ENVIRONMENTAL SCANNING ELECTRON MICROSCOPE-SOME CRITICAL ISSUES

G. D. Danilatos

ESEM Research Laboratory98 Brighton BoulevardNorth Bondi (Sydney)NSW 2026, Australia

Phone No.: (+61 2) 302837FAX No.: (+61 2) 3650326

Abstract

In the environmental scanning electron microscope(ESEM), the gas flow around the main pressure limitingaperture establishes a density gradient through which theelectron beam passes. Electron beam losses occur in thistransition region and in the uniform gas layer above thespecimen surface. In the oligo-scattering regime, the electrondistribution consists of a widely scattered fraction of electronssurrounding an intact focussed probe. The secondaryelectrons are multiplied by means of gaseous ionization anddetected both by the ionization current and the accompanyinggaseous scintillation. The distribution of secondary electronsis governed by the applied external electric and magneticfields and by electron diffusion in the gas. The backscatteredelectrons are detected both by means of the gaseous detectiondevice and by solid scintillating detectors. Uncoated soliddetectors offer the lowest signal to noise ratio especiallyunder low beam accelerating voltages. The lowest pressure ofoperation with uncoated detectors has been expanded by thedeliberate introduction of a gaseous discharge near thedetector. The gaseous scintillation also offers the possibilityof low noise detection and signal discrimination. The"absorbed specimen current" mode is re-examined in theconditions of ESEM and it is found that the current flowingthrough the specimen is not the contrast forming mechanism:It is all the electric carriers in motion that induce signals onthe surrounding electrodes. The electric conductivity of thespecimen may affect the contrast only indirectly, i.e. as asecondary, not a primary process. The ESEM can operateunder any environment including high and low pressure, lowor rough vacuum and high vacuum; it also operates at bothhigh and low beam accelerating voltage, so that it may beconsidered as the universal instrument for virtually anyapplication previously accessible or not to the conventionalSEM.

Key Words: Environmental scanning electron microscope(ESEM), low vacuum, low voltage SEM, electron detectors,electron diffusion, secondary electrons, backscatteredelectrons, gaseous detection device, scintillating detectors,detector efficiency, noise, resolution, contrast, charging.

2

Introduction

The environmental scanning electron microscope(ESEM) is applied to diverse fields of science and has beendescribed in numerous reports. The possibility of examiningspecimens under the electron beam inside a gaseousenvironment has introduced novel ideas and a betterunderstanding of the electron beam physics. The evolution ofthis field has been slow, since early attempts date back to thebeginnings of electron microscopy. The longest part of thehistory of this development involved the transmission electronmicroscope and, during the last two decades, work hasconcentrated around the scanning electron microscope (SEM).The literature of these developments can be traced through anumber of extended surveys (Moretz, 1973; Parsons et al.,1974; Danilatos, 1982, 1988, 1991a).

The practice of operation of ESEM has turned out toinvolve some relatively simple methods but the underlyingphysical phenomena have proved, on many occasions,difficult to understand as they involve some complex physicalprocesses. Therefore, it is not surprising that some erroneousideas appear from time to time. Some misunderstandings ofthe conventional SEM have contributed to these difficultiesand there is a continual need to update our ideas as new workand results appear from the use and theory of ESEM.However, a few authors still seem to overlook or not properlyunderstand certain issues that were thought to have beenadequately dealt with previously. In this context, the presentsurvey aims at examining a selection of topics that areimportant in our understanding for the further development ofthe field:

Gas dynamics science is mandatory to apply to thesestudies. The flow of gas around the vicinity of the finalpressure limiting aperture (PLA1), in particular, can affect theperformance of the instrument.

The way contrast and resolution is affected by theintroduction of gas must be clarified. This is done byexamining the electron beam scattering and distribution in thegas and how this affects the spot size and current. The role ofdetectors in contrast and resolution is surveyed.

New detection methods, such as gaseous detection andthe notion of "specimen absorbed current" are re-examined.The use of uncoated scintillating materials is shown to be agood choice for low keV work, as they demonstrate the best

3

signal-to-noise-ratio (SNR). Some new results are integratedwith the discussion of these topics and an attempt tosummarize our present understanding is made.

The bulk of this paper deals with each of these topics insufficient detail. New results and ideas are included in thecourse of this examination. It is only towards the end of thepaper, in the Discussion section, that some of the issues areconnected with work of previous authors in a critical way.

Mass Thickness of the Gaseous Environment

A high pressure environment can be maintained in thespecimen chamber of ESEM by use of differential pumpingthrough a series of apertures. The electron beam can freelytravel along the electron optical axis but, as it approaches thefinal pressure limiting aperture (PLA1), losses of electronsdue to scattering with the gas can start becoming significant.Hence, there has been a need to determine the flow field ofgas especially around the vicinity of the PLA1, i.e. bothbelow and above it (Danilatos, 1991b).

The direct simulation Monte Carlo (DSMC) methoddeveloped by Bird (1978) for gaseous flows has been used inthis study. This method allows the determination of the flowfield around various shapes of PLA1. One particular case isshown in Fig. 1 with a conically shaped aperture and a flatspecimen placed one diameter below the aperture(D=400µm). The iso-density contours of the field are drawnonly on half of the plane containing the aperture axis, because

2.40E+23 2.10E+23 1.80E+23 1.50E+23 1.20E+23 9.00E+22 6.00E+22 3.00E+22 2.00E+22 1.40E+22 1.00E+22 8.00E+21 6.00E+21 4.00E+21 3.00E+21 2.00E+21

( / cubic meter)Number Density

Stag. n=2.47E+23

Axis

Specimen

PLA1

Fig. 1. Number density contours of argon flowing throughconical pressure limiting aperture (PLA1) with a sharp tipand with a flat specimen placed one diameter below theaperture. Stagnation number density value corresponds to1000 Pa and diameter of aperture is 0.4 mm. The contourvalues decrease monotonically in the direction of the flow.

4

the flow is axially symmetric. A depletion zone is formedbelow the aperture, and the gas jet formed above it has asignificant density up to a certain point. The effects of flowon the specimen surface can be seen by plotting the pressureat the specimen level along a direction normal to the axis as isshown in Fig. 2. At the specimen distance, the pressure hasdecreased only by about 4% from the stagnation specimenchamber pressure of p0=1000 Pa (measured with a pressuregauge at the chamber wall). This decrease takes placedirectly under the area of the aperture; the ragged variationof the curve is due to statistical fluctuation and is smoothedout as we increase the computation time or the number ofstatistical samples. The variation of density (in particles perunit volume) along the axis of the aperture is shown in Fig. 3.For this plot, the number density n has been normalized overthe stagnation value n0 corresponding to pressure p0. Fromthis type of information, we can calculate the mass thicknessof gas through which the electron beam has to pass. Thedefinite integral:

D

dz

n

znh

z∫=

0

)(ζ (1)

gives the normalized number thickness ζ of a gas layer alongthe axis from a point z up to a maximum distance h above theaperture, for which data is available and at which the pressurehas decreased to a sufficiently low value. Above this point,scattering should not be significant for a properly designedinstrument, but the number thickness can be calculated fromthe uniform background pressure p1 prevailing at thispumping stage and the distance H between the two pressurelimiting apertures. Similarly, the number thickness can befound from the constant pressure relationship for any gaslayer below one diameter from the aperture. The importantthing is to establish the parameter ζ for various lengths l=h-zin the transition region of pressure. For the example

Radial distance in PLA diameters

Pre

ssur

e x0

.001

, Pa

0.0

0.2

0.4

0.6

0.8

1.0

0.0 0.5 1.0 1.5 2.0 2.5

At specimen surface

Fig. 2. Variation of pressure at the surface of specimen inFig. 1 along radial distance from the system axis.

5

produced here, we plot the numerical evaluation of theintegral vs z/D in Fig. 4, taken in the transition region fromz=-1D up to a fixed value of z=h=3D. We note that ζ=0.48for l=h, i.e. for the mass of gas above the plane of theaperture with 400 inside cone angle; this is a little above thevalue of 0.4 found in the case of a flat aperture. The conicalshape of aperture is chosen for various reasons: It allowsbetter specimen movement and ventilation, better positioningof scintillating detectors and better efficiency in the detectionof pure secondary electrons (SE) with the gaseous detectiondevice. At the specimen surface (z=-1D), the numberthickness is ζ=1.4 and, therefore, the thickness above theaperture alone represents about 34% of the total amount. Thesignificance of these numbers in absolute terms will becomeapparent as we examine the electron beam scattering below.The values of ζ are applicable to all mechanically similarflow fields, e.g. we can vary p0 and D but keep their productp0D constant. For practical purposes, a small deviation fromthe constancy of this product can be allowed, e.g. we can keepthe aperture constant and vary the pressure from a few mbarto about 20 mbar. As we depart from this range, we can seethe effects of the transition from free molecule flow tocontinuum flow. Also, for the purposes of this paper only, weconsider the same number density characteristic to beapplicable for nitrogen, used in some examples below. Moredetailed information on different gas flow fields will bereported in a specialized paper.

Electron Beam Scattering

The electron beam scattering process in the gas isgoverned by the Poisson distribution probability function,which gives the probability for an electron to undergo xnumber of collisions:

P xm e

x

x m

( )!

=−

(2)

Axial distance in PLA diameters

Nor

mal

ized

num

ber

dens

ity

0.0

0.2

0.4

0.6

0.8

1.0

-1.0 0.0 1.0 2.0 3.0

ArgonD=0.0004 m p=1000 Pa

Fig. 3. Variation of gas density along the axis of pressurelimiting aperture for argon at 1000 Pa specimen chamberpressure.

6

This means that for each electron there is a probability withwhich it may undergo no scattering event, or a singlescattering, a double scattering, etc. If the average number ofscattering events per electron is m, then there is a fraction ofelectron beam I/I0 that is transmitted without any scattering atall, the intensity of which decreases exponentially with m:

I

Ie qm

0

= ≡− (3)

The parameter m is generally given by

∫= dzznm T )(σ (4)

where σT is the total scattering cross-section for a given gas.For the total travel distance between specimen and PLA2 (i.e.the PLA above PLA1) we distinguish three terms below,namely, the first m0 for the uniform gas layer between thespecimen and the beginning of the transition region, thesecond term mt for the entire transition region and the thirdterm m1 for the uniform layer from the end of the transitionregion to PLA2:

m m m mo t= + + 1

)()()( 10 hHndzznDLnm T

h

D

TT −++−=∴ ∫−

σσσ (5)

where L is the distance of the specimen from the PLA1, H thedistance between PLA1 and PLA2 and n1 the backgrounduniform (or average) number density between the twoapertures. For the middle term of the transition region we get

m n Dp D

kTt T T= =σ ζ σ ζ00 (6)

where k is Boltzman's constant and T the absolutetemperature. From this we find that scattering is directlyproportional to the diameter of PLA1 for a fixed specimen

Axial position in PLA diameters

Inte

grat

ed n

umbe

r th

ickn

ess

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

-1.0 0.0 1.0 2.0 3.0

ArgonD=0.0004 mp=1000 Pa

Fig. 4. Integrated number thickness from a given point onthe axis up to three diameters above the pressure limitingaperture.

E/p

ε

0

20

40

60

80

100

120

140

0 10 20 30 40 50

Nitrogen

Fig. 5. Variation of parameter ε vs E/p

7

chamber pressure. Thus, if we want to maintain the specimenas close as possible to the PLA1 with minimum gas flowinfluence (i.e. at one diameter), then Eq. (6) is the maingoverning relationship for the beam scattering.

The next most significant factor to consider is theelectron scattering cross-section, which strongly depends onthe accelerating voltage for a given gas. We still do not haveprecise values for this parameter for the commonly used gasesin the range of accelerating voltages applicable to ESEM.One attempt to calculate it for several molecular gases wasmade by Danilatos (1988) and some of these values arechosen below to help us illustrate the present work.Scattering cross-sections are still due to be determinedexperimentally once and for all. There are indications that thetheoretical values may be an overestimate of the real situationbut the present analysis will still be applicable for when wefinally obtain the correct values. For now, let us considernitrogen for three cases of beam keV as illustrated in Table 1.

