A review of photodetectors for sensing light-emitting reporters in … · 2008-02-22 · A Review...

288 IEEE SENSORS JOURNAL, VOL. 3, NO. 3, JUNE 2003

A Review of Photodetectors for SensingLight-Emitting Reporters in Biological Systems

Rachel A. Yotter, Member, IEEE,and Denise Michelle Wilson, Member, IEEE

Abstract—A review of photodetectors for optical detection inbiological applications is presented. The intent is to provide anoverview of the performance metrics and trade-offs among pop-ular photodetectors in order to facilitate an easier match amongthe photodetector, biological stimulus, and optical pathway. Thecharacteristics and nonidealities of fluorescent and phosphores-cent reporters, and the properties of optical components such asfilters, lenses, and light sources, are reviewed. By accounting forsources of noise in these components, it is shown how to deter-mine metrics for the optical system that can then be convertedto photodetector metrics. Defined photodetector metrics includethe quantum efficiency, responsivity, noise-equivalent power, de-tectivity, gain, dark current, response time, and noise spectrum.The operating principles and performance trade-offs of photode-tectors are reviewed, and emphasis is placed on photodetectorsfor integrated compact systems. Top commercial candidates forphotodetectors for detecting light emitted from reporters are thephotomultiplier tube, avalanche photodiode, and charge-coupleddevice. Focus is placed on new developments in complementarymetal-oxide-semiconductor structures that can provide low-cost,low-power, and low-voltage alternatives to traditional approachesto biological imaging. Reviewed device structures are presented inthe context of supporting the development of laboratory-based in-struments and compact fully-integrated systems.

Index Terms—Avalanche photodetectors, biosensors,charge-coupled devices, complementary metal-oxide-semi-conductor (CMOS) photodetectors, dark current, detectivity,fluorescence, integrated optical systems, noise, noise bandwidth,noise spectrum, noise-equivalent power, phosphorescence, pho-todetector, photodiodes, photomultiplier tubes, power spectraldensity, quantum efficiency, response time, responsivity, review.

I. INTRODUCTION

PHOTODETECTORS used to detect light-emitting re-porters operate under a different set of restrictions than

those used for other common applications. Sensitivity plays apredominant role in the selection of photodetectors or arraysof photodetectors for interpreting events, molecules, proteins,DNA, or other biological segments tagged with fluorescentor phosphorescent probes. Secondary to sensitivity, biologicalsensing requirements include the wavelength range of thelight, response time, and the intensity or number of photonsgenerated in a unit time. Matching the spectral peak of thephotodetector to the application is often an assumed require-ment to maximize application sensitivity. Often, a significant

Manuscript received May 16, 2002; revised November 12, 2002. Thiswork was supported by the National Institute of Health under Award P50HG002360-01. The associate editor coordinating the review of this paper andapproving it for publication was Dr. Gerard L. Coté.

The authors are with the Department of Electrical Engineering, Universityof Washington, Seattle, WA 98195-2500 USA (e-mail: [email protected]).

Digital Object Identifier 10.1109/JSEN.2003.814651

portion of the light intensity from the fluorescing agent iseliminated prior to photodetection because of emission filtersand other elements of the optical train that, in the processof removing interfering light contributions, also remove asignificant portion of the signal. In contrast to imaging forbiosensing, conventional visual imaging requires high spatialresolution, wide dynamic range, and moderate sensitivity inorder to produce the most (spatially) detailed image across awide range of background illumination conditions. Very highsensitivity is usually not required and individual photodetectorresponse times in an imager are often limited by the abilityto scan an entire image within the 30 frame per second limitof human visual capability. Desired spatial resolution is un-bounded in these imagers, limited only by the space consumedby the imager, the amount consumers are willing to pay, andthe frame-level response time incurred by ever-increasingnumbers of pixels on the focal plane. This paper focuses on thedesign issues relevant for optical systems and detectors used todetect light-emitting reporters, including design trade-offs andmethods to match photodetector performance with the opticalsystem. The operating principles of common photodetectorsfor these systems are reviewed, and emphasis is placed on newalternative methods that are especially suited for low-power,fully-integrated biological systems.

II. BACKGROUND

Photodetectors fundamentally operate on the transition of anelectron from a lower energy state to a higher energy state asa result of the absorption of a photon. The energy transitioncan usually be classified into one of the following categories.1) Photoconductive or photovoltaic: The electron undergoes atransition from the valence band to the conduction band. In pho-toconductive devices, photons generate carriers that lower theresistance of the device. Photovoltaic devices include a metal-lurgical junction in which photons generate a voltage across thedepletion region. 2)Photoelectric (photoemissive): The electronundergoes a transition from the conduction band to a vacuum.One electron is released into the vacuum per photon of sufficientenergy. 3)Polarization: The electron undergoes a transition to avirtual energy state (as in index of refraction changes and otherpolarization effects). 4)Phonon generation: The electron under-goes a transition to midgap states and back to an initial relaxedlevel. This is equivalent to heat generation. 5)Other: The energyis converted to other forms via mechanisms such as excitons.

In addition to classification related to the nature of electrontransitions, electronic photodetectors can be also classified intotwo broad categories based on the efficiency of transition: direct

1530-437X/03$17.00 © 2003 IEEE

YOTTER AND WILSON: REVIEW OF PHOTODETECTORS 289

TABLE ICOMPARISON OFPHOTODETECTORS(T = 25 C)

and indirect. Direct photodetectors convert the incident photonsdirectly into an electronic signal without any intermediatetransduction stages. These detectors include photoemissive,bulk photoconductive, or junction-based (photovoltaic) devices.Direct detectors all operate using three basic processes: carriergeneration by incident light, carrier separation, transport,and/or multiplication, and collection of the current by anexternal circuit [1]. Indirect optical sensors, on the other hand,involve at least two transduction stages and generally do notprovide the sensitivity in the visible light range to be applicableto biosensing applications.

Systems that detect light-emitting reporters rely predomi-nantly on the use of the (photoemissive) photomultiplier tube(PMT), cooled charge-coupled devices (CCDs), or the (junc-tion-based) avalanche photodiode (APD) for high sensitivitydetection. Table I compares the performance of these threepopular choices with other common photodetectors. To providea broader perspective on photodetectors for biological applica-tions, we review these three devices as well as some alternativedevices in the context of sensing in a biological application.Alternative choices for biosensing applications include theconventional p-n and p-i-n photodiodes and several specializedcomplementary metal-oxide-semiconductor CMOS structures.The devices reviewed in this paper all demonstrate detectionlimits compatible with detecting small numbers of fluorescingor phosphorescing reporters in bioanalysis applications.

III. OPTICAL SYSTEMS FOR DETECTING

LIGHT-EMITTING REPORTERS

For biological applications that use light-emitting reporters,the optical system is designed based on the characteristics ofthe reporter. Light-emitting reporters are characterized by theirabsorption and emission spectra and quantum efficiency. Thecomponents of the optical system, such as filters, lenses, lightsources, and photodetectors, are then chosen based on the char-acteristics of the reporter. It is beyond the scope of this paper tothoroughly review optical systems or reporters, and the systemis only briefly described to show how the characteristics of the

optical components can be used to calculate a set of metrics thatcan be related to metrics used to describe photodetectors.

A. Light-Emitting Reporters

A light-emitting reporter functions by emitting light when ex-cited by a stimulus, and it is usually a fluorescent or a phos-phorescent agent. An incoming photon of a wavelength withinthe absorption spectrum is absorbed by a reporter to excite anelectron to a higher energy level. The electron eventually re-turns to a lower energy state and emits one or more photonswhile doing so. The difference between fluorescent and phos-phorescent agents is that fluorescent agents emit almost imme-diately after absorbing a photon while a noticeable delay occursbetween absorption and emission for phosphorescent agents.Since the emitted photons are lower in energy than the absorbedphotons, the emitted photons have a longer wavelength, whichis reflected in the difference between absorption and emissionspectra (Fig. 1). More details on this process can be found in [6]and [7].

The light output of reporters is characterized by the absorp-tion and emission spectra, by the quantum yield, and by theextinction coefficient. An example of absorption and emissionspectra is shown in Fig. 1. The extinction coefficient is thephoton absorption per molecule and characterizes how wellthe reporter absorbs photons. Quantum yield characterizes howwell the excited reporter emits photons, and it depends on thewavelength of the incident light, temperature, and other factorssuch as pH or the molecules to which the reporter is attached.By knowing the concentration of the optically active reporter,the intensity of the incident light, the extinction coefficient, andthe quantum yield, the number of emitted photons per unit timecan be roughly calculated. Other factors that will affect thiscalculation include photobleaching (irreversible destruction ofthe fluorescence process which occurs more rapidly for highintensities of incident light), quenching (chemical interferencewith the light-emitting process), and the absorption of emittedphotons by other reporter molecules. The optical system shouldbe designed so that the incident light on the light-emittingreporter is of high enough intensity to generate a detectable


Fig. 1. Absorption and emission spectra of a light-emitting reporter. Shownis a typical example of the absorption and emission spectra of a light-emittingreporter. (dashed line) The absorption spectrum curve is at the shorterwavelengths, and (solid line) the emission spectrum displays the characteristicshift to longer wavelengths. The overlap between the two spectra is common,and a source of signal loss includes emitted photons being reabsorbed by aneighboring reporter molecule. The intensity of the two spectra arc normalizedin this figure. The emission spectrum is usually lower in intensity than theabsorption spectrum.

signal at the end of the optical train, yet of low enough intensityto reduce photobleaching effects to an acceptable level. Theincident light should also be within a wavelength range thatmatches the absorption spectrum.

