Ultramlcroscopy 47 (I 992) 80-100 North-Holland
IIH'lP~ltni~l
High-resolution scanning electron microscopy D a v i d C J o y 1,2 a n d J a m e s B P a w l e y 3 1 E M Facthty, Untt~erslty of Tennessee, Knox,,die, TN 37996.0810, USA 2 Oak Rtdge National Laboratory, Oak Rtdge, TN 37831, USA S lMR, Unwerstty of Wtsconsm, 1675 Observatory Drue, Madison, WI 53706, USA Received at Editorial Office 11 May 1992
The spatial resolution of the scanning electron microscope is limited by at least three factors the diameter of the electron probe, the size and shape of the b e a m / s p e c i m e n interaction volume with the solid for the mode of imaging employed and the Polsson statistics of the detected signal Any practical consideration of the high-resolution performance of the SEM must therefore also revolve a knowledge of the contrast available from the signal producing the image and the radiation sensalvlty of the specimen With state-of-the-art electron optics, resolutions of the order of 1 n m are now possible The optimum conditions for achieving such performance with the m l m m u m radiation damage to the specimen correspond to beam energies in the range 1-3 keV Progress beyond this level may be restricted by the delocahzatlon of SE production and ultimate limits to electron-optical performance
1. Introduction The scanning electron microscope (SEM) builds up an tmage by sampling contiguous subvolumes near the surface of the specimen A fine electron beam selectively excites each sub-volume and then the intensity of some resulting stgnal ts measured The spatml resolution of images made using such a process is limited by at least three factors Two of these determine the size of the interaction volume the s~ze of the electron probe [1] and the extent to which detectable s~gnal is excited from locations remote from the beam impact point [2,3] A third hmltatlon emerges from the fact that the probing beam is composed of a finite number of discrete particles and therefore that the accuracy with which any detectable signal can be measured is limited by Polsson statistics apphed to this number (or to the number of events actually detected if this is smaller) [4] As in other microscopical techniques, the hmitmg signal contrast required to recognize a morphological structure ~s constrained by this statistical consideration [5] The only way to overcome it is to increase either the contrast of the
measured signal [6] or the number of b e a m / specimen interactions detected [7] Unfortunately, these interactions involve ionizing radiation that may damage the very structure under investigation As a result, any practtcal consideration of the htgh-resolutlon performance of the SEM must consider not only the size of the interaction volume but also the contrast available from the signal producing the image and the radiation sensitivity of the specimen [8] This ltmltaUon is even more important in SEM than it ts in TEM where the image acquisition is parallel rather than serial, and where the image contrast mechamsm is coherent with the result that spatial averaging can be used to improve the statlsttcal accuracy of the image In the two decades following the introduction of the SEM, the main hmit on the size of the interaction volume was set by electron-optical considerations that prevented the production of a small electron beam [9] Recent improvements in the design of SEM columns, particularly the use of both short-focal-length lenses [10] and fieldemission electron sources [11] in a single mstrument [12], have made it posstble to produce probes
0304-3991/92/$05 00 © 1992 - Elsevier Science Publishers B V All rights reserved
D C Joy, J B Pawley / Htgh-resolunon scannmg electron microscopy
containing useful amounts of current down to 0 5 nm at 30 keV and 3 nm at 1 5 keV [13] An analysis of the secondary electron (SE) imaging results possible on robust, electron-transparent, inorganic specimens in the high-voltage range ( ~ 100 keV) are discussed elsewhere in this volume [14] Here, we will concentrate on the theoretical and practical limitations that affect the spatial resolution obtainable in topographic images made at lower b e a m voltages (1-30 keV) on bulk specimens that are less tolerant of ionizing radiation Though in many ways the perfect excitation source for scanned image microscopy, an electron b e a m also has a number of disadvantages Because obtaining the optimal image of the specimen is often dependent on avoiding the hmitations associated with these disadvantages, we will begin with a brief consideration of them We will limit our consideration to contrast mechanisms that affect either the SE or the backscattered electron (BSE) signal because only these are produced in sufficient quantity to satlsfy the statistical constraints without producing excessive radiation damage
2. Limitations associated with the use of electrons as the probing radiation
Electrons are a convenient type of excitation for scanned probe microscopy because (1) Sources which are mono-energetlc, high-intensity, inexpensive, and of short wavelength, are readily available, (2) they can be focused, and scanned, with relative ease and (3) they interact efficiently with solids producing a number of outcomes that are useful, interesting and easily detectable However, they also have some disadvantages that are not associated with other focusable quanta such as Ions [15,16], light photons [17], or X-rays [18,19] Though so well known and accepted that their existence often passes without comment, these disadvantages are worth considering more explicitly because the limitations that they place on resolution in the SEM are just as