We see that even with a low beam voltage of E=5 keV,we still get 25% totally unaffected beam through the nitrogengas at the typical pressure of 1000 Pa (=10 mbar). Dependingon the initial beam current, it has been shown that theun-scattered fraction of beam can be used in the normal wayfor imaging. The main remaining question is whether thescattered fraction of electrons interferes with the imagingprocess. This question has been thoroughly examined byDanilatos (1988). It has been shown that the scatteredfraction forms a broad "skirt", orders of magnitude larger thanthe useful spot. It is because of this separation of the twodistributions, namely, that of the focussed spot and that of thesurrounding skirt that ESEM has become possible. Inscanning transmission electron microscopy, we find aneffective beam spot spread, simply because the scatteredfraction closely overlaps with the un-scattered fraction duringthe very short travel distance in the thin-and-dense specimensections. This results in the well known "top-bottom" effect(sharper image at the top than at the bottom) but this must beset aside from the different scattering regime of our case. InESEM, the beam travels through a layer of gas which isorders of magnitude thicker and orders of magnitude lessdense than the solid specimens of transmission microscopy.This has prompted us to define the "oligo-scattering" regimein ESEM, which incorporates the single scattering and the

TABLE 1Electron beam transmission in nitrogen.

E σΤ×1021

I

I0

keV m2 p=100 Pa p=1000Pa

p=2000Pa

5 10.1 0.87 0.25 0.06

10 5.7 0.92 0.45 0.20

20 3.1 0.96 0.65 0.42

8

early part of the plural scattering regime. Plural scatteringincludes between 1 and 25 scattering events and multiplescattering more than 25 scattering events (Cosslett andThomas, 1964). Strictly speaking, a single scattering regimeshould be defined as that where most electrons (say 95%)undergo either one (single) or no collision at all. From Eq.(2) we find that this corresponds to m<0.35, whereas for m=1,we find that 37% of electrons undergo no collisions, 37%undergo a single collision, 18% two collisions, 6% threecollisions, etc. Clearly, it is possible to practice ESEM for mup to 2 or, perhaps, a little more; m=3 corresponds to 95% ofthe original beam removed and has been defined (arbitrarily)as the upper limit of the oligo-scattering regime.Concomitant with this definition we should add that the traveldistance in the gas should be set such as to make theseparation of the skirt from the useful spot feasible.

If the resolving power of the instrument is identified (orrelated) with the beam spot diameter, then we can state thatthis is not affected by the presence of gas. With a resolutiontest specimen producing sufficient contrast, we haverepeatedly confirmed this (Danilatos, 1990b; 1991a). In fact,the smaller the original spot the better separation we obtain.Lanthanum hexaboride electron gun sources are better thantungsten and field emission guns are even better in thisregard. However, resolution for a general specimen isintimately connected with the available contrast. Thepresence of the broad electron skirt affects the contrast of theimage by way of adding a "white" background noise level,and this negative effect is exacerbated by the weakening ofbeam intensity. The contrast can be recovered by simplyincreasing the beam current by an appropriate amount, andthe quantitative relationships for this to occur have beenpresented previously (Danilatos, 1988). An increase of beamcurrent is, of course, accompanied by a larger beam spot, andto this degree the resolution will deteriorate. However, with alarge class of applications such high resolutions are notneeded or we may be prevented from reaching a very highmagnification by other overriding considerations, such as

d

S

r

+V

E=V/d

Fig. 6. Uniform electric field E between two planeelectrodes having a potential difference V and a pointelectron source at S.

9

beam specimen damage. Finally, contrast and resolution atthis level is intimately connected with the detection systemsused, as will be outlined below.

Secondary Electron Diffusion

The presence of gas in the ESEM has necessitated thedesign of novel detection systems as well as the appropriateadaptation of conventional ones. The conventional secondaryelectron detector is not suitable for operation in the gaseousenvironment of ESEM. However, the secondary electronsignal can still be retrieved, amplified and used. There havebeen many works on this novel development and reference tothem will help in the better understanding of the presentsurvey (Danilatos, 1983; 1990a,b,c).

The motion of secondary electrons inside a gas isgoverned primarily by diffusion and by the electric andmagnetic fields applied or formed around. It is well knownthat these electrons have an energy around 2 eV and only asmall fraction of them have sufficient energy to ionize andexcite gas molecules directly. Depending on the nature ofgas, there is also a small probability for an electron to attachitself to a gas molecule to form a negative ion. Thesepossibilities and others have been surveyed in detailpreviously (Danilatos, 1990a). In the main, the electronsremain free to diffuse among the surrounding gas molecules.If an electric field is present, then they will also move in thedirection of the field at the same time.

By considering the electron/gas collision process, it canbe easily shown that the SE (with energies around 2 eV) loseonly a very small fraction of their energy with each elasticcollision (mass of electron being much smaller than mass ofgas molecule). The electrons behave like a gas and in theabsence of an external electric field, they would come inequilibrium with the host gas. However, even in the presenceof a weak field, they immediately acquire a few eV energy.

r/d

R

0.0

0.2

0.4

0.6

0.8

1.0

0.0 1.0 2.0 3.0 4.0

4003216

8(v)421

Nitrogen

pd=1 Pa-m

Fig. 7. Variation of R vs r/d for different fixed valuesof electrode bias V at constant pd=1 Pa-m.

10

This energy of electrons expressed as kinetic energy mv2/2 isfar beyond the kinetic energy of the gas, which is of the orderof 3kT/2. Because of the weak exchange of energy betweenthe electron gas and the host gas, the electrons have their owndiffusion pattern, which is different from the pattern of otherbulky ions that may form around. In the calculation ofelectron diffusion, the ratio of the electron energy over thegas energy enters as a parameter ε:

ε = mv kT2

232

(7)

The variation of ε vs E/p (E being the intensity of theelectric field) has been measured by several workers forvarious gases and their results have been compiled byDanilatos (1990a). One particular example is considered inFig. 5 for nitrogen. The combined effect of thermal diffusionand field attraction can easily be seen in the case of a uniformfield between two parallel electrodes as shown in Fig. 6.

At the bottom electrode there is a point source ofelectrons S like the SE generated at the specimen surface.For this simple case, the distribution of electrons, as theyarrive at the top electrode, has been derived by Huxley andZaazou (1949). When the potential difference is increasedbeyond a certain level for a particular gas, the electronsacquire sufficient energy to excite and ionize the gasmolecules. The electrons then multiply by an avalancheprocess but it can be shown that the distribution of all theelectrons together remains the same as in the case of lowfield. The fraction of electrons R arriving within a radius r atthe top electrode is given by

+−

+−=

− 2/1

2

22/1

2

2

112

exp11d

r

kT

eV

d

rR

ε (8)

We note that R is a function of the ratio r/d vs which wecan plot the result at fixed values of the parameter ratio V/ε.To better illustrate the situation, we can choose different fixedbias V in Volts for the typical case where pd=1 Pa-m (e.g. atp=1000 Pa and d=1 mm). The results are shown in Fig. 7.Unfortunately, we still do not have data on ε for high valuesof bias and the dotted line has been calculated from anarbitrary extrapolation for ε according to ε=36+1.73(E/p).For very low bias, below 1 Volt, all curves coincide with theone for V=1 Volt, because the variation of ε can beapproximated by a straight line passing through the origin ofthe axes in Fig. 5. For low values of bias, the diffusion playsa major part in the distribution of electrons. As we increasethe bias, their distribution becomes significantly narrower.According to Wilkinson (1950), the electrons tend to followthe lines of force for high values of E/p.

The distribution of ions generated at high fields issimply that of SE, because the ions do not diffuse muchfurther: They are practically as bulky as the neutralmolecules and quickly come to thermal equilibrium after eachcollision, i.e. they lose their energy acquired from the externalfield between successive collisions (there is a strong couplingand ε≈1). The ions do not create additional ionization, astheir energy is converted to thermal movement.

11

The equations of electron and ion distribution in theESEM and the corresponding induced signals have beenderived previously. The backscattered electrons have theirown corresponding distribution of charge and their owninduced signal. It has been shown that the two signals can beseparated by use of appropriate electrode configurations(Danilatos, 1990a,b,c).

Noise Propagation in Detection Systems

The theory, practice and literature of signal-to-noise-ratio for various detection systems in SEM have beensurveyed elsewhere (Jones, 1959; Wells, 1974; Reimer,1985). Following the same principles, we can outline thesituation for two basic detection systems used in ESEM,namely, the gaseous detection device and solid scintillatingdetectors.Gaseous Detection Device.

The gaseous detection device (GDD) is based on thedetection of products of interaction between various signalsand gas. Initially, the ionization of gas was used to produceimages and later it was demonstrated that the gaseousscintillation could also be used for the same purpose. Weconsider both of these approaches.Ionization. For a given beam current I0 and pixel dwell timeτ, we get n0 electrons striking the specimen at each pixelelement when the specimen chamber is in vacuum. For agiven signal-to-noise-ratio K and M gray levels on therecorded micrograph, we find the following relationship(Wells, 1974):

222

0 4

++=

A

BBAeMKI

δδδδ

τ (9)

where e is the electron charge, δΑ is the fraction of theelectron beam that is converted to useful signal (or feature)after the beam-specimen interaction and δB is the fraction thatis converted to background noise. When we introduce gas,the above equation is modified as follows (Danilatos, 1988):

2

0022

0

)1(

4

++−+++=

A

BABA

q

qeMKI

δδδδδδδ

τ (10)

where δ0 represents the ionization electrons in the gasgenerated by the beam before the beam strikes the specimen;this constitutes one of three terms adding to the backgroundnoise level. The second term is simply qδB from the usefulspot. The third term is generated from the skirt fraction 1-qof electrons striking the specimen (the primary electronsback-scattered from the gas are neglected because theyconstitute an extremely small fraction under theoligo-scattering condition). For the skirt electrons striking thespecimen, the conversion coefficient δS depends on theprecise specimen nature, the magnification used and theextent of the skirt. A conservative value of δS=δΑ+δB has beentaken for the derivation of Eq. (10), but this can be adjustedaccordingly; for a general purpose analysis of the detectionsystem, the above assumption can be satisfactory. For a

12

graphical presentation below, the last two terms relate to thespecimen nature and, lumped together, are designated asδBB≡qδB+(1-q)δS=(1-q)δΑ+δB.

We can now depict the signal propagation in thedetection system as in Fig. 8. Let us consider a case with 20keV beam, PLA-specimen distance d=0.0004 m, and nitrogengas with 2000 Pa pressure. In vacuum, the incident beamdelivers eIn /00 τ= particles at each pixel element, which

can be normalized to unity (stage No. 0). In the presence ofgas, we get to stage No. 1 just before the beam strikes thespecimen; for this we plot first the number in the useful spotq=0.42, on which we add the skirt 1-q, on which we finallyadd the primary beam effect δ0. We find that δ0=S0pd, whereS0, the ionization efficiency of the beam, depends on theaccelerating voltage; for the present case we take S0=0.13ions/Pa-m and find δ0≈0.1. In the following stage No. 2(beam-specimen interaction), we first plot the useful signalqδΑ, on which we add δBB for the point above, on which wefinally add δ0 to obtain the top point. We have taken, as anexample, δΑ=δB=0.1. Following this, we consider the gaseousamplification to arrive at stage No. 3. The gaseous gain canbe calculated for each component of electrons at stage 2. Wecan readily get an amplification factor eαd-1 for all the SEoriginating at the specimen surface and an amplificationfactor (eαd-1)/αd for δ0 (Danilatos, 1990b), where α, the firstTownsend coefficient, is given by

−= pdV

BAp expα (11)

with A and B constants tabulated for each gas. For nitrogen,with electrode bias V=400 Volts, we get for the firstTownsend coefficient α=9479 ions/m. Thus the gain factorfor δ0 is 11.9, whereas the "spot+skirt" signal has a gain factor43.7. Therefore, the beam effect tends to be suppressed

Stage No.

Nor

mal

ized

# o

f ele

ctro

ns

0.01

0.10

1.00

10.00

100.00

0 1 2 3 4

spot

spot +skirtspot+skirt+beam

Fig. 8. Gaseous detection device (ionization). Relativevariation of number of electrons at various stages for theuseful "spot", the "skirt" (interacting with specimen) and"beam" (interacting with gas before it strikes the specimen).The noise bottleneck is at the beam-specimen interaction.

