an Ocean Optics compact CCD spectrometer the detector and spectrometer are an integrated unit and changes cannot be easily made. In other cases information about the spectrometer may be unavailable. As a result of these factors, it is more appropriate to develop definitions which represent LIBS equipment as it is typically used. In the following discussion and equations, roman characters (L, D, Φ) represent overall characteristics of the spectrometer or detector units, while script characters (L; T ; R) refer to internal characteristics which combine to yield the external properties of the systems under study. Integrated spectral plasma radiance is used to represent the plasma emission, as LIBS detectors typically integrate over a set period of time and thus report energy rather than power. Integrated spectral plasma radiance, Φ(λ), has units of J sr 1 cm 2 nm 1. It is well known that the spatial profiles of ions and excited atoms vary within a LIBS plasma plume. While these effects are ignored in this paper, it should be possible to include them in this factor if spatially resolved measurements are made. Such dependencies could be rather complex, but in principle an effective emission size as a function of wavelength could be ascertained experimentally. In addition, any effects from the image relaying optics should be included in this factor. Spectrometer luminosity can be thought of as a figure of merit for the spectrometer, and is well summarized in [9]. Spectrometer luminosity is synonymous with étendue. The entrance aperture and solid angle of the spectrometer define the maximum amount of plasma emission that can be coupled into the detector. The dispersing effects of the spectrometer are added to convert to spectral luminosity, L(λ), which has units of sr cm 2 nm μm 1, where the addition of μm describes distance along the exit focal plane of the spectrometer. Spectral luminosity can be written as LðkÞ ¼ LðkÞT ðkþ ð1þ where L(λ) represents only the geometric action of the spectrometer, and T (λ) describes the optical losses from the nonideal reflection and transmission of optical components within the spectrometer. Detector response includes factors such as gain, quantum efficiency, and effective pixel size of the CCD at the photocathode or other sensing surface. Detector response, D(λ, G), is a function of wavelength and gain setting and has units of counts μm J 1 channel 1, and can be written as Dðk; GÞ ¼ D 0 Rðk; GÞ ð2þ where D 0 is a geometric constant describing the pixel size for a given detector, and R is the responsivity of the detector alone. Note that gain includes not only signal amplification via a microchannel plate, but also factors such as pixel integration time, relative response of the various pixels and other signal gain mechanisms. Depending on the system used, different component settings will have to be included in the responsivity. In the limiting case, only the quantum efficiency of the detector will be used. Using the functions defined above, the signal obtained using a LIBS system can be expressed as IðchannelÞ ¼ Uk ð ÞLðkÞDðk; GÞ ð3þ where Φ(λ) is the integrated plasma radiance including the effects of any relaying optics, L(λ) is the spectrometer spectral luminosity and D(λ, G) is the detector response. Once these factors are combined, the result I(channel) is the observed counts as a function of channel number, with units of counts channel 1. This corresponds to the experimental observable for typical LIBS detector systems used today. The responsivity of a optical system is typically defined in terms of the measured signal divided by the input optical power [7,9,10]. The input optical power will be the integrated plasma radiance multiplied by the geometric factors of the spectrometer and detector, and the measured signal is I (channel). The resulting factor will have units of counts J 1, or counts per photon. While counts are not directly a physical quantity, it is typically the output unit used by the LIBS experimentalist. In this paper the responsivity of a LIBS system will be defined as Rðk; GÞ ¼ IðchannelÞ ¼Rk; ð GÞT ðkþ: ð4þ Uk ð ÞLðkÞD 0 As can be seen with this definition, it is possible to characterize the responsivity of a LIBS system by characterizing the combined behaviour of the spectrometer transmissivity and detector response. However, care must be taken to ensure that the radiation from the calibration source is completely within the acceptance angle defined by the spectrometer's geometric luminosity L(λ). For the work in this paper, a single channel signal to noise ratio (SNR) will be used, defined as: SNR ¼ IðchannelÞ ð5þ N where I(channel) is the background corrected signal, and N represents all noise sources, and will fully discussed below.

