Você está na página 1de 17

Micron, Vol. 27, No. 3-4, pp.

247-263, 1996
Pergamon Copyright 1996ElsevierScienceLtd Printed in Great Britain. All fights reserved 0968-4328/96 $32.00

PIh S0968-4328(96)00023--6

Low Voltage Scanning Electron Microscopy


D A V I D C. J O Y * t and C A R O L Y N S. J O Y t

*EM Facility, University of Tennessee, Knoxville, TN 37996, U.S.A. "~Oak Ridge National Laboratory, Oak Ridge, TN 37831, U.S.A.

Abstract--Low voltage scanning electron microscopy (LVSEM) is the application of the SEM at beam energies below 5 keV. The fall in electron beam range compared to its magnitude at higher energies leads to significant changes in the beam interaction volume and in the secondary and backscattered electron yields. The topographic and beam penetration contrast effects which dominate images at high energies are replaced by detector collection etliciencycontrast effects giving images which are less three dimensional but which contain more detailed information on the surface morphology and, in some circumstances, the surface chemistry of the specimen. In order to observe non-conducting specimens a state of charge balance must be obtained to obviate imaging artifacts. This requires an optimized choice of the incident beam energy, sample tilt, beam current and magnification for each sample. The high stopping power of electrons at low energy can result in enhanced radiation damage. However, because of the small electron range such damage is confined to a thin, near surface, region of the specimen. The combination of a field emission gun and a high performance lens allows the probe size of the instrument to be made almost independent of the chosen beam energy over the range 1-30 keV and probable advance in electron sources and electron optics promise still better levels of performance for the LVSEM. Copyright 1996 Elsevier Science Ltd. CONTENTS Introduction ....................................................................................................................................... The effects of the choice of electron energy .................................................................................................... Contrast formation in low voltage images ..................................................................................................... The observation of non-conducting samples at low voltage .................................................................................. A. Introduction ................................................................................................................................... B. Procedures to minimize charging ............................................................................................................ V. Radiation damage at low beam energies ....................................................................................................... VI. The electron optics of the low voltage SEM ................................................................................................... VII. Conclusions ........................................................................................................................................ Acknowledgements ............................................................................................................................... References ......................................................................................................................................... I. II. III. IV. 247 247 250 254 254 255 260 261 262 262 262

I. I N T R O D U C T I O N Even before the scanning electron microscope (SEM) existed as a practical instrument it was realized that operation at a low incident b e a m energy (which for the purposes o f this p a p e r will be taken to be below 5 keV) would be advantageous. This was because it was h o p e d to see image c o n t r a s t resulting f r o m the changes in the secondary electron coefficient fi f r o m different elements, and this effect which w o u l d be largest when the SE yield was at its m a x i m u m , i.e. at low b e a m energies (Knoll, 1935a; von Ardenne, 1940a). The first, mechanically scanned, S E M built by K n o l l (1935) was therefore designed to operate between 0.5 and 4 keV, while the pioneering electronically scanned S E M o f Z w o r y k i n et al. (1942) o p e r a t e d at 800 eV. T w e n t y years later, when the first commercial S E M was already under construction, T h o r n t o n (1960) summarized a n o t h e r expected advantage o f low energy observation by noting that "if the b e a m voltage is reduced until the secondary emission coefficient is equal to or greater t h a n unity, the surface potential will be automatically stabilized at that o f its surroundings . . . U n d e r these conditions the effective secondary emission coefficient o f an insulator is unity regardless o f the angle o f incidence o f the p r i m a r y b e a m and the picture c o n t r a s t is controlled only by collector

m o d u l a t i o n " . Despite these insights and expectations it has required a n o t h e r thirty years o f instrument develo p m e n t before the promise o f the low voltage S E M ( L V S E M ) could be routinely exploited. This paper discusses why, and how, low voltage scanning m i c r o s c o p y might be expected to differ f r o m high voltage observations. The f o r m a t i o n o f image contrast in the L V S E M , and the p r o b l e m o f specimen charging, are discussed in detail and the question o f radiation d a m a g e is considered. Finally, the special features o f the electron optics o f the L V S E M , and the future o f this technology are examined.

II. T H E E F F E C T S O F T H E C H O I C E O F ELECTRON ENERGY Reducing the energy o f the electrons incident on a solid has a p r o f o u n d effect on the interactions that occur within that material. Figure 1 shows the variation in the stopping power, i.e. the rate at which the electron is giving up its energy to the s u r r o u n d i n g solid, o f an electron traveling t h r o u g h c a r b o n as a function o f its energy ( L u o et al., 1991; Luo, 1994). As the energy falls the stopping power increases f r o m a b o u t 1 e V / A to a m a x i m u m value close to 10 eV/A and then falls a w a y 247

248
10 "t:: ...... ; ..... i ...... ;---++

D . S . Joy and C. S. Joy [ i t"4+l+l.......... I


10
i i ! i i i I iii ) [ ii i

8
L O

,l. I/.
4
l

7 ]u,t:ut+
s | i

44 t
i i

o
f,O

.,1tt Ilil""
4:::::t:::::t::::

\ iI -

Xl

i ;i~;;;
i [

ll

II

I I

I I IIII I I IIIIII

ii l l s I IIII

l__l_i_!!i]
i i
I

II

I_.LI i i i ;
I I II | l

I I IIIII
I I IIIII

N
+~lIH I IIIII

Cu

+..l"

I 1

I I IIIIII
+: : l
I

I I

I I

l ]'Vr~
I Illll

11
:: ::: : i I : i !

IIIIII
i illJl

"t--I
I I

L.I_J_I t Ili "-'LI I 11111

I I

I I

I I

I fill I IIII

I II I I I

II IIIII II IIIIII
I

i
.01

I I II I ' , I
! IIII

] ::tl:l

] J ..L i J_l.lt i.. I......I. I II I I I I

l_k11111
.01

H! !!!1

: !_.LH..!LL! I

i +++!
lO

I lllU I IIIII

I IIIH

.001

1
.1

. . . . . . . .

i 1

. . . . . . . . 10

.1 1 E l e c t r o n E n e r g y (keV)

E n e r g y (keV) Fig. 2. Variation of electron range as a function o f energy. D a t a obtained by numerical integration o f experimental stopping power data.

Fig. 1. Stopping power variation as a function of energy for a n electron traveling t h r o u g h carbon. Unpublished experimental data from L u o (1994).

again at still lower energies. Electrons in the energy range between about 100 eV and a few keV therefore interact much more strongly with the material through which they are traveling than is the case at higher energies. The first consequence of this is that the distance which the electron will penetrate will fall rapidly as the incident energy is reduced. A convenient measure of the distance traveled is the Bethe range Rn, defined from the equation

R~=J:+,(._(I~_dE))dE
(1)
1 keV 3 keV

where, for an electron of energy E traveling along a trajectory of length s, - (dE/ds) is the electron stopping power, Eo is the incident energy and Emi, is some suitable lower energy limit for the integration, typically between 30 and 100 eV corresponding to the energy at which the stopping power reaches its maximum value.

Figure 2 shows the calculated electron ranges as a function of incident energy for four elements, obtained by numerical integration of experimentally measured stopping powers such as that shown in Fig. 1. It can be seen that the range decreases rapidly (as about E 1"6)with energy, falling from a micrometer or more at 10 keV to only tens of nanometers for energies below 1 keV. Note that while there are significant differences in the range at higher energies for the various targets, at the lowest energies all materials have approximately the same range. By performing a Monte Carlo simulation of the electron trajectories in the specimen (Joy, 1995a) the variation in beam interaction volume with energy can be displayed (Fig. 3). When plotted at a constant scale of magnification the enormous change in the relative interaction volume is evident, with the 1 keV volume being barely visible compared to the 10 keV case. This confirms that while scanning microscopy at conventional energies (10-30 keV) can observe the bulk of a specimen, operation at low beam voltages restricts the

I0 keV

,Z,,

........

1.,

0.5 ~Lm

Fig. 3. Interaction volume o f electrons in carbon at 1, 3 a n d 10 keV beam energies. M o n t e Carlo simulation using experimental stopping power data and M o t t cross-sections.

