Dark-field imaging based on post-processed electron backscatterdiffraction patterns of bulk crystalline materials in a scanningelectron microscope

Nicolas Brodusch n, Hendrix Demers, Raynald GauvinMcGill University, Mining and Materials Engineering Department, Montréal, Québec, Canada H3A 0C5

a r t i c l e i n f o

Article history:Received 8 July 2014Received in revised form8 September 2014Accepted 21 September 2014Available online 22 October 2014

Keywords:Dark-field (DF)Electron channeling contrast imaging (ECCI)Electron channeling pattern (ECP)Electron backscatter diffraction (EBSD)DeformationScanning electron microscope (SEM)

a b s t r a c t

Dark-field (DF) images were acquired in the scanning electron microscope with an offline procedurebased on electron backscatter diffraction (EBSD) patterns (EBSPs). These EBSD-DF images weregenerated by selecting a particular reflection on the electron backscatter diffraction pattern and byreporting the intensity of one or several pixels around this point at each pixel of the EBSD-DF image.Unlike previous studies, the diffraction information of the sample is the basis of the final image contrastwith a pixel scale resolution at the EBSP providing DF imaging in the scanning electron microscope. Theoffline facility of this technique permits the selection of any diffraction condition available in thediffraction pattern and displaying the corresponding image. The high number of diffraction-basedimages available allows a better monitoring of deformation structures compared to electron channelingcontrast imaging (ECCI) which is generally limited to a few images of the same area. This technique wasapplied to steel and iron specimens and showed its high capability in describing more rigorously thedeformation structures around micro-hardness indents. Due to the offline relation between thereference EBSP and the EBSD-DF images, this new technique will undoubtedly greatly improve ourknowledge of deformation mechanism and help to improve our understanding of the ECCI contrastmechanisms.

& 2014 Elsevier B.V. All rights reserved.

1. Introduction

In transmission electron microscopy, a dark-field (DF) image isobtained when a specific diffraction reflection is excited. This canbe accomplished either in the conventional transmission electronmicroscope (CTEM) or the scanning transmission electron micro-scope (STEM) by collecting the signal from the diffracted beamcorresponding to the selected reflection. Practically, this is per-formed by placing an aperture (CTEM) or selecting a particularcollection angle (STEM) to collect electrons scattered through theBragg angle corresponding to the specific lattice planes selected.

In the scanning electron microscope (SEM), DF imaging can beachieved in STEM mode as in a dedicated STEM when thin speci-mens are used [1]. However, no DF imaging has been reported onbulk specimens. Only electron channeling contrast imaging (ECCI)provides a DF type contrast based on the electron channeling

pattern (ECP). This Kikuchi-like pattern is an angular distributionof the backscattered electron (BSE) yield obtained when theprimary electron beam is scanned over a large specimen area orrocked around the optic axis of the microscope at a specific point ofthe specimen surface [2]. In fact, when the angle between the beamand the lattice planes is close to the Bragg angle, the BSE yield isproportional to the probability of the electron to be backscatteredinside the matter in addition to the Z2 dependence of Rutherfordscattering. This probability is based on the Bloch wave theory [3]and is the square of the coefficient of a Bloch wave contributiondivided by the sum of all the square of each Bloch wave contribu-tion. To simplify the calculations, only two Bloch waves aregenerally used to describe this probability. Bloch wave of type Ihas its maximum at the atom sites while type II has its maximabetween the atom rows. At the exact Bragg angle the contributionof the two Bloch waves is equal. When the scan angle is small,typically at magnification higher than 100 times [4], the intensity ateach pixel of the ECCI image is equal to that at the centre of thecorresponding ECP at the same pixel position. This acts like a virtualaperture at the centre of the ECP. However, this virtual aperture,used to select the channeling pattern (ECP) area for the signal

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/ultramic

Ultramicroscopy

http://dx.doi.org/10.1016/j.ultramic.2014.09.0050304-3991/& 2014 Elsevier B.V. All rights reserved.

n Correspondence to: Mining and Materials Engineering Department, WongBuilding, McGill University, 3610 University Street, Montréal, Québec, CanadaH3A 0C5. Tel.: þ1 514 398 7182; fax: þ1 514 398 4492.