3 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) Table 1 Summary of detector characteristics System Number of pixels Pixel dimensions [μm] A μm by 2.5 mm B μm by 26 μm C μm by 13 μm Maximum gain [counts/ photoelectron] Manufacturer 4 Princeton Applied Research 270 Andor Technology 800 Andor Technology Model no. 1455R- 700-G istar DH720-25F-03 istar DH734-18U-03 D ~24 μm N/A Ocean Optics Prototype This definition is simple, but will serve to illustrate the differences in the detectors studied here. With more sophisticated signal analysis techniques the effect of noise can potentially be reduced. On the other hand, this definition does not include the effects of signal shot noise. Overall, these two competing factors will tend to cancel each other out, and the current results should be close to the final expected behaviour. In general LIBS practice, both the signal and noise will be functions of channel number, since noise sources such as the continuum radiation vary with wavelength. However, it should always be possible to define a local noise for regions around and including the line of interest. Thus, SNR can be calculated with a single spectrum, provided a spectrally quiet neighboring region can be found, or the noise estimated. There are many sources of noise that are relevant to LIBS detection systems. For an unintensified CCD the primary noise sources are dark current, σ Dark, and readout noise, σ Readout [6,7]. Howell also considers the noise from the A/D converter, σ A/D,in [6]. Many more sources of noise exist for the unintensified CCD [7,10], but they are not included separately in our current model. In practice, any such effects can be integrated into one of the noise terms that is already included. Each of the σ x is one standard deviation of the distribution of noise for the relevant quantity x, with units of electrons. Once again, to keep close to the experimental observables in a typical LIBS experiment, these quantities will be expressed in units of counts. As a result, it is also necessary to consider the gain of the A/D converter, G A/D, with units of counts per electron. Using these values, a single pixel from an unintensified CCD with no exposure to external light will have a noise level, N, defined in Eq. (6). N 2 ¼G 2 A=D r2 Dark þ r2 Readout þ r 2 A=D : ð6þ This form of the noise would be sufficient for unintensified systems, but will not describe the full behaviour of intensified systems. Intensified systems also exhibit cathode noise spikes, which will be discussed further below. An additional term, σ Cathode, must be added to Eq. (6) to represent this behaviour. The average noise for a single pixel in an intensified system, N intensified, with no exposure to external light will be as given in Eq. (7). Nintensified 2 ¼G2 A=D r2 Dark þ r2 Readout þ r2 Cathode þ r 2 A=D : ð7þ The quantity σ Cathode must be treated with care, as the underlying distribution is not Gaussian [11]. This term is a strong function of the gain of the intensifier. The shot noise due to the signal I must also be taken into account. The dominant source of shot noise as the signal is converted from photons to counts will occur at the point where the fewest countable particles occur. In the case of an intensified system, this will correspond to the number of photoelectrons in a single channel; in a CCD, to the number of electrons in a single channel. The observed signal I enters into the noise once it has been scaled by the A/D gain, G A/D, and the intensifier gain, G I. The SNR produced by a single channel in an intensified system observing a signal I can now be written as given in Eq. (8), which is a variant of the CCD Equation [6]. I SNR ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi G 2 A=D r2 Dark þ r2 Readout þ : r2 Cathode þ r 2 A=D þ IG I G A=D ð8þ Noise-equivalent power or noise-equivalent input is defined as the incident power or energy, respectively, required to generate a SNR of unity at the detector, with signal shot noise neglected [9,10]. However, these definitions do not take into account the limited luminosity of the spectrometer. By following the same approach, quantities which do include the effect of the spectrometer can be defined. Analogous to noise-equivalent power will be noise-equivalent spectral radiance (NESR), and analogous to noise-equivalent input will be noise-equivalent integrated spectral Table 2 Summary of spectrometer characteristics System f/# Focal length [m] Grating [lines/mm] Reciprocal linear dispersion [nm/channel] Spectral resolution [nm] Manufacturer Model no. A m a Acton Research Corporation SpectraPro 500 B m a Oriel Instruments MS 260i C m b Andor Technology Mechelle 5000 D m Varies b Ocean Optics Prototype a Experimental measurement of instrumental limit. b Manufacturer specification.