13

relative to all SE electrons originating from the specimensurface, which receive a preferential amplification.

From the scheme in Fig. 8 we see that the noisebottleneck is at the beam-specimen interaction. The mainconsideration after this is the noise introduced by theoperational amplifier at stage 4. If the equivalent-input noiseof this amplifier is greater than the noise at stage 3, then thesystem is limited by the amplifier. Therefore, every effortshould be made to choose an amplifier with the best possiblecharacteristics and also to try and obtain the maximumpossible gaseous gain with the GDD. The gaseousamplification is associated with very low noise and is to bepreferred over the subsequent operational amplifier's gain.Future development should concentrate on extracting a highergain from the GDD.

To simplify the above analysis, the ionization caused bythe backscattered electrons has not been mentioned. The BSEfrom the specimen also create a primary ionization in the bulkof the gas, which is amplified by the external field. However,the latter can be separated out by proper electrodeconfiguration, and the remaining component adds only asmall fraction of additional signal which has been omittedfrom the present scheme (see details in Danilatos, 1990a,b,c).Scintillation. It has been shown that we can use the gaseousscintillation from the signal-gas interactions to form images(Danilatos, 1986). When we apply a strong electric field, thecascade electrons in the gas, apart from ionization, alsoliberate photons which multiply in an analogous way. Thishas been used to produce secondary electron images bysimply collecting the gaseous scintillation with a suitablelight-pipe/photomultiplier (PMT) system (Danilatos, 1992).The gaseous scintillation thus produced is usually muchstronger than the specimen cathodoluminescence (CL) whichdoes not present a problem. However, in some cases, a strong

Stage No.

Nor

mal

ized

# o

f qua

nta

0.01

0.10

1.00

10.00

100.00

0 1 2 3 4 5 6 7

Fig. 9. Scintillation gaseous detection device. Propagationof the total signal through various stages. Two noisebottlenecks may compete with each other. Dotted line isone improvement starting with same gaseous gain as withthe ionization GDD.

14

CL may be superimposed, especially when a simple PMT isused. In future work, spectroscopic methods and other meanscan be used to separate out the various sources of signal. Theuse of known CL methods in the gaseous environment ofESEM can lead to alternative microanalytical techniques.

For the present, the use of a PMT is a best replacementfor the operational amplifier used with the ionization GDD,because a PMT adds practically no background noise (typicalanode dark current is a few nA at a gain of 106). The signalpropagation in the scintillation GDD is shown in Fig. 9.Here, we plot the total signal, namely, that of"spot+skirt+beam". For illustration purposes, we may assumethat for each cascade electron we also get one photon (theactual situation varies with the applied bias and the nature ofgas). Thus, the stages up to No. 3 have been drawn identicalwith those of Fig. 8. Stage 4 is the collection of photons fromthe gas by an appropriate system. We may assume that wecan collect at least one quarter of the total. Of these, 40%may be transmitted through the light pipe (stage 5) and 15%are converted to photoelectrons at stage 6 (Wells, 1974). ThePMT can then produce a gain up to 106 at stage 7. The dottedline shows the case where we manage to collect twice asmany photons in the gas and improve the photoelectronconversion also by a factor of two. It is evident that a secondnoise bottleneck can appear and every effort should be madeto improve the photon collection and transmission efficiencyof this detector. After this, the main play is with the gaseousgain on which this system mainly relies. Once we shift thesecond "dip" clearly above the level of stage 2, we obtain oneof the simplest and most powerful detection systems forESEM.Solid Scintillating BSE Detectors.

The signal propagation through various stages for asolid scintillating detection BSE system is depicted in Fig. 10.The situation is qualitatively the same up to stage 2 as in theprevious case. The total signal, here, is that of "spot+skirt".The electrons from the initial beam-gas interaction (δ0 factor)are omitted, because they are of very low energy and do notexcite the (unbiased) detector. In this figure, we analyze andcompare a typical SEM case with an improved situationaimed at in ESEM. Usually, not all of the BSE from stage 2are collected by the detector, and it is not uncommon tocollect about half of the total. The loss is represented with alower value at the "collection" stage No. 3. It is alsocustomary in SEM to coat the detector with an aluminumcoating. This layer can absorb a significant amount of signalenergy, especially when we wish to operate the microscope atlow accelerating voltage, say below 5 keV. This situation caneasily account for a loss of more than half of the total BSEsignal energy. The aluminum coating has a beneficial effectby eliminating the low energy BSE, when we use higheraccelerating voltages, in the high magnification range; then,the high energy fraction of BSE is practically transmittedthrough, minus a small percentage of it that is itselfbackscattered from the coating out of the detector. The latterreason alone is sufficient to make us plot the number oftransmitted BSE with a slightly decreased value at the nextstage 4 (e.g. 10% less). The main loss of signal as a result ofthe coating will appear in the next stage (5) in the form of a

15

lesser number of photon quanta that would otherwise beproduced. Stage 5 (scintillation) represents a net gain by theconversion of the remaining electron energy to photons. If2% of the electron energy is converted to photons and if eachphoton takes 3.1 eV (Wells, 1974), then we expect anamplification of about 10 in this process, for an electronenergy around 1.5 keV (presumed to pass through thealuminum coating by use of a 5 keV incident beam).Following this, we expect to have some serious losses again:In the light pipe we may be left with about 40% of the initiallight (stage 6), which is considered a good transmission rate.About 15% of the photon quanta produce photoelectrons atthe PMT photocathode (stage 7). From this point we expect ahuge gain by the PMT of the order of 106 (stage 8).

In ESEM, it is possible to have a much improvedsituation simply because we can dispose of the aluminum orother conductive coatings on the detector and because we canimprove the BSE collection angle to its maximum possible.The mass thickness of a 70 nm aluminum coating is 189µg/m2, whereas that of a nitrogen layer 1 mm thick at 100 Pais only 1.15 µg/m2. This gas layer, or even much less, issufficient to neutralize the accumulated negative charge onthe detector, as will be shown in the next section. This allowsus to use a significantly lower keV beam. Because chargingis not a problem, we can bring the tip of the detector close tothe edge of the PLA1. The dotted line in Fig. 10 shows stage3 with 90% of the signal collected. Stage 4 is omitted.Improved scintillation efficiency (about double) can begained by use of YAG/YAP crystals (Autrata et al., 1983) andsuch an increase is incorporated in stage 5. There is alsoscope to optimize the shape of the detector to increase thelight transmission to the maximum possible (Danilatos,1985); in this example we use 40% transmission rate again.Last, the coupling between detector and PMT can also beimproved to increase the signal conversion and a factor twice

Stage No.

Nor

mal

ized

# o

f qua

nta

0.01

0.10

1.00

10.00

100.00

0 1 2 3 4 5 6 7 8

Improved (ESEM)

Typical (SEM)

Fig. 10. Solid scintillating detectors in conventional SEMand a possible improvement in ESEM. Signal propagationat various stages. Two bottlenecks can appear at low keVoperation.

16

as high has been used in this plot. This and other measures toimprove the efficiency of the detector will be reported inmore detail separately.

The analysis of signal propagation clearly indicates thatthere is good scope in making every effort to improve thecollection and propagation of signal in the detector itself.This allows us to operate the ESEM at low beam keV.

Charge Neutralization, Low Vacuum SEM, Low VoltageSEM and Universal ESEM

Occasionally, concern is expressed about the capabilityof the ESEM to operate at low vacuum or at usual vacuumSEM condition. It is thought that the GDD ceases to operateat low vacuum and an uncoated scintillating detector wouldcharge up. Until recently, every effort was consumed withmaking the ESEM to operate under as high a pressureenvironment as possible. However, it can be shown both inpractice and in principle that the ESEM can also operateunder vacuum or low vacuum condition.

The operation of GDD does not depend on the gaspressure alone but rather on the product of(pressure)x(distance) or pd. Therefore, when the pressuredecreases, we can increase the specimen/electrode distance ininverse proportion to maintain the maximum detector signal.Depending on the design of GDD, the distance ofspecimen/electrode may be separate from the distance ofspecimen/PLA1. Of course, other effects, such as electrondiffusion, should be taken into account and properlycounteracted in various designs. If the GDD electrode issmall, an increasing fraction of electrons can be lost as weincrease the specimen distance, according to Eq. (8) Also, asharp tip PLA, used as an electrode, produces a non-uniformelectric field and can result in losses by diffusion. Thisproblem can be remedied with the use of a multi-electrodeGDD (Danilatos, 1990a,b,c). A second concentric electrodearound the tip of the PLA can be used for the separation anddetection of BSE electrons at sufficiently short specimendistance from PLA1. When the distance is increased, thissecond electrode becomes a SE detector also. A specialconfiguration of a multi-electrode scheme is shown in Fig.11; the first electrode is the PLA1 and the second is madefrom a metal grid or grids attached (or deposited) onto ascintillating detector. The same arrangement is repeatedaround the PLA2. This electrode configuration caters for anyspecimen distance. Differing types and fractions of signalsbetween BSE and SE are detected as we vary the specimendistance from a relatively long to a relatively short distancefrom the PLA1. Each electrode is biased independently withV1, V2, V3 and V4 Volts. At the cost of some electronicscomplexity, we can gain valuable flexibility by splitting eachgrid in two separate electrodes biased with V2' and V4' Volts.

17

The gaseous gain of GDD has been analyzed and foundto exhibit a maximum for some value of pd, which is usuallyaround 1 Pa-m. This maximum depends on the precise gascomposition and electrode configuration. Depending on thegas used, for a specimen distance of 10 mm, the pressure canbe lowered down to around 50 Pa without changing theelectrode bias. Theoretically, we can further increase thedistance and decrease the pressure, but this is generallyundesirable because the electron beam aberrations alsoincrease. Alternatively, we can fix the specimen distance andincrease the bias to achieve sufficient gain. This may not beat the characteristic maximum of gain curve observed as wevary the pd. Eventually, this parameter (bias) is alsoexhausted as the number of ionizing collisions becomespractically very low or zero. At this point, we can turn toother imaging modes, namely, to scintillating detectors and tobiasing the attached electrodes in the keV range so as toaccelerate the slow electrons as in the conventional SEM. Avery high bias on the grids will also act as electrostatic lensfor the incident beam and, hence, we may require additionalscreening grids not shown in Fig. 11.

As we decrease the pressure, we need to inquire atwhich point we lose all the benefits of ESEM and fullyconvert to conventional SEM conditions. As we decrease thepressure, the first benefit that is given up is the wet specimencondition (below 609 Pa). However, there is still greatinterest to observe other gaseous reactions at lower pressuresor simply to use the gas as charge suppression agent. Let us,therefore, inquire about the limits of charge neutralization inESEM. With a conductive specimen, the final limiting factorwould be the presence of the uncoated scintillating detectors,the benefits of which were outlined in the previous section.Scintillating detectors integrated with the GDD and PLAs areshown in Fig. 11. The shapes are those calculated byDanilatos (1985) and at least one pair of such detectors are

Specimen

PLA1

PLA2

V1

V2

V3

V4

YAG/YAP/Quartz

YAG/YAP/Quartz

V2'

V4' Grid

Grid

Pump

Beam

Fig. 11. Universal ESEM. Integrated GDD with solidscintillators and PLAs to operate from vacuum to highpressure environment, at low and high beam voltage, with avariety of detection modes.

18

integrated with PLA1 and another pair with PLA2. Ideally,four detectors at each plane would greatly increase signalcollection and manipulation, and system flexibility. Theelectrode grids are sufficiently thin to allow the maximumpossible free area of detector exposed to the incidence ofBSE. At very short working distance, a large fraction of BSEescapes through the PLA1 and is detected by the system ofdetectors at PLA2.

