MOLECULAR SPECTROSCOPY

Indium Arsenide NIR Array Spectroscopy

G EORGE A. GASPARIAN AND H ARTMUT L UCHT

pectrometers measure the light en- plifier. Longer arrays having hundreds to ergy absorbed or reflected by a ma- thousands of pixels are primarily oper- Electronic terial. Prism spectrometers are the interface ated in a serial-out arrangement, in oldest (~ 2000 years) instruments which each pixel is amplified off-chip Swhereby color measurements were and sequentially read-out. The photodi- made using the eye as the photoreceiver. ode operates in the integrating mode Diode array William Herschel used a prism spec- Grating rather than direct detection, wherein a trometer in the 18th century to examine capacitive transimpedance amplifier is Optical light beyond the red with the first pho- fiber input the first-stage amplification circuit. Fig- ton detector: a blackened thermometer, ure 2 illustrates the detection format. now known as a bolometer. Since Her- These self-scanned linear arrays in- schel and Newton’s time, spectroscopy Slit clude on-board signal analysis electron- has evolved as a separate field of sci- ics that perform operations such as cor- ence. An essential factor to this growth related double sampling, a method to has been the advancements in photore- Figure 1. Diagram of a reflection-grating improve noise performance. Figure 3 il- ceivers. This paper concentrates on the spectrometer with a photodiode array. lustrates an InGaAs photodiode array hy- near- (near-IR) region of the bridized to a silicon readout integrated spectrum between 800 and 2200 nm (ap- circuit. Hybridization is the principal ar- proximately where silica-based optics be- ray fabrication technique for nonsilicon come opaque). Indium materials such as indium antimonide, (InGaAs) is the best material for a solid- VSS mercury cadmium telluride, indium ar- state detector, offering the highest sensi- senide, and InGaAs. tivity in the near-IR. It has become the Light The array operation is straightfor- Reset detector of choice for near-IR spec- ward: a predetermined exposure time is S/H1 troscopy. specified via a clock pulse, and the inci- Scanning monochrometers (prisms dent photons produce electron-hole and diffraction gratings) with single- S/H2 pairs in the InGaAs absorption layer. element have been the VREF These photocarriers generate a pho- mainstay in spectroscopy. Using silver- tocurrent that is processed by the off- emulsion film technology, a complete chip capacitive transimpedance ampli- spectrum (visible–short wave–near-IR) Figure 2. Schematic of an integrating fier into a measurable voltage. After was obtained without the need of a mov- transimpedance amplifier. exposure, the array circuit serially out- ing dispersive element (hence, the name puts each element. A detailed descrip- spectrograph was coined). This allowed tion is beyond the scope of this note. Ju- the spectrometer to be moved into re- in agricultural, food, pharmaceutical, and dicious selection of the source light and mote locations and initiated what be- petrochemical firms. Figure 1 illustrates spectrometer (specifically, diffraction came known as multichannel spectral a reflection grating spectrometer with a grating and optics) is necessary to opti- analysis. With technology photodiode array. mize the detection process. The photo- advances, linear and matrix photodiode Linear detector arrays are available diode array’s mechanical, optical, and arrays are becoming the detector of with thousands of individual photodiode electrical parameters are interdepen- choice for remote monitoring. Real-time elements (pixels) in a convenient, small dent and affect system performance. measurements of key on-line parameters package. Arrays having less than 64 ele- The remainder of this article will high- are performed in process control, using ments can be operated in the parallel out light the performance parameters and diode-array spectroscopy. Uses of re- configuration, in which each pixel is indi- operating techniques for a grating- mote diode-array spectroscopy are found vidually addressable with a separate am- based InGaAs photodiode array...... MOLECULAR SPECTROSCOPY WORKBENCH ......

