SPECIAL FEATURE: TUTORIAL Fundamentals of Focal · PDF fileSPECIAL FEATURE: TUTORIAL Fundamentals of Focal Plane Detectors ... variation which approximates to a Gaussian variation

JOURNAL OF MASS SPECTROMETRY, VOL. 32, 795È806 (1997)

SPECIAL FEATURE: TUTORIAL

Fundamentals of Focal Plane Detectors

Keith Birkinshaw*Department of Physics, University of Wales Aberystwyth, Aberystwyth, Dyfed SY23 3BZ, UK

Spatial dispersion of ions in one dimension is a well established means of analysing ion mass and focal planedetectors (FPDs) allow ions of a wide range of masses to be recorded simultaneously. This paper is concerned withthe principles governing the performance of FPDs and the types of FPD available. It is focused on magnetic sectormass spectrometry but is relevant to all applications in which spatially dispersed particles can be detected using amicrochannel plate electron multiplier, e.g. ions, photons of wavelength Æ200 nm, electrons and energetic neutrals.Although it has proved possible to produce mass spectra with a high resolution, this has not been matched by anability to detect them efficiently. Given that highly resolved spectra are available at the detector but are inaccessi-ble efficiently, it is in the development of high-performance FPDs where there are enormous gains in efficiency tobe achieved. Limitations of FPD performance of two fundamental types are discussed : the position of impact of anion on the FPD cannot be measured exactly, and the upper and lower count rates of the FPD are both restricted.These limitations are not simply characterized but are sometimes determined by the electron multiplier stage,sometimes by the properties of the array and sometimes by the data acquisition system. 1997 by John Wiley &(

Sons, Ltd.

J. Mass Spectrom. 32, 795È806 (1997)No. of Figs : 20 No. of Tables : 1 No. of Refs : 26

KEYWORDS: focal plane detectors ; magnetic sector mass spectrometry

INTRODUCTION

In mass spectrometry, ions of di†ering mass may bemeasured by (i) dispersion in space (e.g. magnetic sector)or time (e.g. time-of-Ñight), (ii) Ðltering (e.g. quadrupole)or (iii) absorption of electromagnetic energy [e.g.Fourier transform ion cyclotron resonance (FT-ICR)].Focal plane detectors (FPDs) are used only whenspectra are spatially dispersed. Electron and photonspectroscopy also normally involve spatial dispersionand have been a driving force in FPD development.Although magnetic sector mass spectrometer design andion optics have advanced over the years and enabledthe production of very highly resolved mass spectra, thiscapability has not been matched by an ability to detectthe dispersed ions efficiently. The basic problem is thatthe high resolving power of a single-slit detector isincompatible with high collection efficiency. This isillustrated in Fig. 1, which shows that the measuredpeak proÐle depends on the slit width (it is the convolu-tion of the incident proÐle and the slit width). The mea-sured proÐle more accurately represents the incidentproÐle when the detector slit is narrow [Fig. 1(a)] butclearly a narrower slit means that a smaller fraction ofthe spectrum is measured at any one time and the entirespectrum takes much longer to measure.

Consider the hypothetical ion beam with the prob-ability distribution P(x) shown in Fig. 1(a). If the dis-

* Correspondence to : K. Birkinshaw, Department of Physics, Uni-versity of Wales Aberystwyth, Aberystwyth, Dyfed SY23 3BZ, UK.

persed ion beam falls across a narrow detector slit ofwidth a small section of the beam is sampled asW N , INshown. The measured ion peak intensity is given to agood approximation by :

IN \ WN[P(x@) ] P(x@] WN)]/2

When is very small, this givesWNIN \ WN P(x)

and measurement of gives the beam proÐle to a con-INstant factor. Note that the probability distribution P(x)has units of counts s~1 m~1 and has units of countsINs~1.

Consider the wide detector slit (2 ] FWHM) in Fig.1(b). Normally the beam is scanned across the detectorslit. For convenience of illustration we imagine a sta-tionary beam with the slit moving from left to right. The

Figure 1. Measurement of an incident particle beam using (a) anarrow slit and (b) a wide slit.

CCC 1076È5174/97/080795È12 $17.50 Received 19 February 1997( 1997 by John Wiley & Sons, Ltd. Accepted 9 May 1996

796 K. BIRKINSHAW

measured ion peak intensity proÐle is found by inte-grating the area under the proÐle falling above the exitslit. For example, the area represents the section ofIWthe beam passing through the slit in position (i). This isplotted at the position of the slit centre along withintensities measured at other positions in the lower partof the Ðgure and it is seen that the measured peak iswider.

Unfortunately, narrow slits sample a mass spectrumvery inefficiently. It is the objective of FPDs to detections over a much greater fraction of the focal plane thanpossible with a single slit. It is in the area of FPD devel-opment where large gains in efficiency are to be made.The emphasis here will be on one-dimensional multi-event discrete FPDs (see below). These devices have ahigh data accumulation rate as particles can be detectedat many sites simultaneously. Other types of FPD arealso mentioned below.

Ion detection

All ion signals result from the combined e†ects of oneor more discrete ion events. Where the Ñux is low theseevents may be counted (digital measurement) or theire†ects may be integrated (integration). For high ionÑuxes, direct measurement of the ion current (DCmeasurement) is possible. In general, non-digital mea-surements are referred to as “analogueÏ measurements.

Digital. In the case of ion (or electron or photon)counting, the occurrences of events are recorded [Fig.2(a)]. Measurement system noise can be almost elimi-nated to the point where the Ñuctuation in the observedcount rate is controlled by ion arrival statistics and israndom. At low counts the peak height shows a Poissonvariation which approximates to a Gaussian variationat high counts with a precision of where N is theJN,number of counts accumulated. Very low ion currents

Figure 2. (a) Measurement of an electron multiplier pulse. Theelectron multiplier may be preceded by a defining slit. If a detectorarray is used then the electron multiplier is an MCP and there is nodefining slit. (b) Measurement of ion flux by integration of freeelectrons created at photosites by photons emitted at the phosphorscreen. (c) Direct measurement of the ion current for large ion flux.

can therefore be measured by counting under condi-tions where DC current measurements would introducea measurement error greater than the signal. However,it is beyond the capacity of pulse counting systems tocount ions at rates comparable to the switching times1of the transistors as the sensing and counting circuitrycannot respond sufficiently rapidly. The upper countrate limit may also be determined by the response time(below) of the electron multiplier (cf. Fig. 2).