The negative charging artifacts in vacuum SEM usuallyarise from the excess negative charge retained by insulatingobjects. In ESEM, this charge accumulation is effectivelysuppressed by the ionized gas and in particular by the positiveions in the gas. We may distinguish between the negativecharges created by all the fast electrons that "stick" on theinsulating surfaces and the "mobile" slow electrons that caneasily be repelled and diffuse away. The fast electrons (beamand BSE) forcefully accumulate on the neighboring surfaces,which would resist any further contribution from the slowelectrons. The slow electrons are further assisted to disperseaway by their high mobility in the hot "electron gas".Conversely, the positive ions are much less mobile, theyconstitute a gas in near equilibrium with the host gas and areattracted by the negatively charged areas. Hence, if there is asufficient number of ions around, they will effectively"neutralize" those areas. With some exceptional specimens,the ions generally would just balance the negative charge,because any additional accumulation of positive chargeswould tend to repel new ones from approaching. In the end,we can have effective charge suppression and any positive ornegative charges in excess for this suppression move away tofinally dissipate on the nearest conductive surface. Theexception to this general process occurs when we deal withvery extended insulating specimens within a very restrictedregion and in the presence of confining electric and magneticfields; then, positive charge accumulation occurs on thespecimen. Here, we shall deal only with the usual and

pd, P-am

Nor

mal

ized

ioni

zatio

n cu

rren

t

0.0

0.2

0.4

0.6

0.8

1.0

0.0 0.5 1.0 1.5 2.0

100 V=0200300

E= 5 keV

Fig. 12. Variation of ionization current normalized overbeam current vs pd at different electrode bias.

19

general case where negative charging occurs by the "sticking"fast electrons.

The general movement just described above takes placein the absence of any biased GDD electrodes. When weintroduce such electrodes, as in Fig. 11, they act as "sinks" forvarious charges. If we bias the electrodes positively, themobile slow electrons need not travel far, as they are directedtowards these electrodes by the field. The ions would stilltravel towards the negatively charging areas to neutralizethem and any excess will diffuse away to the nearest groundsurface. What we need now is to find the relationship ofvarious parameters for the generation of ionization current inthe gas. This current (Ii) is caused by the incident beam, theBSE and SE as in the three terms of the following equation(Danilatos, 1990b):

( )120

0 −

++= dBSEi e

pSpSII αδ

αη

α (12)

where η is the BSE coefficient, δ the SE coefficient and SBSE

the ionization efficiency of the BSE. We can normalize theionization current by dividing by the beam current and,considering Eq. (11), we obtain

−

−

+

−

+= 1expexpexp

20

0

pdV

BApd

pdV

BA

SS

I

I BSEi δη

(13)

The above equation incorporates the additional ionizationcaused by the external field through bias V, in the absence ofwhich we have only the primary ionization from the beamand BSE. The latter case can be directly derived, or reducedfrom Eq. (13) to the simple equation

pd, P-am

Nor

mal

ized

ioni

zatio

n cu

rren

t

0.0

0.2

0.4

0.6

0.8

1.0

0.0 1.0 2.0 3.0 4.0 5.0

100

V=0

200

300

E= 30 keV

Fig. 13. Variation of ionization current normalized overbeam current vs pd, at different electrode bias.

20

pdSSI

IBSE

i )2( 00

η+= (14)

We note that the ionization current is a function of pdagainst which we plot the result for two cases of acceleratingbeam voltage in Figs. 12 and 13. For comparison purposeswe set δ=η=0.1 and, for nitrogen, we use A=9 1/Pa-m andB=256.5 V/Pa-m. For the case of 5 keV beam, we useS0=0.43 and SBSE=0.64, and for 30 keV we use S0=0.09 andSBSE=0.15 (see Danilatos, 1990b). The electrode bias is fixedat 100, 200 and 300 Volts when we apply Eq. (13), and at 0Volts when we apply Eq. (14). It should be stressed thatthese values are realistic but some of these parameters may bedifficult to verify by experiment. The ionization parametersare very sensitive to the nature of gas and gas compositionwhich are not fixed in the ESEM. The parameters A and Bhave limited validity only for a specified pd range and onlyfor pure nitrogen. Any quantitative comparison betweentheory and experiment also requires very accuratemeasurement of pressure and distance, for which we needwell calibrated equipment. Also, a small component of theionization current arises from the γ-processes, i.e from theions liberating additional electrons from the cathode (orspecimen). When the latter component is significant, weoperate near the breakdown point and instabilities occur, sothat it is better not to seek gain from these processes (at leastuntil we learn more about them); these greatly depend on thenature of the cathode and they constitute a special topic forfurther research (see discussion in Danilatos, 1990a). Thepurpose of using and evaluating the above equations is not toobtain precise numerical answers. The great value of thesederivations is the fact that they explain well the interplay ofthe many parameters that are involved in our system and, withthis understanding, we discuss the results obtained. In Fig 12,we find that at pd=1.8 Pa-m enough ion current is producedto neutralize the total beam current in the absence of any bias(V=0 Volts). In fact, we need less ion current if we assumethat the SE diffuse away and even less ion current toneutralize the negative charge of the BSE on the detectoralone. First, let us consider no bias on the electrode (V=0):For η=0.1 we find that we need only pd=0.2 Pa-m (see Fig.12), which is consistent with results on image distortion dueto charging published previously (Danilatos, 1988). For the30 keV case (Fig. 13), we find that pd=8.1 Pa-m is needed toneutralize the total beam current and only pd=0.8 Pa-m forthe detector alone.

The most important finding from the graphs in Figs. 12and 13 is that charge neutralization is achieved much easierwhen we apply bias to the GDD electrode. From Fig. 12 wefind that pd=0.06 Pa-m only is needed to neutralize thedetector when we apply 100 Volts to the electrode; onlymarginal improvement is found with higher bias. Thesituation is similar in Fig. 13, where we get pd=0.09 Pa-mwith 100 Volts and a slightly less value for higher bias. Theprecise values will vary with actual specimen nature, the gascomposition and the electrode geometry. The importantfinding here is that we can benefit greatly by simplyaccompanying our uncoated BSE detectors with a biasedelectrode in order to generate and supply additional ions, over

21

and above those produced by the primary ionization of thebeam and BSE alone.

Initially, one simple approach to achieve the abovebenefit is to employ the BSE detectors together with thePLA1 (or PLA2) electrode alone. When we bias thiselectrode positively, most of the mobile slow electrons will bedissipated on it. Thus, we will be left mainly with the BSEand beam electrons that "stick" on the nearby surfaces, whichcan be neutralized by the generated ions. When the electrodegrids on the detectors are also present, they may be groundedor slightly biased negatively to attract positive ions in theirdirection. It may be preferable to bias the PLA1 electrodewith a minimum positive voltage required only to suppresscharging and thus keep the gaseous scintillation to aminimum. Usually, the solid detector is producing muchmore intense light, and the gas is not expected to interfere.Alternatively, the gaseous scintillation can be controlled bythe gas composition used. With the use of various electrodesand biases, the shaping of the electric field can vary toachieve a desired result. There are many parameters that wecan control, which is an advantage, as each application'sneeds can be catered for accordingly. It is beyond thepurposes of this paper to exhaust all the possibilities nowopen.

The main conclusions and observations above havebeen confirmed by experiment. For example, sharp tipelectrodes have been successfully used and they can coexistnext to and in contact with plastic materials. Detection abovethe PLA1 has also been reported (Danilatos, 1985, 1990d). Inthis case, the PLA1 electrode can act as a control "grid" tomanipulate the fractions of signal passing through theaperture and also as a "sink" for the positive ions formingabove the aperture. In the ElectroScan ESEM, wheredifferential pumping is incorporated within the objective lens,we expect to achieve additional gaseous gain as the chargestend to move in helical paths (due to magnetic field) and theireffective path is lengthened in the low pressure region abovethe PLA1.

Until recently, we have placed emphasis on making theESEM operate at as high pressure as possible. Currently, wehave extended our research work to cover the region of lowvacuum and vacuum regime in order to cater for thosespecialized applications that still need those conditions. Thebehaviour of an uncoated BSE detector with a suitable gridelectrode has not yet been fully tested in conventional SEMvacuum. It is envisaged that by the right choice of griddimensions and attachment, this alone would preventdetrimental charging in vacuum. This possibility will free theions from being used on the scintillator to being used oninsulating specimens and thus extend the low vacuum limitfor such specimen applications. Alternatively, the use of acompromise thickness of aluminum or other type ofconductive coatings can still be considered to help us create adetection system that would cover the complete pressurerange. In vacuum, we can apply bias in the keV range toaccelerate the SE as in the E-T detector. This can be betterachieved with the grid at PLA2 (with possibly an additionalscreening grid in front of it, not shown). The passage throughthe hole of a pole piece and the detection of the SE above the

22

hole in SEM has been reported in numerous papers by variousworkers. Clearly, this possibility is also open to ESEM and,by the appropriate specimen positioning, PLA1 size and bias,we can achieve a very good separation of the BSE from theSE signal. In conclusion, the composite detectionconfiguration of Fig. 11 can operate in vacuum, low vacuumand high pressure in the specimen chamber and thus make theESEM a versatile, highly efficient, universal instrument.

In Fig. 11, it is shown that instead of scintillatingmaterial we can use quartz to detect cathodoluminescence orgaseous scintillation. Sapphire, like quartz, also transmits inthe ultra-violet region, but any other material with a desiredlight transmission or other physical properties may substitutethe scintillating detectors. Therefore, the possibility of usinguncoated materials has freed the microscope to be fitted withalternative and new detection designs not previously possible.

Having explained the operation of ESEM in the lowvacuum regime, we can now see that it is also possible to usea low voltage beam as well, i.e. to work towards 1 keV. Lowvoltage SEM has become increasingly popular in recentpublications. Low voltage ESEM clearly offers all theadvantages without any of the limitations of the SEM. Asmall amount of the appropriate gas present will eliminate theremaining charging artifacts of difficult specimens as thosewith re-entrant surfaces.

Standardization of BSE Detector Efficiencies

The new possibility of operating an uncoated andunbiased BSE detector at low keV beam creates the need forobjective measurement of the quality or efficiency of suchdetectors. Such a measurement should be done in a way thatdifferent detectors can be objectively compared. Thoseworkers involved with the development of scintillating BSEdetectors are usually faced with the task of evaluating thecapability of a given detector and of comparing it with others.It is not uncommon to see graphs of the signal "strength" for aparticular detector but this really tells us little about how thiscompares with work in other laboratories. Differentphotomultipliers even of the same kind have differentcharacteristics and ultimately the only meaningful quantityfor comparison purposes is the amount of noise that aparticular detection system generates for a given signal. Eachdetection system has a characteristic which is determinedbelow.

Baumann and Reimer (1981) have analyzed andmeasured the quality of different detectors (see also Reimer,1985; Oatley 1985; McClure, 1990). Following a similarprocedure, we get that the relative variance υ(nr) at stage r isa function of the variances at the previous stages as follows:

2211

1

11

2

1

11

)()()()()(

−

−++++=r

rr nnn

nnδδδ

δυδδυδυυυ

LL (15)

where the conversion factors δr for each stage relate with thenumber of quanta nr between successive stages as

11 −−= rrr nn δ (16)

23

The test of the detector efficiency should be done at aslow pressure as possible and only enough gas should be usedto neutralize the detector. Thus stage 0 and stage 1 arelumped into stage No. 1. The shot noise in the beam followsa Poisson distribution and the conversion of the beam to BSEfollows a binomial distribution. These first two stagescombined together follow a Poisson distribution (Reimer,1985). Taking into account the laws of these statistics andsetting δ1=η, we find for the first two terms

ηδυυ

11

11

1)()(

nnn =+ (17)

Because the relative variance relates to the SNR as

22 )(

11)(

OUTrr SNRK

n ≡=υ (18)

we find from Eq. (15) for the total system

ηδδδυδυ

122

122

1)()(1

1

nK r

r

r

++=

−

−

LL (19)

One of the conversion factors, namely, at the scintillationstage (i.e. δ3 for an uncoated and δ4 for a coated detector) is afunction of the beam voltage, so that the above expression canbe written as

)(/0

2

1

2

EeI

K

n

K rr ∆==τηη

(20)

where the function ∆(E) is the inverted bracketed factor ofEq. (19). This function can be used as the characteristic ofthe BSE detector. The denominator of Eq. (21) is simply thetotal number of BSE coming out of the specimen, which, onaccount of its Poisson statistics, can be written as

eI

KE

SNR

SNR r

BSE

OUT

/)(

)(

)(

0τη=∆= (21)