81
important as that associated with probe size Furthermore, it is because many of these effects vary strongly with b e a m voltage (V 0) that this parameter becomes the major determinant of SEM performance [20]
3. The finite size of the interaction volume
The crucial factors affecting SEM resolution were debated in the discussions following McMullen's first p a p e r in 1953 [21] Obviously, the primary limitation on resolution is the size of the probe, and McMullen discussed the factors that limited its diameter at that time However, Halne and G a b o r claimed that "electron diffusion" (1 e scattering within the specimen) would eventually place a more fundamental limit on SEM resolution In a scanning probe instrument such as the laser confocal fluorescence microscope [17] in which the detected interaction occurs between a 1 - 2 eV photon from the b e a m and a single atom or molecule of the specimen, the interaction volume is essentially zero and the spatial resolution IS almost totally defined by the size of the exciting probe However, when the probe is a b e a m of kilovolt electrons striking a bulk specimen, a large number of inelastic collisions are required to absorb the energy of each b e a m electron Between each of these inelastic collisions, elastic collisions produce large changes in trajectory causing the b e a m to spread out As a result, the absorption of the energy from one b e a m electron involves a large number of e l e c t r o n / s p e c i m e n interactions, occurring over a significant region of space When the b e a m diameter IS small compared to this volume, the spatial resolution of the image produced is limited more by the dimensions of the volume within which detectable interactions take place than by the b e a m diameter Because of this complication, there is no standard method of measuring SEM resolution and it has therefore always been an elusive concept [22,23] Unlike the TEM, where measurement of the contrast transfer function of a thin-phase object provides at least a standard method and a useful basis for the comparison of Instrumental
D C Joy, J B Pawley / High-resolution scannmg electron microscopy
82
performance [24], the search for an ideal, bulk, SEM test specimen has been long [25-29] but unsuccessful Most high-resolution test specimens used today produce contrast by changes in specimen composition or signal collection efficiency rather than by changes in topography [30] As a result, even when Fourier transform techniques are employed to analyze the image [31-33], the result is only an estimate of the beam dmmeter rather than the size of the smallest topographic feature that might be visible on the surface of a flat, bulk specimen (See refs [29,34-37] for possible exceptions ) The discussion that follows refers only to bulk specimens and not to specimens that have been kept so thin that subsurface scattering has been eliminated by virtue of the fact that there is no subsurface [38] In this case the interaction volume is essentially defined by the area of the probe times the thickness of the specimen [39] For bulk specimens, the problem is that, for most detectable b e a m / s p e o m e n interactions, some of the signal is produced by interactions taking place at some distance from the area of the specimen initially struck by the primary beam [5,40] The secondary electron signal IS comprised not only of SE produced where the beam strikes the specimen (SE1), but also of SE excited by BSE as they re-emerge through the surface (SE2)
1 5keV
5keV
or strike the lens poleplece (SE3) As a result, the signal from the SE detector always has a signific a n t BSE-mduced component of from 10% to 50% if the mean atomic number of the target is high [35,41-44] The m o m e n t u m of the probe electrons carries them into the specimen, exciting a volume within the specimen The total volume excited is roughly hemispherical, while that in which most of the energy is absorbed is smaller and more pearshaped The exact shape can only be discussed from the standpoint of an energy-deposition probability distribution and in terms of the way that the rate of energy disposition decreases with radial distance from the beam impact point The shape and size of this volume have been thoroughly investigated for simple geometries using Monte Carlo simulation techniques [6,20,22,4549] Fig 1 emphastzes the great reduction in this volume with beam energy While, at 1 keV on a Pt specimen, the dimensions of this interaction volume may be only a few tens of nm, these dimensions scale with V ~ / ~ / p [39,50] (where p = density) and may reach hundreds of /zm at 30 keV on samples such as dried lung (p = 005 g / c m 3) Fig 2 shows how the diameter of the area from which BSE emerge varies with V o on a bulk specimen At best, on relatively flat specimens, signals derived from interactions remote
20keV t
t
-
2p~m
Fig 1 Monte Carlo plots of electron scattering m carbon at nominal density of 1 g/cm 3 and for beam energies Vo = 1 5, 5 and 20 keV
D C Joy, J B Pawley / High-resolution scannmg electron microscopy
83
101
4. The characteristics of electron lenses
10 9
All available electron lenses are converging and hence, m practice, the effect of lens aberrations can only be hmtted by reducing the lens aperture angle The dominant aberrations are chromatic and spherical The former produces a blurred spot of diameter