E-mail address: nicolas.brodusch@mcgill.ca (N. Brodusch).

Ultramicroscopy 148 (2015) 123–131

collection, is only limited to the centre of the ECP [4]. Then, thespecimen needs to be tilted and rotated to change the region of theECP that will be located at the centre point.

Payton and Nolze reported the use of diodes on top of the EBSDcamera combined to EBSD scan to improve phase identification [5]following the work initiated by Prior et al. on using semi-conductordiodes attached to the EBSD camera [6]. Previous studies were alsoreported where large areas of the EBSPs were selected to reconstructthe image from an EBSD scan [7,8] and a similar technique wasrecently commercialized by EDAX researchers [9] during the courseof our study, named PRIAS for Pattern Region of Interest AnalysisSystem. However, in these techniques, because of the large regions ofthe EBSPs used to reconstruct an image, the intensity of one region isthe average of several different diffraction conditions (area in theEBSP) and the original diffraction information present in the EBSPs isthus lost. In this work, we report on an innovating technique thatprovides controlled DF imaging in the SEM. At each pixel, thereported intensity is related to a specific diffraction condition witha better angular accuracy contrary to the previous works cited above.This technique is based on the post-processing of electron back-scatter diffraction patterns (EBSPs) and may open interesting appli-cations in the SEM.

2. Materials and instrumentation

The Si specimen used in this work was a 1�1 cm² [001] (001)silicon wafer prepared by the cleavage technique. The iron samplewas cut from a polycrystalline pure iron rod with a diameter of7.6 mm (Alfa Aesar, Ward Hill, MA, USA). The small disc was annealedin vacuum at 800 1C for 24 h followed by slow cooling in the furnace.The thickness of the sample was reduced by 4.7% using a compressionmachine. The compressed sample was then ground with 800 and1200 grit papers, followed by polishing using 3 and 1 mm diamond

suspensions with etching steps (5% and 2% nitric acid in water) inbetween. The sample was then electro-polished in 20% perchloric acidand 80% ethanol for 5 min (30 s interval) at room temperature. A finalpolishing step was manually performed with a 50 nm colloidal silicasuspension. Indentations were carried out using a Clark Microhard-ness Tester CM-1000AT (Sun-Tec Corporation, Novi, MI, USA), with apeak load of 50 g. The steel sample was a stress relief annealed non-oriented Si–Fe electrical steel (NOES) cut in a 1.5�1.5 cm2, groundwith 800 and 1200 grit papers and polished with 3 and 1 mmdiamond suspensions. Final polishing was performed with a 50 nmcolloidal silica suspension for 1 h followed by Arþ ion beam millingusing a Hitachi IM3000 flat milling system (Hitachi High-Technolo-gies, Rexdale, Canada). The accelerating voltage was 3 kV and theangle of incidence in regard to the surface normal was 801.

ECPs were recorded with a Hitachi SU-3500 thermo-ionicemission gun SEM equipped with a tungsten filament (HitachiHigh-Technologies, Rexdale, Canada) and EBSPs were recorded witheither a Hitachi SU-70 or a SU-8000 field-emission SEMs (HitachiHigh-Technologies, Rexdale, Canada). Both were equipped with aHKL Nordlys electron backscatter diffraction (EBSD) system (OxfordInstruments, Concord, USA) controlled with the Channel 5 package.The EBSD camera screen resolution was 640�480 and 1344�1024pixels for the SU-70 and SU-8000, respectively. Pattern acquisition,indexing and simulations were carried out with the Flamencosoftware which is part of the Channel 5 package. The EBSPs wererecorded and stored following the same procedure as for standardEBSD acquisition, i.e., background subtraction and flat fielding wasapplied. The ECP was acquired with an accelerating voltage of 20 kVat normal incidence with a solid state semi-conductor backscat-tered electron detector placed on top of the specimen surface andnormal to the beam (PD-BSE). The ECP image resolution was1280�960 pixels. The EBSPs were recorded with acceleratingvoltages of 20 and 30 kV as specified in the text and a tilt angleof 701, except in Fig. 1 where the tilt angle was 801. The distance