Integration. Two-dimensional semiconductor arrays arewidely used in video cameras for visible photon detec-tion. These can be adapted for use in ion detection byÐrst converting ions to photons, as shown in Fig. 2(b).Photons reaching a photosite excite electrons to theconduction band. Electrons in the conduction band arecalled “free electrons.Ï Other events cause the excitationof more electrons and the total number of free electronsis a measure of the number of ion events incident abovethe photosite. At room temperature free electrons aregenerated thermally and this introduces thermal noise,which can be reduced by reducing the operating tem-perature. This is in addition to noise introduced duringthe measurement of the integrated free electrons. Thesedevices can be used to count events, although they arebetter suited to the integration mode. They can beadapted to measurements in one dimension by averag-ing the signals in the dimension perpendicular to disper-sion.

DC. A sufficiently high ion Ñux can be regarded as acontinuously variable current and measured directly asshown in Fig. 2(c). In the measurement of continuoussignals the measurement system noise and the signal areadded. There is no clear upper limit to the measurementof a DC current. In practice, the minimum currentdetectable by DC measurement is about 10~15 A (about104 ions s~1) but only with a slow ([1 s) responsetime.2 This can be greatly improved using an electronmultiplier.

Focal plane detection. Examples of focal plane detectorsare presented later. The job of an FPD is to measure awhole section of spectrum simultaneously. Figure 3shows a multi-event discrete detector FPD consisting ofa one-dimensional array of metal strips (detectorelectrodes) which form the detector inputs. An ionimpacting on the MCP initiates an electron pulse at the

Figure 3. A multi-event discrete detector FPD.

( 1997 by John Wiley & Sons, Ltd. JOURNAL OF MASS SPECTROMETRY, VOL. 32, 795È806 (1997)

FUNDAMENTALS OF FOCAL PLANE DETECTORS 797

Figure 4. Peaks taken from the spectrum of methyl pentafluorop-ropionate.

MCP output which falls on the electrodes and iscounted. If the electron pulse is wide a single count maybe recorded on more than one electrode. A device ofthis type, an integrated array of 192 detectors on asingle silicon chip, has been successfully produced atAberystwyth.3h6 It is a small, low-power, high-resolution device which realizes the advantages of afully parallel detector system. Each detector operatesindependently and consists of a metal strip (detectorinput electrode) exposed to the MCP output, a chargesensor, an eight-bit counter, control logic and a businterface. The detector electrode pitch (the distancebetween the centres of adjacent electrodes) is 25 lm.

In the 1970s, FPDs were developed following MCPdevelopment at a time when custom integrated circuitswere not easily accessible. Therefore, arrays were devel-oped which required little electronics. Unfortunately,they were limited to low particle Ñux where the positionof arrival of each ion is computed and registered beforearrival of a subsequent ion (see later). Fully parallel par-ticle counting is only possible with a dedicated sensorand memory at each detector site. Timothy and Bybeehave produced a one-dimensional discrete detectorFPD containing 64 detectors7 and a two-dimensionaldevice8 containing 100 detectors. Each FPD consistedof detector electrodes deposited on a ceramic substrateand these were connected to external (non-integrated)charge ampliÐers and other data-handling circuitry. Thenumber of detectors contained by these FPDs waslimited by the large amount of external electronicsrequired. Integration of a small number of detectors atlow resolution is today a simple problem but a high-resolution device on a silicon chip requires carefuldesign to accommodate a high density of electronicswith high performance and functionality, an acceptableyield and robustness. This has been achieved as dis-cussed below.

Figure 4 shows an example of a mass spectrum mea-sured using the Aberystwyth array. Each bar of the his-togram gives the number of counts measured by one ofthe detectors on the array.

Figure 5. Similarities and differences between a single-slit detec-tor (left) and the FPD (right). The FPD contains no defining slitand therefore and detector array sites are exposed to the high-gainMCP output of remote events.

The ideal FPD

An ideal FPD would record the exact location of everyion in the spectrum. Why cannot this be achieved?There are two reasons : the point at which an ion hitsthe electron multiplier cannot be recorded exactlybecause of limitations of both the MCP and the array ;and both the MCP and the array place limits on therate at which ion Ñux can be accurately measured.

In this deÐnition of ideal, we are assuming that thedata can be transferred rapidly to the data acquisitionand display system. If this were not the case then thedata transfer rate itself might determine the upper limitof the Ñux. We are also temporarily ignoring otherissues of importance to a user such as size, complexityand cost. The reason for these omissions is that we arefocusing on fundamental performance issues which limitthe ideal measurement of the exact location of every ionin the spectrum.

Currently, single ions cannot be detected withoutampliÐcation and some form of electron multiplier isessential. Figure 5 shows that in the case of the singleslit detector the part of the spectrum to be measured isdeÐned by the slit before multiplication, which isachieved with a conventional single-stage electronmultiplier. There is necessarily no slit present when anFPD is used and the spectrum falls directly on a multi-element electron multiplier known as a microchannelplate (MCP). Therefore, there is no deÐnition of the partof the spectrum falling on the multiplier and a detectorsite of the array is exposed not only to MCP pulsescentred directly above it but also to the high gain pulsesof more remote events (Fig. 5). This limits the per-formance in di†erent ways depending on the type ofFPD, as will be discussed below.

A single slit has a signiÐcant advantage in that theposition of arrival of an ion can be known to an accu-racy determined by the slit only. When an event islocated from an MCP pulse measured using an array,then the position of arrival cannot be known to betterthan the MCP channel diameter as all information onthe position of arrival of the ion within the channel islost. However, because an ion peak is an accumulationof many events occurring over the length of one ormore electrodes the (normally) random registration ofMCP channels above an electrode can result in a mea-sured peak whose centre is given more accurately thanthe MCP channel diameter.9

The high resolving power of a narrow single slitunfortunately means that a small fraction of the spec-trum is measured at a given time, i.e. there is a lowercollection efficiency. The driving force behind the devel-opment of FPDs is the need for a greater collection effi-ciency. The problem is to achieve higher collectionefficiency and match the cost and performance advan-tages of a single slit.