4 528 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) radiance (NEISR). For the case of NESR, SNR can be expressed as SNR ¼ I NESRðchannelÞ ¼ 1: ð9þ N intensified At the detector we can express the required signal as I NESR (channel), which corresponds to NESR through Eq. (3) and the consideration of an integration width I NESR ðchannelþ ¼ NESRs gate LðkÞDðk; GÞ: ð10þ have a lower sensitivity than the previous three. System D has been optimized for harsher operating conditions for use in the field or industrial settings. LIBS plasma emission is imaged onto a multi-fiber bundle, which couples the light to the different CCDs. The system is externally triggerable and gateable, with a minimum gate width of 10 μs. Detector information is given in Table 1, and a summary of the spectrometer characteristics is given in Table 2. For systems A and B, the gain data is the combined results of a large number of calibration experiments. In the case of systems C and D, the Solving algebraically yields a definition for NESR which includes the detector luminosity, which is more applicable to LIBS than a more conventional definition concerned only with the detector itself. NESR will depend on wavelength and any factors affecting the gain of the detector, and should be written as given in Eq. (11). Unlike the responsivity of a LIBS system, the NESR requires inclusion of the spectrometer luminosity. Noiseequivalent spectral radiance has units of W sr 1 cm 2 nm 1. NESRðk; GÞ ¼ N intensified : ð11þ LðkÞDðk; GÞs gate The use of NEISR is more appropriate when the noise in a system does not scale with integration time, but remains constant. In this case, it is the total integrated energy which is relevant, and τ gate does not enter into the definition. Following the approach used above to derive NESR yields a definition of NEISR which includes the detector luminosity. NEISRðk; GÞ ¼ N intensified LðkÞDðk; GÞ : ð12þ 3. Experimental methodology 3.1. Systems studied A total of four spectrometer/detector combinations are studied in this paper. System A, consisting of an IPDA and Czerny Turner spectrometer, was a standard LIBS detection system in the early phase of LIBS development, until the advent of ICCD technology. System B is very similar to System A, with an ICCD and a Czerny Turner spectrometer. Czerny Turner spectrometers are valuable instruments for LIBS, due to their low f-number and flexible grating configurations. However, there is always a tradeoff between spectral resolution and bandwidth. System C, consisting of an ICCD and an Echelle spectrometer, offers high spectral resolution and bandwidth simultaneously, but at the cost of a larger f-number than the Czerny Turner systems. System C is especially useful for samples with complex spectral features spread across a very wide spectral range, allowing a full LIBS measurement in a single laser shot. Finally, System D is a prototype field portable broadband spectrometer system composed of eight compact CCD spectrometers optimized for different spectral regions. Unlike the other systems, System D is not intensified, and is therefore expected to Fig. 1. Calibrated detector responsivity for (a) System A, (b) System B and (c) manufacturer data for System C. The solid lines are best exponential fits to the data for gains below (a) 6, (b) 150 and (c) 150.