Low Voltage Scanning Electron Microscopy

249

interaction to the near surface regions of the material. We would thus expect the low voltage image to be more sensitive to the chemical nature and topographic form of the specimen surface. Since all of the signal in the low voltage image will come from close to the entrance surface of the electron beam even very thin specimens will behave as if they were bulk. This is demonstrated in Fig. 4 which shows secondary electron (SE) images of a 200 A thick carbon film stretched over a copper mesh grid. At 20 keV, in Fig. 4a, the film is almost invisible because the majority of the incident electrons traverse the carbon without producing any interaction, but at 1 keV (Fig. 4b), all of the electron interaction volume is contained within the thickness of the carbon, and the

resultant SE and BSE emissions from the film make it appear solid and opaque. This variation in the range and interaction volume affects both the secondary electron (SE) and backscattered electron (BSE) signals produced by the specimen. Figure 5 shows the variation in the secondary electron yield of silver as a function of the incident electron energy using data from references in Joy (1995b). As the incident energy falls the secondary yield rises, passes through a broad maximum and then slowly falls, a behavior which can be understood from Figs 1 and 2. The rate at which secondary electrons are produced is proportional to the stopping power of the electron (Bethe, 1941) so the production of SE rises

20kV

I kV
Fig. 4. Images o f a 200 .~ thick carbon film stretched over a copper grid at 20 keV and 1 keV. Image recorded on a Hitachi S-800 F E G SEM.

250
10

D.S. Joy and C. S. Joy between 1 and 10 keV. The yield rises for a light element such as carbon or silicon, stays approximately constant for a transition element such as copper, and falls for heavier elements such as gold. At energies around 1 keV the effective backscattering coefficients of all materials are thus approximately in the range 0.20.3. For energies below 1 keV the backscattering yields are not well known, but significant variations with energy are likely because of the effects of Mott scattering (Reimer, 1993; Frank et al., 1994). A consequence of the effects discussed in this section is that the usual relationships between SE and BSE images may be reversed for low voltage operation. At typical SEM energies (20-30 keV) the contrast information in the SE image comes from a region about 4 2 deep beneath the surface, i.e. about 15-20 nm in extent. By comparison the average BSE electron is emerging from a depth of about 0.2 Rs, i.e. perhaps 0.25/zm or more. The SE therefore carry surface information while the BSE can be said to image the bulk of the sample. At low voltages, however, while the depth of information for the SE is still 15-20 nm because the energy and distribution of the SE does not change with incident beam efiergy, the corresponding depth for the backscattered signal of 0.2 Rs may now only be 3-5 nm because of the fall in electron range. Thus at low energies SE are emitted from the whole interaction volume while the BSE only emerge from a thin surface region and it is the SE which perform the task of 'bulk' imaging.

co

~mml [] mm

[]
c~ o

m &

m i

em @

.1
.01

.......................
.1 10 100

Incident Energy (keV)

Fig. 5. Variation with incident energy of secondary electron yield coefficient6 for silver. Data taken from referencesin Joy (1995b). rapidly in the low energy range. In addition the probability of an SE, produced at some depth z beneath the surface, escaping from the specimen is proportional to exp(-z/2) where 2 is an 'escape depth' (Dwyer and Matthew, 1985) and is typically 3-6 nm for most elements. As the range of the incident electron falls (Fig. 2) more of the secondaries generated are able to escape, further increasing 6 the yield of SE which finally reaches a maximum value of the order of unity or greater when the electron range RB is about 2.5 2 (Seiler, 1984). If the incident energy is lowered still further then the yield of SE begins to fall again because of the decrease in the stopping power of the incident electron and the consequent fall in SE production. The corresponding variation in the backscattering yield r/is shown in Fig. 6 for a low, a medium, and a high, atomic number specimen. Although the backscattering yield is almost constant for energies above about 10 keV, it may show significant variations
0.5

HI. CONTRAST FORMATION IN LOW VOLTAGE IMAGES The changes in the nature of the electron-solid interaction with the incident beam energy discussed above affect the formation of contrast in images. The most significant effects are the reduction, and eventual loss, of topographic image contrast and the contrast from beam penetration. Figure 7 shows four secondary electron images of the same area of a sample at 1, 2, 5 and 20 keV, recorded with a standard EverhartThornley SE detector positioned to one side of the specimen. The 20 keV micrograph shows all of the expected features of a typical SE image, strong contrast from the topography of the sample surfaces, edge brightness highlighting the shape of features, and a pronounced three-dimensional appearance which makes the image easy to interpret even for an untrained observer. As the beam voltage is reduced, however, the image becomes 'flatter' and more two dimensional in appearance as the previous contrast is replaced by a simpler directional shading in which one side of a feature is always brighter than the other side. Clearly, interpreting the 1 keV micrograph in the same way as is done for the equivalent higher voltage image would lead to a misleading result. It is also evident that surface contamination marks, and detail such as scratches and

Platinum (Z=78) r III T


0,4
a c

~' o r~ I ~

0.3

[]

i I14 ii

Immilul

III Ill III


Ill
i

ml [] _.

Silicon (Z-&14) ~1 0.1

0,0
I0

I I
100

Energy (keV)
Fig. 6. Variation in the backscattering yield coefficient ~/ for silicon, nickel a n d p l a t i n u m as a function o f the incident energy.

Data taken from referencesin Joy (1995b).

Low Voltage Scanning Electron Microscopy

251

"o "~

2.

//
/11
lkeV
i i

E
Z
0

20

40

60

Tilt Angle (degrees) Fig. 8. Variation of SE yield for Si with angle of incidence at 0.2, 1, and 5 keV computed using Monte Carlo simulation.

Fig. 7. 23SE images of an aluminum interconnect on silicon sample at 1, 3, 5 and 20 keV showing the change in image contrast. Images recorded on an Hitachi S-800 FEG SEM.

dust particles, which are barely visible at 10 keV become dominant contrast producers in the 1 keV image. The improvement in the visibility of surface detail is to be expected because of the rapid drop in the electron range, but the significant change in the form of the image contrast is also associated with the fall in the electron range and the consequent rise in the SE yield. As discussed in the previous section at high energies most of secondary electrons generated within the interaction volume are too far beneath the surface to escape, and so do not contribute to the observed signal. If the sample is tilted, however, then on average more of the secondaries will fall within the escape depth from the surface and the yield of SE will rise (Fig. 8). It is this effect which gives rise to the familiar 'topographic' contrast SE image in which surfaces at a high angle to the beam are brighter than those at lower angles. As the beam energy is reduced then the maximum depth of the interaction volume falls, and a greater fraction of the SE that are produced escape from the surface even at normal incidence. Consequently the increase in SE yield with tilt angle becomes less (e.g. the 1 keV data on Fig. 8), and eventually disappears when all of the interaction volume fails within the SE escape distance (e.g. the 0.2 keV data on Fig. 8). A second effect which contributes to SE image contrast is the penetration of the beam into the sample. At high beam energies this occurs over a scale of many micrometers and features raised above the general surrounding surface are 'filled' by the incident beam

irradiation, leading to enhanced SE emission from all the free surfaces of the feature and enhancing the characteristic three-dimensional appearance associated with SEM images. At low energies the beam penetration is much smaller than the typical dimensions of a feature and in this case only the edges remain bright. A final effect, which also contributes to SE image contrast, is the efficiency with which the SE detector collects the secondaries from the specimen. The angular distribution of SE from a specimen is approximately a cosine law about the surface normal (Seiler, 1984), and numerical calculations of the electrostatic field and corresponding SE trajectories associated with an Everhart-Thornley SE detector under typical conditions show that about 50 % of the emitted SE, essentially all of those with a component of velocity towards the detector, are collected (Bradley and Joy, 1991). A surface tilted towards the SE detector will emit more secondary electrons with a velocity component towards the detector and so will appear brighter than a corresponding area tilted away from the detector. It is this 'collector efficiency contrast', which through the principle of reciprocity is analogous to Lambert's cosine law for reflected light (Goldstein et al., 1992), that generates contrast in low magnification SE images. At high beam energies the topographic contrast, the beam penetration contrast, and the detector efficiency contrast mechanisms all contribute to the image, so a face tilted towards the detector is bright both because it emits more SE and because more of these SE are collected. A face tilted away from the detector still emits more SE than at normal incidence but a smaller fraction of these are collected than for the opposite facing surface. The topographic contrast is thus modulated by the detector contrast effect to produce the characteristic shading that makes the topography of the surface readily interpretable. At the same time surface relief features comparable to, or smaller than, the beam range are filled by the interaction volume producing enhanced SE emissions that contribute to the strongly three-

252

D.S. Joy and C. S, Joy

dimensional appearance. At low voltages, however, all faces emit about the same number of SE, and the beam penetration is small so that features raised on the surface no longer display enhanced SE emission, thus only the detector contrast efficiency effect remains. Topography is now only visible in as much as it places the local surface normal towards, or away from, the detector. The image is thus flatter, and less dramatic, in appearance and the true surface nature of the specimen is harder to discern. Another result is that the form of SE image contrast will now be dependent on the position of the SE detector relative to the specimen. Figure 9 shows two views of the same area of a specimen imaged at 1 keV using (a) an SE detector placed directly above of the sample and (b) a conventional Everhart-Thornley detector mounted to one side of the sample. While the standard detector image illustrates the detector efficiency contrast which provides topographic information by shadowing, the image from the overhead detector contains no directional shading information and is therefore flat and lacking in impact, with no clear sense of three-dimensionality. The differences between these images demonstrates that the effect of varying the detector collection efficiency with respect to any point on the surface is significant, and consequently when interpreting low voltage images an essential first step is to determine the position of the detector relative to the sample.