ARRAYS ON SILICON

Each detector of a discrete detector array requires itsown charge sensor and a counter large enough to avoid

( 1997 by John Wiley & Sons, Ltd. JOURNAL OF MASS SPECTROMETRY VOL. 32, 795È806 (1997)

798 K. BIRKINSHAW

overÑow before reading. Given the large amount of cir-cuitry needed, it is the case that a high-resolution arrayof ion counters cannot sensibly be achieved any otherway than by integration. It is conceivable that someother semiconductor (e.g GaAs) or fabrication tech-nology (e.g. silicon on sapphire) could be used, butcurrently CMOS (complementary metalÈoxideÈsemiconductor) technology is the most cost e†ectiveroute. Therefore, silicon technology is likely to domi-nate in high-resolution one-dimensional arrays of eventcounters because (i) a high resolution array of discretecounters requires a great deal of associated electronicsand there is no case for non-integrated electronics, and(ii) continuing high investment in silicon technology willlead to increasingly high speciÐcations and lower costs.

It is normally extremely difficult or impossible tomake circuit changes after a chip has been fabricatedand therefore all design features must be identiÐed atthe outset and incorporated in the design. The majoraim is then to achieve all objectives on a single device.The ability to recognize all important design featuresinevitably requires a multidisciplinary approach. Theissues mentioned below represent some of the grossissues involved in the development of a high per-formance array detector.

Sensitivity

The central issue is whether a silicon chip based detec-tor can be made sensitive enough to detect a singleMCP pulse and the answer is a very clear “yesÏ. Thetypical capacitance of a metal layer (from which detec-tor electrodes are formed) on a silicon chip is about2 ] 10~17 F lm~2. The MCP output is around 107electrons (1.6] 10~12 C)10 spread over a circle ofradius about 50 lm (area about 8000 lm2). The voltageinduced on a circular metal electrode of radius 50 lmby 107 electrons is

V \ Q/C\ 1.6] 10~12/(8000 ] 2 ] 10~17)\ 10 V

The capacitance can be reduced by introducing athicker insulating layer to separate the metal electrodefrom underlying circuitry. An electrode of size 2mm] 18 lm on the Aberystwyth array has a capac-itance of around 0.2È0.4 pF. Thus easily detectablevoltage pulses ([0.2 V) are induced by the MCP pulse.The probability of thermal noise of this magnitude isextremely low, giving essentially zero noise for thearray. The MCP itself generates less than three “darkcountsÏ per cm2.11 Given that each detector electrodehas an area of about 4 ] 104 lm2, there are about 10~4counts s~1 for each electrode.12

Silicon area

The second key issue is whether a high-resolutiondevice with high performance and built-in testabilitycan be accommodated on a small silicon chip. As adetector array is a repeat of many identical units, asingle unit must Ðrst be designed. For a spatialresolution of 25 lm there must be 400 detectors per cmand all the circuitry must Ðt within the boundary of the

Figure 6. Schematic diagram of the layout of the Aberystwythdetector array.

chip. Figure 6 shows a plan of the array developed on asilicon chip at Aberystwyth. The detector electrodes arearranged along one side of the chip. All associated cir-cuitry is placed as shown.

The dimension d is Ðxed by the spatial resolution, i.e.the more detector electrodes per unit length (length is inthe direction of l) the more circuitry must be accommo-dated within the distance d. In the case of theAberystwyth array d was 14 mm and l was 5 mm, givingan acceptable yield.

In spite of the large size and dense circuitry, theenergy dissipation is so low that there is no need forcooling even in a vacuum. Low heat dissipation is char-acteristic of CMOS technology.

Robustness, lifetime

Clearly, an FPD will be of little value if it is extremelyfragile or has a short lifetime. The focal plane of a massspectrometer appears to be in a hostile environment fora silicon chip, especially given the proximity of a high-voltage MCP. The possibility of a high-voltage dis-charge would deÐnitely pose a potential hazard. Thisdanger to the Aberystwyth array was minimized byadding extra diode protection as an integral part of thechip to short circuit high-voltage Ñuctuations and byincorporating a high resistance in the high-voltageMCP power line. Most arrays survive numerous dis-charges.

Apart from an accidental discharge, the only wearand tear on the array is due to impact by MCP elec-trons. These emerge from the MCP with a kineticenergy of about 30È100 eV,3 and there is no obviousreason why the bombardment with low-energy electronsshould reduce the lifetime of the array. Two arrays havebeen in intermittent use for over 3 years without anoticeable change of performance.

Physically, the detector array chip is not undulyfragile except in two respects : the surface of the siliconmust not be scratched, and bond wires connecting thechip to the ceramic substrate are very fragile. The Ðrstproblem is avoided by careful handling. Encapsulationof bond wires in a bakeable epoxy resin has proved asuccessful solution to the second problem.

Technology

Silicon technology is now a major force in electronics,but small-volume devices do not usually attract the



attention of designers and hence the value of this tech-nology is often not exploited as users are unable todesign the devices themselves. This is slowly changingand important devices are being produced, such aschemical, physical and medical sensors.

What does silicon have to o†er in array design? TheÐrst point is that there is no other sensible way toproduce a straightforward device containing a greatdeal of electronics than by integration. However, thereare other signiÐcant advantages to be had from integra-tion on silicon :

sensitive circuitry is an integral part of the arrayI Allinside the vacuum system in a shielded environment.A low-impedance bus to the outside is insensitive tonoise.

capacitance of the detector electrodes. TypicalI LowMCP output pulses induce voltage pulses of [0.2 Von electrodes and hence the pulse height discrimi-nation level (PHDL) can be set well above thethermal noise threshold.

device allows the elimination of cable and instru-I Thement clutter and inherits many of the characteristics(e.g. cheapness, reliability, increased functionality,small size and weight) which have led to the domina-tion of integrated circuits (ICs) in many areas.It is a characteristic of CMOS technology that very

little power is consumed. The Aberystwyth array con-sumes about 5 mW for 192 detectors and could bepowered by a 5 V battery with no cooling necessary.The integration of a complete high-resolution discretedetector array system, including computer interface, ona silicon chip is in line with developments in “smartÏsensors, and although the research and developmentcosts are high the production cost of a 192 detectorarray is low. Therefore, the prospect of disposablearrays exists.