0 1 2 3 4 5 6 7 8 9

0.0

0.2

0.4

0.6

0.8

1.0

3

2

Nor

mal

ized

sig

nal

Position (degrees)

ECPEBSP + FFTEBSP1

1°

Fig. 1. Comparison between an electron backscatter diffraction pattern (EBSP) and an electron channeling pattern (ECP) of a [001] (001) silicon wafer. (A) Raw and(B) digitally filtered EBSP and (C) ECP. Both were recorded with an accelerating voltage of 20 kV. The working distance was 10 mm for the ECP and the detector distance was80 mm for the EBSP. Tilt angles were 01 and 801 for the ECP and the EBSP, respectively. (D) Line profiles extracted from (A–C) show the higher resolution obtained with theECP compared to the EBSP.

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131124

from the EBSD detector and the specimen surface was varied from16 mm to 80 mm and will be specified in the text when necessary.The ECP and EBSPs were recorded as 8-bit images, i.e., with a greyscale range of 256 levels. For display purposes, the EBSP imagesdisplayed in this report were noise-filtered by segmenting the fastFourier transform (FFT) images of the original images and then byapplying an inversed FFT to obtain the final images. The bottomdiodes of forescatter detectors (FSDs) attached to both EBSDcameras were used to acquire orientation images. Band contrast(BC) images were extracted from the EBSD scan in Fig. 5 using theTango software which is part of the Channel 5 package. In thisimage, the BC value is related to the relative contrast of the Kikuchibands compared to the overall contrast of the whole EBSP [10] andreflects, more or less, its sharpness.

The signal S extracted from the line profiles in Fig. 1 wasnormalized using S¼(I� Imin)/(Imaxþ Imin) and in Fig. 3 the contrastC was calculated using C¼(Imax� Imin)/(Imaxþ Imin), I being the greyvalue at each pixel of the profile and Imin and Imax the minimumand maximum grey value of the profile.

3. Experimental procedure

As reported by Joy [4], channeling contrast images are obtainedby reporting on each pixel of the final image the intensity from thecenter of the ECP that would be obtained from the material underinvestigation at this specific pixel position. The magnification mustbe high enough (higher than �100) for the beam to specimensurface angle to be considered as constant over the whole scannedarea. Then, if an ECP is recorded, a specific reflection can be selectedby rotating and tilting the specimen in order to bring a particularpseudo-Kikuchi line or a zone axis at the center of the ECP image[11–14] to obtain a new image related to the new area brought atthe center of the ECP. Unfortunately, this facility being only availableby tilting and rotating the specimen in the SEM, this limits greatlythe applicability of such a technique. However, when EBSD is usedto acquire and store EBSPs, an image can be generated where eachpixel is given a specific value derived from the EBSP from thisspecific pixel position on the specimen surface. This value can beeither an orientation related value, a misorientation value or apattern quality value [15–18].

In the present work, the method used for ECCI is transposed tothe EBSD case with the ECP being replaced by the EBSP. In thismethod, instead of moving the specimen to bring a specificreflection at the center of the ECP (optic axis of the SEM), a virtualbeam is selected onto the EBSP by choosing a cluster of pixels andthe final image is then reconstructed by reporting the intensity ofthe area defined by the virtual beam at each pixel position of theEBSD map. The advantage here is obvious: whereas the conven-tional ECCI technique provides an image from one reflection at atime, the EBSD-based technique can provide several images withdifferent diffraction reflections at the same time.