When performing high resolution microscopy at high beam energies it is necessary to distinguish (von Ardenne, 1940b; Drescher et al., 1970) between secondary electrons generated by the incident electrons (SE1), and those produced by backscattered electrons traveling through back the surface (SE2), since these two components have very different spatial distributions (Fig. 10a). The SE1 signal is localized within a few nanometers of the beam impact position, but the SE2 signal comes from a region about RB/3 in radius around the beam point which (as can be deduced from Fig. 2) is a region hundreds ofnanometers in size. Since, typically, the SE2 signal is two to three times the SE1 signal in total intensity, the high resolution information carried by the SE1 electrons is only a small fraction of the magnitude of the low resolution information carried by the SE2 electrons, and consequently the signal to noise ratio of the desired SE1 component is poor (Joy, 1991). At low voltages, RB falls (Fig. 2), and the emission volume for the SE2 becomes of the same order as that for the SE1 for energies of a few keV giving a spatial distribution of emission as shown in Fig. 10b. In this case both components contribute high resolution contrast detail, thus high resolution microscopy is substantially more efficient at low energies. Although, as discussed in a later section, there are electron-optical constraints on the level of performance that can be obtained, on modern instruments using immersion or field projection ('snorkel') lenses and field emission guns there is little penalty to be paid for reducing the beam energy and thus low voltage microscopy may be the method of choice for high resolution scanning microscopy. As shown in Fig. 11 high quality images, showing the resolution of detail at the nanometer level, are now routinely achievable even at energies as low as 1 keV. This means that, for the first time, the resolution of an SEM is substantially independent of the beam energy. The accelerating voltage of the instrument can therefore be used to select the magnitude of the interaction volume, or to optimize a contrast mode, without paying any penalty in terms of resolution.

~
Beam range Fig. 9. -SE images of the same area of a specimen at 1 keY, imaged with the SE detector at (a) directly above the sample and (b) in the normal take-off position. Images recorded from Hitachi S-4500 FEG SEM. /xm

SE1 (few nm wide)

SE2

"-

High energy Fig. 10. Schematic representation of the spatial distribution of SE1 and SE2 electrons at (a) high energy (20 keV) and Co) low energy (1 keV).

Low Voltage Scanning Electron Microscopy

253

Fig. I1. High resolution SE image from gold decorated magnetic recording tape. Recorded at 1 keV from Hitachi S-4500 FEG SEM. Since the SE2 electrons are generated by backscattered electrons it can also be expected that low voltage BSE images would be capable of the same spatial resolution as SE images, unlike the situation at higher energies where the BSE resolution is markedly worse. Few backscatter detectors sensitive enough to operate efficiently at 2 keV or below are currently in use and so experimental verification is lacking, but the anticipated resolution, together with the enhanced surface sensitivity, the relative freedom from charging artifacts, and the sensitivity of the signal to atomic number ('Z') contrast effects, may make BS imaging the method of choice for high resolution operation at low energies. However it must be noted that, unlike the situation at high beam energies where the backscattering coefficient is usually higher than the secondary yield coefficient, at low energies the magnitude of the backscattered signal is much lower than that of the secondary signal (e.g. Figs 5 and 6). Consequently a degradation of the signal to noise ratio is inevitable even before the relatively poorer geometric collection efficiency of BSE relative to SE detectors is taken into account. The original goal of the SEM was to produce contrast between materials based on the differences in secondary electron yield which were known to occur (Matthes, 1942). In practice 'true' SE contrast was never achieved because the SE yield expected from a clean surface is usually modified by the presence of a film of surface contamination (Bruining and de Boer, 1938). However, recent work on low voltage SEMs employing 'through-the-lens' SE detectors (Kimura and Tamura, 1967) has shown what is evidently chemical SE contrast i.e. contrast that does not result from the rise in SE2 yield caused by an increase in the mean atomic number of the specimen. For example, an examination of the images in Fig. 9 shows that there are interesting differences between the micrograph recorded on the upper ('through-the-lens') and the lower detector. The region labeled 's' is dark in Fig. 9a but the same level as the background in Fig. 9b. Similarly, dark dots are visible in region 'd' of Fig. 9a, but invisible in Fig. 9b. Most dramatic is the

region labeled 'r', which is black in Fig. 9a but is the same contrast as its neighbors in Fig. 9b. The most controlled examples of these effects have been found in doped semiconductors. Figure 12, provided through the courtesy of Prof. D. Perovic, shows the SE image of p- and n-doped multi-layers in a GaAs sample (Bleloch et aL, 1994; Perovic et al., 1994). The p- and n-doped regions appear bright and dark respectively compared to background. Since the doping corresponds to only about a 1 ppm change in composition, the contrast cannot be due to variations in the backscattering coefficient resulting from changes in the mean atomic number (Venables and Maher, 1994). Similar effects have also been observed on quantum well structures (Krause and Joy, unpublished), and in specimens of SiC containing impurities such as a few parts per million of aluminum (Ness and Page, 1986). Although these effects may also be visible at high energies they are most readily observed at low energies where the SE yield is also maximized. The visibility of this 'true' SE contrast in the low voltage SEM appears to rely on a number of factors. Firstly, when secondary electrons are collected through the lens the combination of the magnetic field of the lens and the electrostatic extraction field of the detector acts as an energy filter. For example, on the Hitachi S-4500, used for several of the observations cited above, measurements show that the 'through the lens' detector has a response which is sharply peaked at about 4 eV, and is essentially zero for SE energies greater than about 10 eV. The peak of the emitted SE energy distribution occurs at an energy which depends on the work function of the material and is typically 3-5 eV (Chung and Everhart, 1974). Changes in the doping of a semiconductor also change the work function (Beadle et al., 1985) and will thus vary the position of this peak relative to the energy response of the detector so producing contrast variations. Perovic et al. (1994) have also suggested that band pinning effects associated with the

Fig. 12. SE contrast from p- and n-dopedmulti-layerregions in GaAs recorded on a Hitachi S-4500 FEG SEM. Micrograph courtesy Dr D. Perovio,Universityof Toronto.

254

D.S. Joy and C. S. Joy

chemisorption of oxygen at the surface may actually enhance the yield variation between p- and n-doped regions. The fact that this type of contrast is visible even on less than ideally clean samples is also probably due to the energy selecting nature of these detectors. As can be deduced from Fig. 1 the inelastic mean free path for secondary emission reaches many nanometers at low energies (i.e. below 20 eV or so), although the elastic mean free path remains pinned at about 1-2 nm. Therefore, provided that the surface contamination layer is not too thick, as should be the case for a modem dry-pumped LVSEM system, SE generated at the real specimen surface in the critical energy range around 4 eV will have a high probability of diffusing through any surface layer with little change in either their energy distribution or intensity. More conventional SE detectors collect their secondary electrons over a much wider energy range, including those higher energy SE which have much shorter MFPs, and consequently more of the signal comes from the surface layer rather than from the underlying specimen. Finally, throughthe-lens SE detectors are highly efficient collectors for the SE emitted from the specimen but do not collect the spurious SE3 and SE4 secondaries produced by the impact of backscattered electrons on the polepiece, chamber walls, etc. of the SEM. The visibility of contrast effects from the sample is therefore not diluted by any extraneous background. In summary, image contrast formation in the low voltage SEM is in many respects different to that of the conventional SEM. More of the image information now comes from closer to the surface and so may be affected by the presence of contamination and surface defects. On the other hand the low voltage image is less sensitive to the surface topography and its form is dependent on the orientation of the detector to the region being observed. Because of the fall in the size of the interaction volume all of the SE generated in the sample, whether by incident or backscattered electrons, are capable of carrying high resolution information giving a useful improvement in the contrast of such images. Through the lens SE detectors are well suited for low voltage operation, because they efficiently discriminate against stray, scattered, and higher energy secondaries, and because they can respond to changes in the energy distribution and number of secondary electrons resuiting from changes in the chemistry of the sample so providing a chemical SE contrast mechanism.

