Low voltage scanning electron microscopy

The scanning electron microscope (SEM) is usually operated with a beam voltage, V0, in the range of 10–30 kV, even though many early workers had suggested the use of lower voltages to increase topographic contrast and to reduce specimen charging and beam damage. The chief reason for this contradiction is poor instrumental performance when V0=1–3 kV, The problems include low source brightness, greater defocusing due to chromatic aberration greater sensitivity to stray fields, and difficulty in collecting the secondary electron signal. Responding to the needs of the semiconductor industry, which uses low V0 to reduce beam damage, considerable efforts have been made to overcome these problems. The resulting equipment has greatly improved performance at low kV and substantially removes the practical deterrents to operation in this mode. This paper reviews the advantages of low voltage operation, recent progress in instrumentation and describes a prototype instrument designed and built for optimum performance at 1 kV. Other limitations to high resolution topographic imaging such as surface contamination, the de‐localized nature of the inelastic scattering event and radiation damage are also discussed.


I N T R O D U C T I O N
The resurgence in interest in the SEM after 1948 described by Oatley (1982) can be said to have started in 1953 when McMullan published a description of his early SEM (McMullan, 1953). The record of the next 30 years shows a steady increase in the types of study to which the SEM was applied, the manner in which it was used in these applications, and in the availability of equipment designed to meet these new needs. This process is exemplified by the recently renewed interest in operating the SEM at low beam voltage (LVSEM).
As the beam voltage, VO, is reduced in the range 30-1 kV, three physical parameters relevant to specimen damage, surface charging and topographic contrast also change. The secondary electron coefficient, 6, increases to unity while the electron range ( R = k V O~.~) and the energy deposited per electron (eVo) both decrease. When 6 is large there is less surface charging on insulating samples and there is more signal per beam electron, while a smaller R means the beam,'specimen interaction is more localized and the image contrast is higher. This was early recognized by Thornley (1960) but widespread use of the SEM in the 1-3 kV range was delayed by technical limitations that can be grouped in four categories: (1) low source brightness, (2) increased effect of chromatic aberration, (3) increased sensitivity to stray fields, (4) defocusing of the probe by the secondary electron collection field. With a few exceptions (Welter & Coates, 0 1984 The Royal Microscopical Society 1974) these disincentives prevented significant efforts to improve low voltage SEM (LVSEM) performance. Recently, however, SEM studies of semiconductors were found to be limited by the damage caused by the beam and the only way to reduce this damage was to use VO = 0.5-1.5 kV (Keery et al., 1976;Miyoshi et al., 1982). The fact that the semiconductor industry, which presently represents more than half of the SEM market, urgently needed to monitor production procedures and final performance without damaging the specimen, provided a new impetus for the development of equipment optimized for low voltage operation (Tamura et al., 1980;Todokoro et al., 1980;Buchanan, 1982Pomposo & Coates, 1982).
It is the purpose of this paper to draw attention to these developments in the belief that the low voltage capability of this new equipment will find widespread application outside the field of semiconductor research. The advantages of low-voltage operation will be discussed particularly in regard to the possibility that it may eventually provide the ultimate in high resolution topographic images of biological samples. This will be followed by a description of both the instrumentation problems associated with operating the SEM at high resolution with VO z 1 kV and the strategies that have been developed or proposed to overcome these problems. A prototype instrument designed to produce a 1-2 nm beam at 1 kV will be described. Finally there is a brief discussion of the limitations posed to the ultimate topographic resolution of an ideal instrument by radiation damage, sample-derived surface contamination and the de-localized nature of the inelastic scattering event.

THE ADVANTAGES OF LVSEM
Most of the advantages of using an SEM with V o z 1 kV derive directly from the fact that electrons impinging on the surface of a solid with less energy, penetrate into it a shorter distance and also have a higher crossection for producing secondaries near the surface where they can escape (Kotera et al., 1981). As a result, 6 approaches unity, charging artefacts are less pronounced and the signal/beam-electron is increased. Also less energy (eVo) is deposited in the sample and, on insulating samples, charge is not injected and trapped so far beneath the surface All of these features are important in the study of uncoated resist patterns or oxide layers on semiconductor devices. Relative freedom from charging artefacts is an obvious advantage on the non-conducting photoresist and passivation layers, but even more important is the reduction in beam penetration because the trapped charges distort the energy band structure of the device, degrading and possibly destroying its function. At 1 kV, such damage is restricted to the outer 0.02 pm or so, rather than 130 times that depth at 30 kV-a crucial difference in devices only a few micrometres deep. Finally, both the silicon and the resist layer on its surface are composed of materials having relatively low atomic number (Z). At high beam voltages very little surface detail can be seen on uncoated low Z samples but at 1 kV the energy is deposited nearer to the beam entry point and so contrast produced by topographical variations is proportionately larger. Other aspects of the study of semiconductors in the SEM are reviewed in Scanning Electron Microscopy/l981/I(SEM Inc., Chicago), in Scanning, Vol. 5 (1983) and in Journal of Microscopy, Vol. 118, Part 3 (1980). Applications that emphasize the utility of low voltage Operation are discussed by Todokoro et al. (1983) and Yau et al. (1981) and low voltage electron lithography is described by Varnel (1981), Polasko et al. (1983) and Pfeiffer (1982) Although the subjects covered by these authors are in large part responsible for recent instrumental improvements in the LVSEM, they will not be discussed further here. Contrast and charging will now be considered in more detail.