The correspondence between ECP and EBSP has never beenstudied in details, although some authors broached the topic[19,20]. It is generally accepted that both techniques are related toeach other by the reciprocity theorem [21–23]. According to thistheorem, the detector distance (DD) between the EBSD camerascreen and the beam impact point on the specimen, that determines,with the EBSD camera screen size, the EBSP collection angle, isequivalent to the scan or rock angle necessary to generate the ECP.Similarly, the beam divergence angle in ECP is equivalent to thephosphorescent screen angular resolution used to record the EBSP.Based on these requirements, an EBSP recorded from the sameorientation in the same angular configuration as for an ECP shouldexhibit the same contrast and shape. Fig. 1 shows a comparison of anEBSP (Fig. 1A) and an ECP (Fig. 1C) from a [001] (001) silicon wafer at

801 and 01 tilt angle, respectively, both recorded close to the [001]zone axis. The EBSP was recorded at the maximum detector distanceavailable (80 mm) on the SU-8000 microscope to magnify thepattern to have the same dimensions as the ECP. However, it wascropped to the same area as that of the ECP because it could not bemagnified further more. Consequently the image resolution of theEBSP displayed was 470�366 pixels leading to a pattern resolution2.7 times smaller than the ECP. Obviously, the angular resolution ofthe ECP looks much higher than that of the EBSP. The ECP wasrescaled to the same image resolution than that of the EBSD (notshown) and mostly shows the same resolution than the fullresolution image. This demonstrates that the loss of angular resolu-tion is not due to its lower pixel resolution. However, the energy cut-off due to the conductive layer at the BSE detector and EBSD cameraphosphorescent screen surfaces were experimentally estimated to0.5 and 2 keV, respectively. Hence, there is a significant higher lossof signal in the EBSP than in the ECP at the same acceleratingvoltage. However, the main reason for the loss of resolution mayreside in the fact that in semi-conductor type BSE detectors, thesignal generated by the detection system is proportional to theelectron energy and thus the high energy BSEs contribute more tothe ECP [22]. Because the channeling contrast is handled by highenergy BSEs [2], the contrast and resolution of the ECP are thenenhanced by using this type of detectors. Recently, a higher EBSPquality was achieved by using a new type of EBSD detector based onthe direct detection of the diffracted electrons [24]. However, theefficiency of charged couple device (CCD) camera is not well knownin this range of energies and may explain this large loss of resolutionand contrast in the EBSP compared to the ECP. It is known also thatrecording EBSPs on photographic emulsions permits to improveresolution and contrast to a certain extent [22]. To reduce the noiseof the image, the fast Fourier transform (FFT) of the EBSP wasdigitally filtered using FFT and threshold features of ImageJ [25].A mask created by segmenting the FFT image so that only thefrequencies corresponding to the Kikuchi bands in Fourier spacewere selected. Then, the mask was applied on the complex FFTimage and was inverse-fast Fourier transformed. The resulting imageis displayed in Fig. 1B. It can be seen that the noise reduction did notimprove the contrast significantly but enhanced slightly the angular resolution of the EBSP although it was still poor compared tothe ECP. Line profiles were extracted from the three images inFig. 1(A–C) normal to the (220) plane and were plotted in Fig. 1D. Inorder to compare ECP and EBSP images on the same basis, the lineprofile was averaged from 10 lines in the ECP and 3 lines in theEBSPs to cover approximately the same area on the image. Theseresults confirm the higher resolution deduced from visual observa-tions, especially when comparing profiles around (220) first, secondand third orders at points 1, 2, 3. Clearly, more pseudo-Kikuchi lineswere resolved in the ECP than in the EBSP even when a digital filt-ering procedure was used. However, based on this comparison, thecontrast of an EBSP, although it showed a significant loss of angularresolution, was close to that of an ECP.

The procedure used in this work was as follow: EBSPs wererecorded and stored in a raster motion, like in the conventional EBSDmapping process. Then, the EBSPs from each pixel of the map werepost-processed without any image adjustments or manipulation toextract the intensity from a specific cluster of pixels determined bythe user in a square fashion. Different pixel clusters, named virtualbeam collection area in the following text, could be selected leadingto square areas for which an average grey level was calculated andreported in the final map at each pixel. The final DF images werecomputed using an automated code written with the Pythonprogramming language (www.python.org). The images, generatedusing this procedure, were further labeled EBSD-DF images. Thebrightness and contrast of all final EBSD-DF images where digitallyadjusted to cover the full range of the 256 grey levels.