B. Optical System

Optical systems for detecting light-emitting reporters can becategorized as either imaging systems or light-collecting sys-tems. Light-collecting systems are designed to collect the lightsignal with a minimum of loss. Although imaging systems areusually designed to efficiently collect light as well, the optics inthese systems are more complex since spatial information alsoneeds to be preserved with a minimum of distortion. A thor-ough handling of imaging systems is beyond the scope of thisarticle, and only the issues of the light-collecting system willbe reviewed. Note that issues involved in designing a light-col-lecting system often apply to imaging systems as well.

The layout of a typical light-collecting system consists of atleast a light source, filters to select particular wavelengths or re-duce the light intensity, lenses to focus the light onto the sampleor detector, lenses to collimate the light, screens to shield thesystem from stray light, a sample holder, and one or more pho-todetectors. When designing an optical system to detect biolog-ical reporters, the elements need to be chosen in considerationof the characteristics of the reporter.

For compact, low-power implementations, the light sourceoften has a continuous spectrum that emits at optical frequen-cies within the absorption wavelength range of the reporter. Ingeneral, light sources either emit a continuous spectrum or mayonly emit at discrete wavelengths. Sometimes the light sourceincludes a monochronometer to convert a continuous spectruminto discrete wavelengths, but this increases the cost and size ofthe source. Furthermore, the light output can be designed to ei-ther be continuous or to emit short bursts of light [8].

Fig. 2. Selection of absorption and emission filters. The cutoff frequencyof the absorption and emission filters are chosen so that there is no overlapbetween the transmission spectra. Shown is (dotted lines) the originalabsorption and emission spectra and (solid lines) a theoretical choice in thetransmittance spectra for the excitation and emission. The cutoff of a filter isdefined as the point where light is at 50% of maximum transmittance. Becausethe filters require a wavelength gap between their cut-offs due to a slightlysloping cutoff line and to account for shifts due to temperature or other effects,the wavelengths between the two regions are filtered out and are, thus, a sourceof signal loss.

A bandpass or low-pass filter, called an excitation filter, canlimit the wavelengths from the light source to only those arewithin the absorption spectrum. To reduce photobleaching, thelight intensity may also need to be reduced with a filter. Gener-ally, the function of filters is to modify the intensity of radiationbased on the wavelength of light. The three common types offilters are those that have a sharp cutoff between the wavelengthranges that are transmitted and those that are blocked, filtersthat have a gradual change between regions, and neutral densityfilters that reduce the intensity of light for all wavelengths [9].Sharp cutoff filters, the most common filter in optical systemsused to detect reporters, can be further classified as either beingshort-passif it transmits light at shorter wavelengths,long-passif it transmits light at longer wavelengths, orbandpassif it trans-mits or blocks a limited range of wavelengths. These filters arecommonly either colored glass or interference filters. The ad-vantage of colored glass filters over interference filters is that thefilter properties are insensitive to the angle of incidence, whilein interference filters, the cutoff shifts to shorter wavelengthsas the angle of incidence is moved away from the normal. Thespectral properties of both filters have a slight dependence ontemperature [10].

The sample (held in a cuvette or other holder) is excited withthe light emitted from the excitation filter. The reporter absorbsphotons and emits light in all directions, and only part of thislight can be gathered [6]. It is passed through a long-pass orbandpass filter, called an emission filter, that filters out the ex-citation light from the light source. Fig. 2 shows the transmit-tance of absorption and emission filters suitable for the reportershown in Fig. 1. Along the pathway, the light may be collimatedusing a collimating lens or focused using a focusing lens. Theselenses change the optical pathway of the light to reduce loss be-tween optical elements or optimize sensitivity. To travel longer


distances with minimal loss, fiber optics are often used betweenoptical elements. More information about lens design and theprinciples of fiber optics can be found in [11].

The final emitted signal must be properly coupled into thephotodetector. Antireflective coatings at the photodetector/airinterface can reduce scattering effects due to reflection and re-fraction. The size of the light beam must also match the activearea of the photodetector. Theactive areaof a photodetector isthe region of the photodetector that is capable of detecting light.Any light that falls outside this region will not be detected.

In addition to the losses due to the properties of the light-emit-ting reporter, primary sources of signal loss from the opticalsystem arise from filtering and scattering. Because there canbe no overlap between the absorption and emission bands, theemission filter usually filters out a small part of the emissionspectrum, and all light that is emitted within this range is lost.Filters also slightly reduce the intensity of the light that is withinthe transmission wavelength band. Scattering is a general termfor light that has deviated from the intended path [12]. Onecommon source of scattering is reflection, which is caused by adifference in refractive indices between elements along the op-tical train. Proper coupling between elements can reduce thiseffect.

Sometimes noise sources introduce excess light rather thanattenuate it. Stray light is an obvious noise source and can bereduced through proper shielding. Also, many materials exhibitfluorescent properties, or theyautofluoresce[5]. Since the spec-trum of the light arising from autofluorescence is shifted, it maynot be effectively filtered before reaching the detection stage.

C. Metrics for the Optical System

In order to properly select a photodetector, a set of metricsneeds to be calculated based on the choice of optical systemand reporter. To calculate this metric, the properties of both theoptical components and the reporter must be known. First, thespectral characteristics of the light incident on the reporter canbe calculated by accounting for the spectral characteristics ofeach element along the excitation optical train

(1)

where is the intensity as a function of wavelength for thelight striking the sample, is the transmission of the exci-tation filter, is the emission spectrum of the light source,

accounts for scattering effects along the excitationoptics, and accounts for noise such as stray light andautofluorescence along the emission optics. The extinction co-efficient and quantum yield can be used to calculate how muchlight is emitted by the sample

(2)

where is the spectrum of the emission from thereporter, is the quantum yield of the reporter (whichdepends on temperature and other factors), and is the ab-sorption coefficient of the reporter. Finally, the spectral charac-teristics of the emission optics can be accounted for

(3)

Fig. 3. Output spectrum. (Solid line) Shown is a theoretical output spectrumof an optical system. This is the light signal that reaches the surface of thephotodetector. The spectral range matches the emission spectrum of thelight-emitting reporter, but the signal has been attenuated due to filters in theoptical pathway. Other effects such as autofluorescence and scattering hasslightly modified the wavelength content of the signal. The output spectrumcan be used to determine an appropriate photodetector for the system.

where is the transmission of the emission filter,accounts for scattering along the emission optics, ac-counts for noise along the emission optics, and is the finalspectrum that reaches the photodetector. Note that these calcu-lations have been somewhat simplified by ignoring angular de-pendencies. The final result is a curve showing the intensity oflight over a range of wavelengths, like that shown in Fig. 3.

Two factors must be considered in choosing a photodetectorto best match the output spectrum, once known. First, the re-sponsivity or quantum efficiency should be matched to the shapeand magnitude of the output spectrum. Second, the minimumdetectable signal of the photodetector (determined by the noise-equivalent power or dark current metrics) must be smaller thanthe peak of the output spectrum. The next section reviews per-formance metrics used to describe photodetectors and showshow to use the information in Fig. 3 to predict the performanceof a photodetector using these metrics.

IV. PERFORMANCEMETRICS FORPHOTODETECTORS

This section reviews the metrics used to characterize theperformance of a photodetector. The objective is to translatebetween the output spectrum of the optical system (e.g., Fig. 3)to common photodetector metrics. Briefly, responsivity andquantum efficiency determine how well light is transducedinto an electrical signal based on wavelength; gain determineshow much the incoming light signal is amplified by inherentmultiplication of the photodetector; noise-equivalent power,detectivity, and dark current account for noise in determiningthe final transduced signal; response time characterizes thespeed of detection; noise spectrum characterizes the frequencycontent of noise. In this section, the optical spectrum calculatedin the previous section will be related to quantum efficiency.Then the responsivity, noise-equivalent power, and detectivitywill be defined in terms of their relation to quantum efficiency.


Finally, additional information provided by gain, dark current,response time, and noise spectrum are reviewed.

A. Quantum Efficiency

Quantum efficiency is defined as the number of carriersgenerated per incident photon by a photodetector. The internalquantum efficiency is the number of pairs created divided bythe number of photons absorbed and is usually very high in puredefect-free materials. The external quantum efficiency accountsfor only the photon-generated carriers collected as a result oflight absorption. Since it reflects the useful portion of signal cre-ated by interaction of light and photodetector, external quantumefficiency is a more relevant parameter for photodetector char-acterization than internal quantum efficiency. External quantumefficiency is defined as the number of carriers collected dividedby the number of incident photons

(4)

where is the photodetector current generated in response toincident light, is the electron charge, and is the incidentoptical power in watts. The quantity is the energy per photonin Joules, where is Planck’s constant and is the frequency.The frequency is related to wavelength through the well-knownrelationship , where is the wavelength in a vacuumand is the speed of light. Note that if light is propagating in amedium other than a vacuum, the frequency remains the samewhile the velocity and wavelength change, and the change in ve-locity is determined by the refractive index, or , where

is the velocity and is the refractive index [13].The external quantum efficiency depends on the absorption

coefficient of the material and the thickness of the absorbingmaterial. Assuming no reflection of light from the surface of thephotodetection material, if the photon flux density at the surfaceis , then the photon flux at depthis given by Beer’s law:

(5)

where is the absorption coefficient, in cm. Note thatcan be related to the optical power, , by the relation

, where is the active area of the photodetector.is astrong function of wavelength and varies greatly depending onthe material. The electron-hole pair generation rate at a depthfrom the illuminated side of the semiconductor layer (the deple-tion region in a junction-based device) is given by

(6)

in cm s , where is the internal quantum efficiency. Byknowing the depth to which photons are likely to penetrate, thedetector can be optimized for particular wavelengths by phys-ically matching the detection region to regions of highest ab-sorption. For example, to optimize a silicon structure for de-tecting blue light, the detection region should be shallow andclose to the surface. For red light, though, better performancemay be achieved by creating a wider lightly doped depletion re-gion deeper in the device. Furthermore, if there is an overlyinglayer above the photodetection device (e.g., the passivation layer

often included for CMOS devices), the absorption characteris-tics of this layer will result in signal attenuation. This attenu-ation depends on the wavelength of the incident light and candrastically reduce the expected performance of a device.