POSITION MEASUREMENT

The objective of Ðnding the location of an event hasbeen achieved in a number of ingenious ways. Thesehave been reviewed by several authors.12,14,15 In thefollowing, methods of event location are arranged inclasses and an example in each class is discussed. Afuller coverage of individual FPDs can be found in thereviews. It should be noted that many branches of massspectrometry, such as time-of-Ñight, quadrupole andFT-ICR, do not rely on position measurement for massanalysis.

Discrete sites

Discrete detector array. It is generally considered that thediscrete detector array is the most straightforward, con-ceptually the simplest and the best approach if countingstatistics are required at every site. A 1D discrete detec-tor array can be viewed as many independent single-slitdetectors placed side-by-side and events can be record-ed at each detector simultaneously.6h8 Such a devicemay be called a “multiple-eventÏ array to distinguish itfrom “single-eventÏ arrays described below, which can

only measure events sequentially. The barrier to thedevelopment of a multiple event discrete detector arrayis that for particle counting applications there must be adedicated sensor and counter at each site. Such a largeamount of non-integrated electronics is expensive,complex and limits the performance by increasing thecapacitance of detector sites as well as site-to-site capac-itative cross coupling. However, integration of this cir-cuitry on a silicon chip allows the minimization ofcomplexity, cost, capacitance and cross-coupling in onego. A later section focuses on the performance of thediscrete detector array.

Integrating FPD. In this device (Fig. 7) the MCP outputpulses are accelerated on to a phosphor screen and theresulting pulse of photons is transmitted to a photo-diode array16 or a CCD17,18 (e.g. by channelling downan optical Ðbre bundle). Resolution is lost at each inter-face and the photon pulse width measured is aminimum of 100 lm. This resolution combined with agood dynamic range has allowed the use of thesedevices in mass spectrometers.

A useful characteristic of this type of array is that inapplications where bunches of particles arrive at asingle detector site in a very short time (e.g. 1 ns) thee†ect of the resulting MCP pulses would be correctlyintegrated (assuming the number of particles does notsaturate the MCP or the photosite) whereas the discretedetector array would register only one event.

Charge division

Events can be located by charge division by dividingthe MCP output pulse between two or more detectorsand computing the event position in terms of the chargereceived by the detectors, as illustrated below. Thus theposition of each event is computed and the electronicsrequired are minimized. However, the ion Ñux must below enough to ensure that there is little chance ofarrival of a second ion before an event location is com-puted and logged. A high-intensity peak would “blindÏthe FPD and would have to be moved o† the FPD orreduced in intensity if much smaller peaks are to bemeasured.

The resistive strip. In its simplest form (Fig. 8) theresistive strip FPD consists of a thin layer of resistivematerial on the surface of an insulator with electrodesattached to each end. The path resistance seen by anelectron pulse to each end of the strip depends on theposition of arrival of the pulse and this position can becomputed simply from the magnitudes of the chargesarriving at each end. A “spatial location errorÏ of about21 lm over 25 mm has been obtained with this device.19

The resistive strip can be represented as a resistor. Apulse landing at a distance x@ from the left (Fig. 9)

Figure 7. Electro-optical FPD.


800 K. BIRKINSHAW

Figure 8. Resistive strip FPD.

induces a small potential and current Ñows to both endsin the ratio determined by OhmÏs law:

iBiA

\ RARB

\ c

Therefore,

iB\ ciAPt1

t2iB dt \ qB\ c

Pt1

t2iA dt \ cqA

qBqA

\ c\ RARB

\ x@x [ x@

and

x@\ qB

qA

] qB

x

Clearly, for accurate event location the resistive layermust be uniform. The main sources of error are in non-uniformities of this layer, electronic noise and thermalnoise. Other important devices exist in which the chargeis divided by capacitative coupling between discreteelectrodes or by varying the dimensions of electrodesacross an array such that the charge division is relatedto the event location.12,14,15

Coincidence

A family of “multi-anode microchannel arraysÏ(MAMAs) have been described by Timothy and Bybee8for one- and two-dimensional imaging. The location ofan event is determined by the simultaneous detection ofa charge pulse on two sets of anodes deposited on aceramic substrate (Fig. 10). The spatial resolution of theanodes was 25 lm in the device described by Timothyand Bybee.

Figure 9. Measurement of the location of an event using a one-dimensional resistive strip.

Figure 10. Multi-anode coincidence detector.

Electrodes are connected alternately to coarse andÐne position encoding ampliÐers as shown. The ambi-guity in the location of an event when measured by theÐne position encoder is resolved by the coarse positionencoder. The Ðnal ambiguity of whether the event iscloser to the Ðne position electrode or the coarse posi-tion electrode is resolved by determining the electrodereceiving the greater charge.

The position of an event in two dimensions has beenfound using a grid of crossed wires. With a grid ofn ] m wires the positions of arrival at n ] m points canbe found using n ] m ampliÐers. Coincident signals onthe orthogonal wires indicated real events. A resolvingpower of 10 lm has been achieved using a crossed gridof 100 lm wires on a 200 lm pitch. The device cancount at rates up to 104 s~1.

THE MCP OUTPUT

In understanding the FPD performance it is essential tounderstand the factors which inÑuence the gain and thespreading of the electron pulse output by the MCP. Theperformance of MCPs has been studied exten-sively.20h23 We divide the FPD into three regions (Fig.11). In the Ðrst region electron multiplication occurs, inthe second the pulse widens and in the third the pulse ismeasured. Measurement of the MCP output is per-formed in di†erent ways by di†erent arrays. A com-prehensive survey is beyond the scope of this paper andthe discrete detector array is considered here. The issuesraised have relevance for other types of array.

Electron multiplication

The MCP provides ampliÐcation and also retainsspatial information. A schematic diagram of an MCP is

Figure 11. Diagram showing the division of an FPD into regionsfor discussion purposes.