The function ∆(E) is known as the detective quantumefficiency (DQE) with the (SNR)BSE taken as the SNR at theinput of the detector. In the latter case, the DQE incorporatesthe collection stage (efficiency) of the detector, which ispertinent for our case. The theory and practice on DQE canbe traced through elsewhere (Jones, 1959). For us, thepractical steps to take in the evaluation of Eq. (21) are asfollows. We need to experimentally measure (SNR)OUT andcalculate the (SNR)BSE from a beam current measurement. Forthis, we need to use a standard (or reference) material ofknown η. Carbon can be chosen for the role of a referencematerial, for one reason because it could help in thereproducibility of measurements on account of possiblecontamination during measurements. A contamination layer,when present, is generally carbonaceous in nature and couldalter the backscattering coefficient of any other referencematerial. A thin contamination layer could then introduce asignificant experimental error, especially at the very lowbeam voltages where we wish to detect differences in detectorperformance. For each value of beam voltage we measureKOUT, I0 and τ. For this, it is helpful to use as a feature the

24

hole on a polished carbon surface over which we scan thebeam at normal incidence. A smooth carbon surface can bemade by depositing a carbon layer of appropriate thickness ona polished beryllium (Z=5) surface (with a deep hole on it).With this, we can measure both the beam current (Faradaycage) and the output signal (corresponding to the carbonsurface) from the detector with appropriate means. For thisfeature, we define only two gray levels (M=2), one for thehole and one for the surface. It is better to use lowmagnification and measure the signal away from the edge ofthe hole so as to avoid edge effects. We should use severalvalues of beam current of sufficiently low level to make thenoise visible and measurable, and find an average value of∆(E) for the fixed E. The noise level can be measured eitherfrom a micrograph or from the electronic signal output of ourdetector. This measurement presents the main difficulty and,perhaps, the main reason for which DQE measurements are soscarce in the literature. If the noise is measured from amicrograph, this could be done by optical means to re-tracethe noise and measure the true r.m.s value of it, making surethat frequencies corresponding to spatial detail smaller thanthe pixel on the micrograph is rejected (i.e. variation smallerthan the average resolution of the bare human eye); the timeconstant is that corresponding to the pixel on the micrograph.If the noise is measured by an electronic meter directly fromthe output of the detection system, then care should be takento establish the time constant of the system (i.e. of the meteror the detector electronics, of both in combination). In thelatter case, time constant τ is the integrating time over whichthe signal is being built. This time constant is the inverse of afrequency bandwidth ∆f, to which some authors also refer. Arelationship between these parameters is (Reimer, 1985):

f∆=

2

1τ (22)

The same bandwidth or time constant must be used for thecalculation of (SNR)BSE in Eq. (21). Perhaps, a practical wayfor this measurement might be to use a filter with knownbandwidth and measure the r.m.s. with a meter of a muchwider bandwidth.

One small complication is, however, the fact that ηvaries with beam voltage in the low keV range in which weare interested. All materials seem to have this variabilityexcept for copper, which has η=0.31 (see book by Reimer,1985). We note that for the low atomic number materialsthere is an increase of the coefficient as we decrease the beamvoltage, and we have to decide either to use the actual valueof the coefficient for each beam voltage or to use an averagevalue by convention. In the latter case, we could agree thatthe measured quantity ∆(E) is a "figure of merit", not theDQE.

The measurement of SNR is generally a difficult subjecton account of other complications. For example, a "low loss"BSE detector is not covered appropriately by the above"standard" scheme, as the parameter of specimen tilt has notbeen considered. Also, we have assumed that our PMTs areof good quality and operate in a bias range where its variancechanges little with bias (usually above 500 Volts). Generally,we assume that all other parameters are optimum and only the

25

beam energy varies. Even with these restrictions, our presentscheme covers a broad class of scintillating BSE detectors asthese are likely to be used in ESEM and SEM. The oneadditional parameter that must be monitored, nevertheless, isthe specimen distance from the PLA1 (or from the detector,in general). The efficiency of detectors in Fig. 11 shows amaximum at some optimum distance. For very short distancefrom the PLA1, the detectors at PLA1 have decreasedcollection efficiency as most of the BSE escape through thehole. For a very long distance, the collection efficiency isalso low because of the small subtended solid angle.Therefore, there is an optimum maximum at some pointbetween those two positions (usually around 1 or 2 mm fromPLA1).

On account of the complexity of the SNR theory andDQE measurements, the above suggestions are an attempt todiscuss a practical procedure for introduction in thedetermination of the efficiency characteristic of variousdetectors.

The value of ∆(E) starts from zero and approaches unityat some high value of beam keV for a good scintillatingBSE/PMT detector. It significantly departs from unity below5 keV and large deviations are observed around 1 keV, wherevarious detectors are expected to compete. For the E-Tdetector, the fixed 12 keV bias is thought to maintain thesecond "dip" in the SE signal propagation chain well abovethe noise bottleneck at the specimen (this depends on theefficiency of the particular design of detector at hand).ESEM has now ushered the possibility of using solidscintillating BSE detectors with a low voltage beam withoutthe need to accelerate the BSE in the keV range, as is usuallydone with low voltage SEM in vacuum. In SEM, analternative approach has been the use of the converted BSEsignal to SE at the pole piece (CBSE), a detection method thathas produced very good results at low keV operation(Baumann and Reimer, 1980). It should be pointed out that avariation of the CBSE mode is also possible in ESEM and,indeed, with certain advantages: This variation consists inusing the GDD with reverse bias so that the SE from theconverter plate are amplified in the gas. It remains to be seenhow a well designed uncoated BSE detector in ESEMcompares with alternative detection systems in the low beamvoltage mode.

Discussion

Some critical aspects of the principles and operation ofESEM have been surveyed in this paper. It has been shownthat significant pressure levels can be tolerated in thespecimen chamber of ESEM for operation with a beamaccelerating voltage of 5 keV or less. The minimumsaturation water vapor pressure at 0° C is 609 Pa, a theoreticalscattering cross-section at 5 keV is 7x10-21 m2, and for atravel distance of 0.5 mm after a 0.4 mm PLA diameter, wefind about 50% un-scattered transmission beam rate. Inpractice, the cross-section is found to be smaller and the spotshould be even better than that. Therefore, fully wetspecimens at 5 keV, or lower, can be examined (Danilatos,1988).

26

With regard to electron beam distribution and scatteringin SEM, an attempt to investigate the situation was made byMoncrieff et al. (1979). From single scattering theory, theyconcluded that the effect of the gas was to deflect a certainproportion of electrons out of the original beam. However,this important statement was totally negated by acorresponding measurement of the beam diameter. In theirsame paper, Moncrieff et al. say: "The beam diameter wasalso measured from the rise-time of the transmitted signal asthe beam was scanned across a sharp edge (Joy, 1974)... Itwas observed that a 50 nm beam, after 20 mm flight path,changes little up to a pressure of ∼10 Pa. The increase inbeam diameter, up to 100 nm at 133 Pa, above this pressure isindicative of the changing shape of the beam distribution.The change in slope for the onset of the beam maximum isobserved experimentally as an increase in the rise-time of thetransmitted signal, and this is reflected in the larger beamdiameter." They also measured the scattered electrondistribution with a Faraday cup and found good agreementbetween experiment and theory. However, this goodagreement referred only to the tail of the scattered electrons,not to the immediate neighborhood of the useful spot. Itcould then be speculated that the single scattering regimetheory could not account for the plurally scattered electrons,which could, presumably, alter the shape of distribution at thebeam diameter level. As a result, this important issue relatingto the ultimate resolution of ESEM had remained unresolveduntil a comprehensive study was published by Danilatos(1988). In the light of that study, we can now establish theconditions under which Moncrieff et al. conducted theirexperiment for the beam diameter measurement: From thevalues of pressure and distance travelled, we find m=0.125 at10 Pa (with σT=2.54x10-21 m2 at 25 keV) and m=1.25 at 100Pa. Clearly, these values are within the oligo-scatteringregime and no beam spread should have been observed. Infact, the use of a long specimen distance at 20 mm results in awider skirt by one order of magnitude than when we use 2mm, and the separation of useful spot from the skirt should bemuch more distinct. In addition, since the real cross-sectionfor nitrogen is expected to be smaller than the theoreticalvalue used here, the values of m for the Moncrieff et al.experiment should be even less (i.e. they definitely operatedin the oligo-scattering regime). Their observed beam spreadwas probably due to contamination of the sharp edge theyused, as this type of artifact was also observed by Danilatos(1988). The experimental solution to this problem was toheat the edge at a high temperature to stop a contaminationfinger developing.

In connection with the electron beam spread, somemisunderstandings have also been published by Farley andShah (1990a). They have suggested that, when the averagenumber of collisions per electron equals unity (i.e. at onemean free path), then we have a 100% (total) electron beamloss. This has led them to believe that the limit of imaging inESEM occurs when m=1 (presumably thinking every electronis scattered out of the beam). However, under this condition(m=1), we have 63% of the electrons removed from the beamand 37% of original electrons still remaining in the originalspot (see Eq. (3)). Our practice has shown that this can be

27

quite adequate to operate the instrument. Therefore, theirsuggested limits of pressure operation are in error. They alsoincorrectly claim that their results agree with Danilatos(1988). However, their paper closely repeats the work byMoncrieff et al. (1979).

Farley and Shah (1990a) claim that the inelasticallyscattered electrons influence the beam current density profileand, hence, they deteriorate resolution. They generallybelieve that "in high-pressure SEM…the beam profile andelectron current distribution on the specimen surface arealtered", and that this affects the resolution of the image.

Shah and Beckett (1979) used an environmental cell forthe study of wet botanical specimens. Notwithstanding thevalue of that publication along with those of other workersthat preceded it, we are compelled to make a reappraisal ofthe early ideas put forward, especially in view of theircontinual repetition until recently by the same group. In theirfirst paper, they used a differentially pumped environmentalcell in conjunction with the "absorbed specimen current"mode for imaging. That system was named "moistenvironment ambient temperature scanning electronmicroscope" or MEATSEM. Moisture was perceived as anecessary and sufficient condition of MEATSEM, in order tomaintain the specimen conductivity and, hence, make thespecimen current mode feasible. This condition was clearlyspelt throughout the paper as, for example, they say: "Thestage essentially incorporates differentially pumped chamberswhich allow the specimen to remain conducting, during theoperation of the microscope, for a comparatively long period,keeping it at ambient saturated vapour pressure of water"; andfurther "MEATSEM avoids this type of damage because thehigher electrical conductivity of the moisture content of thespecimens eliminates the need for metal coating". However,in our present day understanding, this is not necessary(Danilatos, 1983; 1990a,b,c). The presence of gaseousionization was totally overlooked, or its role mistaken byShah's group for many years later: Shah (1987) says, forexample: "…Forming an image under these conditionspresents formidable difficulties. The conventional techniqueof constructing an image by secondary electrons does notwork because secondary electrons, primary electrons andback-scattered electrons ionize water or gas molecules closeto the specimen and produce additional electrons. Theseelectrons, which do not carry any information about thespecimen surface, have a similar energy range to that of thesecondary electrons emitted from the specimen, so theycannot be separated easily from the secondary electronsreleased from the specimen surface. Without such separation,there is a severe deterioration of the secondary emittedimage…". Clearly, such views are not helpful and caution isrequired when referring to these works. The use of ionizationto suppress charging artifacts was known and used by severalauthors previously (Pfefferkorn et al., 1972; Parsons et al.1974, Moncrieff et al., 1978). The use of ionization forimaging purposes was first introduced by Danilatos (1983).This was first applied to the commercial ElectroScan ESEMfor secondary electron imaging in early 1986.