5 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) gain of the detectors was provided by the manufacturer. Detector gain for the intensified systems is given in Fig Radiometric calibrations Four types of calibration sources were used to characterize the overall responsivity, R(λ, G) (combined detector response and spectrometer transmissivity, R(λ, G)T (λ, G)), of the systems studied here. These consisted of: a tungsten blackbody source (Eppley Laboratories), a combination deuterium/halogen broadband source (Ocean Optics DH2000), a Hg lamp (Oriel 6035) and diffusely scattered pulsed laser light sources. The manufacturer data was used for the tungsten and Hg lamps. The diffuse light source was pulsed laser light scattered off a barium sulphate surface, with pulse energy monitored with a calibrated photodiode. The tungsten lamp, Hg lamp and diffuse scattered light source were used as absolute sources, while the deuterium/ halogen source was used as a relative source to extend the calibration into the ultraviolet. The detector responsivity calibrations were carried out with different subsets of the available calibration sources. During calibration procedures, the tungsten lamp was operated with the filament of the lamp placed with its normal pointed to the entrance aperture of the spectrometer, at a distance of several meters. This distance ensures that the 60 mm 2 size of the filament can be considered as a point source as viewed by the spectrometer. The deuterium/halogen lamps, which overlapped the range of the tungsten lamp, were fiber coupled to systems C and D. Using this overlap, the calibration of systems C and D was extended down to 200 nm. The calibration for system A was performed a limited number of times. As a result, absolute accuracy is estimated at a factor of 2 [12]. For the other systems, the absolute accuracy is estimated at ~40% based on a large number of calibration experiments performed for system B Cathode noise spike characterization To characterize the cathode noise spikes, the location, area and width distributions of noise spikes were determined by a peak finding and fitting routine that was used to process blank spectra. The blank spectra were taken with the entrance aperture covered from any incident light, and in the case of the Czerny Turner systems, shutters and entrance slits closed. It is therefore expected that the observed spikes must be a characteristic of the intensified detectors. This algorithm finds peaks by looking for channels more than 4σ above the single channel background noise. When no cathode noise spikes are present, the background has a Gaussian distribution [11]. Peak location, height, width and area for each region were found using a least squares Gaussian fit. Further details regarding the algorithm can be found in [13]. Fig. 2. Responsivity of the (a) IPDA/Czerny Turner system with a 150 l/mm grating, (b) ICCD/Czerny Turner system with a 600 l/mm grating, (c) ICCD/Echelle system and (d) Multichannel CCD spectrometer prototype. Intensified systems were run at maximum gain.

6 530 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) Experimental results Using the theory and methodology presented above, the four systems are characterized for responsivity, cathode spike distributions, and noise-equivalent signal levels LIBS system responsivities The absolute responsivity results for the systems at maximum gain are presented in Fig. 2. System B was the most responsive of the systems studied, with a peak responsivity of ~27 counts per photon. The next most sensitive system is C, with a peak Fig. 4. Characteristics of the log-area distribution for the (a) system A, (b) system B and (c) system C. responsivity of ~ 6.5 counts per photon. System A falls below both ICCD systems in terms of overall responsivity, with a maximum of ~0.4 counts per photon. Finally, system D, which does not have an intensified detector, was the least responsive, with a peak responsivity of ~10 2 counts per photon LIBS system noise characterization Fig. 3. Mean and standard deviation of the cathode spike arrival rate for the (a) system A (b) system B and (c) system C. The solid lines are a guide for the eye. The cathode spike distributions were characterized as functions of detector gain for systems A, B and C. As system