~--" 10 8
o~
f. 10 7 ,
IIH'lP~ltni~l
High-resolution scanning electron microscopy D a v i d C J o y 1,2 a n d J a m e s B P a w l e y 3 1 E M Facthty, Untt~erslty of Tennessee, Knox,,die, TN 37996.0810, USA 2 Oak Rtdge National Laboratory, Oak Rtdge, TN 37831, USA S lMR, Unwerstty of Wtsconsm, 1675 Observatory Drue, Madison, WI 53706, USA Received at Editorial Office 11 May 1992
The spatial resolution of the scanning electron microscope is limited by at least three factors the diameter of the electron probe, the size and shape of the b e a m / s p e c i m e n interaction volume with the solid for the mode of imaging employed and the Polsson statistics of the detected signal Any practical consideration of the high-resolution performance of the SEM must therefore also revolve a knowledge of the contrast available from the signal producing the image and the radiation sensalvlty of the specimen With state-of-the-art electron optics, resolutions of the order of 1 n m are now possible The optimum conditions for achieving such performance with the m l m m u m radiation damage to the specimen correspond to beam energies in the range 1-3 keV Progress beyond this level may be restricted by the delocahzatlon of SE production and ultimate limits to electron-optical performance
1. Introduction The scanning electron microscope (SEM) builds up an tmage by sampling contiguous subvolumes near the surface of the specimen A fine electron beam selectively excites each sub-volume and then the intensity of some resulting stgnal ts measured The spatml resolution of images made using such a process is limited by at least three factors Two of these determine the size of the interaction volume the s~ze of the electron probe [1] and the extent to which detectable s~gnal is excited from locations remote from the beam impact point [2,3] A third hmltatlon emerges from the fact that the probing beam is composed of a finite number of discrete particles and therefore that the accuracy with which any detectable signal can be measured is limited by Polsson statistics apphed to this number (or to the number of events actually detected if this is smaller) [4] As in other microscopical techniques, the hmitmg signal contrast required to recognize a morphological structure ~s constrained by this statistical consideration [5] The only way to overcome it is to increase either the contrast of the
measured signal [6] or the number of b e a m / specimen interactions detected [7] Unfortunately, these interactions involve ionizing radiation that may damage the very structure under investigation As a result, any practtcal consideration of the htgh-resolutlon performance of the SEM must consider not only the size of the interaction volume but also the contrast available from the signal producing the image and the radiation sensitivity of the specimen [8] This ltmltaUon is even more important in SEM than it ts in TEM where the image acquisition is parallel rather than serial, and where the image contrast mechamsm is coherent with the result that spatial averaging can be used to improve the statlsttcal accuracy of the image In the two decades following the introduction of the SEM, the main hmit on the size of the interaction volume was set by electron-optical considerations that prevented the production of a small electron beam [9] Recent improvements in the design of SEM columns, particularly the use of both short-focal-length lenses [10] and fieldemission electron sources [11] in a single mstrument [12], have made it posstble to produce probes
0304-3991/92/$05 00 © 1992 - Elsevier Science Publishers B V All rights reserved
D C Joy, J B Pawley / Htgh-resolunon scannmg electron microscopy
containing useful amounts of current down to 0 5 nm at 30 keV and 3 nm at 1 5 keV [13] An analysis of the secondary electron (SE) imaging results possible on robust, electron-transparent, inorganic specimens in the high-voltage range ( ~ 100 keV) are discussed elsewhere in this volume [14] Here, we will concentrate on the theoretical and practical limitations that affect the spatial resolution obtainable in topographic images made at lower b e a m voltages (1-30 keV) on bulk specimens that are less tolerant of ionizing radiation Though in many ways the perfect excitation source for scanned image microscopy, an electron b e a m also has a number of disadvantages Because obtaining the optimal image of the specimen is often dependent on avoiding the hmitations associated with these disadvantages, we will begin with a brief consideration of them We will limit our consideration to contrast mechanisms that affect either the SE or the backscattered electron (BSE) signal because only these are produced in sufficient quantity to satlsfy the statistical constraints without producing excessive radiation damage
2. Limitations associated with the use of electrons as the probing radiation
Electrons are a convenient type of excitation for scanned probe microscopy because (1) Sources which are mono-energetlc, high-intensity, inexpensive, and of short wavelength, are readily available, (2) they can be focused, and scanned, with relative ease and (3) they interact efficiently with solids producing a number of outcomes that are useful, interesting and easily detectable However, they also have some disadvantages that are not associated with other focusable quanta such as Ions [15,16], light photons [17], or X-rays [18,19] Though so well known and accepted that their existence often passes without comment, these disadvantages are worth considering more explicitly because the limitations that they place on resolution in the SEM are just as