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131 125

4. Results

4.1. ECCI simulation

To compare the EBSD-DF images contrast with a conventionalECCI image, a 200�200 pixels EBSD map of a Si–Fe NOES wasrecorded with a 10 mm step size and an accelerating voltage of20 kV. Fig. 2A shows the ECCI image of the same area obtained at20 kV and 01 tilt with the PD-BSE detector. The contrast observed isthat expected from a polycrystalline material, i.e., different greylevels for different crystal orientations although two distinctoriented crystals can exhibit the same grey level. In Fig. 2(B–E)are shown EBSD-DF images with random reflections selected on theEBSD map first pixel EBSP (reference EBSP) and using a 2�2collection area (0.221 collection angle). The image contrast is ofthe same type and it is clearly seen, as expected, that the contrastchanges greatly when different reflections are used. Because thetechnique described in this work is only based on one small area ofthe pattern, it is assumed to be more sensitive to small angle crystalrotations than EBSD orientation calculations. Indeed, the latter isbased on the measurement of angles between several Kikuchibands and we assume that the very small pattern shift due to thecrystal rotation in the subgrain will fall into the error of the anglemeasurement. However, when a very small cluster of pixel is used,the shift of the pattern immediately manifests as a change ofcontrast, making the subgrain visible. Hence, EBSD-DF images canbe used for grain boundaries imaging with high angular resolution.As an example, over 4000 EBSD-DF images obtained from the sameset of data were used to generate the image shown in Fig. 2F. EachEBSD-DF image was processed using the ImageJ software [25] todetect the edges in the image. As mentioned above, due to theorigin of the contrast, some grains of different orientation can givethe same contrast and to overcome this issue, the average of the4000 images after edge detection was calculated to obtain the finalimage of Fig. 2F. This permitted to enhance the contrast at grainboundaries and have thus the realistic image of grain and subgrainboundaries network. The area encircled in Fig. 2G highlights aregion of the EBSD map where the two upper grains, misorientedby only 31, were barely visible in the IPF map. However, they weredetected while displaying grain boundaries (GB) (dark GB4101 andwhite GB411) but this resulted in an increase of the noise on theGB image (superimposed with the IPF map in Fig. 2G). This showshow the grain detection process from the EBSD software may beaffected at these low misorientations, the grain size detection beingbased on the GB misorientations. This could be circumvented ifsmoothing techniques were applied to the original data, leading to apossible loss of information. In contrast, the image obtained withthe EBSD-DF images (Fig. 2F) clearly separates the grains, even witha low misorientation, without additional noise on the image andwithout any further data manipulation.

In order to support this statement, grain size distributions fromthe inverse pole figure map (Fig. 2G) and the image of Fig. 2F werecompared. The grain distribution was measured with the Tangosoftware for the EBSD data (based on the GB detection) and by usingthe GrainsJ plugin (available at https://bitbucket.org/hendrix_demers/grainsj) in ImageJ software for the image of Fig. 2F (imageanalysis technique, i.e., not related to orientation). Both normalizeddistributions are displayed in Fig. 2H. The trend of the two distribu-tions is mostly identical except for the low side of the histogramswhere an inversion is observed for grains smaller than 50 mm. Thistrend seems to confirm that this method provides a higher efficiencyin detecting small grains compared to the EBSD-based grain dis-tribution measurement. However, a more systematic study shouldbe conducted to evaluate the reproducibility of the technique andmonitor its precision on a wide range of samples. We also bring tothe reader's attention that this comparison was conducted with two

different techniques, one based on GB detection (orientation) andthe other based on segmentation of the grey scale image. However,the goal here was to show the benefit of our technique compared tothe classical EBSD grain size detection technique.

However, it has to be noted that because of the limited EBSPsaving capabilities (maximum of 40,000 EBSPs) with the Flamencosoftware, the grain boundaries appeared enlarged due to the largestep size used. Consequently, the grain distribution calculatedfrom Fig. 2F was underestimated compared to that based on theEBSD data where the grain boundaries are sharper due thesmoothing procedure. A higher resolution map would be neces-sary to compare the two sets of data.