7 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) D is not intensified, no cathode noise spikes were observed, and this system is therefore not characterized in this section Cathode spike rate The number of cathode spikes per unit channel and unit time was investigated for the systems as a function of gain. The arrival rate and density was normalized using the intensifier gate width, and the number of active channels. In the case of the Echelle, a channel does not relate directly to a column on the ICCD, but rather a smaller area defined by the manufacturer's readout program. We refer to the channels reported by the software as virtual channels. The number and location of these channels were observed to vary with wavelength calibration. For the experiments reported here, the number of virtual channels was 29,147. The mean and standard deviation of the cathode spike arrival rate as a function of gain for the intensified systems are presented in Fig. 3. For the systems A and C the small number of spectra and the very quiet nature of the detector, respectively, limit the precision of the characterization. However, the data set enables a comparison of the detector/spectrometer performance in high and low gain regimes for the various systems. At maximum gain, system A has a cathode spike rate (CSR) of ~2000 s 1 channel 1, system B a CSR of ~25 s 1 channel 1, and system C a CSR of ~0.03 s 1 channel Area distribution of cathode spikes The area distribution of the cathode noise spikes was characterized for the different systems under study. In all cases, the cathode spike area followed the log-normal distribution, which occurs when the logarithm of a variable is normally distributed [14]. In our case, log base 10 is used. The log-normal mode ( p) and deviation (σ) as a function of gain for the intensified systems is given in Fig. 4. The solid lines are a best linear fit to the data. For system A data was only obtained at high gains. For system B characterization was carried out for gains settings of 100 and more corresponding to typical operating conditions. For system C no cathode noise spikes were observed for gains below 200, despite the 20 ms gate width used for this data Width distribution of cathode spikes The mean and standard deviation of the best fit to the cathode spike width distribution are reported in Fig LIBS system noise equivalents There are two regimes of operation for systems A, B and C: with cathode spikes at higher gains, and without cathode spikes at low gain. In the case where cathode spikes are observed, an increase in gate width will increase the number of spikes which are observed. In this regime, it is appropriate to quote the noise equivalent in terms of noise-equivalent spectral radiance (NESR). In the case where no noise spikes are observed, it is more appropriate to quote the noise equivalent in terms of noiseequivalent integrated spectral radiance (NEISR). Fig. 5. Characteristics of the cathode spike width distribution for the (a) system A, (b) system B and (c) system C. NESR and NEISR are quoted at the entrance slit of the spectrometer in terms of spectral radiance [W sr 1 cm 2 nm 1 ] and integrated spectral radiance [J sr 1 cm 2 nm 1 ], respectively. In order to calculate the NESR and NEISR, Eq. (11) is used. The quantity N intensified is approximated by the standard deviation of the background signal when no light enters the LIBS detection system. This approach should include the effects of all of the noise terms identified in Eq. (7). In order to calculate the spectrometer luminosity, the f-number and entrance aperture must be clearly defined. For all spectrometers, the full available f-number is used. To convert from

8 532 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) Fig. 6. NEISR for the (a) IPDA/Czerny Turner system (150 l/mm grating, slit area of cm 2 ), (b) ICCD/Czerny Turner system (600 l/mm grating and slit area of cm 2 ), (c) ICCD/Echelle system (entrance aperture of cm 2 ) and (d) Multichannel CCD prototype (slit area of ~ cm 2 ). For the Czerny Turner systems, a different grating or entrance slit would affect these values. The intensified systems are run at minimum gain. channels to nm, the channel width of the detectors was used. For the Czerny Turner spectrometer systems, it is possible to alter these values significantly by changing the grating. In the case of systems A and B, the slit width and slit height can be changed. For the noise study performed earlier, the full height of the detector was used. To remain consistent, the full available height of the detector will be used to calculate the spectrometer luminosity. The effective height of system A is ~2.5 mm, and 4.1 mm for system B. A typical slit width for our μlibs experiments is 100 μm, and is used to define the entrance area for the Czerny Turner spectrometer systems. In the case of system C and D, the entrance aperture of the system is well defined by the optical fibers used and cannot be changed during use Noise-equivalent integrated spectral radiance at minimum gain This analysis was performed for the case where no cathode noise spikes were observed. Acquisition times were kept short enough that the contribution of dark current is negligible. This is representative of the typical operating conditions used in LIBS. System D can be included in this analysis. In all cases, the minimum gain for the intensified systems was used. The standard deviation of all available low gain noise spectra was calculated, and the average used to evaluate NEISR. This NEISR for the various systems is presented in Fig. 6. The minimum NEISR, luminosity and maximum responsivity at minimum gain are tabulated in Table 3. In the