Figure 12. Multiplication of electrons by an MCP. (a) A particleentering an MCP channel emits one or more electrons which areaccelerated along the channel to give further collisions and multi-plication of the signal. (b) A single event leads to a wide electronpulse at the MCP exit with further spreading on travelling to thearray. The channels are arranged to slope in opposite directions toreduce ion feedback and optimize, the incident ion detectionefficiency.

shown in Fig. 12(a). It consists of a plate typically ofthickness 0.5 mm and composed of many small tubes(typically straight tubes of diameter 12 lm) each ofwhich is at an angle of typically 10¡ to the vertical andhas an internal coating of a low work function material.A voltage (typically O1 kV) is applied between themetallized front and rear surfaces of the MCP. An ionentering a MCP channel causes the release of secondaryelectrons, which are accelerated down the channel, eachelectron releasing further electrons resulting in a pulseof G (gain) electrons being emitted at the MCP exitwithin 1 ns. The Ðring of an MCP channel is analogousto the sudden discharge of a capacitor and the capacitormust be recharged by current in the channel wall. Typi-cally the recharge time is 10~2 s. In order to increasethe gain, two MCPs may be stacked together as shownin Fig. 12(b).

Of most importance here is to understand the proper-ties of the electron pulse output by the MCP, as this isthe evidence of the arrival of an ion to be measured bythe array. Here we will consider the output of a singleMCP and the output of a stack of two MCPs.

Pulse gain. Single plate : gain distribution. The gain dis-tribution of a single MCP plate is typically a quasi-exponential decay (average gain about 104) as shown inFig. 13. As the supply voltage is increased the distribu-tion shifts to higher gains but the e†ect known as “ionfeedbackÏ limits the maximum operating voltage. Athigh supply voltages (typically [1 kV), ions formedfrom residual gas are accelerated back up a MCPchannel, collide with the channel wall and initiate sec-ondary electron pulses. This degrades the performance

Figure 13. Schematic diagram of the MCP pulse gain distributionf(G) of a single MCP plate. The pulse voltage distribution f(V)induced on the detector electrodes by the MCP electron pulsesmirrors the gain distribution and is plotted on the same axes. Apulse is counted by a detector only if the induced voltage on thedetector electrode is ¿PHDL.

and may lead to a continuous discharge. It can be seenthat all pulses are counted (i.e. they are above the pulseheight discrimination level (PHDL) of the detectors)only if the PHDL is set very low, in which case noisewill degrade the data.

MCP stack : gain distribution. If two MCPs arearranged as in Fig. 12(b), ions formed from residual gasin the second MCP are stopped at the interface betweenthe MCPs before they acquire sufficient energy torelease secondary electrons, thus reducing ion feedbacke†ects. The input of each channel of the second MCPreceives sufficient electrons to drive it to saturation.When the channel output is saturated it is independentof the initiating event and the gain distribution ispeaked. The importance of this for event counting canbe seen from Fig. 14. The PHDL can be placed abovethe noise level but below the signal voltage pulse dis-tribution induced by the MCP output. It can be seenthat in this case all events are detected (i.e. they areabove the PHDL) and that for small Ñuctuation of thePHDL or MCP gain (which shifts the pulsedistribution) there is little variation of the measuredsignal. This is the region of stable operation.

Capacitative cross-coupling between detector elec-trodes gives an image of the signal voltage distributionon adjacent electrodes. The dotted distribution in Fig.14 shows such an image. If cross-coupling is low thetwo distributions may not overlap and the PHDL maybe set at a level which detects all events without mea-suring “imageÏ pulses on adjacent electrodes or thenoise. Thus the measured noise level would be theminimum attainable, inherent in the random time ofarrival of the ions, the purely statistical. There would beno region of stable operation if the signal and noiseoverlap signiÐcantly.

Typically the FWHM of the gain distribution isabout ^50% of the gain. If the MCP supply voltage isreduced then the peak moves to lower gain and theproÐle eventually changes to a quasi-exponential fall o†.

Pulse spreading

An MCP pulse is output in \1 ns. The electrons fall ondetector electrodes and induce voltage pulses. The Ðrstquestion to be addressed is how many electrons fall oneach electrode. This is determined by the gain and theproÐle of the pulse which overlaps the electrodes.

Pulse proÐle. A key issue in understanding the per-formance of a multievent discrete detector FPD is the

Figure 14. Schematic diagram of the gain distribution of a stackof two MCP plates. The dotted distribution shows pulses capac-itatively induced on an electrode by MCP pulses falling on adja-cent electrodes.


802 K. BIRKINSHAW

shape of the electron pulse landing on the electrodes.Figure 12 shows that the electron pulse initiated by asingle event can become very wide on reaching thearray. The proÐle is not known exactly but shouldapproximate a Ñat-topped peak with Gaussian fall-o†.The factors which inÑuence the width of the measuredpeak include the following.

Separation between MCPs in a stack. A number of chan-nels of the second MCP are activated by one channel ofthe Ðrst MCP. Even when the plates are in intimatecontact, it has been observed24 that the electron pulseemerging from the second MCP has a diameter ofabout 100 lm, corresponding to the activation ofaround 50 channels. This may be due to non-planarityof the MCP plates. An electric Ðeld between the stackedMCPs helps to reduce the electron pulse spreading.

MCP gain. There is greater spreading for a larger gainMCP output pulse as spaceÈcharge repulsion (mutualelectrostatic repulsion of the electrons) is greater.

Separation between the MCP and the array. The greaterthe separation, the greater is the spreading of the elec-tron pulse. This is due not only to geometric factors butalso to the increased time during which spaceÈchargerepulsion is active.

Field between the MCP and the array. An attractive Ðeldwill help to conÐne the electrons and give less spread-ing.

Capacitative coupling between electrodes. A voltagepulse will be coupled to adjacent electrodes due tocapacitative coupling between the electrodes thusspreading the e†ect of the pulse.

A computer model of the Aberystwth FPD includesall the above factors and has given good agreementwith experimental results.24

Curved channel MCPs25 provide a high gain from asingle channel, thereby avoiding the need for a stack oftwo MCPs. An electron pulse of high gain emergingfrom a single channel would have a small diameter butwould experience higher spaceÈcharge repulsion.Curved channel MCPs are currently not available com-mercially except by special order.