Reflection of light from the surface of the material is anothersource of signal loss. Most semiconductors have a high refrac-tive index, typically 3 to 3.5. The difference in refractive indicesacross interfaces between the photodetector and the surroundingenvironment can cause significant refraction and reflection. Ofthe light flux falling onto any air-semiconductor interface, usu-ally 30 to 40% is reflected. A well-designed antireflective (AR)coating can reduce or even eliminate this loss [13]. In summary,these sources of loss relate the external quantum efficiency tothe internal quantum efficiency, and this relation can be mathe-matically expressed by

(7)

where is the optical reflectivity at the interface [14].In addition to the above losses, the detection efficiency of a

photodetector will also be affected by poor matching betweenthe active area and the size of the light signal beam, materialdefects, and the band structure of a semiconductor material.The band structure of a semiconductor can either have an in-direct bandgap or a direct bandgap. A direct bandgap impliesthat no phonon (change in carrier momentum) is necessary foran electron to complete the energy transition from the valenceband to the conduction band. The transition is then a one-par-ticle process, and the probability of the transition is increased.If a phonon is required, the transition becomes a two-particleprocess and the transition probability drops dramatically [15].

For both direct and indirect bandgap materials, the absorptioncoefficient drops to zero if photons do not have enough energyto excite an electron to the conduction band. For direct bandgapmaterials, the minimum energy of a photon is equal to the energygap of the material. The situation is more complicated forindirect bandgap materials. Since the photon absorption processcan absorb a phonon, the minimum energy is the energy gapminus the phonon energy. A more thorough discussion of bandtheory can be found in [15].

In summary, the limit of absorption at long wavelengths is de-termined by the energy gap (or absorption edge) of the material.At short wavelengths, absorption is limited by the incident op-tical energy being absorbed near the surface of the device whereit is more difficult to detect. Because of fabrication constraints,it is usually difficult to place the detection region precisely atthe surface of the device. For example, silicon forms a thin sil-icon dioxide layer at surfaces exposed to air, and photons at shortwavelengths can be absorbed in this layer and result in no signalgeneration.

Because of these sources of loss, no device has perfect ex-ternal quantum efficiency. A typical range is between 20% and95%, depending on the device material, photon wavelength, andsensing mechanism. Gallium arsenide, a direct bandgap semi-conductor, can have quantum efficiencies higher than 70% [4].On the other hand, silicon, an indirect material, has significantlylower quantum efficiencies within the range of 20% to 40% [4],[5].


In a biological system, if the output spectrum of the opticalsystem is known (shown in Fig. 3), the external quantum effi-ciency can be used to determine the magnitude of the photo-generated current in response to a light signal. Fig. 3 is the op-tical intensity as a function of frequency. For each opticalfrequency over the frequency range of interest, the current thatflows in response to light at that frequency is .The total current can then be calculated by summing all currentcontributions from each optical frequency. For imaging devicesthat collect charge, it is more useful to calculate the number ofphotogenerated electron-hole pairs. Ifis the integration timeof the device, then the number of electron-hole pairs generateddue to light at a specific frequency is . Thetotal number of photogenerated charges is the sum of the con-tributions at each frequency.

Other metrics which describe the sensitivity of a photode-tector, including responsivity and noise-equivalent power, de-pend directly on the external quantum efficiency. If the quantumefficiency is known, the optical frequencies over which the de-tector performs well can be determined, and peak sensitivity canideally be matched to the optical signal produced by the biolog-ical application.

B. Responsivity

Responsivity is defined as the output current divided by theincident light power, or

(8)

where is the output current of the photodetector in responseto light and is the optical power incident on the photode-tector. Note that the equation for responsivity strongly resem-bles that for quantum efficiency, and the two metrics are relatedby a factor of . Thus, responsivity depends directly on theexternal quantum efficiency of the device, and it can be usedwith Fig. 3 to determine the photogenerated current in responseto light emitted from biological reporters. Thespectral responsecurveis a plot of responsivity as a function of wavelength. Thepeak of this curve should match the peak of the emission spec-trum of the optical system for highest sensitivity.

C. Noise Equivalent Power

Noise equivalent power(NEP) is the amount of light requiredto collect a signal equivalent in power to that of the noise in aphotodetecting device. In other words, it is the amount of lightthat results in a signal-to-noise ratio (SNR) equal to 1. In termsof other performance metrics, NEP can be defined as the rmsnoise current divided by the responsivity, and it has the units

Hz. The Hz component arises from the dependency ofthe rms noise current on the noise bandwidth. Note that noisebandwidth and the common 3-dB bandwidth used to describeelectronic circuits are different. Section V contains more detailsabout noise current and noise bandwidth, and a thorough han-dling of the subject of noise can be found in [16].

Usually photodetectors are characterized by a single NEPvalue, but NEP is actually dependent on the frequency of light.The NEP given in datasheets usually corresponds to the respon-

sivity peak of a photodetector. Although NEP is useful for com-paring photodetectors, the lack of spectral response informa-tion means that this value will only give a best-case approxi-mation to how the photodetector will respond to a light signal.If the rms noise current is known, the peak responsivity is

NEP. An upper limit of the photogenerated signalcan be calculated by assuming that the responsivity is constantover the entire operating wavelength range of the photodetector.

D. Detectivity

NEP is proportional to the square root of the detector’s ac-tive area, or . The detectivity, , corrects for the propor-tionality of NEP and is the reciprocal of the NEP [17].In other words, the detectivity provides a representation of thenoise level in a photodetector, independent of the detector area.Mathematically, is given by

NEP(9)

Like NEP, detectivity is a metric that is used to compare pho-todetectors, but is not well-suited for determining the responseof a photodetector to an intensity spectrum.

E. Gain

The gain characterizes any inherent signal multiplication of aphotodetector. Gain is defined as the ratio of the total current thatflows in response to incoming light to the current that flows indirect response to impinging photons (the primary photocurrent)

(10)

where is the gain, is the output current, and is thephoton-generated current. Although the symbolis used here,this is by no means standard and the symbol used for gain varieswidely. In many photodetectors, such as the p-n junction, themaximum possible gain is 1, since no inherent carrier multi-plication scheme exists in the device. In other devices, such asthe PMT and APD, inherent carrier multiplication can result ingains much higher than one. High gain is not without disad-vantages, however, as inherent carrier gain amplifies both pri-mary photocurrent and the noise in the device. The best trade-offbetween gain and noise performance ultimately determines theusefulness of the photodetector device for a particular applica-tion.

F. Dark Current

Photodetectors are usually operated under an applied biaswhich generates an electrical signal even in the absence of light.The generation of this signal arises from material imperfec-tions and operating temperatures above absolute zero, both ofwhich generates carriers under no-light conditions. This elec-trical signal is termed thedark currentof a photodetector. Bydefinition, the dark current is a small current that is present inthe absence of light. In general, a larger applied bias voltageacross the photodetector increases the dark current. Dark cur-rent places a limit on the ability of subsequent detection elec-tronics to detect small signals from the photodetector.


G. Response Time

The response timeis a time constant (in seconds) that ex-presses the time required for a photodetector to go from 10%to 90% of its final response state. Another name for responsetime is rise time. In biological applications which detect events,the response time should be much shorter than the time it takesfor the light-emitting biological event to occur. If the responsetime is longer, the system may not be able to resolve separateevents.

For photodiodes (both p-i-n and APD), the response time isusually dependent either on the size of the active area or theRC time constant of the circuit. The RC time constant is the ca-pacitance C of the photodiode junction multiplied by the loadresistance R (usually a small resistor placed in parallel with thedetector or the small-signal resistance of the photodetector de-vice itself), and can be represented as RC. Response timeis also dependent on how long it takes a photogenerated elec-tron to travel from the active area to one of the output terminals;the larger the active photodetecting area, the longer the responsetime. Overall response time is limited by the slowest process inthe photodetecting process.

For photoemissive devices, the response time is determinedby the electron transit time through the device or the external cir-cuit. The response time for these devices tends to be extremelyfast. Other devices, such as the CCD, integrate the light signaland as a result have a response time that is much slower than thetime required for an incident photon to become a current carrier.In these devices, incident photon effects are accumulated overtime and the response time of the devices is set by how long thedevice accumulates carriers before sweeping them out to detec-tion circuitry. For this reason, for integrating devices it is moreappropriate to describe theframe rateof the array rather thanthe response time. The frame rate describes how long it takes tomove the integrated signal outputs of an entire array of pixelsoff the focal plane.

H. Noise Spectrum

Often the noise of a photodetector depends on the frequencyof operation. Thenoise spectrumis the noise voltage or cur-rent plotted as a function of electrical frequency. By knowingthe electrical frequency at which the photodetector operates, thenoise spectrum can be used to determine the magnitude of thenoise. It can also be used to calculate the dependence of the SNRto frequency [18]. The noise spectrum is not directly related tothe noise bandwidth.

The termpower spectral densityis used to describe the noisecontent as a function of frequency. The power spectral densitycan determine the noise power of the system by integrating overthe (electrical) bandwidth

(11)

where is the noise power over a bandwidth range, isthe power spectral density, and is the bandwidth.

V. PHOTODETECTORNOISE AND ELECTRONIC

NOISE REDUCTION

In biological sensing applications, the light signal can pushthe detection limits of a photodetector, and because the inputsignal is at the detection threshold, both the noise of the pho-todetector and attenuation of the signal from the optical systemplay an important role determining whether a photodetector issuitable for a given application. Noise can be defined as anyunwanted disturbance that obscures or interferes with a desiredsignal [16]. The performance of a photodetector depends on theinherent electrical noise in the detector, the noise in the opticalpathway (Section III), and nonidealities of the photodetectormaterial (Section IV-A). Major sources of electrical noise inphotodetectors include thermal noise, shot noise, and low-fre-quency noise. In this section, the major sources of electronicnoise for biological applications are defined, and methods forreducing these noise sources are reviewed.