Topographic contrast in the SEM
In the discussion which follows, topographic contrast should be understood to mean that component of the variation in the detected secondary electron signal arising solely from the variation in the angle of incidence of an electron beam impinging on a rough surface. This is discussed specifically by Catto & Smith (1973) and it gives rise to the 'real' appearance of images of three-dimensional objects viewed at low magnifications in the SEM. It does so because there is a rough equivalence between the secant laws relating the apparent brightness of a diffusely-illuminated matt surface and its angle with the line of sight on the one hand and the variation of 6 with incidence angle on the other (Everhart et al., 1959). This near equivalence is easily appreciated when viewing an object such as an insect in the SEM, but it is not perfect and contains a number of hidden assumptions. Generally these assumptions are only satisfied when the size ofa pixel in the image, referred to the sample, is larger than the beam penetration area. As beam penetration can be tens of microns on dried biological material, this assumption becomes increasingly invalid as magnification is increased. This can easily be seen in Fig. 1 which shovits a hair on a flour beetle imaged at 2, 5 and 10 kV. In the 2 kV image topographic detail is visible on the hair and, more importantly, it is the same shade of grey as nearby surfaces of similar orientation on the bulk sample. At 10 kV the hair appears far brighter than neighbouring areas of similar angle because a large number of secondaries are being produced from its far side. Clearly the 'topographical' coding has broken down. The 5 kV image shows an intermediate result. been sputter-coated with gold and imaged with V O = 2,5 and 10 kV. I n (a), details on the surface of the hair can be seen and the near side is approximately the same shade of grey as comparably oriented surfaces on the bulk of the specimen. In (c) the hair appears much brighter than these adjacent areas because considerable signal is generated as the beam emerges from the far side of the hair and the coding of the image is no longer strictly topographic. Figure l(b) shows an intermediate condition. George & Robinson (1975) performed a Monte Carlo simulation of both the secondary electron and the backscattered signals from a 20 nm, 30 kV beam scanning across gold cubes sitting on a solid gold surface. These show the degradation of topographic signal very well: at 30 kV the vertical cube edges which should appear as white lines surrounding a grey square, merge to produce a single white dot when the cube becomes less than 100 nm. A more recent study by Joy (1984) is more complete in that it uses a more accurate algorithm, and simulates more specimen geometries and probe diameters (0-5 nm) using VO from 5 to 30 kV. Even here, however, topographic resolution is found to be substantially less than beam diameter when the total secondary signal is used.
The detected secondary electron signal from a given pixel on the sample is a function not only of 6 and the surface angle, but also of the average, 2, of the volume of beam penetration (Everhart et al., 1959;Seiler, 1976;Ball & McCartney, 1981), the crystallographic orientation (LeGressus et al., 1983), the surface potential (Oatley & Everhart, 1957;Oatley, 1969;Banbury & Nixon, 1970;Kursheed & Dinnis, 1983), the presence of nearby surface features which may affect the collection efficiency (Everhart et al., 1959;Pawley, 1972), the presence of second or third surfaces of the sample within the penetration volume from which additional electrons may be produced and collected (Wells, 1978), the efficiency with which high energy backscattered electrons are converted into collectable secondaries by collisions within the sample chamber (Oatley et al., 1965;Reimer & Volbert, 1979;Peters, 1982) and, finally, various arcane variables such as the presence of subsurface charge that may effect 6 on uncoated insulators (Shaffner & Hearle, 1976).
The complexity of the interaction of these variables as they affect the high magnification image of a typical sample is considerable and is perhaps the reason contrast and resolution in the secondary electron mode have been the object of so much research and interest, for example: Everhart et al., 1959;Oatley et al., 1965;Pease & Nixon, 1965;Clarke, 1970;Oatley, 1972;Catto & Smith, 1973;Wells, 1974a, b;Haggis & Bond, 1979;Peters, 1979Peters, ,1982Reimer, 1978;Joy, 1984. Secondary electrons have low energy (0-50 eV) and can travel only a few nm in solids. The conventional theory (Oatley et al., 1965;Catto & Smith, 1973;Reimer, 1979;Oatley, 1983) postulates that the signal collected by the standard scintillator/photomultiplier secondary electron detector (Everhart & Thornley, 1960) is generated by three mechanisms: Type 1: interaction of the beam with the surface near the point of entry; Type 11: interaction of high energy (50 eV to e Vo) backscattered electrons generated perhaps microns below the surface, as they re-emerge through the surface some distance from the entry point; Type 111: interaction between these same backscattered electrons and the walls of the specimen chamber, primarily the lower surface of the final lens pole-piece. Of these three mechanisms, only the first is expected to give high resolution information characteristic of the surface in the immediate vicinity ( & a few nm) of the beam impact because of the short range of these secondary electron?.
The distribution of secondary electrons leaving the sample as a function of the distance from the point of impact is assumed to have a small peak within a few nm of the beam axis and a long 'tail' extending many micrometres in all directions and corresponding to the probability of a secondary electron being produced by a re-emergent backscattered electron (Joy, 1984).
For fairly high Vo, such a distribution has been directly observed by using the surface of a tilted SEM sample as the source of an emission microscope (Hasselbach & Rieke, 1982;Hasselbach et al., 1983). In visualizing the relative dimensions of these two parts of the distribution, it is important to keep in mind the large magnitude of the difference between the range of a 20 kV backscattered electron and a 4-50 eV secondary electron. On metal-coated, dried biological material (density 0.2 g/cm3, Z= 7) the former may be 100 pm and the latter 0.002 pm (Joy, 1984). To obtain high resolution topographical information it is necessary to somehow separate the relatively small peak signal (Type I) from the much larger signal produced by the tail (Types I1 and 111).
Initially, it was thought that this cmld be done simply by raising the beam current and treating the tail signal as a DC noise signal that could be electronically removed by analogue subtraction. This approach was not very successful, probably because of the difficulty of increasing the beam current in a small spot sufficiently and because, even though the average value of the DC offset could be removed, the noise associated with statistical variations in the number of electrons that this signal represented could not be removed and soon this noise swamped variations in the Type I signal (Wells, 1974a, b).
To reduce the Type 111 and to a smaller extent the Type I1 signal, Peters (1982) has recommended placing backscattered electron absorbers below the polepiece and coating the surface of the sample with very thin layers of low Z metals. His results, using a field emission SEM at 30 kV, show a clear improvement over normal operation but the approach does not tackle the problem of removing the Type I1 signal very directly. On the other hand, Crewe & Lin (1976) recommended detecting the backscattered signal independently, using a semiconductor detector attached to the pdepiece, and then subtracting some fraction of this from the signal derived from the normal scintillator-photomultiplier detector. The logic is that the Type I1 and Type I11 signals should be proportional to the backscattered detector output and subtracting this from the normal detectw output should leave only the Type I signal. As the backscatter detector used in this work covered about T steradians, this is a reasonable analysis but the correspondence is not perfect because the detector has a large hole in the middle to allow the beam to pass through and the energy and angle of a backscattered electron may its chance of producing a collectable secondary in a way not proportional to the signal it produces in the semiconductor detector. None the less, these authors also show a clear improvement (pp. 236) and the technique has been used by others (Volbert, 1982a, b).
Finally, there is a large group of investigators who, untroubled by theoretical misgivings, have made images of a variety of samples that appear to demonstrate topographic resolution far in excess of that which would be possible if the Type I signal is indeed likely to be swamped by Types I1 and 111. Some of these studies have used scanning attachments on the transmission electron microscope (TEM/SEM) operated at 20-80 kV (Koike et al., 1971(Koike et al., , 1973Arro et al., 1981;Haggis & Bond, 1979;Haggis, 1982;Haggis et al., 1983). In the TEM/SEM, the sample is immersed in the lens field and the secondary electron signal consists of those electrons that spiral up the field lines and out through the upper pole piece. This process may preferentially exclude the Type I11 signal and definitely provides a distinctly different image of the sample than does a conventional detector (Buchanan, 1982. Other workers have utilized field emission SEMs (Lin & Lamvik, 1975;Watabe et al., 1978;Sawada, 1981; which in some cases were modified to permit secondary electrons to be collected from a sample located in the lens field (Tanaka, 1980(Tanaka, , 1981. A third approach involves looking at what might be considered the inverse of the Type I signal, namely the signal derived from backscattered electrons that have only lost a small amount of energy in a glancing collision with a steeply tilted sample. This low-loss backscattered signal (LLE) can be detected in a normal SEM with an appropriately placed backscattered electron detector (Wells, 1979) or from a sample immersed in the field of a short focal-length condensor-objective lens which also serves as an energy filter (Wells et al., 1973;Joy & Maher, 1976;Kokubo et al., 1975). The latter method is capable of producing very high resolution images of metal-coated samples because electrons that have lost only 200-400 eV have only participated in interactions near the sample surface (Broers et al., 1975).
It is not clear that any of these signals is really a topographic signal rather than a Z signal that chiefly responds to variations in the granularity or the effective thickness of the metal coating on such samples and is further modified by differences in the signal collection efficiency from point to point on the sample. In fact Wells points out that a layer of carbon contamination, artificially produced to cover the surface of such a sample, is barely visible using the LLE mode (Wells, 1979, pp. 213). A similar lack of fine details on flat surfaces can be seen in the images of carbon black particles shown in the TEM/SEM by Koike et al. (1973). Though these images appear to be topographic, they are not topographic in the same sense as is the case at low magnification. They may indeed provide useful information about the sample, but it is important to realize that they are in fact analogous to TEM images of shadowed replicas and should likewise be viewed with caution. They may, for instance, reveal more about the nucleation of metal particles than they do about sample topography and they discriminate against small features on flat surfaces and in favour of similar features suspended over the cavities. Our acceptance of Z-contrast images as 'topographic' can be traced to the need of manufacturers to demonstrate real improvements in instrument performance. In the early 1970s resolution in images made using the signal from the secondary electron detector fell below about 20 nm and the criterion ceased to be the smallest discernible surface object and became instead the smallest discernible object. Subsequently, the probe diameter was greatly reduced and the test objects were chosen to demonstrate this improvement, rather than to demonstrate that smaller surface features could be resolved (Ballard, 1972). As a result the 'resolution' in the secondary electron mode is often quoted to be 1.5-3 nm while the best results show images of biological objects such as intermediate filaments and ribosomes in the size range of 10-25 nm (Tanaka, 1981;Haggis et al., 1983;Peters, 1982;Peters & Green, 1983).
Many popular test objects can be modelled as a series of heavy metal particles covering the surface of a low Z substrate such as a carbon film or a dried biological sample. In this case, the main contrast is Z contrast, either between high Z metal grains and low Z inter-grain spaces, or, In the case of uniform coating on a bumpy surface, variations in the effective thickness of this coating as the beam traverses the coating at different angles. On a highly convoluted surface this signal may also reflect variations in coating thickness and the large variations that exist in the efficiency with which electrons emerging from a given area are collected. These effects are diagrammed in Fig. 2 which shows a hypothetical coated surface and a corresponding image shaded solely in response to changes in effective coating thickness. shows the total thickness of the coating at each point. Figure 2(c) shows what appears to be a topographic image of an extended specimen having the cross-section shown in (a) but which is actually the result of coding image lightness with coating thickness rather than surface angle. (The image may be more easy to interpret if viewed from a distance.) The most effective approach to the problem of low contrast, however, involves reducing the beam voltage as suggested by many early authors (Thornley, 1960;Kosuge et al., 1970;Boyde, 1971;Catto & Smith, 1973;Welter & Coates, 1974;Wells, 1974b, p. 127;Dilly, 1980). At Vo < 10 kV there is a sharp increase in Type I signal (Joy, 1984). In addition R at 1 kV is only about 2% of that at 20 kV and so the area of sample from which Type I1 secondaries are produced is substantially smaller. As the backscatter coefficient diminishes only slightly with voltage, the number of Type I1 electrons may still be significant (Niedrig, 1978;Reimer, 1979, Fig. 8;Kotera et al., 1981). There have been few attempts to produce high resolution SEM images using beam energies near 1 kV because of the electron-optical constraints mentioned in the Introduction and discussed in the next section, so it is still not certain that, on the finest scale, LVSEM has any clear advantage. On the other hand, results at somewhat lower resolution show a clear increase in the contrast of small details (Fig. 3) and so there is reason to expect that the same will hold if LVSEMs with smaller probes can be made.