beam, or can even damage the specimen. The control, or elimination, of charging is thus an important goal of scanning microscopy. If the specimen is an electrical conductor and is connected to ground, then a current balance equation can be written in the form
IB = 61n + vv ln + Isc

(2)

where 1B is the incident beam current, 6 is the SE yield, r/ is the BSE yield, and l s c is the specimen current flowing between the sample and ground. If the path to ground is now broken, or if the specimen is not electrically conducting, then Isc is zero and the excess current causes a charge buildup AQ in the specimen at a rate AQ/second = ls(1 - (6 + v)) (3)

IV. THE OBSERVATION OF NON-CONDUCTING SAMPLES AT L O W VOLTAGE A. Introduction

A specimen becomes electrically charged when it is receiving more electrons than it is emitting, or emitting more than it is receiving. The presence of this charge generates an electric field which may interfere with the collection of secondary eleotrons, deflect the incident

At high beam energies (e.g. see Figs 5 and 6) the total electron yield (6 + r/) is less than unity, so from eqn. (3) charge is being injected into the specimen which therefore acquires a negative charge. This negative charge produces several effects. Firstly it repels the SE from the surface so raising the yield. Secondly the landing energy of the incident electrons is reduced which again leads to an enhancement of the SE yield. Finally the field between the surface and the SE detector is increased so improving the detector collection efficiency. Thus, for all of these reasons, negatively charged regions appear bright in the SE image. At low energies (6 + r/) may become greater than unity, so that charge is being removed from the specimen which then acquires a positive charge (since initially it was electrically neutral). Positively charged areas can re-collect their own SE emission, and so they appear dark in the SE image. As a special case if (6 + ~/) equals unity then there is a dynamic charge balance and no net charging occurs even though the sample is not conducting. This will clearly be the optimum choice of operating energy in most cases. In general, because of the rise and subsequent fall of SE production with accelerating voltage discussed earlier, there are two incident beam energies at which this condition might be met, a lower value El--typically between 50 and 150 eV--and a higher value E2 which is typically between 0.5 and 3 keV (Frank et al., 1994). The fact that E2 is in the few keV range for materials of technological importance such as semiconductors, polymers, and ceramics, has been an important factor in the development of low voltage SEM techniques. The value of E2 can be deduced from a knowledge of the variation of secondary and backscattered yields with beam energy (e.g. Joy, 1995b) and Fig. 13 show such values for a variety of elements. Although the data is scattered it is evident that the value of E2 rises with the atomic number of the target, from below 1 keV for the lightest elements to as much as 10 keV for the heaviest elements. A regression fit suggests that E2~0.12*Z where Z is the atomic number of the sample. E2 also seems to vary with the density of the target, being lower for low densities and higher for higher densities, but the

Low Voltage Scanning Electron Microscopy

255

available data is too sparse to have much predictive value. In practice, however, most samples do not consist of a single element so the data of Fig. 13 is of limited value and the need is to be able to determine E2 experimentally for the specimen of interest. This can be done using the following procedure, illustrated in Fig. 14. (1) Set the SEM to a fast scan rate and to a magnification of about x 100. (2) Wait for five seconds to allow the image to stabilize. (3) Raise the magnification to x 1000 and wait for five seconds. (4) Drop the magnification back to x 100 and observe the scan square on the viewing screen. If the area scanned at high magnification appears bright in the low magnification image (Fig. 14c) then the specimen is charging negative and E0 is higher than E2. If the area scanned at high magnification appears dark in the low magnification image (Fig. 14d) then the specimen is charging positive and the incident beam energy E0 is lower than E2. If no scan square is visible, or if it disappears very quickly, then E0 is close in value to E2. Because the incident energy E0 can be adjusted in quite small steps on modem SEMs, E2 can quickly be determined to an accuracy of +100 eV or so for any given sample using this technique. Tables 1 and 2 give some typical E2 values for materials used in semiconductor device fabrication, and for common polymers, determined using this technique. These values range, without any clear trend, between 0.5 and about 3 keV for most materials although for polymers there is evidence (Butler et al., 1992) that E2 depends systematically on the electronic state of the original monomer, as well as on the temperature of the polymer. If more than one material is present then the charge balance must be computed for each component of the system. For example, in the case of a poorly conducting thin film on a substrate if the film is thin enough for the
14
12 []

II

Increase magnification

(a) Scan area at low magnification :#:i~~':'i~@~:::~@~i "':~i~::~#@~i '''''"

(b) Count to 5 then go back to original magnification

/ I " '"' ~ (d) Dark scan square sample charging positive (c) Bright scan square sample charging negative

Fig, 14, Procedure for the determination of E2 using variable magnification scanning.

10
m

[]

beam to pass through it then in addition to the SE and BSE yields from the entrance (top) surface of the film, there is a transmitted signal passing out through the bottom surface. There will also be backscattered and secondary electrons from the substrate coming back into the film from the substrate beneath. When the incident energy is high enough that the film is only a small fraction of the beam range in thickness then the transmitted signal is almost equal to the incident signal, and when the SE and BSE yield are included it is seen that the film is emitting more electrons than it is receiving and so will charge positively. At a lower voltage, where the film transmits less of the incident beam the net emission will again become less than unity and the film will charge negatively. At still lower voltages the film acts like a bulk sample and its BSE and SE yields will again rise to give a yield greater than one and the film will charge positively once more. In this case there are three energies El, E2 and E3 at which the film is in charge balance. Depending on the nature and thickness of the film, and on the composition of the substrate, E3 can be from 0.5 to 25 keV. For example, surface contamination layers of hydrocarbons on a typical substrate have an E3 value of around 1 keV and will charge positively at usual low voltage SEM energies. The resultant darkening of the SE image is often mistakenly identified as contamination build up, but will disappear if the beam is removed or if the energy is raised or lowered.

10

20

30

40

510

~ 60

70

80

90

B. Procedures to minimize charging

Atomic Number Z

Fig. 13, Variation of E2 energy with atomic number Z for pure elements. The solid line is a least squares regression fit, giving E2~0.12 Z. Data from Joy (1995b).

The problem of observing poorly conductive, or insulating, specimens can be made easier by following a systematic procedure, outlined in Fig. 15. Each step involves some trade-off in performance or convenience,

256

D.S. Joy and C. S. Joy Table 1. E2 Data, semiconductors and other inorganics Semiconductors Compound Low density resist Resist on Cr substrate Resist on oxide Resist on poly-Si PMMA resist Other inorganics Compound NaCI KCI LiF E2 (keV) 0.55 0.70 0.90 1.10 1.6 E2 (keV) 2.0 1.6 1.9 Compound Cr on glass Glass passivation SiO2 (quartz) Alumina Al203 High res. GaAs Compound Pyrex glass CaF2(fluorite) E2 (keY) 2.0 2.0 3.0 2.9 2.6 E2 (keV) 1.9 1.9

so the intent is to use the least objectionable strategies first. The initial step is to decide on the lowest 'acceptable' operating voltage, Emin, of the microscope to be used. In the case of a modern, field emission gun SEM this would usually be the lowest voltage available; for an older machine it may be higher because of inadequacies in the gun, or lenses, or shielding, of the machine and the judgment will depend on the level of resolution that is being required of the system. Observation of an unknown sample is always commenced at this lowest energy so that, even if charging does occur, the charge will not be deposited deep in the sample but remain close to the surface. The charging state of the specimen is now checked using the procedure of Fig. 14. If the sample is charging positive then Emin < E2 for this specimen and stable imaging will be possible by increasing the beam energy in small steps until E2 is reached. If the sample is charging negatively then Emi, is already higher than E2. Since the SEM energy cannot be lowered it is now necessary, instead, to try and raise E2 to match Emin. It has been demonstrated experimentally (Sugiyama et al., 1986; Reimer, 1993), and shown by Monte Carlo simulations (Joy, 1987) that if the E2 energy for a given specimen at normal beam incidence is E2(0), then when the angle of incidence is the corresponding E2 value, E2(), is approximately