A. Thermal Noise

Thermal agitation of the carriers that occurs in resistive ele-ments gives rise to thermal noise. This is one of the most preva-lent sources of noise. The mean square thermal noise poweris given by

(12)

where is Boltzmann’s constant, is the absolute temperature,and is the noise bandwidth [18]. Another term for thermalnoise iswhite noise. For any given bandwidth, the noise contentis equivalent, or in other words, the magnitude of the thermalnoise power does not depend on frequency. To reduce thermalnoise, the device can be operated at lower temperatures.

B. Low Frequency Noise

Low-frequency or 1/f noise is any noise whose spectraldensity has a 1/f dependence (noise increases as frequency de-creases). Other names for low-frequency noise include flickernoise, excess noise, pink noise, semiconductor noise, andcontact noise. The major cause of 1/f noise in semiconductordevices is traceable to the surface properties of the material[16].

Circuit techniques such as autozeroing (AZ), correlateddouble sampling (CDS), and chopper stabilization (CHS) areeffective for reducing flicker noise. AZ and CDS operate bysampling the noise and then subtracting the noise signal fromthe input or output of the circuit. Because AZ and CDS reducenoise through sampling, these circuits are most applicable tosystems that already use data-sampling. For CHS, the signal ismodulated to a higher frequency, amplified, then demodulatedback to the original frequency band. By amplifying the signalat high frequencies, the low-frequency noise is not amplifiedalong with the signal. For continuous-signal systems with alarge low-frequency noise component, CHS is the preferredmethod [19].

C. Shot Noise

Current flowing in devices is not smooth and continuous, butis rather the sum of pulses of current caused by the flow of car-


riers, each carrying one electronic charge. The variations in thispulsing flow are referred to as shot noise. The rms value of theshot noise current is given by

(13)

where is theelectroncharge, isdirectcurrent inamps,andis thenoisebandwidth inHz.Shotnoise isassociatedwithcurrentflow across a potential barrier. Such a barrier exists at every p-njunction in semiconductor devices and at the cathode surface in avacuum tube. Since a simple conductor has no barrier, it also hasno associated shot noise. To reduce shot noise, the device shouldbe operated at a lower applied bias, if possible.

D. Noise Bandwidth

Noise bandwidth is defined as follows. A circuit is character-ized by a power gain curve, or the amplification of a signal versusfrequency.Theareaunderneath thiscurve isequal toarectangularareawhoseheight is thepeakof thepowergainandwhosewidth isdefined as the noise bandwidth. Mathematically, the noise band-width can be represented by the following equation:

(14)

where is the power gain as a function of frequency andis the peak power gain [16]. The power gain can be deter-

mined by analyzing the behavior of the circuit. For a photode-tector operating under a constant applied electrical voltage, thenoise bandwidth is directly related to the inherent noise proper-ties of the photodetector. Changes in measurement strategy canreduce the noise extracted from the photodetector during highsensitivity measurements.

The dependency of NEP on noise bandwidth can be under-stood through the following qualitative example on reducing theelectronic noise contribution. A photodetector is to be integratedwith an amplifier circuit characterized by some frequency re-sponse. If the amplifier is designed to respond to a wide rangeof frequencies, then it will also respond to the noise at those fre-quencies, and the total noise signal will be larger, requiring alarger optical signal to obtain a signal-to-noise ratio equivalentto 1. But if the amplifier circuit passes only a very narrow bandof frequencies, the output noise signal will be reduced. Relatingthis to (14), the narrowband amplifier tends to have less areaunder its power gain curve since it amplifies the signal over asmaller frequency range. To reduce noise, the amplifier can bedesigned so that it passes as narrow a band of electrical frequen-cies as possible while still obtaining a signal from the photode-tector. Since the noise component is smaller, the device will bemore sensitive. Finally, the calculation of the power curve de-pends on the actual design of the circuit, so simply reducing theelectrical bandwidth of the amplifier may not decrease the noisebandwidth, although it often does.

E. Summary

Sources of electronic noise in photodetectors include thermalnoise, flicker noise, and shot noise. Because the contribution ofthese noise sources depend on noise bandwidth, proper choice

Fig. 4. Structure of a PMT. The PMT consists of a photocathode to convertincident photons to electrons, focusing optics to focus the electrons onto thefirst dynode, a cascade of dynodes that multiply the electron, and an anode tocollect the now numerous electrons. In order for the electrons to not be scatteredfrom gas molecules, the inside chamber is at a high vacuum. To assist withfocusing, the photocathode is placed at a large negative potential. Dynodes areat subsequently lower potentials.

of bandwidth can reduce the effective noise in the system. Fur-thermore, dependencies on temperature for thermal noise andbias for shot noise pose alternative solutions to reducing noise,and circuit techniques are available for reducing low-frequencynoise sources. The operating principles behind different classesof photodetectors will change the noise characteristics of the de-vice. In the following sections, popular photodetector devicesare overviewed, with an emphasis on applications to integratedmicrosystems for biological applications and noise sources spe-cific to a photodetector architecture.

VI. PHOTOMULTIPLIER TUBES

Photomultiplier tubes (PMTs) are the most sensitive photode-tectors for the visible light range. They are also fragile, largerthan semiconducting detectors, and more expensive. Because oftheir superior performance, they are a popular choice for bio-logical sensing applications, especially in a laboratory setting,but PMTs are not the best choice for fully integrated microscalesystems because of their high operating voltages, fragility, size,and cost.

PMTs detect light through the photoelectric effect and typi-cally operate in the following way: 1) Photons enter through aninput window, 2) the photocathode absorbs photons and emitselectrons into a vacuum, 3) the electrons are accelerated and fo-cused onto the first dynode, 4) the dynode multiplies the elec-tron through secondary electron emission, 5) the electrons traveldown the dynode chain and are multiplied by each dynode, and6) the electrons are finally collected by the anode. The typicalstructure of a PMT is shown in Fig. 4. The following sectionsbriefly overview the components of the PMT, and more detailsare available in [14].

A. Photocathodes

The principle of operation for photocathodes involves ex-citing an electron from the conduction band into the vacuum.This process is termed the photoelectric effect, the photoemis-sive effect, or the Einstein effect, and it is represented by theequation

(15)


where is the energy imparted to the electron,is Planck’sconstant, is the frequency of the light, and is the binding en-ergy of the electron to the atom (the work function for metals orthe electron affinity for semiconductors). When a photon strikesthe photocathode material and the energy of the photonexceeds the binding energy of the electron, the electron will beexcited from the conduction band of the photocathode to thevacuum.

Ideally, photocathode materials have a low electron bindingenergy and a high quantum efficiency (photons are moreefficiently converted to electrons). Photocathode materials areusually either alkali and group V metals (e.g., Na, K, Cs, Sb)evaporated onto a metal electrode or III-V materials [20]. Theadvantage of III-V photocathodes over conventional cathodes isthat, when activated by cesium, it is possible to produce mate-rials with negative electron affinity (NEA) in which the electronbinding energy is negative [21]. Most photocathode materialsmust remain in a vacuum to retain their photoemission proper-ties.

The probability that an electron will be emitted is givenby the quantum efficiency, which depends on the reflectioncoefficients of the window-air, window-photocathode, andphotocathode-vacuum interfaces; the absorption coefficientof the photocathode material; the probability that the photonexcites an electron to an energy level above the electron bindingenergy ; the distance that excited electrons can diffuse beforereturning to a lower-energy state; the probability that electronsreaching the photocathode surface will be released into thevacuum [14].

Photocathodes either reflect or transmit the photon-gener-ated electron. Reflection-mode photocathodes are metal platescoated with a photoemissive material, and transmission-modephotocathodes are usually transparent plates coated with a thinphotoemissive film. Of the two types of PMTs, transmissionmode tubes are most often used for biological sensing applica-tions. In these devices, the electron travels the same direction asthe initial photon.

B. Secondary Gain Mechanisms

The two most common secondary gain mechanisms of pho-toemissive sensors are gas ionization and dynode multiplication.Of the two types of secondary gain, dynode multiplication pro-vides the highest sensitivity. Because it is less applicable to bio-logical sensing applications, gas ionization multiplication is notreviewed here, and recent advances in these types of sensors canbe found in [22].

Dynode multiplication devices have a dynode chain that elec-trons strike as they travel through the device. A dynode is a typeof cathode whose surface is coated with a low work function ma-terial. By choosing an appropriate dynode geometry, electronsare focused to avoid a large spread in transit path lengths. For

dynodes with an average electron gain, the overall gain is.

There are a variety of dynode configurations which result indiffering response time and gain. For biological sensing appli-cations, the box-and-grid and linear-focused configurations aremost common because of their high collection efficiency. An-other dynode configuration is the micro-channel plate (MCP).

TABLE IICHARACTERISTICS OFDYNODES [14]

The MCP is constructed by bonding capillaries together to forma parallel array, pulling the capillaries axially, then cutting thearrays at an angle. A low work function material is then de-posited inside each channel and a voltage is dropped along thechannels. The final plate is around 1-mm thick and exhibitsultra-high speed detection. These devices can also provide spa-tial resolution of the signal. An overview of the characteristicsof each structure is shown in Table II.

The response time of a PMT is largely determined by theflight path of the electrons through the dynode structures.Within limits, increasing the supply voltage increases theelectron transit speed and shortens the transit time.