Specimen charging in the LVSEM
Many objects of microscopic interest are electrical insulators. When the surface of such a sample is scanned by a 1-30 kV electron beam, collisions in the layer immediately adjacent to the surface cause it to become somewhat depleted in electrons, while the layer immediately underneath becomes negatively charged because beam electrons become trapped as they reach the end of their range. The deposition of a net charge in the sample depends on 6, which in turn depends on the type of material, and the local surface angle. Around 10-30 kV, 6 for most samples is less than unity and the sample accumulates a negative charge which may degrade the image by defocusing the beam or by distorting the collection field so as to produce the anomalous changes in apparent brightness familiar as the most common form of charging artefacts (Pawley, 1972). As higher surface potentials are reached ( > tens of volts), other, more extreme phenomena are recorded as described by Shaffner & Hearle (1976). Other variables that exacerbate the charging problem are high specimen resistivity (so-called insulators vary in resistivity over a range of 15 orders of magnitude), low specimen dielectric constant and slow scan speeds (Welter &. McKee, 1972).
The situation is somewhat different in the LVSEM because on a variety of samples 6 becomes greater than unity in the range of approximately 05-3 kV and so the sample charges positively. This is a far more stable situation because low energy electrons are constantly being evolved from the surface and so even a small positive charge imbalance can attract an appropriate neutralizing charge without the necessity of developing a surface potential higher than a few hundred millivolts. Because this self-regulating process is so efficient, it is often claimed that charging artefacts do not exist when 6 2 1 and the rapidity with which TV-rate images of such samples stabilize is offered as proof of this contention (Welter & McKee, 1972). However, this analysis is only strictly true for the trivial case of a flat, featureless sample with V O adjusted for 8 = 1. More topographically interesting samples show contrast and hence, 6 = 1 cannot be satisfied everywhere. In practice, areas where the beam incidence approaches normal may become slightly negative while areas of glancing incidence, or where the beam penetrates porous surface features will tend to become positive. Lateral electrostatic fields will exist between neighbouring charged areas and vertical fields will exist between the electron-depleted surface layer and the trapped charge below. A small amount of current flows between these areas using free subsurface electrons, ionized by the beam, as charged carriers (Bresse, 1982). Clearly the situation is far more complicated than is implied by the simple statement that there are no charging artefacts whenever VO is set so that S 2 1. The stability of an image scanned at T V rate is only evidence that a particular charge distribution is stable, not that there is an absence of charging.
The details of this process are of interest here because they involve the establishment of surface and subsurface potentials which, though small when compared to those found with higher VO, may still be capable of defocusing or deflecting a beam of only 1 kV and thereby degrading the image. Surface potentials of the same size as AV (0.2 V for FE guns) will defocus the beam and even smaller potentials could deflect it a few nm, perhaps in an erratic manner. In fact some investigators claim less trouble with charging at high voltage in the TEMI'SEM than at lower voltages in the SEM (Haggis, 1982). This can be attributed to increased beaminduced conductivity at higher VO and a partial immunity of the TEM/SEM detector to voltage contrast.
At present, the magnitude of these effects has not been investigated in the range of resolution and VO discussed here. It can be expected that when V O E 1 kV, charging effects on insulating samples scanned at TV rates should be insufficient to produce large variations in signal ~01lection efficiency, but not that they will be totally absent. Very thin (1-1.5 nm) layers of coating material such as those used by Peters (1979Peters ( , 1982 may be necessary. Fortunately, the procedures for applying these coatings have greatly improved in the past few years particularly with the introduction of ion-gun based sputter sources (Adachi et al., 1983;Evans & Franks, 1981;Kemmenoe & Bullock, 1983). As a result, problems with decoration artefacts should be less and the pseudo-topographical contrast caused by the coating and discussed above, should be minimized by the use of very thin coatings.