m(0)
E2() ~ cos" (4) where the exponent n is typically 2 for low atomic number materials such as polymers and between 1 and 1.5 for heavier materials (Reimer, 1993) (Note that an implication of this result is that on a surface with significant topographic detail even though one region of

the sample may be compensated, other areas at different angles of incidence to the beam may simultaneously show either positive or negative charging). This result is used in step two in the decision table. Here the sample is tilted to 45 degrees or so, which will increase E2(0) by a factor of between 50 and 100 %. The state of the surface charge is then examined again by the procedure of Fig. 14. If the area is charging positively then the tilt can be reduced to make E2()= E~in, and the sample is ready for observation. If, however, the sample is still charging negatively then E2() is again less than EminAlthough, in principle, the tilt could be further increased to give a bigger value of E2() the foreshortening of the image and the limited depth of field usually make this an undesirable option. At this point it is therefore necessary to accept that the specimen will continue to be negatively charging, and the aim must therefore be to minimize the effects resulting from this. Understanding how to do this firstly requires a knowledge of what happens when a specimen charges negatively. The simple view (e.g. Thornton, 1960) is as follows; when a surface charges negatively to some potential Es,rra~ it reduces the effective landing energy of the electrons from E0 to E0--Esurfac~, this in turn increases the net yield of electrons, and so the surface continues to float up in energy until Eo--E~urra~ = E2 when the surface potential will stabilize. In practice this is not what is found to happen. Figure 16 shows how E~,rr~~ varies for alumina as a function of the incident beam energy. This measurement was made by observing the high energy cut-off ('DuaneHunt' limit) of the X-ray bremsstrahlung emitted from the sample, since no X-ray can have an energy exceeding the landing energy of an electron. Below an energy E of about 2.8 keV Esurface is zero, so the E2 value is also

Table 2. E2 data, polymers Compound Polypropylene copolymer Polypropylene Polystyrene Polyethylene Polyethylene acrylic acid Polyaniline Polycaprolactone Polysulfone E2 (keV) 0.9 1.5 1.3 1.5 1.04 0.95 1.14 1.1 Compound Acrylamide Acetal Kapton Nylon Nylon/5% PBT PVC PTFE Teflon E2 (keV) 1.3 1.65 0.4 1.2 1.4 1.65 1.8 1.8

Low Voltage Scanning Electron Microscopy

257

Set SEM to lowest usable operating energy

Test charge state of sample

Negative charging energy too high

Positive charging - increase keV ready to operate

Tilt the specimen to 45 degrees

Test charge state sample

Negative charging

Positive charging - reduce tilt ready to operate

Cannot reach E2 - sample will be charging


. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Reduce the incident beam current

Stable image - ready to operate

Reduce themagnification

Stable image - ready to operate

Go to high scan rate

Stable image - ready to operate

..................
. . . . . . . . . . . . . . . . . . .

I ...................
. . . . . . . . . . . . . . . . . .

Sample cannot be stabilized

Coat the specimen Surface modifications Heat the sample Fig. 15. Decision table for observing charging specimens.

2.8 keV. (This is because positive charging is inherently a self-limiting procedure since by recollecting the secondary electrons that it is emitting a surface can always stabilize its potential to bring it to within a few volts of E2.) As E is increased beyond E2,//surface rises quite rapidly reaching 1 keV for a nominal incident energy of 5 keV, and about 3 keV for an incident energy of 10 keV. It is this high surface potential which produces the electrostatic fields which deflect and distort the incident beam and the emitted electrons. Figure 16 shows that when the sample is tilted, E2 shifts upwards in value as discussed earlier. It can also be seen that the surface potential of the tilted sample is always less than for the corresponding case at normal incidence, but the shape of the charging profile, and the magnitude of Es~rrao~, remains about the same. Because of the limited penetration of the electrons at these energies, the surface potential Esurrace is separated

by a distance of only a fraction of a micrometer from the rest of the sample which is at earth potential. The resultant field may thus easily exceed 106 volts/cm leading to dielectric breakdown and permanent damage in the sample (Cazaux, 1986). Associated with this field is a leakage current IL which can be written as Esurraee/R, where R is an effective leakage resistance to ground through the sample. If the potentials on the sample are not constant, then there is also a current required to charge the capacitor represented by the specimen. Incorporating these additional terms into the current balance eqn. (2) gives.

IB = (6 + rl)E,.ls + ~

Esurfaee + C ~Esurface

6t

(5)

where the first term represents, as before, the re-emission of electrons from the sample by secondary and backscattered processes at a rate determined by the value of

258 i j

D.S. Joy and C. S. Joy

Alumina~

01o..o!
j1

.~
.~

/
/ /
I /~*j /

/
45 degrees, lnA

==
=.,
o J O

2
r.,/3

/
1 0 degrees, ~.3nA

/i J-/"
0 5
0 15 20

Incident Energy (keV) Fig. 16. Experimental measurements of E~f~e vs beam energy for alumina at zero and 45C, angles of beam incidence at 1 nA incident current; and at zero incidence for 0.3 nA current.

(6 + r/) at the effective incident beam energy E', where E' = E0-- Esu~fa~. The second term represents the leakage current across the sample to earth, and the third term is the charging current flowing into the capacitor C from the virtual electrode formed by the incident beam. As shown by Cazaux (1986), the time taken for the specimen to fully charge up is usually of the order of microseconds or less, a period which is small compared with the typical dwell time of the electron beam on the surface of the specimen. Thus the last term in eqn. (1) can be neglected without any loss in accuracy for all cases except that of TV rate scanning. We can then solve for the surface potential Esurraee to get:
esurfaee "~- IsR(1 - (fl + r/))z,

(6)

In order to explicitly calculate Esurra~ we require a value for 1 - (6 + r/)e,, To a good first approximation the variation of (1 - (fl + r/)g, against E is linear so giving the relation:
1 -

(6 + q)e, = 1 - (6 + ~/)e0E0 -EoE=urfa=- E2 E2-

(7)

This value can now be substituted into eqn. (6) to give an estimate for E=nrface.

IsRfo(Eo - E2)
E=~fa~ = ( ( e 0 -- e 2 ) + IBRf0)

(8)

where fo = 1 - (6 + r/)g0. Notice that this expression involves not only E0 and E2, but also l s and the leakage resistance R. Fitting the data shown in Fig. 16 gives R ~1013f~ which is consistent with the bulk resistivity of alumina. Note that only in the case when IsRfo > >(E0--E2) e.g. for a high quality insulator does Es become approximately equal to (E0--E2) as predicted by Thornton (1960). In every other case Esurface will be lower so that the effective landing energy of the electrons E' will always be greater than E2. Since

E2 varies with the angle of incidence, and hence with surface topography, Esurrae e from eqn. (8) will also vary from point to point which is why complicated surfaces can exhibit such challenging problems with charging. Since the surface potential Esurra~ depends on I , the next step of the decision table therefore requires reducing the beam current to minimize the charging. As shown in Fig. 16, or as can be deduced from eqn. (8), reducing the incident beam current by a factor of 3 from 1 nA to 0.3 nA, reduces the surface potential Esurfaee by a factor of 3 to 4 times, for example reducing Esurrace at 20 keV nominal beam energy from 3.75 kV to 1.2 kV. A further factor of 3 x drop in the incident beam current would further reduce Esurface at 20 keV to about 300 eV, a value which would cause little or no problem for either imaging or analysis. Thus, when E0 must be chosen to be higher than E2, careful control of the incident beam current can often minimize, or eliminate, charging problems. Experimentally the current can conveniently be reduced by using a smaller final aperture, or a smaller spot size, with Is being reduced until charging is no longer a problem or until the signal to noise ratio, or count rate, become unacceptably low. If charging is still a problem then the next step calls for reducing the magnification of the image. None of the preceding analysis indicates that this should be the case, since changing the magnification of the image does not alter lB. Nevertheless, experimentally it can readily be demonstrated that changing the magnification at which is area is being examined very often alters the magnitude of the charging, giving an effect which is usually called 'dynamic charging'. It is, in fact, only because of dynamic effects that the charging test illustrated in Fig. 14 works at all. Figure 17 shows how the surface potentials of Teflon and alumina change as a function of the incident current density. The measurements were made by keeping the beam current constant but varying the size of the scanned area by the magnification. At 5 keV on alumina the surface potential rises steadily as the incident current density is increased, rising by about 200 volts for a current density increase of 100x. A further increase of x 10 in the current density raises the surface potential by a further 200 volts, but increasing the density still more has no significant effect and the potential stabilizes. The behavior of the Teflon at 5 keV is also quite similar to that for alumina although the absolute rise in potential is smaller. For the 10 keV Teflon case, however, the behavior is quite different since, within experimental error, the magnitude of the charge-up is constant and independent of the applied current density over the whole experimental range. An analysis of the data shown here (Joy and Joy, 1995), and other comparable results, demonstrates that the important variable seems to be the ratio between the nominal incident beam energy E0 and the E2 energy for the specimen. Here the E2 energy for alumina is 2.9 keV, while the E2 for Teflon is 1.8 keV. When E0 is equal to E2 then no charging occurs, but when Eo is above E2 then both static and dynamic charging will appear. If E0 is close to E2, as is the case for the 5 keV irradiation of