tinction must be made between linear eight (relative to 20 °C) and the dark response and saturated response. As the noise is reduced by a factor of 2.83 (or photoelement approaches saturation, it 883-e noise level). However, if the readout can produce a nonlinear response noise is 2000 e, the root-sum-square noise (supraresponsivity). The preferred mode is 2186 e, a 9% noise reduction. The extra of operation of the array is radiometric, effort required to cool the array for a 9% common to most spectrometers. improvement might not be cost effective. The lowest signal level is defined by Techniques are available to reduce the ef- the silicon multiplexer’s readout noise. fects of dark signal such as offset correc- Neither the photon shot noise nor dark tion, dark field subtraction, and multiple Figure 3. An InGaAs photodiode array hy- signal shot noise is considered here. scans at shorter integration times. bridized to a silicon readout integrated The latter exclusion may seem unusual. The charge capacity in a capacitive circuit. Normally, the dark signal limits the per- transimpedance amplifier is determined formance; cooling the array enhances from the integrating capacitance output the dynamic range. Sometimes this voltage product. Diode arrays are avail- The principal mechanical parameters helps, but there is no point in cooling an able with multiple capacitors selectable are array length, center-to-center spac- array such that the dark noise is much to the specific application (for example, ing, element geometry, fill factor, and below the silicon multiplexer’s read high sensitivity and high dynamic operability (number of nonworking pix- noise. The noise is uncorrelated, hence, range). Consider the following example: els). Photodiode-array lengths are the root-sum-square defines the total the amplifier output is 2 V with an inte- paired to the length (and availability) of noise (see the following example). Per- grating capacitor equal to 10.4 pF, the silicon readout integrated circuit (or forming at levels much below the read hence, the maximum charge capacity is multiplexer). Presently, commercial In- noise (temperature independent) does 20.8 pC, corresponding to 130 Me. Be- GaAs photodiode arrays are available not help. Meanwhile, much money can cause we are measuring voltages, the that range in size from 128 elements to be spent on a cooling system. Dynamic transimpedance gain is calculated to be 512 range is also used to determine the level 2 V/130 Me, which equals 15.4 nV/e or elements. The pitch or center-to-center of digitization and whether the operation 11 nV/photon. Thus, the array’s signal- spacing has been standardized to 50 is optimized for high sensitivity or high to-noise ratio is simply the maximum m; hence, array lengths are between dynamic range. Selection will be based charge capacity (sometimes called “full 6.4 mm and 25.6 mm. The pixel geome- on the spectrometer and application. well”) and shot noise level quotient. try is rectangular with available heights The pixel dark signal is defined in units The photodetector spectral response ranging from 50 m to 1 mm. Tradeoffs of volts per second to rapidly determine defines the maximum spectrometer free- between size (area) and signal-to-noise allowable exposure times as well as the spectral range. A single photodiode ar- ratio will determine the element selec- more familiar definition of dark current in ray covering the 800–2200 nm spectrum tion. amperes. It is due to the nonzero input (allowable transmission band for silica- InGaAs photodiode arrays have a 100% offset voltage of the multiplexer’s capaci- based lenses and optical fiber) is not fill factor (ratio of active photoresponse tive transimpedance amplifier. This will ei- available. Two InGaAs compositions area to element area). Thus, maximized ther forward or reverse bias the photodi- have been commercially developed that light collection efficiency is realized ode producing a dark current. cover the 800–1700 nm and 1100–2200 compared with arrays having less than Additionally, this bias level can be differ- nm ranges. Selection of a particular unity fill factor. Zero dropout arrays are ent for each element, which produces wavelength range is dependent on the common with 128 and 256 elements. Sys- dark signal nonuniformity — a compo- application and tradeoffs associated with tem performance depends on the coordi- nent in fixed pattern noise (defined later the instrument performance. nation of the components to match the in this article). In most cases, the dark Exposure time and readout time for array of the spectrometer, based on the signal is the dominant noise generator, es- diode arrays are temporally separated. application. Figure 4 is a photograph of pecially during long exposures. An In- The photodiode array, including the the InGaAs photodiode array illustrating GaAs photodiode element can have a multiplexer, is first exposed (integration a 0.5-mm pixel on a 50-m pitch. dark current of 1 pA, corresponding to time), then deactivated and serially The linear array is characterized by 6.25 Me/s. follow Poisson read-out. Commercially available photo- its opto-electronic performance parame- statistics and the rms noise value is the diode arrays can have second expo- ters: dynamic range, dark signal, maxi- square root of the measured output. sure and readout times. Exposure times mum output voltage (transimpedance Thus, a 1-s exposure will produce a 6.25 are electrically generated (no mechani- gain), spectral responsivity, and maxi- Me dark signal having an rms noise level cal shutter) via the control . mum exposure and minimum (fastest) of 2500 e. Cooling the photodiode array Readout times are specified by the num- readout time. Dynamic range can be de- will decrease the dark current and the ber of samples read per second. Thus, a fined as the maximum usable output noise. InGaAs photodiode dark signal is 512-element array operating at 2.0 voltage divided by the root-mean-square temperature dependent with the dark cur- 107 samples/s has a 26 second read- (rms) multiplexer readout noise. There rent doubling or halving with each out rate. The scan time is the sum of ex- may be a perceived problem with this decade of temperature. Operation at 10 posure and readout time; scan times of definition of usable output voltage. A dis- °C reduces the dark current by a factor of 50 seconds (frame rates of 20 kHz) are ...... MOLECULAR SPECTROSCOPY WORKBENCH ......