T E C H N I C A L L I M I T A T I O N S A N D P O S S I B L E S O L U T I O N S I N LVSEM
The technical difficulties that must be overcome in order to produce high resolution information from an LVSEM are similar to those noted for the low voltage TEM by Wilska (1964Wilska ( , 1965. They were first listed for the SEM by Oatleyet al. (1965, pp. 215-217). They can be lumped into four areas: (1) low source brightness, (2) increased effect of chromatic aberration, (3) increased susceptibility to stray fields, (4) interactions between the beam and the signal ,-ollection field. Though these problems and their solutions sometimes interact, they will be treated separately below.
Brightness of thermionic sources at low voltage It was early recognized that source brightness was the practical limit on the performance of an SEM with a heated tungsten source (Broers, 1974(Broers, , 1982. In principle, the effect of spherical or chromatic aberration on spot size can be minimized by reducing the acceptance angle of the final lens, a, until the lens becomes diffraction limited. However, in instruments with conventional tungsten sources, the image becomes too noisy for convenient use long before 01 is reduced to the diffraction limit. The brightness ( p) of a thermionic electron source is limited by the Langmuir Equation (Langmuir, 1937) for small a.
where jc, = current density at the source surface in amps/cm2, T= source temperature in K, and Vo = beam voltage in volts.
From equation (1) it follows that low Vo operation will produce reduced brightness, but in practice the brightness actually obtained is even lower than we might expect from (1) because this equation is only valid in the absence of space charge near the cathode surface. Such space charge shields the cathode from the accelerating field and further reduces p. Though a given gun geometry may be virtually free of space charge effects near its highest operating voltage (20-30 kV) (Broers, 1974;Oatley, 1975), the fields present at the filament surface are proportionately less at 1 kV and the gun brightness will be limited by space charge unless the geometry is changed.
Practical measures to improve thermionic gun brightness at low kV therefore include changing gun geometry and the use of La& cathodes. La& cathodes are about 6-10 times brighter then tungsten for comparable lifetime, and have a more pointed tip which reduces space charge effects. They also operate at a lower temperature than normal tungsten (T = 1500 K v. 2500 K). Changes in gun geometry may involve simply reducing the gap between the Wehnelt and the anode by using an anode spacer or a mechanism to actually move the anode towards the cathode or it may involve adding additional anodes to the gun. A description of this 'doubleanode' approach was recently given by Yamazaki et al. (1983). This group installed extraction anodes of various shapes and spacings between the Wehnelt and the normal anode. At low Vo this electrode is run a few kV above ground to produce a higher constant voltage between it and the cathode (Vl) and thereby reduce the effects of space charge.
Careful measurements by Yamazaki et aZ. verified that, at 1 and 2 kV, this produced 10 times the brightness of a normal 30 kV gun with tungsten and 8 times the brightness with LaBs. They reached the theoretical brightness at both these voltages using their best geometry, which is diagrammed in Fig. 4. Although it is not specifically pointed out in their paper, the acceleration/ deceleration electrode system acts as a weak positive lens and this is probably why these workers found that P went through a maximum at V , = 1.5 kV. The effect of this lens on the imaging system was unclear but in images of a test sample, a distinct improvement was associated with the use of the double-anode system.

Brightness of jield-emission sources at low voltage
Field-emission (FE) sources have long been identified with high brightness (Crewe et al., 1968(Crewe et al., , 1970Crewe, 1973;Hainfeld, 1977, has a good introductory review), but few commercial instruments have capitalized on this feature because of the stringent requirement for vacuum in the range of 10-8 Pa (10 -10 torr) round the emitter tip and because the current produced in a fine beam is subject to some temporal instability which tends to produce streaky images. Most FE sources utilize a double anode design. The V1 is normally 3-7 kV and is used to adjust beam current, which is otherwise only dependent on the work functions, 4, and the tip radius, TI,.
A secmd supply between ground and the cathode adjusts the beam voltage (Vo) to the desired level and, in the case of the LVSEM, this means reducing it and thereby again producing an electrostatic lens.
In principle, the geometric parameters can be adjusted so that the tip emits efficiently with V1= 1 kV. This would make the second anode unnecessary and avoid the consequent lens action. However, in practice, tips with sufficiently small usually prove unstable and subject to catastrophic failure while a suitable choice of the spacing and shape of the two anodes can reduce the lens effect to a low level.
In the range of voltages discussed here, the brightness of the source depends only on V1 and not directly on V0 except to the extent that the lens effect degrades the source image (Hainfield, 1977).
Several FE guns have been designed to incorporate a magnetic lens which operates on the beam as it leaves the tip (for example Kuo & Siegel, 1976;Ichinokawa et al., 1982). These lenses have superior electron optical characteristics to the electrostatic lenses and are said to produce improved performance especially when operating at high current and low b-am voltage. However, because of their high current these guns are more susceptible to the lateral electronelectron interactions discussed below and so it is not clear they would be suitable to high resolution LVSEM.
Measurements with V l = 3 kV, V. = 1 kV have yielded values of fi = 3-70 x 106 amps cm2, ster or about 1000 times that measured by Yamazaki ei al. (1983) on the LaBG double-anode gun.
Although there are reports that electron-electron interactions within the beam can degrade the expected performance of FE guns both in terms of reducing the brightness and increasing effective energy spread (Bauer & Speidel, 1981;Van der Mast, 1983) these effects seem to be most serious on heated FE sources, sometimes referred to as TF guns, and less serious on temperature FE guns operated at moderate tip currents of about 10 PA. Other workers have observed no such effects as long as high current density crossovers are avoided (Crewe et ul., 1971). Clearly the electron-electron interactions near the cathode surface are reduced by the fact that the FE cathode has a tip radius 20-50 times smaller than LaBI;. This subject is considered further in the next section.