Low Voltage Scanning Electron Microscopy 10

259

When a small area is scanned inside a previously irradiated larger area then the average exit path length between the beam point and the rest of the material will Teflon 10keV be larger, and hence the leakage resistance will increase. An alternative hypothesis is that placing the beam inside Teflon 5keV ;> a region that has already acquired a negative charge "d surrounds the irradiated area with a potential 'wall' which, being negative in sign, will pinch-off current flow :I through it. As shown by Cazaux (1986) every charged region contains within itself a point where the field is g zero. All the other charge field vectors are then directed r~ towards this point so producing a radial potential barrier. In the case of a scanned area whose size is large compared with the incident beam range, symmetry suggests that the zero field point will be located at the center of the scan field, so that successively smaller scan 10 100 1000 10000 Relative Beam Dose (amp/em2) areas will encounter increasingly high barriers to current flow. At beam energies that are large compared to E2 Fig. 17. Experimental surface potential measurements under the field will reach the value determined by the dielectric dynamic charging conditions for alumina at 5 keV, and teflon breakdown strength of the specimen, and so further at 5 and 10 keV nominal beam energy, as a function of relative current density. irradiation will not be able to increase the potential barrier any higher. In summary, it is clear that since alumina (E0~l.5xE2), then the magnitude of the dynamic charging is most evident when the sample has dynamic charging effect is high and even a small surface been set up to the best available condition (i.e. the beam charge can be magnified by increasing the apparent energy close to but somewhat higher than E2) the current density. This effect is readily evident when solution is to keep the magnification as low as is examining a specimen at an energy close to its E2 value. consistent with achieving the necessary pixel resolution Although most of the specimen is at charge balance in the image. local variations in E2 resulting, for example, from If, after working through all the other steps shown in topography or changes in composition produce areas Fig. 15, the sample is still charging to an unacceptable that flare brightly as the magnification is increased degree then the final experimental procedure which is because of the dynamic charging that occurs in these available is to change the scan rate used to acquire the isolated regions. As E0 is raised higher relative to E2 images. When the electron beam strikes the first pixel in then dynamic charging persists, displaying the same the field of view then that region charges up to its character as before, but the magnitude of the differential maximum surface potential, in a period which can range charging is reduced. This is the situation for the 5 keV from nanoseconds to microseconds (Cazaux, 1986), and observation of Teflon for which E0~2.5xE2. At still maintains that potential as long as the beam stays in higher energies, for example for Teflon at 10 keV where place. As soon as the beam moves to the next pixel, the E0~5xE2, dynamic charging is absent and the charge first pixel region begins to discharge, a process which is up of the sample follows the static curve discussed typically hundreds of times slower than the correabove. sponding charge-up. The scanning beam therefore leaves Since all of the other terms in eqn. (8) are constant it is behind it a trail of pixels each of which is slowly clear that dynamic charging must be the result of a discharging in potential. The combined effect of these change in the leakage resistance R. An examination of charges is an electrostatic field, continually varying in this equation shows that increasing R will raise Esurrace both magnitude and direction, which disturbs both the but the effect will become less as E0 is increased further incident beam and the collected SE signal. Therefore, above E2 implying that the surface potential saturates as the difficulty is not that the specimen charges-up, but observed experimentally. The remaining question is that it subsequently discharges. This problem can be therefore to determine how the leakage resistance R overcome by scanning the field of view at such a high can be altered by the beam irradiation conditions. The rate that the beam returns to any given point before that simplest explanation for the observed effect is purely pixel has had a chance to discharge itself significantly. In geometric. The leakage resistance R of some scanned this case the whole viewed area will be uniformly area of the sample will depend on the resistivity p of the charged up, and hence there will be no electrostatic fields material (as modified by the effects of beam induced across it to deflect the beam or the secondary electrons. conductivity and field enhancement) and on the path A stable image can therefore be produced, although length L to ground, and the cross-sectional area A of the refocusing and stigmation may be required because the irradiated volume. The cross-sectional area is propor- landing energy of the incident beam will be different to tional to the depth of beam penetration, while the that on an uncharged region of the sample. For most effective path length depends on the size of the scanned specimens a scan rate of a few frames per second or region. more (i.e. fast visual rate or TV rate) is adequate to
o R ~II -0 . . . . . . . . i . . . . . . . i . . . . . . . . i . . . . . . .

260

D.S. Joy and C. S. Joy

achieve the desired effect. The drawback to this approach is the inevitable reduction in signal to noise ratio. However, on modem SEMs equipped with digital framestores the high speed scan can be integrated for a sufficient period to give a quality image. In such cases the use of rapid scanning is one of the most useful tools to observe charging specimens. It should be noted, however, that the linear distortion of a high speed scan can approach 15--20 %, and so slower scan speeds are always preferable when the most accurate representation of the sample is required. The decision table discussed here makes it possible to obtain quality micrographs from almost all samples, under low voltage conditions, without the need to coat the specimen. Although metal coatings are an essential tool for building contrast in high resolution microscopy (Joy, 1991), for low voltage microscopy even the thinnest film obscures the surface to some degree and so coating is to be avoided if possible. If a coating is to be used then it must be chosen carefully since the task of a metal coating is twofold; firstly it forms a local ground plane, or Faraday screen, at the surface, eliminating fields from any charge stored within the sample; secondly, the additional SE and BSE production at the film reduce the amount of charge retained in the specimen. In light of these criteria it can be seen that, although widely used, carbon coatings--which have poor conductivity and low SE and BSE yields--are not a good choice for low voltage operation. The ideal coating is a thin film of a material such as Cr or W which has a high SE yield, good conductivity, and a low surface energy to ensure a continuous, structureless, film even at nanometer thicknesses. The deposition of such a film requires special equipment and careful attention to detail (Peters, 1980). Other approaches to actively overcoming charging are also possible, although they will not be discussed here in detail. For example, warming a glass or ceramic sample to 50C often generates sufficient surface leakage to eliminate surface potentials; light plasma etching of polymer surfaces produces a nanometer scale roughness which enhances SE yields and again minimizes charging; and a low pressure (10 -6 torr) flow of an inert gas over the sample has also been shown very effective in eliminating charging (Ohlendorf et al., 1991). In summary, although much of the interest in low voltage SEM was motivated by the hope that charging would be eliminated by the use of lower incident energies, in practice charging remains a constant problem. By using a systematic approach, and the capabilities of modem SEMs, the unwanted effects of charging can, however, be minimized. It should also be noted that local charging can often be a useful source of image contrast--although very dependent on beam current and scan rate---and that, under controlled conditions charging can be used to enhance specific contrast mechanisms (e.g. Borchert et aL, 1991; Harker et al., 1994).