case of the Multichannel CCD system, the effective pixel width is approximately 24 μm. Overall, the NEISR is lowest for system B, largely due to the large luminosity of the Czerny Turner spectrometer. The next best performing system is system D, followed by system A. The low responsivity of system A is offset by the high luminosity of the Czerny Turner spectrometer, resulting in competitive overall performance. The system with the highest NEISR is system C, due to the relatively low responsivity at minimum gain and low luminosity of the system. Table 3 Comparison of NEISR results and contributing parameters for minimum operating gain System Minimum NEISR [J sr 1 cm 2 nm 1 ] Luminosity [L] [sr cm 2 nm μm 1 ] A B C D Maximum responsivity [counts per photon]

9 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) Noise-equivalent spectral radiance at maximum gain The same analysis was carried out at maximum gain for the three intensified systems. The analysis for the regime in which cathode noise spikes were observed is presented in Fig. 7. In all cases, the maximum gain was used. The minimum NESR, luminosity and maximum responsivity at maximum gain are tabulated in Table 4. Overall the NESR is lowest for system B. System C is a factor of ~110 noisier, while system A is another factor of ~17 Table 4 Comparison of NESR results and contributing parameters at maximum operating gain System Minimum NESR [W sr 1 cm 2 nm 1 ] Luminosity [L] [sr cm 2 nm μm 1 ] A B C Maximum responsivity [counts per photon] noisier still. In this case, the good performance of the ICCD in the system C offsets the low luminosity of the Echelle system to yield good overall performance. 6. Discussion Fig. 7. NESR for the (a) IPDA/Czerny Turner system (150 l/mm grating, slit area of cm 2 ), (b) ICCD/Czerny Turner system (600 l/mm grating, slit area of cm 2 ) and (c) ICCD/Echelle system (entrance aperture cm 2 ). For the Czerny Turner systems, a different grating or entrance slit would affect these values. All systems were run at maximum gain. The results presented here are expected to be characteristic of the instrumental classes represented. Clearly, the performance of specific detectors will vary, and the characterizations carried out here will be required for any system for which an absolute calibration is desired. In addition, detector characteristics such as photocathode material, maximum gain and pixel size will affect the combined spectrometer/detector performance and must be selected with care. In addition, the Multichannel CCD spectrometer system is a prototype, and as such the performance reported here may not represent the full capabilities of such a system. For the Czerny Turner systems the magnitude and location of the peak responsivity can be controlled by the gratings. With a Czerny Turner system, a good choice of grating is essential. For both the Echelle and Multichannel CCD spectrometer systems, variations in responsivity as a function of wavelength are observed. While the calibration data presented in Fig. 2c and d can be used to remove these variations from spectral data, it may not always be possible to avoid regions of low sensitivity in LIBS experiments. Additionally, the effective dynamic range of the broadband instruments will be a strong function of wavelength, which would require some care during experimental design. It should be noted that the ICCD used on system C has a gain about a factor of 4 greater than that used on system B. As a result, the peak responsivity of the Czerny Turner could be improved by employing a newer ICCD similar to the one used in the ICCD/Echelle system. The difference in the CSR for the two Czerny Turner systems A and B is attributed to the difference in detector technologies, and the fact that system B was cooled to 10 C. System A (IPDA based) is also 15 years older than system B (ICCD based), and photocathode materials may have improved in this time. The photocathode area per channel is cm 2 for system A, and cm 2 for system B. Combining the difference in cathode area with the cathode spike rate, the ICCD/Czerny Turner outperforms the IPDA/ Czerny Turner by a factor of ~ 240 in terms of reduced photocathode spike rate. However, the same argument cannot be made for the 10 3 improvement in cathode spike rates demonstrated by system C, as this system was also cooled to 10 C. For system B the full