Noise in FE sources
If the current present in the final beam of a high resolution FE SEM is traced back to the tip, it is found to arise from an area of only a few nm2. The adsorption and desorption of individual molecules from this small surface can therefore produce a significant variation in its average work function while the etching produced by the collision of a single ion can change the microtopography and hence the local surface field. As a result, the current in the final probe is found to have a noise component unrelated to shot noise of between 3",, and loo,, (Hainfeld, 1977).
This noise drops in magnitude with increasing frequency and therefore is most troublesome at low frequencies. Efforts to stabilize the beam current by measuring the current striking an aperture and using this as a feedback signal to readjust V1 (Nomura et al., 1973) are quite effective but not wholly successful because, due to the localized nature of the disturbance at the tip, the current striking the aperture is not necessarily a good measure of the current passing through it. Also, the changes in Vl necessary to stabilize the current change the optical properties of the electrostatic lens. More recent systems avoid this optical effect by applying the signal from the aperture to an analogue multiplier which directly normalizes the video signal for changes in beam current (Saito et al., 1982).
Another approach involves rapid, multiple scanning of the sample with the idea that low frequency variations will average out (Welter & McKee, 1972). The TV scan rate has other advantages with respect to ease of operation, stabilization of charging artefacts and quasiimmunity to stray field but it requires very high detector bandwidth (40 MHz for a 1000 x 1000 raster, 1/30 s) and careful scan coil design to avoid image distortion.
The matter of source noise can be crucial to the performance in the LVSEM. There will be little net gain in contrast by going to lower Vo if the improved contrast at the sample is swamped by false contrast due to source instability.
Chromatic aberration in the L VSEM disk of confusion due to this defect, dc, is : The dominant lens defect in the LVSEM is chromatic aberration and the diameter of the where Cc is chromatic aberration coefficient and A V is the energy spread of the beam. Clearly, 1 V / Vo increases rapidly at low VO, hence the problem. It can be attacked by lowering C,, (Y or 1 V. Lenses can be designed to reduce C, by shortening their focal length. While it is relatively easy, in terms of the total magnetic flux required, to construct a lens of short focal length at this low energy, the sample is soon immersed in the lens field so it may become more difficult to collect the secondary electron signal. At the ultimate, it would probably be very difficult to design a practical system where C, was much less than 0.2 mm (Pawley & Wall, 1982;Barth & Poole, 1976). This compares with the 5-10 mm found on most commercial instruments and the 1-3 mm found on some sample-in-lens SEMs (Koike et al., 1971 ;Buchanan, 1982).
The semi-angle, 01, can also be manipulated, but because of the relatively long wavelength, A, of 1 kV electrons (37 pm), the diffraction limit, dd, is soon reached.

James l'awley
Finally, there is some control over LV. The energy spread of the beam has many sources: the intrinsic energy spread of electrons leaving the cathode, power supply instabilities and energy broadening caused by lateral electron-electron interactions in high-current crossovers known generally as the Boersch effect (1954).
The expected energy spread at the cathode surface for thermal emitters is KT and this again emphasizes the advantage of LaBs v. tungsten because of its lower operating temperature (kT= 0.13 eV v. 0.2 eV). Intrinsic energy spread in FE sources depends on the shape, the tip material and the crystallographic orientation of the tip but it is usually quoted as about 0.2 eV for tungsten Lateral interactions between electrons are more noticeable when high current beams must form crossovers and this is often the case in thermal sources where large beam currents ere a byproduct of efforts toincrease / 3 by reducing the effect of space charge (Oatley, 1975). The problem is compounded by the fact that this large current is usually focused into a small gun crossover by the effect of the Wehnelt. Under these circumstances, a considerable improvement can be gained by employing pointed cathodes as these permit a high extraction field over only a small emitting area and therefore a lower total current (Wiesner, 1973;Wiesner & Everhart, 1973;Oshita et al., 1978). Measured values of L V from thermal sources usually average about 2-1: eV but the measurements are normally made at voltage smuch higher than 1 kV, where the Boersch effect is likely to be less strong because the electrons move faster and therefore have less opportunity to interact (Pfeiffer, 1972(Pfeiffer, , 1982. The FE gun has clear advantages in this regard. It not only has low intrinsic energy spread but it operates well at low total beam currents and because of its small source size it can, in principle, operate with no crossover before the sample. Taken together with the higher brightness of FE at low kV, the reduced energy spread provides a convincing rationale that any serious effort to produce optimum performance in the LVSEM will necessarily require a FE source and a method for compensating for is temporal instability. Crewe et al., 1971;Hainfeld, 1977).