Shah has recently acknowledged the use of ionization asan imaging means but this is still confused with the notion of

28

the conventional specimen current mode for imaging. Inrecent articles, another acronym, namely, HPSEM for "highpressure SEM" was introduced to essentially refer toMEATSEM (Shah et al. 1990; Farley and Shah, 1990a,b).Those authors still advocate that "specimen currentimaging…can be usefully applied to high-pressure SEM sinceit does not rely on the interception of the emissive electronsor the physical amplification of the signal within thespecimen chamber" (Farley and Shah, 1990b). In otherwords, the emissive modes are thought to be intercepted bythe gas, whereas the specimen current is not. This clearlyexplains their concept of specimen current, which, flowingthrough the specimen, is not affected by the gas (their idea).This, of course, does not explain the last remark in that samepaper that under charge neutralization "no net specimencurrent flows into or out of the surface" and it does notexplain how their imaging is possible with their specimencurrent mode when no specimen current is present. Suchideas seriously overlook the true natural processes occurringin the microscope, and they are clearly set apart from our ownunderstanding. We advocate that contrast is formed byinduction during the flight of all charges between electrodes,i.e. by displacement current (Danilatos, 1990a,b,c). The flowof current through the specimen should be taken into accountwhen we balance the total charge. Accumulation of chargecan still be present in certain cases and the conductivity of thespecimen can influence the final contrast, but the "specimenabsorbed current mode" does not really exist in its own rightas an imaging mode per se. Charging and specimen current,to the extent they occur, are after-effects in the final imageformation.

In Fig. 7, we have considered the effect of diffusion onthe total current collected by two plane electrodes. Othereffects, such as recombination and space charge, have beendiscussed in detail and found not to contribute to anysignificant degree in the conditions of ESEM. However,Farley and Shah (1990a,b) believe that these factors controlthe signal intensity or quality. For space charge, in particular,they write: "In the absence of any electric field the ioniccarriers form an accumulation of space charge above thespecimen which can inhibit or distort the emission of thesecondary electrons or the collection of low-energy ioniccarriers. To counter the action of the space charges, it isnecessary to extract them from the vicinity of the specimen.This can be done by an electric extraction field provided by abiased electrode placed above the specimen". The "spacecharge" notion is heavily promoted throughout that paper(Farley and Shah, 1990b), which, otherwise, merely repeatsthe work by Moncrieff et al. (1978). According to theliterature surveyed and to our own experience, the presentauthor has reported that space charge is of no concern in theESEM, especially at low electrode bias (Danilatos, 1990a).Space charge effects can appear within individual avalanchesat the head of the avalanche, at very high electrode bias onlyunder specialized conditions of very high gaseous gain.However, the positive ions can pose a problem only becauseof a possible accumulation on the specimen surface,especially with very large and extended flat insulatingspecimens below a flat anode electrode. For this problem, we

29

have employed an additional electrode above the specimensurface as an ion controller, or sink, for any excess positiveions. A fraction of positive ions is attracted by the specimensurface tending to become negative by the electron beambombardment (negative charge neutralization) and the rest ofthe positive ions find their way to the nearest conductingsurface (acting as a cathode). If such a surface is very faraway, we should provide one closer to prevent positive chargeaccumulation on the specimen surface, and this will result inan improvement of contrast. The improved contrast is causedby the higher gaseous gain attainable, as the effective field ismaintained high when the positive ions cannot accumulate onthe insulating surface. When they accumulate on the surface,the effective field is reduced and the signal gain deteriorates.The true nature of phenomena ought to be clearly understood,if we are to improve the performance of the GDD.

The signal propagation characteristic of the ionizationGDD in Fig. 8 shows the real and potential advantages of thismethod. For a low gaseous gain the noise bottleneck showncan be superseded by the equivalent-input noise current of theoperational amplifier. With a GDD gain factor up to 100,beam currents well below 100 pA are commonly used inESEM. Several orders of magnitude higher gaseous gain hasbeen achieved with equivalent nuclear devices and, hence, wecan expect significant improvements in future designs ofGDD (see extensive review of nuclear devices by Danilatos,1990a). At present, the main limitation arises from our desireto use water vapor in many applications, but for another classof applications, for which water vapor is not required, highgains should be achievable. Generally, any improvement inthe gaseous gain, even by small amounts, is very significantand desirable for this device.

The scintillation GDD holds great promise because ofthe very low noise level of the associated PMT. If every careis taken to optimize the scintillation of the gas as well as theoptical coupling and light transmission from the light sourceto the PMT, then we can expect some excellent SNRcharacteristics.

The early work on ESEM involved the use of plasticscintillating detectors and this prompted a consciousdevelopment of these detectors. The aim was to increase thedetective quantum efficiency of the detectors in order tocompensate for the loss of signal from the beam-gasscattering. Also, those detectors had to be reshaped andredesigned generally to make them fit in the restricted regionof operation of the prototype machine. It became apparentthat the shape of these detectors could easily become criticaland could result in serious light loss, which, in turn, wouldcreate a second noise bottleneck. That was indeed the casewith early shapes of BSE detectors. Optimum shapes thatwould fit in a particular prototype were reported later(Danilatos, 1985). In Fig. 10, it is clearly shown how criticalthe detector design becomes for low keV operation, a qualitythat is highly sought in ESEM, especially for beam sensitivematerials. The first factor that can be improved is the BSEcollection angle. Here, we refer to that fraction of BSEassociated with particular type of information and withparticular spatial or atomic number resolution; noconsideration is given on how the various fractions of

30

electrons are separated, something that has been the subject ofstudy by the electron microscopy community for a very longtime. Most recent reports have quoted resolutions below 1nm by use of a wide angle BSE (Autrata, 1990; 1992). Theconcept of using a wide angle BSE detector has beensupported by various workers, but we must separate out thisconcept from Robinson's ideas and practice (Robinson, 1973;1974). Robinson advocates that "the complete rediffusedelectron signal must be detected, using a 2π geometricrediffused electron detector" and "…that collection of thetotal rediffused electron signal gives the same resolution asthe secondary electron signal…". In practice, Robinson hasused a large (near hemispherical) piece of bulk scintillatingdetector with a large hole in it for the passage of the beam.This subtends a wide angle at the specimen. However, theimages with this design show directionality of illumination(shadowing), which is indicative of loss of BSE signal fromthe side of the detector across the light pipe. Some workersmask a portion of the more efficient side of this detector tomake the image uniform, but all this shows is that theemployed shape with a single PMT is significantly less thanoptimum. The high resolutions observed recently with theuse of more efficient BSE detectors are now generallyattributed to the "Murata peak" (Murata, 1974). The reasonfor resurfacing this old topic is because there is a need toimprove the efficiency of BSE detection in ESEM and toshow that there is scope for further improvements of BSEdesigns. For example, the hemispherical type of bulkscintillating detector with a hole in it and a single PMT isinefficient (or insufficient) in ESEM. The hole in the detectorcannot become less than the PLA1. In ESEM, the specimenmay be placed close to the PLA1 and most of BSEs areconcentrated in a small region of the order of the PLAdiameter. Small variations (i.e. fraction of mm) in thedetector machining or positioning can result in largedeviations from optimum signal in ESEM. The light pipedesign and the presence of the hole present serious obstaclesand the end result is that we can have large BSE signal lossfollowed by large light loss in the light pipe. The situation isgreatly improved by a calculated shape of detector and by useof two detectors instead of one. Much better results by wayof efficiency and signal manipulation could be achieved by asystem of four detectors. The high resolution "Murata peak"is associated with a low BSE fraction of the total signal andonly an efficient detector would render this signal visible. Alow efficiency detector uses additional BSE electrons fromand towards the tail of the spatial electron distribution and theresolution deteriorates.

The need for efficient solid scintillating detectors inESEM has prompted us to propose a standard way forobjectively measuring the efficiency of these detectors. Thesame method, namely, the measurement of ∆(E), could beused for all BSE detectors in general, except that this can alsobe dependent on the beam current used. The case that weanalyzed in this paper is applicable to detectors with goodPMT which have very low anode dark current. If we useoperating amplifiers instead, then the efficiency characteristicwill also depend on the beam current, generally speaking, andhence we should consider ∆(E,I) as the appropriate

31

characteristic. For our present needs, we need to incorporatethe specimen distance as an additional parameter: ∆(E,L).

One main aspect of the present survey is the advance ofthe universal detection system in Fig. 11. An importantingredient of this detection configuration is the introductionof an "ion generator" by means of a biased electrode in theneighborhood of uncoated scintillating detectors. Thisextends the operation of ESEM down to relatively very lowpressures, much lower than without a controlled discharge. Aself-controlling discharge usually results in an erratic orirregular charge suppression which becomes evident on theimage as an instability as we decrease pressure. For example,the value of pd=0.06 Pa-m found implies that for a specimendistance of 10 mm we can operate down to 0.06 mbarpressure. By carefully choosing the gas mixture, this pressurecould be even lower. Therefore, we can bridge the gapbetween the vacuum of the conventional SEM and the usualhigh pressure environment of the ESEM. This is a novelapproach that has come as a "spin-off" from the developmentof the GDD. Moncrieff et al. (1978) experimented with themeasurement of the ionization current of the gaseousdischarge only for the purpose of determining the effectivenegative bias that automatically forms on insulatingspecimens from the electron beam bombardment. Theyconcluded that a discharge was automatically forming as thespecimen was charging to about -140 Volts by the incidentelectron beam, and the residual gas in the specimen chamberwas sufficient for the purpose. Until the present time, onemember of that group still uses and still considers both thealuminum coating and a metal liner in the large hole of thedetector as necessary elements for wide angle plasticscintillating detectors (Robinson, 1980). The deliberateintroduction of a controlled gaseous discharge at low andintermediate vacuum, as suggested here, can significantlyimprove the performance of detectors.

The main conclusion on the BSE detectors is thatESEM has created the unique opportunity of detecting theBSE signal with uncoated and unbiased solid detectors atbeam accelerating voltages well below 5 keV. Already, thelow voltage SEM (in vacuum) has demonstrated its value butalso its limitations with regard to the type of specimens andrange of beam keV. The possibility of charge suppression atintermediate vacuum implies also the possibility of using lowbeam voltage in ESEM, without the hurdles and limitations ofvacuum SEM. In the near future, practice will show themerits of this new approach in electron microscopy.

We need to clarify that we should not resist using someof the conventional methods of SEM, if some applicationsrequire us to do so; all these methods can be incorporated inthe ESEM. For example, if we wish to image hot specimenswith YAG/YAP detectors, then we have to coat the detectorswith aluminum to prevent the hot stage light from interferingwith imaging. The ionization GDD is, of course, capable tooperate in the presence of light. Also, we can easilyincorporate the conventional E-T detector, should its presencebe required. If some applications have to have long workingdistances, or very high tilt or other manipulation that is usedin SEM, then ESEM can also incorporate these parameters,with the understanding that some of the advantages that the

32

ESEM per se offers may have to be compromized, orsacrificed. Any of the specimen preparation techniques, or amodification of these, can be applied to ESEM also. Inconclusion, the ESEM is in no way lacking over theconventional SEM. The latter is a subset of the former.

All imaging has been omitted from the present report,as this needs to be systematic with each separate topic. Adetailed examination of all these topics, simultaneously,would fall outside our original aim. The different topicssurveyed in this report have only been outlined and discussedin order to facilitate their better understanding and to invitefurther contributions from other workers. There are stillmany questions open, and many aspects require furtheranalysis and experimental support. New results are plannedto be reported in due course.

Conclusions

The electron beam in the ESEM splits in two fractionsas it travels through the gas layer to reach the specimen. Onefraction remains focussed in the same spot that forms invacuum and the other fraction forms a broad scatteredelectron skirt around the focussed spot. This defines theoligo-scattering regime of ESEM. The focussed spot can beused for imaging in the usual way while the skirt adds abackground noise level. The resolving power is limited bythe probe size which remains constant as we increase thechamber pressure. Only when we are forced to increase thecontrast by increasing the beam current do we sacrificeresolution on account of beam diameter increase. However,for a large number of applications in ESEM we rarely needbeam diameter magnifications.

The gaseous detection device has replaced theconventional SE detector. At present, the GDD is used in twomodes, namely, the ionization and the scintillation mode.Both these modes can produce SE and BSE imaging. Inaddition, solid scintillating detectors have been developed toproduce high SNR imaging. The high efficiency is achievedby specially calculated shapes and by their ability to operateuncoated. This ability is greatly enhanced by deliberatelyproviding a gaseous discharge in the neighborhood of theactive surface of the detectors. The signal propagation of allthese detection systems has shown that they produce some ofthe best SNR features. Their efficiency coupled with theirability to image the natural surfaces of virtually any specimenproduces new types of contrast and new information inpractically every field of application of this microscope. TheESEM has become the universal instrument for operationvirtually under any environment including vacuum, lowvacuum and high pressure as well as low and high voltagemicroscopy.