C. Sources of Noise

The major noise source for photoemissive detectors is thedark current, which is due mostly to thermionic emission ofelectrons from the photocathode or dynodes, and is given by theRichardson-Dushman equation

(16)

where is Richardson’s constant ( in avacuum), is the cathode area, is the work function, isBoltzmann’s constant, and is the absolute temperature in K[20]. Other contributions to dark current are 1) leakage currentbetween electrodes inside the tube, 2) scintillation-inducedphotocurrent, 3) field emission current, 4) ion feedback fromresidual gases, and 5) noise current caused by cosmic rays,radiation, and gamma rays. Shot noise is another source ofnoise that arises from fluctuations in the arrival rate of photons.A review of noise in PMTs and reducing the noise can be foundin [14].

Multiplication in a PMT is an inherently less noisy processthan avalanche multiplication in a semiconductor. In an APD,avalanche multiplication is randomly initiated in space, whilein a PMT, the multiplication process occurs semi-coherently atthe dynodes. A coordinated multiplication process is less noisythan a randomly initiated one. In addition, single-carrier multi-plication is less noisy than a bipolar (electron-hole pair) multi-plication process.

D. Photon Counting Mode

Photon counting mode is implemented by setting a lowerand upper threshold for the output signal. Thermally generated


electrons from the dynodes are multiplied by subsequent dyn-odes in the chain, but the electrons undergo less multiplicationthan electrons emitted from the photocathode. By determiningthe typical range of photon-generated signals, any other sig-nals which are not within this range can be filtered out. Photon-counting usually also sets an upper threshold on the signal toaround two photons within the time resolution of the PMT [14].This technique still does not filter out thermally generated elec-trons emitted from the photocathode.

E. Summary

For experiments which produce an extremely low level lightsignal, PMTs are often the best option for obtaining useful in-formation from the experiment. For example, the inherent fluo-rescence (bioluminescence) of a biological sample produces anextremely weak signal which is capable of being detected witha PMT [23]. The extremely fast response time of a PMT canalso be used to obtain time-dependent information for the ex-periment, such as for time-resolved fluorescence spectroscopy[24]. For fully-integrated systems, though, other considerations,such as cost, fragility, and size, may eliminate the PMT as a vi-able choice of photodetector for the biological application.

VII. APDs

The APD is the most sensitive solid-state device available.It is capable of being miniaturized for microscale devices, butdisadvantages of this device include high operating voltages, theneed for complicated supporting circuitry, and a higher cost dueto the difficulty in fabricating the device. Photon counting is alsoavailable for APD devices, and the implementation of this modeof operation is similar to that used for PMTs.

APDs are p-i-n diode structures that are operated at largereverse bias voltages. Signal multiplication is obtained whenphotogenerated carriers gain enough energy from the electricfield to generate secondary carriers through impact ionization(Fig. 5). These secondary carriers are also accelerated by theelectric field and generate other electron-hole pairs. The outputcurrent is the primary photocurrent multiplied by the avalanchemultiplication factor, , which is dependent on bias voltage.

The impact ionization coefficients of electrons and holes,and (in cm ), respectively, are the reciprocal of

the average distance that a carrier will travel for a given electricfield before generating an additional electron-hole pair throughimpact ionization.

Assuming that the ionization coefficients are constantthroughout the depletion region, the avalanche multiplicationfactor is given by

(17)

where is the depletion region width [1]. The avalanche mul-tiplication process also introduces avalanche noise calledex-cess noise. This noise is a result of the random nature of theavalanche multiplication process since every electron-hole pairgenerated at a random location within the depletion region does

Fig. 5. Avalanche multiplication process. Shown is the band diagram ofthe avalanche multiplication process. (left) A photon generates an initialelectron-hole pair and the two carriers are accelerated. Upon impact, theelectron generates a new electron-hole pair, whose carriers are then acceleratedand can generate further electron-hole pairs. When two carriers can generatenew electron-hole pairs, a feedback loop can result in which there is notermination of the electron-hole generation process. The overall multiplicationwill change depending on the impact ionization coefficients and the location ofthe original photogenerated electron-hole pair.

not experience the same multiplication. The excess noise factoris given by

(18)

When , the noise factor is maximized and is equalto . On the other hand, if one of the ionization coefficientsis zero, e.g., , then the noise factor is equal to 2 [25].Complicated APD structures have been developed in order toreduce one of the ionization coefficients and thus reduce theexcess noise factor.

If one of the ionization coefficients is zero, the avalanche mul-tiplication factor for electrons simplifies to

(19)

Another reason to minimize one of the ionization coefficientsis to limit the time over which multiplication takes place. Thetime during which multiplication occurs is equal to the transittime through the depletion region plus the hole transit time. Ifone of the coefficients is zero, then the multiplication time issimply the transit time of an electron plus the transit time ofa hole across the depletion region. But if both coefficients arepositive, then a feedback loop can occur and the multiplicationtime can be infinitely long.

The high reverse bias used with APDs requires very uniformdoping and sometimes elaborate doping profiles to prevent localvariations in the electric field and premature breakdown. Thus,APDs are more expensive to manufacture, and they require com-plex temperature compensation and supply voltage stabilizationcircuitry. Quantum efficiencies in APDs can be greater than 90%at the peak wavelength of response, so the primary photocur-rent generation mechanism is highly efficient. Unfortunately,avalanche gain is accompanied by noise that is much worse


than for photomultipliers, since nonphotogenerated carriers arealso subject to avalanche multiplication. Thorough overviews ofavalanche photodiode technologies, including underlying prin-ciples and devices structures, can be found in [26] and [27].

VIII. p-n AND p-i-n PHOTODIODES

Photodiodes are an attractive photodetector choice for use inbiological applications because they are inexpensive, easy touse, are easily miniaturized, and can be implemented in arrays.But because these devices have no inherent gain mechanism,they are not as sensitive as the APD or PMT. Still, some ofthese disadvantages can be overcome through accompanyingcircuitry, and many biological sensing applications produceenough light to be easily detected by a photodiode.

A photodiode can operate either as a photovoltaic or photo-conductive device, but the intensity of light required to gen-erate a change in voltage large enough to be detected in thephotovoltaic mode makes it unsuitable for low-level light de-tection. Because of its higher sensitivity, only the photoconduc-tive mode of operation is applicable to biological applications.In the photoconductive mode, a reverse bias is placed acrossthe junction and the reverse current increases when photons areabsorbed within the depletion region (Fig. 6). Although elec-tron-hole pairs are generated at any location in the device thatabsorbs a photon, only those absorbed in or near the depletionregion contribute to the signal current [13]. The signal currentrequires that the carriers be transported in order to generate acurrent, and in order to transport carriers, an electric field mustbe present. An electron and hole generated close to the depletionregion may diffuse into the depletion region and be collected.The charge separation results in a potential difference across thejunction that tends to decrease the built-in potential barrier (i.e.,forward-bias the junction). The (electrical)I-V characteristic ofa photodiode is described by

(20)

where is the saturation current, is the current resultingfrom photon generation, is the applied bias, is Boltzmann’sconstant, is the absolute temperature, andis the electroncharge. Illumination shifts theI-V characteristics of a p-n junc-tion downward and therefore results in an increased reverse cur-rent.

A p-i-n photodiode is a p-n photodiode with a lightly-dopedregion at the p-n junction. This architecture effectively increasesthe depletion width, which results in a higher collection effi-ciency since incoming photons are more likely to generate col-lected electron-hole pairs. It also lowers the junction capaci-tance and increases the transit time. Although less junction ca-pacitance reduces the RC time constant of the device, the longertransit time limits the overall time response of the device. Pho-todiode speed is mainly determined by three factors: diffusiontime of carriers generated outside the depletion region that areclose enough to diffuse into it (can be reduced by making thedepletion region close to the surface), drift time in the diffusionregion (reduced by making it only wide enough to absorb maxi-mally but not too thin so that capacitance goes up), and junction

(a)

(b)

Fig. 6. p-i-n photodiode andI–Vcharacteristics. Shown is (a) the structure of ap-i-n photodiode and (b) its current-voltage characteristics. Incident light strikesthe depletion region and photogenerated electron-hole pairs are swept out to becollected at the contacts. As the light intensity increases, the photogeneratedcurrent increases. The p-i-n photodetector acts as a photoconductor in this case.

capacitance (reduced by strong reverse bias and an intrinsic re-gion). An overview of p-n diodes can be found in [25].

When designing a high-throughput microscale biologicalsystem, the samples usually flow through multiple parallelmicrochannels. This architecture is easily integrable withphotodiode arrays, and if the arrays are fabricated in a siliconsubstrate, integration with signal processing circuitry becomestrivial. Recent research has focused on the development ofmicrochannel arrays integrated with photodiodes in silicon[28], and these devices have the potential for achieving highthroughput processing of biological samples.

IX. CCD STRUCTURES

Recently, CCDs have become a popular choice for use in bi-ological applications. The main advantage of CCDs over otherphotodetectors is that they provide spatial information about thesystem. Although CCDs have an extremely slow response time,this is often not a critical factor in biological experiments, andthe devices can be cooled to obtain the sensitivity required todetect the light-emitting reporter. These devices are not easilyminiaturized since they often require both focusing optics to col-lect the signal as well as cooling.

A CCD is an array of metal-insulator-semiconductor (MIS)or metal-oxide-semiconductor (MOS) structures that can detect,store, and transfer photogenerated charge. The basic structureis a metal or polysilicon gate above a dielectric and a semi-conductor substrate below, forming an MOS capacitor which ischarged up from photogenerated carriers (Fig. 7).


Fig. 7. Structure of a CCD. When a voltage is applied to the metal gate, apotential well is formed underneath the oxide in the silicon substrate. Thevoltages(V1 < V3 < V2) can control the size of the potential well, andwith the correct wave-form, the charge stored in one well can be pushed to theadjacent well.