Stray fields
A 1 kV electron takes 5 times as long to travel down a given column as does a 25 kV electron. For this reason, in simple terms, it is 5 times as susceptible to transverse stray electrostatic or magnetic fields. This effect is sometimes exaggerated on large, older instruments because they are often run with less demagnification in the intermediate lenses to compensate for low suurce brightness and, as a result, stray fields acting on the upper column, which usually have no visible effect, begin to be noticeable. Field emission systems are similarly susceptible because they normally operate with little or no demagnification.
Both AC and DC electromagnetic fields can degrade the performance of electron optical instruments. DC magnetic fields are produced by ion pumps, lenses, the earth and any stray ferromagnetic materials that may have been built into the apparatus by mistake. Usually their only effect is to cause misalignment between the mechanical and the electrical axis of the instrument but they can also be responsible for saturating high permeability shielding materials, thereby rendering them ineffective for shielding AC fields. At low voltage, stray electrostatic fields can also be very troublesome. Any insulator which can be 'struck' by the beam will develop a surface charge and this charge will in turn produce a field that deflects the beam. To avoid this, the column must be designed so that all insulators are shielded from the beam and great care must be taken to exclude even the most minute particles of dust or lint from the apparatus. Even the choice of materials is important because many metals commonly used for high vacuum applications such as stainless steel, Mo, Ti and A1 are normally covered with a layer of nonconductive oxide. 'The effects of surface charge accumulation on many of the metals used in electron microscopy is well described by Anger et al. (1983). Particular attention should be given to the fabrication of beam tubes and apertures and these should probably be made of acid-cleaned Pt.
Although there are instances of electrostatic pickup from nearby radio stations producing noise in the scanning circuits, most AC fields of interest to the microscopist are magnetic and are linked to the mains frequency. Their effect can be reduced by sychronizing the scan frequency to the mains which has the effect that the stray field becomes an image distortion rather than a blurring function. Even so, stray AC magnetic fields remain one of the most persistent practical problems associated with operating the SEM at low voltage, especially when smallarea, rapid-scan rates are used for focusing and astigmatism correction. To avoid them, great cafe must be taken in selecting the installation site, in making the column as short as practical and shielding it with several layers of high permeability magnetic materials, especially the gun region and the sample chamber. Several small commercial FE SEMs have been entirely enclosed in a box of shielding material. In addition, it is necessary to ensure that no stray fields are introduced to the column by currents flowing in ground loops through the equipment or by ripple on supplies feeding the scanning, stigmator or alignment coils or the field used to collect the signal electrons. These stray currents may be insignificant when the instrument operates at high voltage and only become noticeable when the magnitude of the current in the deflection coils is reduced to operate at low voltage.
The ability to detect and eliminate stray fields is crucial to successful operation of the LVSEM, but unlike the design of the microscope column, it is at least in part susceptible to improvement by the efforts of a well-informed operator. The techniques by which this may be done are discussed in more detail elsewhere (Pawley, 1984).

Problems of signal collection in the LVSEM
In most SEMs, secondary electrons produced by collisions between the beam electrons and the sample are collected by a field imposed by a grid at about + 300 V, which attracts electrons to the entrance of the scintillator/photomultiplier signal amplifier (Everhart & Thornley, 1960). This works well when the sample is 5-10 mm below the objective lens pole-piece, but as the working distance is reduced in an attempt to diminish C,, the same horizontal field at the sample surface becomes less efficient at collecting signal electrons. On the other hand, this field becomes relatively large compared with the beam energy and it therefore can produce some distortion of the beam.
There have been three strategies to overcome this problem. The simplest is to use an objective lens with a sharply conical lower pole-piece which permits the collection field to penetrate to the electron optical axis more easily (Nakagawa et al., 1982;Pomposo & Coates, 1982). This approach also permits observation of large highly-tilted flat samples but it has the disadvantage that conical lenses often have reduced electron optical properties due chiefly to flux leakage in the region where the conical pole-pieces taper together. The second method is to use the signal collection system employed in the TEM/SEM where the sample is immersed in the lens field and the low energy electrons spiral up the field lines through the hole in the upper polepiece where they are then collected by a small transverse electric field (Koike et d., 1971(Koike et d., , 1973Buchanan, 1982;Tamura et al., 1980). This system has many advantages: (1) it will work with very short focal length lenses, (2) the transverse field occurs in an area where it can be carefully controlled and is not subject to inhomogeneities produced by irregularities in specimen topography; a consideration that becomes more important on samples which are not flat semiconductors. (3) It seems to selectively exclude at least some of the low resolution Type I1 and I11 signal produced by backscattered electrons ). (4) It seems to be somewhat immune to the variations in collection efficiency caused by differences in specimen surface potential that are responsible for most simple charging artefacts. There are also some disadvantages : magnetic samples cannot be viewed and because there is no directional collection field at the sample, the 'shadowing' familiar from normal SEM micrographs is absent.
The last method is similar to that described by Volbert & Reimer (1980) and involves the use of a pair of scintillator/photomultiplier detectors, one on either side of a sample with the result that there is no field on the axis. We have used this approach with a sample immersed in the lens field (Pawley & Wall, 1982). An axial metal tube protects the beam from the effects of the collection field until just before it reaches the sample. This detector will be described further in the next section.

Fig. 5
None of these schemes represents an ideal solution in that all have the potential to degrade the beam before it reaches the sample. Their efficiency, in terms of fraction of emitted secondary electrons actually collected, has not been reported, but, of course, electrons lost at this stage can only be replaced by higher beam current and a larger spot so that this is an important parameter.

AN LVSEM TEST B E N C H
In 1977 we reported on a freeze-fracture chamber directly attached to an SEM with an LaBs source and designed so that the coated fracture surface could be directly viewed using a cold stage (Pawley & Norton, 1978). Though we had hoped that such a system would provide an image similar to that obtained from freeze-fracture TEM, we found that the resolution/contrast at 20-30 kV was insufficient to resolve even the 10-12 nm intermembrane particles normally found on fractured membrane surfaces (Pawley et al., 1980). We then tried to image an actual shadowed freeze-fracture replica suspended over a Faraday cage at room temperature (Pawley et al., 1978). Such a sample should permit very high resolution SEM imaging because, as the sample is so thin, the Type I1 and I11 signal are almost absent. However, images of the replicas showed no trace of the particles. Indeed the signal from the replica was very small altogether, about 5% of that on a solid metal surface and it was to this low signal level that we attributed our failure. The only possible solution seemed to be to go to lower beam voltage and as there was no high resolution LVSEM equipment commercially available at that time, we began a modest programme to develop a prototype instrument in 1980. This instrument was designed to overcome some of the problems discussed above and to produce a 1-2 nm probe at 1 kV in order to determine whether or not images made with it were superior to those made at higher voltage (Pawley & Wall, 1982;Pawley & Winters, 1983).