References

Autrata R, Schauer P, Kvapil Jos, Kvapil J. (1983)Single-crystal aluminates - a new generation of scintillatorsfor scanning electron microscopes and transparent screens inelectron optical devices, Scanning Electron Microsc./1983/II489-500.

33

Autrata R (1990) New configurations of single-crystalscintillator detectors in SEM, Electron Microscopy 1990,Proc. XIIth Int. Congr. El. Microsc., San Francisco Press (Ed.LD Peachey and DB Williams), Vol. 1, 376-377.

Autrata R (1992) Single crystal detector suitable forhigh resolution scanning electron microscopy, EMSABulletin, Vol. 22, No. 2, 54-58.

Baumann W and Reimer L (1981) Comparison of thenoise of different electron detection systems using ascintillator-photomultiplier combination, Scanning 4,141-151.

Bird GA (1978) Monte Carlo simulation of gas flows,Ann. Rev. Fluid Mech. 10, 11-31.

Cosslett VE and Thomas RN (1964) The pluralscattering of 20 keV electrons, Brit. J. Appl. Phys. 15,235-248.

Danilatos GD (1982) The environmental scanningelectron microscope and its applications, Scanning ElectronMicrosc./1982/I, 1-16.

Danilatos GD (1983) A gaseous detector device for anenvironmental SEM, Micron and Microscopica Acta 14,307-318.

Danilatos GD (1985) Design and construction of anatmospheric or environmental SEM (Part 3), Scanning 7,26-42.

Danilatos GD (1986) Cathodoluminescence andgaseous scintillation in the environmental SEM, Scanning 8,279-284.

Danilatos GD (1988) Foundations of environmentalscanning electron microscopy, Adv. Electronics ElectronPhys. 71, 109-250.

Danilatos GD (1990a) Theory of the gaseous detectordevice in the environmental scanning electron microscope,Adv. Electronics Electron Phys. 78, 1-102.

Danilatos GD (1990b) Equations of charge distributionin the environmental scanning electron microscope (ESEM),Scanning Microscopy 4, 799-823.

Danilatos GD (1990c) Mechanisms of detection andimaging in the ESEM, J. Microsc. 160, 9-19.

Danilatos GD (1990d) Design and construction of anenvironmental scanning electron microscope (part 4),Scanning 12, 23-27.

Danilatos GD (1991a) Review and outline ofenvironmental SEM at present, J. Microsc. 162, 391-402.

Danilatos GD (1991b) Gas flow properties in theenvironmental SEM, Microbeam Analysis-1991, SanFrancisco Press (Ed. DG Howitt), 201-203.

Danilatos GD (1992) Secondary electron imaging byscintillating gaseous detection device, Proc. 50th AnnualEMSA Meeting, San Francisco Press (Ed. GW Bailey, JBentley and A Small), 1302-1303.

Farley AN and Shah JS (1990a) Primary considerationsfor image enhancement in high-pressure scanning electronmicroscopy (1. Electron beam scattering and contrast), J.Microsc. 158, 379-388.

Farley AN and Shah JS (1990b) Primary considerationsfor image enhancement in high-pressure scanning electronmicroscopy (2. Image contrast), J. Microsc. 158, 389-401.

34

Huxley LHG and Zaazou AA (1949) Experimental andtheoretical studies of slow electrons in air, Proc. Roy. Soc.Lond. 196, 402-426.

Jones RC (1959) Quantum efficiency of detectors forvisible and infrared radiation, Adv. Electronics and ElectronPhys. 11, 87-183.

Joy DA (1974) SEM parameters and theirmeasurements, Scanning Electron Microsc. I, 327-334.

McLure SG (1990) Backscattered electron imaging,M.Sc. Thesis, La Trobe University.

Moncrieff DA, Robinson VNE and Barker PR (1978)Charge neutralization of insulating surfaces in the SEM bygas ionization, J. Phys. D: Appl. Phys. 11, 2315-2325.

Moncrieff DA, Barker PR and Robinson VNE (1979)Electron scattering by gas in the scanning electronmicroscope, J. Phys. D: Appl. Phys. 12, 481-488.

Moretz RC (1973) Electron microscopy of biologicalmaterials in the natural state, Ph.D. Thesis, State Universityof New York at Buffalo.

Murata K (1974) Spatial distribution of backscatteredelectron in the scanning electron microscope and electronmicroprobe, J. Appl. Phys. 45, 4110-4117.

Oatley CW (1985) The detective quantum efficiency ofthe scintillator/photomultiplier in the scanning electronmicroscope, J. Microsc. 139, 153-166.

Parsons DF, Matricardi VR, Moretz RC and Turner JN(1974) Electron microscopy and diffraction of wet unstainedand unfixed biological objects, Adv. Biol. Med. Phys. 15,161-271.

Pfefferkorn GE, Gruter H and Pfautsch M (1972)Observations on the prevention of specimen charging,Scanning Electron Microsc. 1972; 147-152.

Reimer L (1985) Scanning Electron Microscopy,Springer-Verlag, Berlin, p 137.

Robinson VNE (1973) A reappraisal of the completeelectron emission spectrum in scanning electron microscopy,J. Phys. D: Appl. Phys. 6, L105-L107.

Robinson VNE (1974) The construction and uses of anefficient backscattered electron detector for scanning electronmicroscopy, J. Phys. E:Sci. Instr. 7, 650-652.

Robinson VNE (1980) Electron microscopebackscattered electron detectors, U.S. Patent 4217495, Aug.12, 1980.

Shah J (1987) Electron microscopy comes to life, No.208/1987/SPECTRUM, 6-9, (Published by Central Office ofInformation, London, UK, available by British Embassy orConsulate).

Shah JS, Durkin R and Farley AN (1990) Scattering ofelectrons at high-pressure and restoration of image contrast inhigh pressure scanning electron microscopy, ElectronMicroscopy 1990, Proc. XIIth Int. Congr. El. Microsc., SanFrancisco Press (Ed. LD Peachey and DB Williams), Vol. 1,384-385.

Shah JS and Beckett A (1979) A preliminaryevaluation of moist environment ambient temperaturescanning electron microscopy, Micron 10, 13-23.

Wells OC (1974) Scanning Electron Microscopy,McGraw-Hill Book Company, New York, pp 20-36.

35

Wilkinson DH (1950) Ionization Chambers andCounters, Cambridge University Press, Cambridge, p.38.

Discussion with Reviewers

M. Kotera: In the gaseous detection, a potential is appliedbetween the specimen and the detector. A kind of plasma isexcited in the field. The electric field is not linear in theregion, and the field may be large close to the specimensurface. Then, positive ions, which are ionized by electronbombardment, hit the specimen surface with relatively largeenergy. Is there higher possibility for the specimen to getdamaged by the bombardment?Author: The kind of plasma situation that you refer to doesnot occur in our system. We only have a weakly ionized gaswith the present gaseous gain of the GDD and with the lowelectron beam currents used. In future work, we will attemptto obtain a few orders of magnitude higher gain and then yourquestion could become more relevant. At present, sometimeswe have the opposite effect: With extended flat insulatingsurfaces, in a uniform electric field, we have a net positivecharge accumulation on the specimen surface, which resultsin a decreased total electric field, which corresponds to alower gaseous gain and lower amount of ions. With respectto the mechanisms of specimen damage for when it occurs,we have done little study up to the present [see Danilatos GD,(1986) Beam-radiation effects on wool in the ESEM, Proc.44th Annual Meeting EMSA, 674-675]. The ion-specimeninteraction will be followed with great interest in the future.

M. Kotera.: It seems that images obtained by the gaseousdetector and those obtained by the solid state detector showdifferently, because of the difference in the imagingmechanisms. Is it possible to estimate or evaluate what kindof information can be revealed by the difference? and why isthat?Author: Unfortunately, I have not seen published anycomparison between images obtained by solid state detectorsand GDD. I cannot comment on this yet. Tentatively, I maysuggest that the solid state detector images are BSE images,whereas those that you might have seen from GDD are SEimages.

A. Dubus: You conclude your paper by writing: "The ESEMhas become the universal instrument for operation...". Howdo you look to the future of this particular technique?Author: As I have stated on numerous occasions, the ESEMis the natural extension of SEM; the former is destined toreplace the latter in its traditional applications and, inaddition, it has opened many new areas of application notpreviously accessible to SEM. With reference to othermicroscopical techniques, ESEM is again destined to play itsrole, because it can give types of information not accessible tothose techniques either. The fact that magnification rangesmay overlap between various techniques is irrelevant, becauseeach technique has its own merits, advantages anddisadvantages. I do not see various microscopical techniques

36

as "competing" with each other. I rather see all thosetechniques as complementing each other.

J.M. Cowley: What is the relationship of the detector systemto the objective lens pole-pieces? If the detector electrodedistances are increased, as proposed, will this not decrease theresolution by increasing the focal length of the lens?Author: With reference to Fig. 11, the lower set of BSEdetectors can be placed just below (abutting with the bottomface of the lower pole-piece of the objective lens), as iscurrently done. The upper set of detectors can be integratedinside the objective lens. This suggested configuration is ageneral one and the specific dimensions can be varied toallow integration with the lens. The ESEM requirements areoutlined herewith, first, so as they can be incorporated infuture generations of instruments.

The resolution will decrease by increasing the focallength of the lens. My comment of increased specimendistance is referred to those workers demanding suchincreased working distances in order to accommodate largespecimen tilting or for other reasons. The preferred workingdistance in ESEM is a short one, in order to allow high gaspressure and, fortuitously, better resolutions. In fact, theGDD is ideally suited for such short working distances and,hence, ESEM is promising to produce the best possibleresolution with a given electron optics column.

R. Autrata: In the chapter "Solid Scintillating BSE detectors"you report that 2% of the electron energy is transformed intophotons in the scintillating material. According to Fig. 11,this material is YAG or YAP. The conversion radiationefficiency ξ=2% was reported by O.C. Wells (ScanningElectron Microscopy, McGraw-Hill Book Company, NewYork, 1974, p. 33) for the plastic scintillator, type Pilot. It is,however, known that for YAG it amounts to 4-5% (Takeda etal., J. Electrochem. Soc. 1980, Vol. 127, No. 2, p. 438) andfor YAP ξ=6-7% (Autrata et al., Proc. 8th European Congr.on EM, 1984, Vol. 1, p. 167). Precise values of radiationefficiency depend on the technology of preparation ofscintillators. Can a higher value of ξ influence yourevaluations of efficiency of your detection systems. Andhow?Author: In Fig. 11, I am proposing the use of YAG/YAPcrystals, but the calculations for a typical SEM in Fig. 10were done for plastic scintillator, as per reference given. Forthe ESEM graph in Fig. 10, I have used ξ=4%, as an examplefor possible improvement. You are stating that we can havean even better improvement than this, which is, of course,most welcome. I have tried to demonstrate in this paper that ahigh conversion efficiency material is highly sought for lowvoltage work, and the use of good grade YAP materials areplanned to be incorporated in our system (as in Fig. 11). Ahigher ξ shifts the second "dip" in Fig. 10 upwards, whichallows use of a lower keV beam.

R. Autrata: In the chapter "Solid Scintillating BSE Detectors"you report that 40% of the initial light passes through thelight guide toward PMT. It is known that the light passagedepends on the guide material (index of refraction and

37

absorption spectrum), wavelength of the passing light, shapeand surface treatment of the light guide. What material wasused for your light guide and how was it shaped? What doyou deduce the 60% light absorption in the light guide from?The light loss is extraordinarily high.Author: It appears that, between us, we use some expressionscorresponding to different objects: In your works, I think,you distinguish a "light guide" from a "light pipe", where thefirst leads to the second. In my present paper, I lump both ofthese parts under the "light pipe". The initial part of my lightpipe is what you term as light guide, but I have not seen thepurpose for distinguishing these two parts (at least not in thepresent paper). This may explain why you find a 60% lightloss in the "light guide" as extraordinarily high. In myexample, I consider that 40% of the total photons produced atthe BSE/photon conversion stage reaches the photocathode.The difference is lost on the way (any way). The same figureof 40% transmitted light was quoted by Wells, and I use it forthe typical SEM case. This was done to demonstrate thatthere is ample scope for improving the light pipes (ordetectors), as I state. I have reported special shapes of lightpipes (my term) with light transmission around 50%(Danilatos, 1985). I would have no hesitation to accept bettertransmission rates, whenever these are found and shown to bepossible, and this is the spirit of the present report.