81
important as that associated with probe size Furthermore, it is because many of these effects vary strongly with b e a m voltage (V 0) that this parameter becomes the major determinant of SEM performance [20]
3. The finite size of the interaction volume
The crucial factors affecting SEM resolution were debated in the discussions following McMullen's first p a p e r in 1953 [21] Obviously, the primary limitation on resolution is the size of the probe, and McMullen discussed the factors that limited its diameter at that time However, Halne and G a b o r claimed that "electron diffusion" (1 e scattering within the specimen) would eventually place a more fundamental limit on SEM resolution In a scanning probe instrument such as the laser confocal fluorescence microscope [17] in which the detected interaction occurs between a 1 - 2 eV photon from the b e a m and a single atom or molecule of the specimen, the interaction volume is essentially zero and the spatial resolution IS almost totally defined by the size of the exciting probe However, when the probe is a b e a m of kilovolt electrons striking a bulk specimen, a large number of inelastic collisions are required to absorb the energy of each b e a m electron Between each of these inelastic collisions, elastic collisions produce large changes in trajectory causing the b e a m to spread out As a result, the absorption of the energy from one b e a m electron involves a large number of e l e c t r o n / s p e c i m e n interactions, occurring over a significant region of space When the b e a m diameter IS small compared to this volume, the spatial resolution of the image produced is limited more by the dimensions of the volume within which detectable interactions take place than by the b e a m diameter Because of this complication, there is no standard method of measuring SEM resolution and it has therefore always been an elusive concept [22,23] Unlike the TEM, where measurement of the contrast transfer function of a thin-phase object provides at least a standard method and a useful basis for the comparison of Instrumental
D C Joy, J B Pawley / High-resolution scannmg electron microscopy
82
performance [24], the search for an ideal, bulk, SEM test specimen has been long [25-29] but unsuccessful Most high-resolution test specimens used today produce contrast by changes in specimen composition or signal collection efficiency rather than by changes in topography [30] As a result, even when Fourier transform techniques are employed to analyze the image [31-33], the result is only an estimate of the beam dmmeter rather than the size of the smallest topographic feature that might be visible on the surface of a flat, bulk specimen (See refs [29,34-37] for possible exceptions ) The discussion that follows refers only to bulk specimens and not to specimens that have been kept so thin that subsurface scattering has been eliminated by virtue of the fact that there is no subsurface [38] In this case the interaction volume is essentially defined by the area of the probe times the thickness of the specimen [39] For bulk specimens, the problem is that, for most detectable b e a m / s p e o m e n interactions, some of the signal is produced by interactions taking place at some distance from the area of the specimen initially struck by the primary beam [5,40] The secondary electron signal IS comprised not only of SE produced where the beam strikes the specimen (SE1), but also of SE excited by BSE as they re-emerge through the surface (SE2)
1 5keV
5keV
or strike the lens poleplece (SE3) As a result, the signal from the SE detector always has a signific a n t BSE-mduced component of from 10% to 50% if the mean atomic number of the target is high [35,41-44] The m o m e n t u m of the probe electrons carries them into the specimen, exciting a volume within the specimen The total volume excited is roughly hemispherical, while that in which most of the energy is absorbed is smaller and more pearshaped The exact shape can only be discussed from the standpoint of an energy-deposition probability distribution and in terms of the way that the rate of energy disposition decreases with radial distance from the beam impact point The shape and size of this volume have been thoroughly investigated for simple geometries using Monte Carlo simulation techniques [6,20,22,4549] Fig 1 emphastzes the great reduction in this volume with beam energy While, at 1 keV on a Pt specimen, the dimensions of this interaction volume may be only a few tens of nm, these dimensions scale with V ~ / ~ / p [39,50] (where p = density) and may reach hundreds of /zm at 30 keV on samples such as dried lung (p = 005 g / c m 3) Fig 2 shows how the diameter of the area from which BSE emerge varies with V o on a bulk specimen At best, on relatively flat specimens, signals derived from interactions remote
20keV t
t
-
2p~m
Fig 1 Monte Carlo plots of electron scattering m carbon at nominal density of 1 g/cm 3 and for beam energies Vo = 1 5, 5 and 20 keV
D C Joy, J B Pawley / High-resolution scannmg electron microscopy
83
101
4. The characteristics of electron lenses
10 9
All available electron lenses are converging and hence, m practice, the effect of lens aberrations can only be hmtted by reducing the lens aperture angle The dominant aberrations are chromatic and spherical The former produces a blurred spot of diameter
~--" 10 8
o~
f. 10 7 ,