Design
A diagram of the present version of this instrument, and photos of the entire assembly and the column itself are shown in Fig. 6. The electron source is a cold FE cathode, using single crystal tungsten in the (l,l,l) orientation (F.E.I., Hillsboro, Oregon) and the gun is pumped from above with a 220 I/s ion pump. It has been designed to be rigid, compact and well-shielded from internal and external magnetic fields. The gun block has side ports for a window and to feed through both high voltage and heater currents. It is machined from a single block of stainless steel to avoid the possibility that the welds might become ferromagnetic and the cylindrical part of the first anode is made of mu-metal. The anode itself is made of a thin Mo foil and can be heated by radiation and electron bombardment from a tungsten filament located below it to speed out-gassing. The gun isolation valve and the movable, three-position, aperture are bellows-sealed into the lower part of the gun block. Below the aperture the beam enters a platinum vacuum liner tube outside of which are situated stigmator, alignment and doubledeflection scanning coils and also a small condensor lens. The beam tube is brazed to the specimen chamber which has side-ports for two scintillator/photomultiplier secondary electron detectors, each employing a single crystal Yttrium Aluminium Garnet (Ce) hemispherical scintillator (Pawley, 1974;Autrata et al., 1978Autrata et al., , 1983, a 20 l/s, water-cooled ion pump and the controls for a specimen stage holding two 3 mm grids. A viton-sealed port on the bottom permits specimen exchange. The objective lens is of the pancake type as described by Mulvey (1982). It is excited by a 225 turn tape winding and cooled by laminar water flow past the bottom of the lower lens pole plate. The pole tip radius is 1 mm and the calculated lens parameters are: f = 0 5 4 mm, Cc=0.22 mm. Mechanical alignment of the tip is performed by adjusting set screws in the spider which holds the filament assembly while looking up the axis at the tip. After tip alignment, all of the gun components are clamped rigidly together by the upper threaded ring. All other components are prealigned and clamped by bolts to the gun block except for the objective lens which can be translated f 1 mm, X and Y.
The electronics are a modified version of those supplied for an AMRay 1200 microscope (AMRay, Bedford, Mass.). Separate controls are provided for the acceleration and PMT voltage of each electron detector. The entire column is surrounded by a 400 mm diam. alloy cylinder which is lined with 1 mm thick mu-metal and to which is attached part of the isolation valve mechanism. The cylinder and the column hang from the gun ion pump which is supported on a large steel plate. This is, in turn, suspended from a steel frame using a set of pulleys and a total of twenty-two thicknesses of elastic cordage which provides vibration isolation (vertical resonant frequency = 2 Hz).

Operation
Initially, very sharp cathodes were used to permit operation with V1= VO= 1 kV. However after a few hours of operation these tips would fail and so they were replaced by tips operating at V1= 3 kV for a tip current of 20 pA.
The apertures used are carefully cleaned and then coated with evaporated gold immediately before use. A 1000 pm aperture is used for the coarse set-up and 100 pm is used when operating the condensor lens only. Though the lower lens produces approximately the expected magnetic field (2-3 kgauss, on axis 75 pm above tip at 5 A), we have not yet operated it successfully as part of the microscope. To begin with, the level of the signal is very low when the lens is excited and this is true even when the collection field is increased by applying 12 kV to only one scintillator so that there is no null point on the axis. Secondly, the beam enters the fringe field well before it reaches the sample (7 gauss at the distance of the FE cathode) and this pre-field is so strong that the optical properties of the lens seem not to be as calculated. Similar problems were encountered by Hill & Smith (1982) when they used a similar lens in a conventional SEM.
Unfortunately, no attempt to achieve high resolution images can be made with this instrument until this problem is overcome. Therefore we plan to modify the lens by adding an upper pole piece, thereby making it more similar to the lens used in the TEM/SEM (Fig. 6). The calculated aberration constants of this lens are somewhat larger than those of the present lens but we still hope to obtain a beam of 2.5 nm at 1 kV.

OTHER LIMITATIONS TO HIGH RESOLUTION PERFORMANCE
Spatial resolution in the topographic image from the SEM is so central to one's assessment of the instrument's capabilities it has been much studied and discussed (Oatley et al., 1965;pease & Nixon, 1965;Wells, 1974a, by Broers, 1982Catto & Smith, 1973;Watabe et al., 1978;peters, 1979. AS has been mentioned above, many of these analyses, when evaluating the h a 1 images, have mistaken the topographically modified Z-contrast produced by the coating for true topographic contrast. Catto & Smith (1973) avoid this but their theoretical analysis deals strictly with the information theory aspects of the beam/specimen interaction given certain ideal conditions. Their analysis uses basic electron scattering theory to calculate the signal-to-noise ratio of the signal from small Gaussian-profile asperities on a solid gold sample versus the radius of and distance between these asperities, The analysis is performed at 10, 20 and 30 kV and assuming a probe diameter of 0, 0.5 nm and 5 nm. Not surprisingly, their results show the best performance at the lowest voltage where a 1.0 nm asperity should just be visible using a 0.5 nm beam.
Comforting though it is to know that such resolution is not impossible from the point of view of scattering and information theory, it should be kept in mind that there are several practical limitations to actually obtaining this performance on real samples besides probe diameter and current. Specifically these include (1) surface contamination, (2) radiation damage, (3) the delocalized nature of inelastic scattering, and (4) beam tailing.

Contamination
Layers of organic contamination accumulate on surfaces subjected to electron beam bombardment and the problem is more severe when small, high-current probes are used (Fourie, 1981). The presence of such a film is much more noticeable in the secondary electron image if a low Vo is used. Figures 7(a) and 7(b) show two high magnification micrographs of the same area of Type I1 cell on a lung alveolar wall. These micrographs were intended as a stereo pair but the surface is seen to be obscured in Fig. 7(b). The contaminated area can be distinguished at lower magnification in Fig. 7(c), made with Vo= 1 kV though it is not evident with V O = 10 kV in Fig. 7(d).
It has been assumed that any effort to produce the ultimate in topographic resolution will entail a FE source and therefore an oil-free, generally bakable, vacuum system where these problems would be less serious. In such an instrument the sample itself becomes the major source of contamination. Even using the cleanest possible grids and support films, a layer of contamination rapidly builds up on biological samples unless they and their surroundings are cooled sufficiently (about -6O'C) to arrest the process of surface diffusion (Wall et al., 1977;Voreades & Wall, 1979). Therefore any effort to obtain high resolution surface images from organic materials. rather than from a metal such as gold, will probably require a cooled sample and lens assembly.

Radiation damage
The kilovolt electrons impinging on an SEM sample undergo inelastic collisions that may result in the transfer of more than a few electron volts of energy. As such, they constitute an intense source of ionizing radiation. The damage caused by this interaction to covalently bonded samples viewed in the TEM has been widely studied (Glaeser, 1971(Glaeser, , 1975Cosslett, 1978) and found to seriously limit structural information retrieval below 2 nm. The situation is even more serious in the SEM because the entire beam energy is absorbed in the sample. Even with Vo= 1 kV and a 10-11 A beam the power of the beam is I Vo= W. If we assume that one half of this energy is absorbed in the upper 10 nm of a sample with density 1, scanned with a raster 1000 nm on a side, the dose rate, D , is where 1 rad=10-5 J depositedg. This is a very high dose rate and, assuming a 100 s scan, it is more than 103 times that common in the low-dose T E M studies designed to preserve structures below 2 nm. It is reasonable to assume that biological samples exposed to this flux of ionizing radiation will be rapidly reduced to a highly traumatized carbon skeleton of the original structure. The image obtained will be an image of ashes and the relation it bears to the original structure will be unknown. Certainly any structure of less than 2-3 nm should be initially treated very sceptically. On the other hand, the acceptable level of damage depends on the end-point. The 100 electronsInm2, thought acceptable for low dose TEM, is far more than that required to inactivate all enzymes while gross molecular shape is sometimes preserved at much higher doses (Ottensmeyer et al., 1978). As we are not seeking atomic resolution, it is not unreasonable to expect that the original number and position of specific features in the original sample may be retained as lumps on the surface of the ash and, of course, the situation is less severe if the sample is really a thin metal coat covering the organic material of interest.