Pulse measurement

The voltage pulses induced on the detector electrodesmust be measured. If the pulse height on an electrode isabove a set threshold (PHDL), then a single count isadded to the counter associated with the electrode. InFig. 15(a), most of the MCP pulse falls on a single elec-trode and a count is registered at only one site. In Fig.15(b) a wider MCP pulse gives voltage pulses higherthan the PHDL on three electrodes, resulting in therecording of a single count at three sites.

Uniformity

It is immediately obvious that in the case of an FPD anunwelcome factor presents itself which is not an issuefor the single-slit detector, namely that any non-

Figure 15. Voltage pulses induced on detector electrodes by asingle event. Pulse heights are represented by vertical lines. (a)Charge lands on a single electrode. Small pulses are induced onadjacent electrodes due to capacitative coupling but fall below thethreshold for measurements set by the PHDL. (b) A wide electronpulse falls on several electrodes giving voltage pulses ¿PHDL onthree electrodes.

uniformities of the multiplier or array will inÑuence themeasured spectrum. The quality of MCPs is generallyhigh. When properly commissioned, low dark countsover the entire MCP are normally observed. Non-uniformities of the MCP quality can take the forms ofvariations in the channel to channel average gain andnon-planarity of the MCP plates. The latter might givea variation of electron pulse spreading (see above)because of the variation of the distance of separationbetween MCPs in a stack and also the distance betweenthe MCP output and the array. However, as manyMCP channels service a single detector electrode, varia-tions of channel to channel performance should be aver-aged out. On the other hand, the e†ects of longer rangevariations in planarity may not be averaged out.

Dynamic range

The duration of a single MCP pulse is of the order of10~9 s and it may appear that a single MCP channelcould deliver around 109 distinguishable pulses persecond. However, as pointed out above, each channelrequires a certain time to recover (typically 10~2 s) andthe maximum pulse rate which can be delivered by anMCP channel without signiÐcant gain reduction istherefore of the order of 102 s~1. Up to about 1000MCP channels can service a 2 mm] 18 lm electrode,giving a maximum count rate of the order of 105 s~1.

However, the random time and position of arrival ofthe ions also inÑuence the maximum Ñux accuratelymeasurable. Even when the ion Ñux is below themaximum calculated above, some ions will by chancearrive at channels within the recovery time and may notbe counted owing to the reduced gain of the channel. Ifthe total incident ion Ñux on N MCP channels(Ilim)were N/(recovery time), then by chance 26% would fallon unrecovered channels. At a Ñux of this would0.4Ilimfall to 6% and at a Ñux of to 2%.0.2Ilim

FPD PERFORMANCE

A complete detector system incorporating an integrateddiscrete detector array includes the array, an MCP, acomputer (a PC) and associated power supplies. Thearray contains all the sensing circuitry, eight (or more)bit data bu†ers and a computer interface. A completespeciÐcation would be lengthy and therefore only threeof the most important performance issues, uniformity,



dynamic range and resolving power, will be consideredhere. These issues are relevant to all FPDs but forreasons of space the discussion here is restricted to thediscrete detector FPD.

Uniformity

Figure 15 shows a uniform PHDL across an array. Ifthe PHDL is not the same for all detectors, then thosedetectors with lower than average PHDL will onaverage record more than their fair share of pulses, andthe reverse is true for the detectors with higher thanaverage PHDL.

Uniformity is not an issue for a single slit where par-ticles land on the same multiplier with the same mea-surement electronics. In the case of a discrete detectorarray on a silicon chip each detector has its own chargepulse sensing circuitry and small di†erences between thePHDLs of the detectors will arise due to manufacturingtolerances, etc. Non-uniformity is reÑected in the accu-racy with which peak heights and positions can befound from the quantized position information. Digitalsignal processing yields both the height and position ofpeaks and as the PHDL variation across the array canbe measured, corrections for this can be incorporated.For the Aberystwyth array processed measurements ofa spectrum have shown that a position accuracy of theorder of 0.2 lm (as deÐned by the standard deviation ofthe distance between two peaks as they were sweptacross the array) is achieved and the relative peakheight variation approximately equals that expected sta-tistically.9 The raw data used in this analysis are shownin Fig. 16. The larger peak was moved in steps of 12.5lm so that it was alternately directly above an electrodeand then between two electrodes hence the alternatehighÈlow variation of peak height. It should be notedthat a constant peak height (but lower resolving power)is obtained with a lower PHDL.12

Quantization. The location of an event cannot be knownexactly whether a single slit or an FPD is being used. Inthe case of a slit, the width limits the position accuracy.In the case of a silicon detector array, the positionalaccuracy is limited by any non-uniformity of the chargepulse sensor performance and quantization of the posi-tional information. Both the MCP and the array posi-tional information are quantised.

Figure 17 is an approximate scale diagram of asection of a discrete detector array preceded by an

Figure 16. Two ion peaks scanned across the Aberystwyth FPD.(a) The peaks were positioned alternately directly above a detectorelectrode and between electrodes. (b) The ion peaks at oneposition.

Figure 17. An approximately scale diagram showing relative sizesof detector electrodes and MCP channels.

MCP. The registration of MCP channels and electrodesis normally unknown.

Because many MCP channels service a single detec-tor electrode, variations of channel to channel per-formance should be averaged out. The peak measuredby a discrete detector array is in fact a histogram andalthough the height and centroid of such a peak cannotbe read o† by eye as easily as in the case of a contin-uous outline, they can be rapidly computed using digitalsignal processing techniques.

Dynamic range

A spectrum often contains many peaks of greatly di†er-ing height and one objective is to measure the height ofeach one as accurately as possible. The dynamic rangeof an FPD is the range over which the ion Ñux can beaccurately measured. At very low ion Ñuxes measure-ments become inaccurate when the noise contributionto the measured Ñux is comparable to the ion signal.The lower limit of the dynamic range may be deÐned asthe point at which the signal and the noise are equal, i.e.the lower limit is the noise level. At high ion Ñux therate of arrival of ions exceeds the limit of a countingsystem and/or FPD and the measured count rate fallsbelow the incident event rate. The upper limit of thedynamic range is the ion Ñux above which the measuredÑux becomes non-linear. This point is somewhat arbi-trary as very small deviations from linearity will bepresent over most of the range of measurement. In someapplications a deviation of 5% from linearity would betolerable. This would imply that the height of a largepeak would be underestimated by 5% when the error inthe height of a small peak would be random.