4.2. Effect of virtual beam collection area

The influence of the virtual beam collection area on the EBSPwas measured on EBSD-DF images reconstructed around a micro-hardness indent in the same material as in Fig. 2. The virtual beamcollection area was set to 2�2 (0.221 collection angle) (Fig. 3A),4�4 (0.361 collection angle) (Fig. 3B), 8�8 (0.651 collection angle)(Fig. 3C), and 16�16 (1.221 collection angle) (Fig. 3D) around a pixelon the band edge corresponding to the (121) plane in Fig. 3F (whitearrow). The line profile along the white arrow in the inset of Fig. 3Ewas extracted and the contrast C of the whole profile was calculatedand plotted in Fig. 3E as a function of the collection area.

As intuitively expected from the images in Fig. 3(A–D), thecontrast was increased at small collection area on the EBSP but witha higher level of noise. On the contrary, the SNR was significantlyimproved when higher collection area values were used but thecounterpart was a large drop of the image contrast due to the pixelcluster averaging. Hence, this post-processing technique allows themanipulation of one dataset to get the best optimized image. A2�2 virtual beam collection area was used in the rest of the workreported here.

4.3. Dark-field imaging of micro-hardness indents

The main advantage of the technique reported in this work is theability to choose a particular reflection for imaging, in an interactiveway after the complete set of EBSPs has been recorded and stored.This is illustrated in Fig. 4 where an enlarged view of an EBSPrecorded from the first pixel of the image of Fig. 4(B–E). The EBSDmap was recorded with an accelerating voltage of 30 kV and a pixeldwell time of 50 ms with 3 frames averaged. No camera binningwas used and the EBSP image resolution was 640�480 pixels. Fourpositions, labeled B, C, D, and E, were selected along the white linedrawn in this figure. This line is normal to the tilt axis of themicroscope (with the tilt axis parallel to the camera plane) and thensimulates the tilting of the specimen just like in an ECCI experi-ment. The line crosses the (1–10) plane. It has to be noted that inthis particular case, the specimen was compressed and hence thegrains exhibit some visible deformation which renders the choice ofthe reference EBSP delicate. However, in the images showed inFig. 4(B–E), it clear that the contrast of each image changes drasti-cally depending on the reflection used. Although each of the imagesshow a typical indent ECCI contrast, the comparison between thempermits to show that a single image could be misleading in describingthe deformation structures around the indent. In Fig. 4B and C, whenthe virtual beam is close to the (1–10) and (3–10) band edges, thecontrast is low and mostly monotone inside the bulbs around theindent, except the bottom one. On the contrary, when the virtual beamis placed inside the (1–10) band, more contrast are observed inside thebulbs, indicating that a more complicated deformation mechanismoccurs in these regions. Also, the bottom bulb, that exhibited a moredistinctive contrast compared to the other bulbs in Fig. 4B and C, ismuch more described in Fig. 4D and E. Especially in Fig. 4E, the bulb

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131126

0 50 100 150 200 250 300 350 4000.0

0.1

0.2

0.3

0.4

Nor

mal

ized

f req u

ency

Diameter (µm)

EBSDEBSD-DF

Fig. 2. Comparison between electron channeling contrast (ECCI) and electron backscatter diffraction dark-field (EBSD-DF) images and an electron backscatter diffraction(EBSD) map from the same region of a stress relief annealed non-oriented Si–Fe electrical steel. (A) ECCI image, (B–E) EBSD-DF images obtained with random reflections fromthe reference EBSP, (F) reconstructed image showing grain boundaries based on 4000 EBSD-DF images (see plain text for more details), (G) inverse pole figure map based onthe same EBSD data as used in (B–F). The EBSPs image resolution was 640�480 pixels. (G) Grain size distribution obtained from the EBSD map (black) and the EBSD-DFreconstructed image (F) (grey) showing a higher number of small grains detected in (F).

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131 127

Con

tras

t

Virtual beam collection area (pixels)

b = 2×2 b = 4×4

b = 16×16b = 8×8

30 µm

Fig. 3. Effect of the virtual beam collection area on the final EBSD-DF imagecontrast around a micro-hardness indent in a stress relief annealed non-orientedSi–Fe electrical steel with collection area of (A) 2�2 (0.221 collection angle),(B) 4�4 (0.361 collection angle), (C) 8�8 (0.651 collection angle), and (D) 16�16(1.221 collection angle). (E) Plot of the contrast C between the points of highest andlowest grey value extracted from the line profile drawn in the top right insert(white arrow). (F) Reference EBSP showing the location of the virtual beamcollection area centre point corresponding to the (121) reflection (white arrow).Optimum collection area was 4�4, corresponding to a square of 141 mm width onthe phosphorescent screen.