10 534 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) height of the ICCD was used, corresponding to a total of 256 pixels, each 26 μm by 26 μm. For system C, the integration height was set to the maximum of 5 pixels. If we assume that a single pixel width corresponds to a virtual channel, a total of 5 pixels are used, each 13 μm by13μm. The estimate for the cathode area used by the ICCD/Echelle for a single channel is then cm 2. Thus the photocathode area used by the ICCD/Echelle for a single channel is about 200 times smaller than that used by the ICCD/Czerny Turner system. It is therefore expected that the true difference in the cathode spike rate per unit area of the two ICCDs is a factor of ~4. Since this factor is associated with the ICCD themselves, a similar improvement in the cathode spike rate should be possible for system B using the same ICCD as system C. One important point about the NEISR values is the influence of intensifier gain. In general, a moderate increase in gain will not lead to a strong increase in the background noise, until the onset of cathode noise spikes. Up to this point, as the gain is increased the NEISR will drop. As a result, the performance of the intensified systems can be improved significantly over the results presented in Fig. 6. The increase in performance can be predicted by combining the intensifier gain results presented in Fig. 1 with the NEISR results presented in Fig. 6. The influence of integration time must also be considered. Under long-lived, but low intensity, light conditions a very long gate at low gain would be the correct choice. However, the NEISR results presented above are not expected to be valid at arbitrarily large gate widths, as dark current will increase with increased CCD integration time. This characterization has not been performed for the intensified systems studied here, but the Multichannel CCD spectrometer system shows no increase in background variability up to a gate width of 100 ms indicating that integrated dark current noise is still less than readout noise up to these gate times. When the current results are combined with absolute measurements of LIBS plasma emission such as presented in [15], a full calculation of LIBS system performance is possible for a wide variety of conditions. Preliminary work towards simulations of full LIBS systems has begun, and will be the subject of a future report. It will then be possible to conduct engineering design studies of LIBS systems, reducing requirements for preliminary experimental work. As the absolute emission from more systems are characterized, LIBS system engineering will become possible. More sophisticated signal analysis techniques will have an impact on the effective performance of the systems. In cases where the cathode noise spikes can be well characterized and discriminated from real signals, the effective noise power of the systems can potentially be reduced. In addition, the use of multiple emission line signatures for a given element will further reduce the effect of detector noise on LIBS experiments. The broadband systems C and D would be well suited to such an approach. The results given here assume that the luminosity of the spectrometers is fully exploited. However, there will be cases where the imaging optics or plasma size does not fully fill the entrance aperture of the system. In such a case, the detector noise will have a greater effect as the angle and area observed is reduced, thus increasing the noise-equivalent signals. It is expected that this will be more of a problem for the Czerny Turner spectrometer systems as the entrance apertures can be much larger for these systems. As a result, the advantage conferred by the increased luminosity of these systems will be somewhat offset if it cannot be fully utilized. Finally, the impact of the simultaneous high spectral resolution and high bandwidth of the ICCD/Echelle systems and Multichannel CCD spectrometer systems on LIBS performance was not examined here. Czerny Turner systems can achieve similar spectral resolutions, but at the cost of a greatly reduced spectral bandwidth. With the broadband systems, the ability to simultaneously observe the brightest emission lines from each species in a LIBS plasma may improve performance for complicated, multi-elemental targets. In the case of overlapping lines these high resolution broadband systems have advantages in being able to simultaneously resolve neighboring lines and observe a multiline signature. These advantages will be especially useful for the case of conventional, mj pulse energy LIBS applications. Whether this advantage is sufficient to overcome the overall lower responsivity of these systems will depend on the individual case of interest. However, in the case of μlibs, the use of such systems to record multiple lines of one species may not lead to an increased sensitivity for detection since these systems have an overall NESR or NEISR significantly higher than an optimized Czerny Turner spectrometer system. Any advantage of multiple line identification may be lost due to the poorer detection of each line. In addition, the very low plasma luminosities observed in μlibs will present a further challenge to the use of the broadband systems. 