The complete detector system can be divided into theMCP, the array and the data accumulation system.


804 K. BIRKINSHAW

The noise level will be determined by the noisiestcomponent and the maximum count rate will be deter-mined by the slowest component. Table 1 presents aguide to the factors limiting the higher (H) and lower(L) count rates of the various types of array. In the caseof the integrating array it is assumed to be operating atroom temperature.

Both single-event and discrete detector arrays operatein the counting mode and the main source of noise isMCP dark counts. Thermal generation of free electronsat photosites dominates the noise level for semicon-ductor integrating arrays at room temperature.

The ability of a single-event array to count events ata high rate (typically 105 s~1) is far less than the abilityof the MCP to deliver pulses. The dynamic range islimited by the time taken to compute and log the loca-tion of events by the data acquisition system. However,for a discrete detector array and for an integratingarray, the MCP gain begins to fall before the limit setby the array is reached.12

Resolving power

The resolution of a mass spectrometer is a measure ofits ability to separate ions of di†erent mass and isusually deÐned as M/*M, where *M is the massseparation of two peaks which overlap to give a 10%valley between the peaks (Fig. 18).

For our purposes, we consider the FPD to beseparate from the mass spectrometer. The spectrometerdelivers the incident ion spectrum and the FPD mea-sures it.

We have seen that in the case of a single slit, the nar-rower the slit the closer the measured peak proÐle is tothe incident beam proÐle. The wider the slit, the wider isthe measured peak and the lower is the resolving power.When the slit width is about equal to the FWHM of anincident peak there is no point in making the slit nar-rower as this simply reduces the measured ion Ñux butdoes not reduce the measured peak width.26

Figure 5 shows that the position and accuracy oflocation of an ion using a slit are physically deÐned bythe dimensions of the slit. In the case of a FPD the

Table 1. The FPD component limiting thedynamic range ; the data acquisitionsystem consists of the electronicsystem used to read the array andprocess the data

Limiting componenta

Data

Array type MCP Array acquisition

Single event H Ã

L Ã

Integrating H Ã

L Ã

Discrete H Ã b

electrode L Ã

a Excessive noise in any component will limit thelower count rate.b A longer array requires a longer reading time andeventually limits the upper count rate.12

Figure 18. Resolving power is usually defined as M /DM, whereDM is the separation between two ion peaks which overlap to givea 10% valley.

event location must be calculated (i) from the measuredcharge in the case of a single event FPD and (ii) fromthe quantized position information in the case of a dis-crete site FPD. At present the precision with which anevent location can be measured using a single slit hasnot been matched by an FPD. However, the precisionof measurement of the centroid of an ion beam can bevery high for a discrete detector array.

In order to focus on the performance of a FPD, wedeÐne resolving power as the FWHM of a measuredpeak when the incident beam is narrow (less than thedetector electrode width). An insight into the FPD mea-surement can be gained by recognizing that the mea-sured beam proÐle is an accumulation of many discreteevents. With a multi-event discrete detector FPD atrade-o† between dynamic range and resolving powercan be made, as explained in the following two sections.

Single-event analysis. If the ion Ñux is low enough then anion peak location can be computed from the observ-ation of single events. A single electron pulse activatesN detectors (i.e. generates a single count on N detectors)in the vicinity of the event where N \ 0, 1, 2, . . . . An ionpeak is measured by computing and recording thecentre of mass of each pulse group. The FWHM of thispeak should be about half a detector electrode pitch(distance between electrode centres).12

Figure 19 shows single events recorded at a low parti-cle intensity. Each event is recorded as a single count onN detectors where N \ 1, 2, 3, . . . . Events with N \ 0

Figure 19. Detection of single events using the AberystwythFPD.



are unobservable. Single-event analysis is, of course, thenormal mode of operation for a single-event FPD. Adiscrete detector FPD working in this mode canmeasure single events at many positions simultaneously.So far little work has been carried out in the singleevent analysis mode.

Event accumulation. The detectors store the counts gener-ated by each event. A peak is accumulated and informa-tion on the location of each event is lost. In this modethe recorded peak is wider than in the single-eventmode and hence the resolving power is lower. Thedynamic range of the array is greater in the accumula-tion mode and this is the normal mode of operation.

Resolving power contour diagram. We have seen that theresolving power of an FPD depends on the gain and thespreading of the MCP output pulse. For a given FPDconÐguration we can plot the resolving power as a func-tion of both MCP supply voltage and PHDL as shownin Fig. 20. This provides a clear and concise means ofrepresenting the resolving power of an FPD and alsoenables a comparison to be made with theory. A modelof the FPD has shown satisfactory agreement with theabove data.24

Type speciÐc limitations

Each device has its own advantages and its own areasof application. However, for each class of FPD there arefundamental factors which limit the performance. It isuseful to examine this list because, although a givenFPD may be under development to increase its per-formance, it is important to recognize the limits beyondwhich it cannot progress.

Single slit : a very low collection efficiency is inherent.It can currently achieve the highest resolving power butonly at the expense of very low collection efficiency.

Single-event FPD : a very low dynamic range is inher-ent. It is a simple, cheap option for low count rates butis blinded by an intense peak.

Opto-electrical FPD with a CCD : serial readout ofpixels of a two-dimensional CCD gives a long readouttime. It must be cooled for low-noise operation. The

CCD can integrate the output from events occurringsimultaneously and therefore is useful in time-of-Ñightmeasurements where bunches of ions may arrive in lessthan a nano-second. However, such events are notcounted.

Discrete detector array : this has a high collection effi-ciency, a high dynamic range, operates in the countingmode and has a resolving power which at least matchesother FPDs. Currently the main feature in need ofdevelopment is the length in the direction of the iondispersion, i.e. more detectors should be added in orderto measure a greater fraction of a spectrum simulta-neously.

THE FUTURE

At the risk of oversimpliÐcation, the trends in detectordevelopment appear to be as follows.

Counting or analogue

Too high ion peak intensity is not often a limitingproblem as there usually exists a simple means ofreducing it. Too low ion peak intensity, on the otherhand, is often a problem. The low noise levels associ-ated with counting means that ion counting devices willbe important in future.