B

CD

E

[001]

[111]

[113]

(1-10) (3-10)

30 µm

111

001 101

Fig. 4. EBSD-DF images of a micro-hardness indent on compressed iron obtainedwith an accelerating voltage of 30 kV and a detector distance of 16 mm as afunction of the virtual beam position on the reference EBSP. (A) Reference EBSPtaken from the first pixel of the EBSD map, (B) Magnified view of the squared areain the reference EBSP in the insert, (B–E) EBSD-DF images with specific reflectionsdisplayed on (A). (F) Band contrast and (G) inverse pole figure maps of the samedata. The EBSPs image resolution was 640�480 pixels. The white arrow representsthe line fromwhich the points B to D were taken as if the sample was tilted towardthe EBSD camera. The movie available in the supplementary data was generatedwith all the EBSPs along the white arrow.

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131128

shows a symmetrical structure that could be related to iso-deform-ation contours occurring in this region. To emphasize the importanceof using EBSD-DF images to monitor deformation structures, a movie(“EBSD-DF movie.avi”) is available as supplementary material on theUltramicroscopy Journal website. This movie is a compilation of all theEBSD-DF images constructed with the virtual beam following thewhite arrow in Fig. 4A. In Fig. 4F and G, the band contrast and IPFmaps from the same data are displayed. The band contrast map showsinteresting features with quite good contrast but which is only relatedto the quality of the EBSP, i.e. not related to any diffraction condition.The IPF map, however, shows orientation related details around theindent, but the contrast is very weak compared to the EBSD-DF imagesin Fig. 4B–D.

Supplementary material related to this article can be found onlineat http://dx.doi.org/10.1016/j.ultramic.2014.09.005.

To take advantage of the capabilities of EBSD for mapping strains inmaterials, an EBSD camera with a better pixel resolution and a longerdetector distance were used to record an EBSD map of a micro-hardness indent and generate EBSD-DF images of the deformationstructures around it. The accelerating voltage was 30 kV and the EBSPimage resolution was 1344�1024 pixels. The dwell time was 500 msand 2 frames were averaged to generate the stored EBSP. The detectordistance to the sample surface was increased from the normal 16mmsetting used in the scan presented in Fig. 4 to 50 mm to magnify theEBSP as described by Wilkinson [19]. However, in our experimentalset-up, due to the geometry of the SU-8000 SEM specimen chamber,the maximum detector distance was only 80 mm compared to140mm used in Wilkinson's work. Finally, 50 mm was chosen tokeep sufficient Kikuchi bands on the EBSD screen to select differentreflections that may have been out of the screen at 80 mm. Despitethis limitation, high resolution EBSPs were recorded as can be seenfrom the reference EBSP shown in Fig. 5A from a pixel located at thefirst quarter of the first row. The FSD (Fig. 5B) and band contrast(Fig. 5C) images are also displayed for comparison purposes. The EBSPdisplayed in Fig. 5A was centered on the (001) zone axis and [001],(200), (020), (031), (200), and (310) reflections were used to generatethe EBSD-DF images in Fig. 5(D–H), respectively.

In this example, a larger contrast between the deformation areaand the rest of the grainwas achieved in EBSD-DF images compared tothat obtained with the standard EBSD settings used in Fig. 4 as well asthat of the FSD image or the band contrast map in Fig. 5B and C.However, because the deformation pattern changes in the EBSD-DFimages depending on the reflection chosen in the reference EBSP, aprecise measure of the contrast for comparison was not possible andonly visual observations could be done. Again, multiple EBSD-DFimages help in monitoring more precisely the deformation structuresaround the indent compared to the FSD image displayed in Fig. 5B. Infact, the FSD image is based on the mean intensity collected by thediode which is several tens of mm2 in size, offering low contrastimages compared to EBSD-DF images and Prior et al. [6] concludedsome time ago that several FSD images with different tilt angles werenecessary to locate clearly the grain boundaries of polycrystallinespecimens. In addition, it is interesting to note the high contrastachieved with EBSD-DF image generated at the center of the [001]zone axis. This might be due to the high level of details observedinside zone axis in Kikuchi-like patterns as described by Joy [4] andMarthinsen and Hoer [26].