V. RADIATION DAMAGE AT LOW BEAM

ENERGIES The impression is sometimes given that low energy electrons are less damaging to a sample than higher energy electrons. That this is not the case is evident from Fig. 1 which shows that as the energy of the incident electron is reduced it transfers increasing amounts of energy to the sample. Damage in biological and polymeric materials occurs through the breaking of bonds within the sample, and the energy required for this is only of the order of a few eV, much lower than the energy of any normal incident electron. A low energy electron beam thus actually damages a sample more rapidly than a higher energy beam because the energy deposited per unit volume in the sample is greater. The magnitude of this effect can be demonstrated by a simple calculation. The range R of an electron of energy E0 varies as about E d6. If the interaction volume is approximated as a cube of diameter R, then the interaction volume goes as about E05. Since the total energy to be deposited is proportional to E0 the energy deposited per unit interaction volume varies as 1 / E 4. A 1 keV beam would thus be 10,000x more potent a source of damage than a I0 keV beam. A more optimistic estimate can be made by assuming that the pixels in the scanned area overlap at all beam energies. In this case the effective interaction volume varies as the change in the depth of penetration and the energy deposition would increase as 1/E 1"6. The fact that either of these results seems to be at variance with conventional wisdom is explained by the fact that, at low voltage, the damage is confined to the top few nanometers of the sample while at energies of 10 keV or higher the damage---although less in magnitude--is spread through a volume a micrometer in radius of more of the sample. Artifacts such as shrinkage are thus much more significant and noticeable at high, rather than low, energies. Other effects may also contribute; for example, charging at high energies produces electrostatic fields of 106V/cm or more which are high enough to cause dielectric breakdown (Cazaux, 1986), mechanical deformation, and changes in the chemistry of the sample (Graham et al., 1984). Even though it is more limited in physical extent, low voltage damage may still have significant effects. For example, in inorganic and dielectric materials, such as the gate oxides used in the semiconductor industry, it is found that while the electrical defects induced throughout the volume of the sample by high energy electrons are readily removed by low temperature annealing, both heating and optical bleaching may be required to remove the damage caused by low energy irradiation. If the LVSEM could be operated at a low enough energy, possibly 25 eV or less, then damage might truly be eliminated completely because there are no high cross-section inelastic scattering events that can result in the transfer of sufficient energy to the sample to break bonds. There is no direct experimental evidence at present to support this hypothesis although observa-

Low Voltage Scanning Electron Microscopy

261

tions in transmission mode of purple membrane at 70 eV energy have shown greatly reduced dose sensitivity (Spence et al., 1994). Instrumentation capable of generating SEM images in this energy range has been described (Frank et al., 1994) and work is currently in progress to build and exploit such devices. In summary, for all except the very lowest energies, low voltage scanning microscopy should be recognized as having the advantage of limiting the physical extent of radiation damage rather than of reducing its magnitude.

depends upon the source brightness r, with the highest value being the most desirable. However, eqn. (9)eqn. (10), although widely used for this purpose, cannot accurately be used to make predictions of the probe forming performance at low beam energies. This is partly because, at low energies, the effects of chromatic aberration cannot be neglected. Including the chromatic term in eqn. (9) produces an expression which is less easy to optimize analytically, although a solution is obtainable (Shao and Crewe, 1989). However, the result is still only of limited value because the effect of spherical and chromatic aberration is to distort the profile of the beam and thus what is required is some VI. THE ELECTRON OPTICS OF THE LOW information about the shape of the beam rather than VOLTAGE SEM just a single number representing its effective size. eqn. (10) is also misleading in that it assigns fixed maximum Without an adequate electron-optical system none of values to and AE although the values experienced by the benefits of low voltage operation discussed above any particular electron will be less than these. To gain can be obtained. The quality of the image depends on both the size of the probe size (which determines the quantitative information about the focussed beam it is necessary to use a numerical ray tracing program which resolution) and on the current that the beam contains can predict probe current profiles on the basis of (which determines the signal to noise ratio). These requirements are the same for both high and low voltage specified energy and angular distributions for a given microscopes, but the factors which limit performance set of lens parameters. One such program is UPDENS are different in the two cases. The final probe size d at (Sato and Orloff, 1991) while another, less rigorous, the sample can be approximated (Goldstein et aL, 1992) approach is the Monte Carlo based DOCALC program (Kenway and Cliff, 1984; Jia and Joy, 1992). as Figure 18 shows typical results of numerical ray tracing simulations using DOCALC. The plots show the d= [ 4Ip /'1 3~2 /,0.612,~2 landing positions, in the plane of optimum focus (i.e. the disc of least confusion), of 250 electrons. The calcula1/2 tions have been made for an LaB6 thermionic electron + defocus terms etc.] source with an energy spread of 1.5 eV, and for a cold (9) field emission source with an energy spread of 0.4 eV, at 30 keV, 5 keV, and 1 keV beam energies. In all four where 2 is the electron wavelength, Cs and Cc are cases the same probe forming lens has been assumed, respectively the spherical and chromatic aberration and the lens parameters were taken to be: focal length of coefficients of the probe forming lens, ~t is the semi2.3 mm, Cs of 1.9 mm, and a Cc of 2.5 mm (essentially angle of the beam convergence at the specimen, E0 is the those of the objective lens in the Hitachi $900 high beam energy and AE is the energy spread of the resolution SEM). The lens is demagnifying an effective electrons. Ij, is the current in the probe, and fl is the electron source size of 5 nm by a factor of 10x, and the brightness of the electron source. At high beam energies convergence semi-angle of the incident beam at the (above 15 keV) it is usual to assume that the effective of specimen is 7 milli-rads. At 30 keV for both sources the chromatic aberration is insignificant, and that astigmaprobe is concentrated tightly around the optic axis and tism, defocus errors and other image defects can be numerical integrations show that the profile of the eliminated. Under these conditions the relationship current distributions in the probe are essentially between the probe size d and the incident beam current Gaussian. By 5 keV, however, the probes are more Ip has the form (Goldstein et al., 1992) diffuse and show much less tendency for electrons at land on, or even close to, the optic axis, and by 1 keV it d= KC~1423/4[I,,~22+ 1] 3Is (1O) is difficult in either case even to discern the optic axis position. This degradation in probe forming ability where K is a constant of order unity, for an optimum comes about because of the effect of chromatic aperture angle ct equal to (d/C~)1/3. These relationships aberration which transfers electrons from the axis to predict that, for a given beam current I~, the probe size d the edge of the probe. Since the energy spread of the 1/2 of the SEM will vary as about 1/E~ because of the 2 3/4 LaB6 source is a factor of three times higher than that of term in eqn. (2). (Note that, since the electron optical the FEG the probe profile is correspondingly worse in brightness increases linearly with energy, the term fl~,2 in that case. The full width at half maximum (FWHM) of eqn. (2) is independent of the accelerating voltage). The the FEG and LaB6 probes at 1 keV differ by only 50 %, resolution therefore becomes worse and the available but while more than 70 % of the current in the FEG probe current falls as the beam energy is reduced. The probe lies within the F W H M width, for the LaB 6 probe extent to which the performance degrades with falling E the corresponding factor is only 40 %. The rest of the

(CcctA 2

262

D.S. Joy and C. S. Joy (Mulvey and Newman, 1973) which, because it produces a strong magnetic flux outside of, and below, the lens combines both small aberration coefficients and unrestricted sample access. It seems probable that further improvements in performance can be anticipated. For example, field emission 'nano-tips' (Fink, 1986; Scheinfein e t al., 1993) could offer a factor of x 100 improvement in brightness at low energy as well as a smaller source size. Further, because these tips only require an extraction potential of 30-70 volts it would always be possible to operate the gun in an accelerating mode, so simplifying the optical design. The susceptibility of low energy beams to external electro-magnetic interference can be reduced by having the sample immersed in the lens field since when the sample is inside a field of several hundred Gauss of more the effect of a few milli-Gauss of external disturbance is negligible. This has the happy result that immersion and snorkel lenses can simultaneously offer both a higher level of performance and less sensitivity to the external environment. The use of high Tc superconductor materials to provide Meissner shielding for the column, could still further reduce the problem of field sensitivity. Finally the application to the SEM specimen stage of some of the principles of design developed for scanning probe microscopes (such as the STM and AFM) should eliminate the present restrictions on performance caused by drift and vibration. The LVSEM will then be able to fully demonstrate its status as the high resolution imaging instrument of choice.

.;:~,'

.:

;:,";r.,, ,~':..

30 keV

.. '. " : ., '..Y~ ~. ~- . : .': .'.'..'...

.' '..",i '

".::',~ i ,"..~:: . .. .'.:::~j~,~.'.. .: ,.:',':.~~i;~ .~.


. ::..::1:.~ :%.. .

"' ' "~::~F':""':" " ' .......:, ~ =.'.'.... :

, ":.....'.~; q..

~ ."."

5 keV

.. ..;.....
. . '." . :.
'

i .
"

'
1 keV

:. .. :.... :.;:~.':.:.;.."
. . . . .... 'A' ", ,, :.~w...*....'

~ ::,ii :::
.