7. Conclusion In this paper, a method for the assessment of LIBS detector systems has been developed, and used to define responsivity, noise-equivalent spectral radiance and noise-equivalent integrated spectral radiance in terms useful for the LIBS experimentalist. Using this treatment, four LIBS detection systems have been characterized for responsivity, NESR and NEISR. A detailed noise characterization of the intensified systems has been carried out with particular attention to cathode noise spikes in terms of arrival rate, area and width. These factors should be considered for the optimum selection of a detector/spectrometer system which include the detector responsivity, the spectrometer luminosity and the NESR and NEISR. NESR should be used when detector acquisition periods are long, and NEISR for short acquisitions. Of the systems studied here, ICCD/Czerny Turner (system B) has the best responsivity and luminosity. In terms of luminosity, the ICCD/Czerny Turner is followed closely by the IPDA/Czerny Turner (system A), with both the ICCD/Echelle (system C) and the Multichannel CCD spectrometer (system D) exhibiting much lower luminosities. In terms of responsivity, the

11 M.T. Taschuk et al. / Spectrochimica Acta Part B 63 (2008) ICCD/Czerny Turner system is closely followed by the ICCD/ Echelle, with the IPDA/Czerny Turner and Multichannel CCD spectrometer exhibiting much lower responsivities. However, the Multichannel CCD spectrometer system is not intensified, so the lower responsivity is not surprising. For acquisitions at low gain, the NEISR was found to range from Jsr 1 cm 2 nm 1 for the ICCD/Czerny Turner system, to Jsr 1 cm 2 nm 1 for the ICCD/Echelle system. For acquisitions at high gain, the NESR was found to range from Wsr 1 cm 2 nm 1 for the ICCD/Czerny Turner system, to Wsr 1 cm 2 nm 1 for the IPDA/Czerny Turner system. A very important trade off between spectral resolution and bandwidth was not investigated in this work. For the ICCD/ Echelle and Multichannel CCD spectrometer systems it is possible to acquire spectra with high spectral resolution and high spectral bandwidth which may give additional advantages for simultaneous identification of multiple species and observation of multiple lines for the identification of individual species. These advantages will be more useful for conventional LIBS applications, where multiple mj laser pulses are used. In the case of μlibs applications, it is expected that the low luminosity of the plasma combined with higher noise levels of the broadband systems may limit their utility. The detector/spectrometer characterizations performed here are only a guide to the performance of the class of instrument, and specific detector/spectrometer combinations will have different results than those reported here. In particular, the Multichannel CCD spectrometer system studied here is a prototype, and with further development the performance of such a system may improve. Acknowledgments The authors gratefully acknowledge financial support for this research from both MPB Technologies Inc. and the Natural Sciences and Engineering Research Council of Canada. References [1] H.E. Bauer, F. Leis, K. Niemax, Laser induced breakdown spectrometry with an Echelle spectrometer and intensified charge coupled device detection, Spectrochim. Acta Part B 53 (1998) [2] S.R. Goode, S.L. Morgan, R. Hoskins, A. Oxsher, Identifiying alloys by laser-induced breakdown spectroscopy with a time-resolved high resolution Echelle spectrometer, J. Anal. At. Spectrom. 15 (2000) [3] V. Detalle, R. Heon, M. Sabsabi, L. St-Onge, An evaluation of a commercial Echelle spectrometer with intensified charge-coupled device detector for materials analysis by laser-induced plasma spectroscopy, Spectrochim. Spectrochim. Acta Part B 56 (2001) [4] P. Fichet, D. Menut, R. Brennetot, E. Vors, A. Rivoallan, Analysis of laserinduced breakdown spectroscopy of complex solids, liquids and powders with an Echelle spectrometer, Appl. Opt. 42 (2003) [5] Y. Talmi, Intensified array detectors, in: J.V. Sweedler, K.L. Ratzlaff, M.B. Denton (Eds.), Charge-Transfer Devices in Spectroscopy, VCH Publishers, Inc., New York, NY, 1994, pp [6] S.B. Howell, Handbook of CCD Astronomy, Cambridge University Press, [7] J.R. Janesick, Scientific Charge-Coupled Devices, SPIE Press, [8] D.C. O Shea, Elements of Modern Optical Design, John Wiley & Sons, [9] J.F. James, R.S. Sternberg, The Design of Optical Spectrometers, Chapman and Hall Ltd., [10] E.L. Dereniak, D.G. Crowe, Optical Radiation Detectors, John Wiley & Sons, [11] M. Taschuk, I. Cravetchi, Y.Y. Tsui, R. Fedosejevs, Scaling to millijoule energies for laser induced breakdown spectroscopy of water samples, Proceedings of the 2002 IQEC/LAT, SPIE 5149, 2002, pp [12] G.W. Rieger, M. Taschuk, Y.Y. Tsui, R. Fedosejevs, Laser induced breakdown spectroscopy for microanalysis using sub-millijoule UV laser pulses, Appl. Spectrosc. 56 (2002) [13] M.T. Taschuk, Quantification of Laser-Induced Breakdown Spectroscopy at Low Energies, Ph.D. thesis, University of Alberta, Edmonton, Alberta, Canada (2007). [14] J. Aitchison, J.A.C. Brown, The Lognormal Distribution, Cambridge University Press, [15] M.T. Taschuk, S.E. Kirkwood, Y.Y. Tsui, R. Fedosejevs, Quantitative emission from femtosecond microplasmas for laser-induced breakdown spectroscopy, J. Phys.: Conf. Ser. 59 (2007)