De-localization of the inelastic scattering event
The secondary electrons that provide the SEM signal are produced by inelastic collisions with electrons in the sample. This process is not highly localized in that it can occur when the probe electron passes some nm away from the electron being excited (Isaacson & Langmore, 1974). Barth & Poole (1976) point out that, as the delocalization is proportional to Vo, better results are to be expected at low voltage. In particular, they predict that 0.6 nm localization should be possible at 1 kV under somewhat optimistic experimental circumstances (G, C$= 0.07 mm, A V S 0.2 V, a = 4 x 10-2 rad).
Monte Carlo calculations of electron scattering in solids usually consider only a small subset of the known interactions such as archetypical elastic and inelastic collisions assumed to occur at highly localized points along the trajectory of the impinging electrons. Recently these calculations have been improved, and the predicted spatial resolution reduced, by including the effects of fast secondaries (Murata et al., 1981;Kyser, 1981;Joy et al., 1982;Joy, 1984) but many other interactions exist which might be important when the results must be applicable to a size scale of a few nm. One such interaction might be the production of Bremsstrahlung or braking radiation which is produced whenever a charged particle decelerates.
Though the cross-section for Bremsstrahlung production is small, in the range 1-10 keV, it increases exponentially with 1,'Vo. Interactions producing 10-1000 eV photons are usually ignored because these photons are absorbed so strongly by the sample that very few of them leave it. Their generation is of interest here because their absorption in the sample can result in the production of a secondary electron a few nm further away from the beam than would have been the case if only electron-electron interactions were considered. We are unaware of quantitative data at these voltages and on this size scale that would permit an accurate estimate of the size of this effect, but it could be an important factor.
As all these effects act independently from all the electron optical blurring functions, they should reduce the actual point-t3-point resolution by at least an additional 0.5-1.0 nm from that theoretically calculated from electron optical and electron scattering considerations and possibly more.

Beam tailing
When speaking of the beam diameter of a focused spot, it is customary to assume that the current density resembles a Gaussian distribution or an Airey disk and to refer to its diameter at half maximum or the distance over which the intensity drops from 80°,, to IOO,, of the peak value. As has been pointed out by Cliff & Kenway (1982), beams in probe-forming instruments are often non-gaussian for various reasons, and in particular, they often have a small central peak surrounded by a much wider 'shadow' of lower but significant intensity. This shadow greatly complicates the criteria for visibility as measured by Catto & Smith (1973). When discussing small probes it is essential to keep in mind the fraction of the total current actually in the central spot.

T H E P R O S P E C T S FOR T O P O G R A P H I C I M A G I N G I N T H E L V S E M
In light of all these problems, what performance is it reasonable to expect under the best possible conditions and what are these conditions ? As discussed above, the ideal microscope should employ a low-current cold-field-emission source and a lens with short focal length which also permits collection of the secondary electron signal. Beyond that, there are many theoretical advantages of operating it at liquid helium temperature. Not only is surface contamination negligible but superconducting materials are also perfect shields against stray electromagnetic fields (Dietrich et al., 1977). Furthermore, the low frequency noise and energy spread of the gun would be somewhat less. Though the primary ionizing event that produces radiation damage would not be eliminated, and chemical reactions would still occur following the accumulation of sufficient beam-produced free radicals, it is still probable that many low molecular weight species produced by the interaction might remain frozen nearly in place at these temperatures. This would not preserve molecular integrity but, to some extent, the lumps would not move.
One disadvantage, aside from considerable complexity, might be increased charging artefacts. Even semiconductors become insulators at these temperatures and it might be necessary to lightly coat all samples. Furthermore, trace amounts of residual gas could create unwanted insulators if they became frozen onto sensitive surfaces.
With such an instrument it would seem that uncoated, cubic, surface features of low Z material as small as 3 nm on a side might be detectable on a flat solid surface as long as they were not destroyed by radiation damage. Two such objects could be distinctly imaged if separated by about 5 nm centre to centre. Information from lightly coated (1-1.5 nm) samples might be somewhat better, assuming the coating material was chosen for low secondary electron meanfree-path (Everhart, 1970) and that we are now referring to the size of the surface features after coating. Though this is the same size range that is covered by the best replica techniques, it is important to remember the benefits of directly imaging the actual sample. Oatley (1982) points out that in the early 1960s commercial introduction of the SEM was delayed by the logic that, as replica techniques had higher resolution, there would be no market for the SEM.
This analysis failed to take into account the extent to which specimen preparation is simplified and the areas of possible application increased by avoiding the necessity of having to produce a replica. Because of a willingness to accept the SEM's lower resolution in order to be able to examine the surface of a larger and more convoluted sample, the instrument came into common use and it has now been improved to the point where it may no longer even have to defer to the replica techniques in terms of resolution. When this happens it will be a considerable achievement.
Apart from ultimate performance at the limit of topographic imaging, important improvements in the low voltage performance of most current instruments should lead to their increased use in the 2-5 kV range by a wide variety of users (Fig. Ib). The important test should be whether or not going to a higher voltage produces more information about the topography of a sample or merely sharper pictures of the metal grains of the coating material.

A C K N O W L E D G M E N T S
Thanks are due to the AMRay Corporation (Bedford, Mass.) and to the Wisconsin Alumni Research Foundation for supplying the electronics and vacuum pumps (respectively) used in building the LVSEM test bench and to the National Institutes of Health for support through Grant RR-00570-14 to the Madison HVEM Biotechnology Resource. Personal thanks are also due to Dr J. Wall (Brookhaven National Laboratory) for his generous assistance and encouragement in the design of the test bench and to Mr M. Winters, without whom it would not have been completed. Dr Autrata (Institute of Scientific Instruments, Brno) kindly provided the single crystal YAG scintillators and Dr E. deHarven (University of Toronto) and Dr J. Bastacky Krewson for their patience with the manuscript and to Dr D. Joy who read the manuscript and showed me a preprint of his recent paper.