, ..5"i;:" "..~'.. 'I".'Y


.. , . ,:

FEG
1 nm

LaB6

Fig. 18. Result of numerical ray tracings for an electron-optical column employingeither a LaBs or a FEG source, and using the same high performanceimmersionlens, at 1 keV, 5 keV and 30 keV, showing the landing position of the first 250 electrons in the probe at the plane of the disc of least confusion. current is contained in the 'skirt' around the beam which extends outwards for a radius o f 10 nm or more. Thus while the nominal F W H M resolution of both the LaB6 and F E G systems is about the same at low energies the transfer of current from the center to the rim of the probe leads to a reduction in high resolution image contrast for the LaB6 system as well as to a loss of spatial resolution when the probe is used for X-ray microanalysis. Because of this behavior low voltage microscopy is now usually performed with a field emission gun. In many cases the F E G can be operated directly at the desired final energy because the widely-used Butler gun geometry (Butler, 1966; Crewe e t al., 1968) remains efficient even when the final operating voltage is less than the tip extraction voltage. Alternatively the gun can be operated at a fixed, higher, energy such as 10 keV and the electrons then decelerated to the required energy using an retarding electrostatic lens (Frank e t al., 1994), an arrangement which has the merit of greatly reducing the effect of chromatic aberration on the final probe (Brodie, 1994). In either case the effective source brightness at 1 keV remains high enough (typically > 106 amp/cmZ/sterad) to ensure an adequate signal to noise ratio even for probe diameters of a few nanometers. The first SEMs capable of high performance at low energies used high excitation immersion lenses to achieve the small values o f Cs and C c that are required (Nagatani e t al., 1987), but such an arrangement severely restricts sample access and size. More recently almost equivalent performance has been demonstrated (Sato e t al., 1994) using a 'snorkel' lens

VH. C O N C L U S I O N S Low voltage scanning electron microscopy has developed to become a major topic in its own right. The many effects associated with the fall in electron range with energy can be exploited to yield information, and to obtain images from specimens, which would not be available using higher beam energies. Even though charging and radiation damage remain as problems careful operating procedures can minimize these effects without compromising the image quality. The use of ultra-low beam energies may also offer some interesting possibilities. authors are grateful to Profs. T. Mulvey,T. Page, D. Perovic,and Drs. M. Sato, and K. C. H. Smith, for valuable discussions. Oak Ridge National Laboratory is managed by Lockheed Martin Energy Research Corp. under contract DE-AC05-84OR21400 with the U.S. Department of Energy.
Acknowledgements--The

REFERENCES

von Ardenne,M., 1940a.Elektronen-Ubermikroskopie. Springer,Berlin. vonArdenne, M., 1940b. Elektronen-Ubermikroskopie, SpringerVerlag, Berlin; reprinted by Edwards Brothers Inc. Ann Arbor., MI, 1943. Beadle, W. E., Tsai, J. C. C. and Plummer, R. D., 1985. Quick Reference Manual f o r Silicon Integrated Circuit Technology. John Wiley and Sons, New York Bethe, H., 1941Phys. Rev., 59, 940.

Low Voltage Scanning Electron Microscopy Bleloch, A. L., Castell, M. R., Howie, A. and Walsh, C. A., 1994. UItamicroscopy, 54, 107. Borchert, A. M., Ve~chio, K. S. and Stein, R. D., 1991. Scanning, 13, 344. Bradley, G. F. and Joy, D. C., 1991. Proc. 49th Ann. Mtg EMSA, G. W. Bailey (cd.), San Francisco Press, San Francisco, 534. Brodie, A. D., 1994. J. Vac. Sci. Technol., B12, 3489-3492. Bruining, H. and de Boer, J. H., 1938. Physica, (Amsterdam J, 5, 17. Butler, J. W., 1966. Proc. 6th Int. Cong. on EM (Kyoto, Japan) 1,191. Butler, J. H., Joy, D. C., Bradley, G. F., Krause, S. J. and Brown, G. M., 1992. In: Microscopy--The Key Research Tool, EMSA, Boston, 103. Cazaux, J., 1986. J. appl. Phys., 59, 1418. Chung, M. and Everhart, T. E., 1974. J. appl. Phys., 45, 707. Crcwe, A. V., Eggenbcrger, E. N., Wall, J. and Welter, L. M., 1968. Rev. sci. Instr, 39, 576. Drescher, H., Reimer, L. and Sciler, H., 1970. Z. agnew. Phys., 29, 331. Dwyer, V. M. and Matthew, J. A. D., 1985. Surface Sci., 152, 884. Fiuk, H. W.. IBM J. Res. Develop., 30, 1986, 460. Frank, L., Mullerova, I. and Delong, A., 1994. Czech. J. Phys, 44, 195. Graham, J., Butt, C. R. M. and Vigers, R. B. W., 1984. X-ray Spectrometry, 13, 126. Goldstein, J. I., Newbury, D. E., Echlin, P., Joy, D. C., Romig, A. D., Lyman, C. E., Fiori C. E. and Lifshin, E., 1992. Scanning Electron Microscopy and X-ray Microanalysis, Plenum Press, New York, 21-47. Harker, A. B., Howitt, D. G., DeNatale, J. F. and Flintoff, J. F., 1994. Scanning, 16, 87. Jia, Y. and Joy, D. C., 1992. Acta Microscopica, 1, 20. Joy, D. C., 1987. In: Microbeam Analysis--1987, R. H. Gciss (ed.), San Francisco Press, San Francisco, 117. Joy, D. C., 1991. Ultramicroscopy, 37, 216. Joy, D. C., 1995a. Monte Carlo Modeling for Electron Microscopy and Microanalysis, Oxford University Press, New York. Joy, D. C., 1995b. Scanning, 17, 270. A copy of this database on Electron Interaction with Solids is available on request from the author. The SE and BSE data may also be accessed and down loaded from the World Wide Web at http://www.nsctoronto.com/ nissei-sangyo. Joy, D. C. and Joy, C. S., 1995. J. Micros. Soc. Am., 1, 109-113. Kenway, P. B. and Cliff, G., 1984. Inst. Phys. Conf. Ser., 68, 83-86.

263

Knoll, M., 1935. Z. tech. Phys., 16, 467. Kimura, H. and Tamura, H., 1967. Proc. 9th Ann. Symp. on Electron, Ion, and Laser Beams, IEEE, New York, 198. Knoll, M., 1935. Z. Tech. Physik., 11, 467-475. Luo, S., 1994. Ph.D. Thesis., University of Tennessee. Luo, S., Zhang, X. and Joy, D. C., 1991. Rad. Effects and Defects in Solids, 117, 235. Matthes, I., 1942. Z. tech. Phys, 22, 232. Mulvey, T. E. and Newman, C. D., 1973. Inst. of Physics Conf. Ser., 18, (IOP: London), 16. Nagatani, T., Saito, S., Sato, M. and Yamada, M., 1987. Scanning Microsc., 1, 901. Ness, J. N. and Page, T. F., 1986. J. Mat. Sci., 21, 1377. Ohlendoff, G., Koch, W., Kempter, V. and Borchardt, G., 1991. Surf Interface Anal., 16, 948. Perovic, D. D., Castell, M. R., Howie, A., Lavoie, C., Tiedje, T. and Cole, J. S. W., 1995. Field-emission SEM imaging of compositional and doping layer semiconductor supcrlattices. Ultramicroscopy, 58, 104-113. Peters, K.-R., 1980. Scanning Electron Microscopy, 1, 143. Reimer, L., 1993. Low voltage scanning microscopy, SPIE, Bellingham, WA. Sato, M., Todokoro, H., Suzuki, T. and Yamada, M., 1994. Hitachi Instrument News, 26, 18. Sato, M. and Orloff, J., 1991. J. Vac. Sci. Tech., B9, 2602. Scheinfein, M. R., Qian, W. and Spence, J. C. H., 1993. Proc. 51st Ann. Mtg. MSA, G. W. Bailey and C. L. Reider (eds), San Francisco Press, San Francisco, 632. Sciler, H., 1984. J. appl. Phys, 54, R1. Shao, Z. and Crewe, A. V., 1989. Ultramicroscopy, 31, 199. Spenc, J. C. H., Qian, w. and Zhang, X., 1994. Ultramicroscopy, 55, 19. Sugiyama, N., Ikeda, S. and Uclxikava, Y., 1986. J. Electr. Micros., 35, 9. Thornton, R. F. M., 1960. Proc. European Regional Conference on Electron Microscopy, (Delft: 1960), Houwink and Spit (eds), Nededandse Vereniging voor EM, Delft, 173. Venables, D. and Maher, D. M., 1994. Proc. 52nd Annual Meeting MSA, G. W. Bailey and A. J. Garratt-Reed (eds), San Francisco Press, San Francisco, 1024. Zworykin, V. A., Hillier, J. and Snyder, R. L., 1942. ASTM Bull., 117, 15.

Você também pode gostar