The gate is biased to place the semiconductor into inversion(positive gate voltage for a p-type substrate). Under equilibriumconditions, minority carriers (electrons in this case) would invertthe silicon underneath the gate. A CCD structure, though, neverallows the MOS structure to reach this state; instead, it transfersthe accumulated carriers from the well before the well fills up.The minority carriers stored underneath the gate are generatedby two mechanisms: thermal generation and photon generation.When a photon is absorbed, an electron-hole pair is created, andthe electron is swept toward the gate where it is stored until thecharge is transferred out.

In order to transfer the charge underneath the gate, the MOSstructure is placed close to other MOS structures. Chargeis transferred in a definite direction by either using threesets of clocked gates or an asymmetric well structure. Foreither transfer scheme, the efficiency of the charge transfer isextremely high ( 99.99%).

To manufacture color CCDs, filters are applied directly to thesensor areas using dyed photoresist [1]. These filters introducean additional source of loss of the optical signal.

A. Signal Handling Limits

The limit of the signal handling capacity depends on the max-imum number of minority carriers that a well can store. When awell is full, the device is under equilibrium and any further pho-togenerated electron-hole pairs are lost through recombination.A full well may also create a noise source called “blooming”,in which the full well overflows and minority carriers flow intosurrounding potential wells, creating an area of saturated pixels.

B. Charge Transfer Inefficiency

Ideally, all charge would transfer from one well to anotherduring a transfer cycle, but this rarely happens in a real device.The transfer inefficiencyis the percentage of charge leftbehind after a transfer. The two main causes of incompletecharge transfer are trapping at interface states and incompletefree charge transfer. At slow clocking frequencies, interfacestate trapping determines the transfer inefficiency (especiallyfor a surface channel device), whereas at higher clockingfrequencies, there are limitations in the speed of transfer of thefree charges. The influence of interface state trapping is verycomplex since the interface states fill nearly instantaneouslywhen exposed to minority carriers and emit more slowly. Incases where the amount of transferred charge is small, as isoften the case for biological applications, the effect of interfacestates is even more significant. A detailed analysis of interfacestate trapping and charge transfer can be found in [30]–[32].

One method to reduce the number of interface states is to storecharge away from the gate by implanting a p-n junction deepwithin the substrate. These devices are called buried-channelCCDs (BCCDs).

C. Sources of Noise

The major source of noise in CCDs is thermal generation ofminority carriers [32]. Because the performance of a CCD de-pends on charge storage, thermal noise is often put in terms ofcharges per pixel per time unit rather than a dark current. Othersources of noise include fluctuations in cosmic rays, transfer in-efficiency, and fluctuations in the thermal generation of minoritycarriers [33].

D. CCD Structures

Most CCDs are BCCDs, in which the charge storage area is faraway from the gate. Because the minority carriers do not comeinto contact with surface states, the devices exhibit more com-pletechargetransferandlowernoise.BCCDscanalsobeoperatedat higher clocking frequencies, but a potential disadvantage ofthesedevices is thewell storagecapacity isa factorof twoor threesmaller than for surface CCDs (SCCDs). In SCCDs, the chargestorage area is at the surface directly below the oxide layer.

CCDs are also either front-illuminated or back-illuminated.When the device is illuminated from the front, surface reflec-tivity increases with decreasing wavelength, and the absorptionof blue light is limited. Backside illumination requires thinningthe substrate from 300m to around 10 m, which increasesthe manufacturing costs and requires complicated manufac-turing processes to avoid undesirable side-effects such as waferbending. Backside-illuminated devices are inherently moresensitive than frontside-illuminated devices, since the lightis not attenuated by passing through the polysilicon gate andoxide regions before reaching the potential well.

The architecture of a CCD generally falls into one of fourcategories: full frame, frame transfer, split frame transfer, andinterline transfer. For full frame, the image is transferred directlyfrom the imaging region to the readout register. However, sinceonly one pixel at a time can be read out, the other pixels arewaiting to be read out and can collect more photogeneratedcarriers. The imaged information can thus be distorted beforereaching the readout register. To preserve the image information,a mechanical shutter can be used to block light during the readoutperiod. But because the mechanical shutter requires that no lightcan be collected for a substantial delay, light-emitting systemsthat produce a continuous light signal may not be compatiblewith thisarchitecture.Theotherarchitectureshave light-shieldedstorage sections: a full frame for frame transfer, two half framesfor split frame transfer, and columns for interline transfer. Fol-lowing the image capture period, the image is quickly transferredto the adjacent storage section. While the next image is beingcaptured, the previous image is transferred to the readout register.Although no mechanical shutter is needed, a large proportion ofthe imaging section is not sensitive to light, reducing the overallsensitivity of the device. For biological systems which produce avery low-intensity light signal, systems which use a mechanicalshutter and cooling will exhibit extremely low distortion of thesignal.


X. ALTERNATIVE CMOS STRUCTURES

CMOS photodetectors have been largely dismissed for main-stream imaging applications due to their poor noise performancecompared to their CCD counterparts. However, decreasing fea-ture size has continued to narrow the performance gap betweenCMOS photodetectors and CCD devices for imaging applica-tions. In fact, commercial CMOS imagers whose system levelperformance can compete with CCD performance are now onthe market. The low cost and ease of manufacturability associ-ated with CMOS photodetector structures has continued to feedresearch and development into making these devices competi-tive with CCD imagers and is also generating interest in usingthese devices for biological sensing and analysis applications.

Many of the primary design constraints associated withCMOS photodetectors shift significantly when consideredfor relatively low (spatial) resolution applications such asthose associated with biological sensing and analysis. Forthese applications, many CMOS photodetector structures offercomparable or better performance than CCDs. Like CCDs, theyalso offer a practical choice for compact or portable analysissystems that is not possible with PMTs due to their large size orwith APDs due to their high operating voltages. In traditionalimaging or in large feature size processes, the CCD offersbetter electrical noise performance because of decreased gatecapacitance. The primary contribution to electrical noise inthese devices occurs when the electrical charge is transferredaway from the source of impinging photons. The electronicnoise is dominated by Johnson noise in the channels of transis-tors that ultimately convert the photocharge or photocurrent toan amplified signal [34]

(21)

where is the number of charge carriers on the transistor gate(the photocharge); is the total gate capacitance of the tran-sistor (generated by the oxide capacitance of all transistors thegate is connected to as well as stray capacitance from intercon-nect lines); is the electronic charge; is Boltzmann’s con-stant, is the absolute temperature;is the bandwidth of thetransistor (total frequency range under which the transistor per-forms at maximum gain); is the transistor’s transconduc-tance (which is proportional to the channel current);is 2/3under typical operating conditions in a MOSFET.

In a given transfer cycle in the CCD, charge is transferredin packets from one pixel to an adjacent pixel. At the end ofthe line of transfer, the amplification circuitry (transistors) seesonly one pixel’s worth of capacitance because of the serial na-ture of photocharge transfer in a CCD. In typical CMOS pho-tostructures, however, the amplification circuitry sees a capac-itance generated by a full row or column of pixels and associ-ated interconnect capacitance. In conventional imagers, ampli-fication occurs only at the end of every row or column, causingthe C term in (21) to be far less (fF) than that of a CMOS pho-todetector structure (pF). Conventional imagers, place amplifi-cation circuitry at the end of every row to conserve space and

Fig. 8. Buried double junction structure. This novel photodetector structureis implemented in a CMOS process and consists of two p-n junctions:shallow and deep junctions which are sensitive to short and long wavelengths,respectively. The sum of the two photocurrents is representative of the totalimpinging optical power; the shallow junction photocurrent is proportional tothe impinging optical power at wavelengths between 400 and 650 nm, and theratio of the two photocurrents provides information about the wavelength ofimpinging light.

fill factor for maximum image resolution. However, in biolog-ical sensing and analysis, high image resolution is usually sec-ondary to high sensitivity. For this reason, and because of de-creasing feature size which enable even smaller circuit size, am-plification circuitry can be dedicated to each pixel in a CMOSphotostructure array rather than to an entire column or row. Pho-todetectors in these active pixel architectures experience a muchsmaller capacitance on the input transistor channel (amplifica-tion circuit inputs) and therefore exhibit significantly reducedelectronic noise. Equivalent noise levels of 20 electrons per pixelhave been demonstrated in research efforts on CMOS arrays[35]. Thus, although the inherent gain of the CMOS photode-tector structure is lower than that of the PMT and the APD, itis apparent that the cost, size, fabrication, power, and operatingvoltage advantages of CMOS in combination with low electricalnoise realized in active pixels structures make them a viable al-ternative to both CCDs and PMT/APD methods for detectingfluorescent and luminescent signals associated with biologicalevents. The spectral sensitivity of a CMOS photodetector struc-ture can be adjusted with the way in which the photodetector isdriven electrically [36], or it can be adjusted by combining pho-tosensor with different spectral sensitivities in standard CMOSprocesses into a single compound photodetection structure toachieve desired spectral sensitivity [37].

Adaptation of CMOS photodetector structures to the low spa-tial resolution, stringent spectral sensitivity, and high signal-to-noise performance required of biological sensing and analysiscan be broken into two broad categories: the fabrication of novelphotodetector structures and the design and demonstration ofnovel transduction circuits suited to these applications. StandardCMOS processes offer limited control of junction depths andmaterials for adjusting spectral sensitivity in the design of novelCMOS photodetector structures. However, the buried doublep-n junction (Fig. 8) [38] uses the ratio of two photocurrentsat two different junction depths to acquire spectral informationstarting at 430 nm, a common spectral range of interest in bio-analysis applications. Spectral responsivity of up to 0.2 A/W(photocurrent divided by impinging optical power) is typical inthe shallow photodiode junction at wavelength sensitivities be-tween 400 and 500 nm, while color information is obtained from


TABLE IIIREPRESENTATIVEPERFORMANCE OFALTERNATIVE CMOS STRUCTURES

the ratio of the shallow junction photocurrent to the deep junc-tion photocurrent. Using the shallow junction photocurrent andthe ratio of the shallow and deep photocurrents, spectral infor-mation across a range of wavelengths can be approximated.