R. Autrata: You say that "by the appropriate specimenpositioning, PLA1 size and bias we can achieve a very goodseparation of the BSE from the SE signal." Can you giveparameters of V1 and V4 voltages (Fig. 11) and otherconditions under which the separation of the SE from theBSE signal occurs?. Can you really obtain the true SEimage? Isn't it a mixture?Author: One example is to use a few tens of volts for V1 anda few hundreds of volts for V4 (both positive) with a fewhundreds of Pa pressure. The SE that get though PLA1receive a preferential amplification over the BSE that also getthough the aperture. I am referring to using the GDD alonefor imaging in this case. It is true that the two signals are inmixture, but if, say, 90% of the intensity is due to SE and10% of it to BSE, then we can safely classify the image as aSE one. We have obtained results of SE imaging above thePLA1, under these conditions, not yet reported. However,there are other electrode configurations by which we havereported good separation of the two types of imagingelsewhere (Danilatos, 1990c). In the present report, we areproposing to integrate the GDD with solid scintillatingdetectors to achieve a much more flexible system, with morepowerful (deliberate) signal mixing, separation andprocessing, in general.

R. Autrata: In the chapter "Standardization of BSE DetectorEfficiencies" you say that "ESEM has now ushered thepossibility of using solid scintillating BSE detectors with alow voltage beam without the need to accelerate the BSE inthe kV range, as is usually done with low voltage in vacuum."However, the need to accelerate BSE at low voltage operationdoes not result from the BSE energy loss which occurs duringthe passage of BSE though the conductive coating of the

38

scintillator. It is possible to prepare conductive layers onYAG with energy absorption less than 20% for the 1 keVelectron beam energy. A more serious problem is thedependence of the light signal of the scintillator on the energyof incident electrons. The light signal of the scintillatorproduced by the BSE impact is very low at the 1 keV electronbeam energy. The difference in the number of photonsincident on the first dynode of PMT is one order for 10 keVbeam energies. The resulting low signal-to-noise-ratio is thereason for accelerating the BSE toward the scintillator. Doyou have some experimental experience in using ESEM lowvoltage operation and an uncoated scintillator for BSEdetection?Author: I agree that the difficulty is caused by the fact that atlow voltage operation we start only with a relatively smallamount of photons at the BSE/photon conversion stage. Thisdifficulty should not be interpreted as the root cause for theinability to use unbiased detectors in conventional SEM at 1keV. The presence of a conductive layer, conventionally analuminum layer, has significantly contributed to signal energylosses at this early conversion stage and, hence, in theprevention of use of the very low keV range, withoutaccelerating the BSE. An aluminum coating will absorb allthe BSE with 1 keV beam. The use of light pipes withoutoptimum transmission and coupling with the PMT, areadditional factors, and all together have prevented theconventional SEM from operating in such a mode. Thespecial conductive layers that you are referring to are not yetwidely known or practiced, and they may help for somespecial detection requirements in our universal ESEM.However, as you say, even these layers absorb 20% of energyand, hence, they must come second to using a minimum gaslayer over the specimen, instead.

For high keV, the noise bottleneck is at the number ofBSE incident on the uncoated detector. For low keV, thenoise bottleneck is at the number of photoelectrons incidenton the first dynode of the PMT. In Fig. 10, I have assignedthe second minimum ("dip") of the signal to the photocathodeitself, as I have tacitly assumed that all the photoelectrons arecollected by the first dynode. I have assumed that the PMTmanufacturer has provided the best possible design ofphotocathode/dynode configuration. Assuming an optimumPMT, Fig. 10 describes which variables are left for us tomanipulate so that we can design an over all optimumdetection system. For low keV work, we both agree that thenoise bottleneck is at the photocathode/first-dynode stage, andthe most critical factor becomes the choice of photocathodematerial and its condition. This relies entirely on the PMTmanufacturer. Given the best choice of PMT, we are thenfaced with optimizing the previous stages in the detectionchain as outlined in the present paper. We have preliminaryresults of operation at low voltage with uncoated scintillatorfor BSE detection and a proper report will be made in duecourse.

D.E. Newbury: Since, as the author himself notes, watervapor is the preferred gaseous medium in ESEM, what are theexpected general trends for the influence of such a polar

39

molecule on the various gas thickness calculations ascompared to nitrogen or argon?Author: The theoretical scattering cross-section for watermolecule is less than that of nitrogen and preliminaryexperimental values are even below the theoretical value.This has been confirmed by practice during imaging, wherebetter results are achieved with water vapor than withnitrogen. A survey of scattering cross-sections for variousgases has been presented previously (Danilatos, 1988). Goodexperimental measurements on many gases that can be of usein ESEM are still lacking, and we would welcome resultsfrom other laboratories that could dedicate some work in thisarea.

D.E. Newbury: I find it difficult to believe that you can placemuch hope that the efficiency of the GDD can be significantlyimproved based upon the experience with nuclear devices: (1)Nuclear particle detection generally involves particle energiesin the MeV range rather than the low keV range with whichwe deal in the SEM. (2) In nuclear particle detection there isno issue of keeping a beam sharply focussed, as there is in theESEM.Author: (1) In nuclear physics, there is also an interest todetect electrons with insufficient energy to produceimmediate ionization (i.e. like our SE). See, for example,"Single electron detection in proportional gas counters" [GenzH (1973), Nuclear Instruments and Methods 112, 83-90] and"Electron multiplication process in proportional counters"[Raymond G and Bennett EF (1966), Physical Review 147,201-213]. (2) I agree that in nuclear particle detection there isno issue of keeping a beam sharply focussed. In thebeginning, it was not obvious that those methods would workin our field. A lot of experimental and theoretical work hadto be done. This is where our contribution lies. One of thenovelties in ESEM is that we have successfully transferrednuclear methods to electron microscopy. It works, and itworks well, indeed!. For a thorough and detailedinvestigation please refer to Danilatos (1990a).

D.E. Newbury: No mention is made of the critical issue ofthe possibility, or lack thereof, of X-ray microanalysis in theESEM. In any discussion of the relative intensities of thebeam and skirt, the utility of such a beam for X-raymicroanalysis should be considered. It seems clear that whenthe ESEM is operated at such high pressures that 67% of theelectrons are located in the skirt, such a beam is useless forany realistic microanalysis applications. It would beinteresting to consider what could be achieved in X-raymicroanalysis with the ESEM operating at the other end ofthe scale, that is, a pressure such that water can just beretained. What is the beam/skirt ratio and how degraded areX-ray spectra obtained from small objects such as 5micrometer diameter inclusions in a matrix? The author mayregard this topic as outside his range of "critical issues", butconsidering the claims made for the ESEM relative to theconventional SEM, the X-ray microanalysis shortcomings ofthe ESEM should be ventilated.Author: I am fully aware of the importance of X-raymicroanalysis and associated complexities in ESEM. Because

40

this topic is quite extensive and relatively little work has beendone in ESEM up to the present, I had opted to leave it out ofthis paper. There are many more critical issues that have notbeen dealt with in the present work. Only some issues havebeen considered. You are correct, though, that I should havementioned it. It is only that I wanted to reduce the amount ofspeculation and restrict the contents only to the more obviouscases. I can only make a few suggestions here: One or moreof the detectors in Fig. 11 can be replaced with an X-raydetector. With a high pressure environment, one way toremedy the interference of the skirt is to make the skirt verysmall. This can be achieved by placing the specimen so closeto the PLA1 that the skirt radius is of the same order ofmagnitude as the beam-specimen interaction volume. Thisapproach is facilitated by use of high keV in conjunction witha small PLA1, whereby the low magnification is traded offfor high magnification. This approach has not been practicedyet, because, for best results, it requires the positioning of X-ray detectors above the PLA1, which implies the integrationof these detectors with the lens system. An alternativesolution to the question of electron skirt has been to subtract acalibrated percentage of a "raster" spectrum from a spotspectrum [Bolon RB, (1991) X-ray microanalysis in theESEM, Microbeam Analysis-1991, Ed. Howitt DG, SanFrancisco Press, San Francisco 199-200]. The idea is tosomehow calibrate out the effects of skirt.

For a fully wet specimen, the effects of skirt remainsevere. However, for insulating specimens, we only need avery small amount of gas to dissipate charging and this is anarea where X-ray microanalysis can be practiced in the usualway in ESEM. It should be realized that ESEM is still at itsinfancy of development and is crying out for contributions byexperts in various fields as those working in X-raymicroanalysis.

I share your concern about the loss of spatialresolution with X-ray microanalysis at high pressure and longworking distance range. At the present, my best answer tothis is that I believe that a solution and a method for thisproblem will be found in the near future.

D.E. Newbury: The comment "Low voltage ESEM clearlyoffers all the advantages without any of the limitations ofSEM." is clearly wrong. Low voltage SEM is limited in itsresolution by the decreased brightness of the source. For FE-SEM, reasonably high resolution can still be obtained becauseof the inherent brightness of the source, but for LaB6 orconventional tungsten sources, the brightness limitation ismuch more significant. If the inevitable ESEM loss of beamelectrons due to gas scattering is considered, the resolutionwill be even poorer. It therefore seems that ESEM has asignificant disadvantage relative to the SEM when resolvingpower is considered. Finally, low voltage X-raymicroanalysis is indeed possible in the conventional SEM, butmicroanalysis is impossible in a microscope where 50% ormore of the beam electrons are found in the wide skirt of thebeam.Author: ESEM, like a SEM, can be used with all three typesof guns, not only with tungsten and LaB6 as the ElectroScanESEM currently operates. Nikon Corporation has recently

41

announced a FE ESEM, which has been adapted as a CriticalDimension Measuring SEM to serve in the electronicsindustry. Furthermore, ESEM, like a SEM, can be used withall ordinary modes of detection. On the X-ray mode, Icommented previously.

The question of resolving power in ESEM requires twosteps of approach. The first step is to establish whether theuseful spot spreads or not. It has been found that theunscattered electron beam fraction is clearly separated outfrom the skirt which is orders of magnitude larger. This isvery significant, because, otherwise, the electron skirt wouldproduce a first order, i.e. a gross deterioration of the spotdiameter (i.e. if it were distributed in the immediateneighborhood of the original spot (see extensive study byDanilatos, 1988)). In this connection we state that theresolving power is the same as in vacuum, and this has beendemonstrated in practice. The second step is to consider theintensity of a fixed diameter spot. This step has beenconsidered quantitatively with a formulation of SNRrelationships (Danilatos, 1988). It has been acknowledgedthat, with a loss of beam electrons, we lose some of theultimate possible resolution but this loss is not gross orcatastrophic under the normal operating conditions of ESEM.Your question is ultimately reduced to quantifying the"losses" and "gains" in the ESEM. The parameter of gaspressure may be considered as the independent variable whichdetermines the limits (or range) of operation of all othervariables (or parameters). Some of those other variables arethe accelerating voltage, beam current, specimen positioning(distance and tilt), temperature and detection efficiency.They, in turn, determine a host of other variables such asbeam spot, contrast and resolution, specimen stability, etc. Inthis work it has been attempted to show that ESEM can bemade to operate in the complete pressure range from highpressure to high vacuum. Formally, then, we can state that

Universal SEMESEM p → →0

The art of establishing the interrelationships andultimate physical limits of operating parameters, as we varythe pressure of gas in the specimen chamber of themicroscope, constitutes the science of ESEM. It is a "give-and-take" situation as we vary the pressure but, moreprecisely, practice has shown that it is much more of a "take"and much less of a "give" situation. Under this light, we canfirmly state that SEM is a partial case of ESEM. The ESEMcan be reduced to a SEM. Therefore, to make the previousstatement unambiguous, we have introduced the modifieruniversal so that we can state: The universal ESEM offers allthe advantages without any of the limitations of SEM (i.e.limitations associated with the vacuum condition).