charge amplifier produces reset noise ple shorter exposures. Reciprocity refers caused by its operation (resetting the to the accumulation of signal charge. capacitor for the next integration). This Care must be taken when analyzing the noise must be eliminated for accurate effect on the various components of the measurements. The design of the multi- noise charge. This is another technique plexer is such that sample/hold circuits to improve signal-to-noise ratio when the are used to hold the signal immediately dark current is large. after reset and then to subtract it from In conclusion, diode-array spec- the signal obtained at the end of the ex- troscopy is becoming the instrument of posure time. Reset noise is superim- choice for sensing in the process con- posed on both the dark signal (sam- trol industry. Its advantages — robust- Figure 4. A linear InGaAs photodiode array. ple/hold 1) and the accumulated light ness on the factory floor, nonmoving plus dark signal (sample/hold 2). Sub- parts, and rapid collection of spectra — traction yields only the light signal. provide the incentive to further look at Fixed-pattern noise is caused by this technology in areas that were pre- possible. Because exposure times affect nonuniformities in the response of the viously the domain of the scanning signal level, dark signal, photon shot photodiode array and multiplexer elec- spectrometer. A real-life application will noise, dark noise, and offset level, the tronics capacitive transimpedance ampli- be presented in the second installment specific exposure time in a spec- fier that distort the electrical output. of this discussion. troscopy application will depend on the This noise is nonrandom and remains experiment. stable from one scan to the next scan. Diode-array spectroscopy uses opti- Two factors contribute to fixed-pattern George A. Gasparian works at Sensors mization techniques to enhance the noise — dark signal nonuniformity and Unlimited, Inc., 3490 US Route 1, performance and reduce the degrada- photo response nonuniformity. Gain and Bldg. 12, Princeton, NJ 08540. tion caused by nonuniformities that offset correction through flat-fielding He can be reached at (609) 520-0610, generate fixed pattern noise, reduce and dark-fielding are the mechanisms for ext. 225, fax: (609) 520-0638, amplifier-based noise through corre- reducing the effects. e-mail: [email protected]. lated double sampling, and use the reci- The principle of reciprocity simply Hartmut Lucht works at LLA Umwelttech- procity law to detect weak signals. Cor- states that the total signal charge is the nische Analytik und Anlagen GmbH related double sampling was product of the photocurrent and the total (Rudower Chaussee 6, Geb. 19.1, 12484 mentioned earlier as a mechanism to exposure time. It does not matter if the Berlin, Adlershof, Germany). He can be reduce the noise caused by the capaci- total exposure time is achieved using a reached at (0) 30 6719 8376, fax: (0) 30 tive transimpedance amplifier. The single long exposure or by adding multi- 6392 4766, e-mail: [email protected].

©Reprinted from SPECTROSCOPY, February 2000 AN ADVANSTAR PUBLICATION Printed in U.S.A.

Copyright Notice Copyright by Advanstar Communications Inc. Advanstar Communications Inc. retains all rights to this article. This article may only be viewed or printed (1) for personal use. User may not actively save any text or graphics/photos to local hard drives or duplicate this article in whole or in part, in any medium. Advanstar Communications Inc. home page is located at http://www.advanstar.com.