An example of effective transduction strategy in an activepixel architecture is a charge amplifier that forces a photodiodeto remain under constant reverse bias, thereby keeping the elec-tronic properties constant even under the influence of changinglight input. The charge amplifier configuration allows a linearresponse of the electrical output to impinging optical power,limiting temperature sensitivity of the device. The constant re-verse bias in this same charge amplifier configuration main-tains a constant electrical current through the device while thephoton-induced current varies in response to light. Signal tonoise performance of this CMOS device is reported to be as highas 66 dB at 10 mWcm optical power [39].

Other representative CMOS photodetector structures andtransduction strategies that demonstrate appropriate perfor-mance levels for biological sensing and analysis applicationsare summarized in Table III. Some research has been doneto include CMOS photodetectors in biological analysis ap-plications. A highly integrated microspectrometer containingphotodetectors, read-out circuits, Fabry–Perot etalons forspectral sensitivity, and dark current compensation has beendemonstrated by Correiaet al. with sensitivities up to 13mA/W [40] although stray-light influences limit the detectionof the integrated device. CMOS photodetectors have also beendemonstrated in an integrated photodetector for luminescencemonitoring by Simpsonet al. [41]. The microluminometer,using both electrode configurations optimized to reduce noiseand current-to-frequency signal processing circuitry to en-hance sensitivity, demonstrates performance comparable toPMT-based detection systems at cell concentrations rangingfrom to CFU m.

In summary, the CMOS photodetector will continue to becompetitive for integration into small, low-power systems forbiological instrumentation and sensing. Traditional shortfalls ofthese structures have been significantly improved through de-creasing feature size, creative photodetector structures and in-novative transduction strategies.

XI. A PPLICATION OFPHOTODETECTORS TOINTEGRATED

BIOLOGICAL MICROSYSTEMS

Fully integrated biological microsystems are often targetedtoward applications either in the medical industry or in thebiologist’s laboratory. For both applications, the advantagesof building a miniaturized integrated system include smallersample size, faster detection, and ideally less technical trainingrequired in order to obtain accurate results. For the medical in-dustry, microscale systems are ideally battery-powered portabledevices that can be used on-site, for example in an ambulance.In life-threatening situations, the time it takes for a medicaltechnician to obtain information about the patient, includingblood type, the presence of drugs in the body, or the extent ofthe injuries, will determine the type and quality of the care ofthe patient [49]. The generally less-invasive sensing processused by microscale devices also reduces any further damagedone to the patient. In the laboratory, microscale devices tendto focus on either automating the sample preparation, sorting,and handling process, as well as increasing the accuracy of theexperiment through conformity of experimental parameters,or on pushing the limits of detection to the single cell level.Through automation, biologists have more flexibility to try newexperiments through the decrease in time needed to performexperiments as well as the ability to obtain more data for agiven sample size. Detecting smaller samples is attractive sinceit provides information about the actual state of the samplerather than a time-averaged value.


For both applications, the main problem with respect to thephotodetection process is transporting the light signal to thedetection window with minimal loss. Laboratory equipmentand large instruments use fiber optic coupling to integrate thefluidic system (microscale or otherwise) to the optical detectionsystem, but as systems move toward fully integrated handheldsystems in which size becomes critical, the interface to theoptical detector will have to be reconsidered. Some researchhas already focused on integrating waveguides on-chip withthe photodetector [50]. Although small in size, the integratedwaveguides have the disadvantage that they cannot be adjustedonce fabricated. Microoptical benches provide an intermediatestage between flexibility and full small-scale integration [51],[52]. In these systems, microscale optical elements are securedusing a material that can allows for later adjustment in case ofmistakes or redesign. Another promising technique is to use theoptical properties of the device to guide the light signal directlyto the photodetector. For instance, the refractive indices of acapillary filled with liquid can be designed so that the entirecapillary tube becomes a waveguide [53]. Photodetectors locatedat the ends of the capillary can collect the light generated fromthe fluorescent sample.

Other factors that should be considered in the selection of aphotodetector for a biological analysis application are the sizeand power requirements of the photodetector, electrical oper-ating parameters of the overall system, fragility of the opticsfor use in rugged environments, and scaling effects on the op-tical system. For example, the high voltages (electrical parame-ters) used in capillary electrophoresis make silicon an unsuitablechoice for substrate material [54]. Similarly, high voltage andpower requirements required to operate APDs and PMTs maymake them unsuitable for use in portable applications wherehigh voltage generation is not possible due to power or weightconstraints. Optical components are often carefully aligned, andsome disturbance of the system can result in a deterioration of itsfunctionality. The photodetectors may themselves be fragile; forinstance, some PMT photocathodes suffer performance degra-dation if exposed to too much light.

These factors, in addition to the basic performance limitationsof the photodetecting device make the selection of the appro-priate system for biological analysis complicated. However, bymatching the optical system metrics with photodetector metrics,and by accounting for trade-offs, alternative photodetectorschosen based on other factors such as size and power can be intel-ligently selected for use in fully-integrated biological systems.

XII. CONCLUSION

We have reviewed photodetectors in the context of detectinglight-emitting reporters for biological applications. Detectorperformance metrics that are most relevant to biological sensingand analysis emphasize, in order of priority, the lower limit ofsignal detection, the resolution, and the speed of the device.Lower limits of signal detection are determined by a combi-nation of gain, noise equivalent power, quantum efficiency,detectivity, and dark current. Resolution is defined by respon-sivity and response time. Secondarily, the type of informationprovided by the photodetector, for example spatial information,

may be an important factor in choosing a photodetector for anapplication. Other factors such as size, power, cost, and ease ofimplementation may also be relevant.

The PMT is the most popular choice for biosensing and anal-ysis applications because it has high gain and responsivity, butthe need for high voltage operation of the PMT and its largesize limit its suitability for integrated applications. The APD isan attractive alternative to the PMT because it can be reduced tomicroscale sizes and still exhibits moderately high responsivityand gain. Drawbacks of the APD are the high voltages requiredfor its operation, custom fabrication steps needed to fabricate anAPD, and dependency on complicated supporting circuitry maylimit its usefulness in integrated modules. The CCD is an at-tractive choice for microscale systems because of its fabricationmaturity, small size, and two-dimensional imaging capabilities.However, high dark currents and limited gain can prevent it frombeing useful for small reporter count measurements. Finally,CMOS structures are inherently less efficient because the ma-terial is an indirect bandgap material (silicon) and the fabrica-tion process is not designed for photodetection applications, butthese devices are also the most compatible with single chip ormicroscale systems. Other advantages include cost, simplicityof operation, and the mature development of circuitry that par-tially overcomes its limitations.

It is clear from the current state of design and research in bi-ological sensing and analysis systems that the full performancelimits of many alternative photodetection technologies are notclearly characterized yet. The success of the PMT and APDmay preclude the expansion of alternative photodetectors intothe biosensing arena except in applications where low voltageoperation and highly compact size are critical to the success ofthe system (such as in the portable or throw-away module). De-tection limits of the photodetector itself are often not well cor-related to the analyze detection limit, and it remains to be seenwhether alternative devices will be selected for use in biolog-ical fluorescence detection systems or will continue to be a sec-ondary choice for full integration.

ACKNOWLEDGMENT

The authors would like to thank Prof. B. Darling and Prof. M.Troll at the Department of Electrical Engineering, University ofWashington, for their assistance.

REFERENCES

[1] G. T. A. Kovacs,Micromachined Transducers Sourcebook. Boston,MA: WCB/McGraw-Hill, 1998.

[2] Hamamatsu Corporation (USA), Hamamatsu, Inc. (2002, May).http://www.hamamatsu.com/[Online]

[3] Digi-Key Product Index: Optoelectronics, Digi-Key, Inc.. (2002, May).http://www.digikey.com/[Online]

[4] Biomedical Imaging, Hamamatsu, Inc.. (2002, May).http://www.hama-matsu.com/[Online]

[5] Life Sciences/Microscopy Cameras, Roper Scientific Photometrics.(2002, May).http://www.roperscientific.com/[Online]

[6] G. R. Price and S. Schwartz, “Fluorescence microscopy,” inCells andTissues, G. Oster and A. W. Pollister, Eds. New York: Academic, 1956,vol. 3, Physical Techniques in Biological Research, pp. 91–111.

[7] J. R. Lakowicz,Principles of Fluorescence Microscopy. New York:Plenum, 1983.

[8] J. L. Rosenberg, “Photochemistry and luminescence,” inOptical Tech-niques, G. Oster and A. W. Pollister, Eds. New York: Academic, 1955,vol. 1, Physical Techniques in Biological Research, pp. 37–39.


[9] P. T. Sharf, “Filters,” inLight: Its Generation and Modification, R.Kingslake, Ed. New York: Academic, 1965, vol. 1, Appl. Opt. Opt.Eng., pp. 111–125.

[10] H. A. Macleod, Thin-Film Optical Filters. Philadelphia, PA: IOP,2001.

[11] R. Kingslake,Lens Design Fundamentals. New York: Academic,1978.

[12] C. Clark, “Infrared spectrophotometry,” inOptical Techniques, G. Osterand A. W. Pollister, Eds. New York: Academic Press, 1955, vol. 1,Physical Techniques in Biological Research, pp. 206–316.

[13] B. E. A. Saleh and M. C. Teich,Fundamentals of Photonics. NewYork: Wiley, 1991.

[14] H. Kume, Ed., Photomultiplier Tube: Principle to Applica-tion. Toyooka, Japan: Hamamatsu Photonics K. K., 1994.

[15] P. Bhattacharya,Semiconductor Optoelectronic Devices. NJ: Prentice-Hall, 1997.

[16] C. D. Motchenbacher and F. C. Fitchen,Low-Noise Electronic De-sign. New York: Wiley, 1972.