Multiple event or single event

The future for 1D single-event devices does not appearto be rosy. They require relatively little electronics buthave a very restricted dynamic range compared with amultiple-event FPD. The electronics could be integratedfor a single-event device as for a multiple-event devicebut there seems to be little point as there would be littleimprovement of the dynamic range.

Figure 20. Contours show locations of constant resolving power. The MCP supply voltage and the PHDL are varied. All other variables areheld constant.


806 K. BIRKINSHAW

Miniaturization

The needs of mass spectrometrists are varied but someof the driving forces in mass spectrometer developmentinclude miniaturization and portability, performanceand, of course, cost. Integrated FPDs can make animportant contribution in these areas. A sector massspectrometer does not require the high-frequencysupply of a quadrupole mass spectrometer or the longÑight tube of a time-of-Ñight mass spectrometer andfuture miniature, low-resolution, portable instrumentsmay consist of small sector mass spectrometersequipped with integrated FPDs.

Integration

There is no alternative to integration if a high-resolution array of ion counters is required. This paperhas set out some of the issues concerning high-resolution integrated arrays and the feasibility of theseimportant devices has been established. The focal planeof a mass spectrometer is not as hostile an environmentas was feared and the apparent complexity of a discretedetector array on a silicon chip is an illusion. Apartfrom the MCP, power supplies and the external com-puter it is a complete measurement system and rep-resents a system simpliÐcation. Investment is needed todevelop longer devices with even better performance.

DEFINITIONS

A number of terms have been used which are collecteddeÐned here.

ArrayÈAn array is preceded by a microchannel plateelectron multiplier (MCP). The term “arrayÏ is used forthe device following the MCP. The term “position-sensitive detectorÏ (PSD) is a more appropriate descrip-tion for devices which do not consist of an array ofidentical detectors. For convenience the generic name“arrayÏ is used.

Focal plane detector (FPD)ÈA general term for adetector which measures ion Ñux at more than onepoint in the focal plane of a spectrometer in one or twodimensions. The term refers here to the combination ofthe array plus the MCP mounted above it.

Pulse height discrimination level (PHDL )ÈA voltagepulse on a detector electrode above this discriminationlevel will be counted. A pulse below this level will not becounted.

Spatial resolutionÈThe pitch (distance betweencentres) of the detector electrodes on an array.

ResolutionÈA measure of the ability of a spectro-meter as a whole to resolve dispersed particles.

Resolving powerÈThis term is used here when refer-ring to a detector in isolation from the rest of thespectrometer. Assuming a narrow incident particlebeam the resolving power is deÐned here as the FWHMof the measured peak (in units of detector electrodepitch for discrete detector arrays). The highest resolvingpower is unity.

ParticleÈA photon of wavelength \ 200 nm, elec-tron, ion, energetic neutral or other body which canactivate an MCP.

EventÈImpact of a particle on the MCP which initi-ates an output pulse.

SensitivityÈThis word is avoided as it means di†er-ent things when used by mass spectrometrists and elec-trical engineers. When used here it is used in the senseof PHDL.

REFERENCES

1. P. Horowitz and W. Hill, The Art of Electronics . CambridgeUniversity Press, Cambridge (1980).

2. Low Level Measurements , 4th ed. Keithley Instruments,Cleveland, OH (1992).

3. K. Birkinshaw, Int . Pat . WO91/00612 (1989).4. K. Birkinshaw,Analyst 117, 1099 (1992).5. K. Birkinshaw, T. M. McGinnity, D. P. Langstaff, M. W.

Lawton and D. M. Forbes, Sensors Technology, Systems andApplications . Adam Hilger, Bristol (1991).

6. D. P. Langstaff, M. W. Lawton, T. M. McGinnity, D. M. Forbesand K. Birkinshaw,Meas. Sci . Technol . 5, 389 (1994).

7. J. G. Timothy and R. L. Bybee, Appl .Opt . 14, 1632 (1975).8. J. G. Timothy and R. L. Bybee, SPIE 265, 93 (1981).9. K. Birkinshaw and D. P. Langstaff, Rapid Commun. Mass

Spectrom. 10, 1675 (1996).10. W. B. Colson, J. McPherson and F. T. King, Rev. Sci . Instrum.

44, 1694 (1973).11. Technical Information Sheet , MCP Assembly . Hamamatsu,

Hamamatsu City (1991).12. K. Birkinshaw, Int . Rev. Phys.Chem. 13, 15 (1996).13. N. Koshida and M. Hosobuchi, Rev. Sci . Instrum. 56, 1329

(1985).

14. L. J. Richter and W. Ho,Rev. Sci . Instrum. 57, 1469 (1986).15. K. Smith, Exp.Methods Phys. Sci . 29, 253 (1995).16. J. A. Hill, J. E. Biller and K. Biemann, Int . J . Mass Spectrom.

Ion Processes 111, 1 (1991).17. H. G. Boettger, C. E. Giffin and D. D. Norris, ACS Sympo. Ser .

102, 291 (1979).18. J. S. Cottrell and S. Evans, Anal . Chem. 59, 1990 (1987).19. C. Firmani, E. Ruiz, C. W. Carlson, M. Lampton and F.

Paresce, Rev. Sci . Instrum. 53, 570 (1982).20. J. L. Wiza,Nucl . Instrum.Methods 162, 587 (1979).21. G. W. Fraser, Nucl . Instrum. Methods Phys. Res. 221, 115

(1984).22. C. Loty, Acta Electron. 14, 107 (1971).23. D. C. Anacker and J. L. Erskine, Rev. Sci . Instrum. 62, 1246

(1991).24. D. J. Narayan, D. P. Langstaff and K. Birkinshaw, Int . J . Mass

Spectrom. Ion Processes 149/150, 439 (1995).25. J. G. Timothy, Rev. Sci . Instrum. 52, 1131 (1981).26. K. Birkinshaw, Trans. Inst .Meas.Control 16, 149 (1994).


Documents

SPECIAL FEATURE: TUTORIAL Fundamentals of Focal · PDF fileSPECIAL FEATURE: TUTORIAL Fundamentals of Focal Plane Detectors ... variation which approximates to a Gaussian variation