[17] R. C. Jones, “The ultimate sensitivity of radiation detectors,”J. Opt. Soc.Amer., vol. 37, p. 879, 1947.

[18] P. W. Kruse, “The photon detection process,” inOptical and InfraredDetectors, R. J. Keyes, Ed. Berlin, Germany: Springer-Verlag, 1977.

[19] C. C. Enz and G. C. Temes, “Circuit techniques for reducing the effectsof op-amp imperfections: autozeroing, correlated double sampling, andchopper stabilization,”Proc. IEEE, vol. 84, pp. 1584–1614, Nov. 1996.

[20] R. L. Bell, Negative Electron Affinity Devices. Oxford, U.K.: OxfordUniv. Press, 1973.

[21] J. S. Escher, “NEA semiconductor photoemitters,” inContacts, Junc-tions and Emitters, R. K. Willardson and A. C. Beer, Eds. New York:Academic, 1981.

[22] R. Chechik, A. Breskin, G. Garty, E. Shefer, B. K. Singh, and S.Shchemelinin, “Advances in single-charge detectors and their applica-tions,” IEEE Trans. Nucl. Sci., vol. 47, pp. 1812–1819, Dec. 2000.

[23] M. L. Heimer, A. Gutman, and A. Del Boca, “Using a photomultipliertube to measure weak bioluminescence,” inIEEE Proc. Southeastcon,vol. 1, 1991, pp. 494–498.

[24] L. Marcu, D. Cohen, J. M. I. Maarek, and W. S. Grundfest, “Character-ization of type I, II, III, IV, and V collagens by time-resolved laser-in-duced fluorescence spectroscopy,”Proc. SPIE, vol. 3917, pp. 93–101,2000.

[25] S. M. Sze,Physics of Semiconductor Devices. New York: Wiley, 1981.[26] G. E. Stillman and C. M. Wolfe, “Avalanche photodiodes,” inInfrared

Detectors II, R. K. Willardson and A. C. Beer, Eds. New York: Aca-demic, 1977.

[27] T. Kaneda, “Silicon and germanium avalanche photodiodes,” inLight-wave Communications Technology, R. K. Willardson, A. C. Beer, andW. S. Tsang, Eds. New York: Academic, 1985.

[28] D. Baechi, R. Buser, and J. Dual, “A high density microchannel networkwith integrated valves and photodiodes,”Sens. Actuators A, vol. A95,no. 2–3, pp. 77–83, Jan. 2002.

[29] W. F. Kosonocky, “Charge-coupled devices: an overview,” inWesternElectron. Show Conv. Tech. Papers, vol. 18, Sept 10–13, 1974, pp.2/1–2/20.

[30] R. J. Strain, “Properties of an idealized traveling-wave charge-coupleddevice,”IEEE Trans. Electron Devices, vol. ED-19, pp. 1119–1130, Oct.1972.

[31] M. F. Tompsett, “The quantitative effects of interface states on the per-formance of charge-coupled devices,”IEEE Trans. Electron Devices,vol. ED-20, pp. 45–55, Jan. 1973.

[32] J. E. Carnes and W. F. Kosonocky, “Noise sources in charge-coupleddevices,”RCA Rev., vol. 33, pp. 327–343, 1972.

[33] A. M. Mohsen and M. F. Tompsett, “The effects of bulk traps on the per-formance of bulk channel charge-coupled devices,”IEEE Trans. Elec-tron Devices, vol. ED-21, pp. 701–712, May 1974.

[34] P. Seitz, “Smart image sensors: an emerging key technology for ad-vanced optical measurement and microsystems,”Proc. SPIE, vol. 2783,pp. 244–255, 1996.

[35] R. H. Nixon, S. E. Kemeny, C. O. Staller, and E. R. Fossum, “128� 128CMOS photodiode-type active pixel sensor with on-chip timing, controland signal chain electronics,”Proc. SPIE, vol. 2415, pp. 505–514, 1995.

[36] Q. Zhu, H. Stiebig, P. Rieve, H. Fischer, and M. Bohm, “A novel a-Si(C):H color sensor array,” inProc. MRS Spring Meet. Symp. A—AmorphousSilicon Technol., San Francisco, CA, Apr. 4–18, 1994.

[37] P. Seitz, D. Leipold, J. Kramer, and J. M. Raynor, “Smart optical andimage sensors fabricated with industrial CMOS/CCD semiconductorprocesses,”Proc. SPIE, vol. 1900, pp. 21–30, 1993.

[38] G. N. Lu, G. Sou, F. Devigny, and G. Guillaud, “Design and testing of aCMOS BDJ detector for integrated micro-analysis systems,”Microelec-tron. J., vol. 32, pp. 227–234, 2001.

[39] J. M. Raynor and P. Seitz, “A linear array of photodetectors with widedynamic range and near photon quantum-noise limit,”Sens. ActuatorsA, vol. 61, pp. 327–330, 1997.

[40] J. H. Correia, G. de Graaf, S. H. Kong, M. Bartek, and R. F. Wolffen-buttel, “Single-chip CMOS optical microspectrometer,”Sens. ActuatorsA, vol. 82, pp. 191–197, 2000.

[41] M. L. Simpson, G. S. Sayler, G. Patterson, D. E. Nivens, E. K. Bolton,J. M. Rochelle, J. C. Arnott, B. M. Applegate, S. Ripp, and M. A. Guil-lorn, “An integrated CMOS microluminometer for low-level lumines-cence sensing in the bioluminescent bioreporter integrate circuit,”Sens.Actuators B, vol. 72, pp. 134–140, 2001.

[42] D. Cristea, F. Craciunoiu, M. Modreanu, M. Caldararu, and I. Cernica,“Photonic circuits integrated with CMOS compatible photodetectors,”Opt. Mater., vol. 17, pp. 201–205, 2001.

[43] U. Hilleringmann and K. Goser, “Optoelectronic system integration onsilicon: waveguides, photodetectors, and VLSI CMOS circuits on onechip,” IEEE Trans. Electron Devices, vol. 42, pp. 841–846, May 1995.

[44] A. J. Makynen, J. T. Kostamovaara, and T. E. Rahkonen, “CMOS pho-todetectors for industrial position sensing,”IEEE Trans. Instrum. Meas.,vol. 43, pp. 489–492, June 1994.

[45] A. Makynen, T. Ruotsalainen, T. Rahkonen, and J. Kostamovaara, “Highperformance CMOS position-sensitive photodetectors (PSDs),” inProc.IEEE Int. Symp. Circuits Syst., vol. 6, 1998, pp. 610–616.

[46] R. H. Nixon, S. E. Kemeny, C. O. Staller, and E. R. Fossum, “128� 128CMOS photodiode-type active pixel sensor with on-chip timing, control,and signal chain electronics,”Proc. SPIE, vol. 2415, pp. 117–123.

[47] Q. Zhu, H. Stiebig, P. Rieve, H. Fischer, and M. Bohm, “A novel a-Si(C):H color sensor array,” inProc. Mater. Res. Soc. Symp., vol. 336, 1994,pp. 843–849.

[48] S. S. Konanki and F. R. Beyette Jr., “Characterization and performanceevaluation of CMOS based photodetectors,” inLEOS Summer TopicalMeet., 2000, pp. 13–14.

[49] M. F. Gourley and P. L. Gourley, “Integration of electro-optical mechan-ical systems and medicine: where are we and where can we go?,”Proc.SPIE, vol. 2978, pp. 197–204, 1997.

[50] O. Leistiko, “Integrated microfluidic-optical detection system on achip,” Proc. SPIE, vol. 2978, pp. 94–102, 1997.

[51] R. Mariella, “Microtechnology for instrumentation,”Proc. SPIE, vol.3258, pp. 216–228, 1998.

[52] K.-H. Brenner, “Development of modules for micro optical integrationand MOEMS packaging,”Proc. SPIE, vol. 4178, pp. 138–140, 2000.

[53] B. Schelle, P. Dress, H. Franke, K. F. Klein, and J. Slupek, “Physicalcharacterization of lightguide capillary cells,”J. Phys. D, vol. 32, no.24, pp. 3157–3163, Dec. 21, 1999.

[54] R. Mariella, “Instrumentation for biomedical and environmental appli-cations based on microtechnology—lessons learned,” inSPIE Conf.Micro- and Nanofabricated Structures and Devices for BiomedicalEnvrionmental Applications II, vol. 3606, San Jose, CA, 1999, pp.74–81.

Rachel A. Yotter (M’01) received the B.S. degreein electrical engineering from the University of Cal-ifornia, Berkeley, in 1998. She is currently pursuingthe M.S. degree in electrical engineering at the Uni-versity of Washington, Seattle.

Her research interests include MEMS, neurophys-iology, sensors, and biological instrumentation. Hercurrent research topic is photonic and chemical sen-sors for single-cell detection in proteomic and ge-nomic applications.

Denise Michelle Wilson (M’89) was born inChicago, IL, in 1966. She received the B.S. degreein mechanical engineering from Stanford University,Stanford, CA, in 1988 and the M.S. and Ph.D.degrees in electrical engineering from the GeorgiaInstitute of Technology, Atlanta, in 1989 and 1995,respectively.

She is currently an Associate Professor in the Elec-trical Engineering Department, University of Wash-ington, Seattle, and from 1996 to 1999, she was withthe University of Kentucky, Lexington, in a similar

position. Her research interests focus on the development of signal processingarchitectures, array platforms, and other infrastructures for visual, auditory, andchemical sensing microsystems. From 1990 to 1992, she was with Applied Ma-terials, a semiconductor capital equipment supplier.

A review of photodetectors for sensing light-emitting reporters in … · 2008-02-22 · A Review...

Documents

Transcript of A review of photodetectors for sensing light-emitting reporters in … · 2008-02-22 · A Review...