A similar treatment as that used in Fig. 2F was applied to this setof images (4000 images) and showed similar structures (not shown)as described in Welsch et al. [27]. However, the physical meaning ofthese structures could not be easily related to the microstructure,especially dislocation structures as described in their work.

5. Discussion

5.1. Dark-field imaging in the SEM

As raised by our results, dark-field images can be extractedfrom an EBSD scan just by selecting specific pixels in each EBSPs ofthe map, and by that way, chose a specific reflection to generate animage. This is, theoretically speaking, the same concept as used in

Fig. 5. High contrast EBSD-DF images of a micro-hardness indent on compressediron obtained using long EBSD detector distance for high angular resolution EBSPswith an accelerating voltage of 30 kV and a detector distance of 50 mm as afunction of the virtual beam position on the high angular resolution reference EBSP.(A) Reference EBSP (see plain text for detailed description), (B) band contrast map,(C) FSD image and (D–H) EBSD-DF images with specific reflections marked byarrows in (A). The EBSPs image resolution was 1344�1024 pixels.

[001]

(310)

(020)

(031)

(200)

(1-10)

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131 129

TEM dark-field imaging. However, the main advantage here is thelarge field of view available in the SEM and the fact that thistreatment of the EBSPs is offline, permitting to generate a largenumber of crystallographic orientation images at the same time.Also, the large collection angle of nowadays EBSD cameras allowrecording large number of Kikuchi bands on only one EBSP, andthus, permits a large range of potential reflections to be used. Ahigh angular resolution was achieved by increasing the EBSDcamera angular resolution and high contrast deformation imageswere generated.

In addition, the virtual beam collection area can be used toadjust the noise and contrast levels of each generated EBSD-DFimage depending on the EBSP angular resolution and contrast asshown in Fig. 3. This parameter might be related to the acceleratingvoltage used in ECCI. In fact, the band width increases in the ECPwhen the accelerating voltage is decreased due to the Braggrelation. Consequently, using a small virtual beam collection areashould have the same effect as reducing the accelerating voltage interms of image contrast, except inside zone axis where it is wellknown that slight changes in accelerating voltage affect the positionand visibility of zone axis fine structures. However, this was verifiedbecause the dwell time at low primary energy was too long toacquire a full EBSD map with the full EBSD camera resolution.

5.2. Limitations

As mentioned above, an important limitation, nowadays, is thelong dwell time necessary for acquiring EBSD maps with the fullresolution of the EBSD camera, although the miniaturization andthe efficiency of the new generation of charge coupled devices tendto reduce dwell times at high EBSP image resolution. This is a keyparameter because a high EBSP angular resolution is necessary toapproach the contrast of standard ECCI images based on ECPs asshownwith the comparison of the ECP and EBSP in Fig. 1. The workreported here should give even more credits in developing fast,efficient, and resolutive EBSD detectors in the future to apply moreroutinely this technique to any deformation characterization work.

In addition to this, and similarly to the work reported byWilkinson and Randman [28], the choice of the reference EBSPfor the selection of the specific reflections to be used is critical. Inboth procedures, a strain free region is required to compare thedeformed structures observed in the EBSD-DF images with a non-deformed area. Also, as reported by Britton [29], the accuratedefinition of the pattern centre in EBSPs is mandatory if quanti-tative interpretation of the strain-induced contrast is to be under-taken. This will be a necessary direction of progress in the futureto bring this technique to a high level of precision.

6. Conclusions

Similarly to ECCI where the intensity at the centre of the ECP isreported on each pixel of the BSE image, a technique was developedto generate offline orientation images with controlled crystallographicconditions. It is based on the stored EBSP at each pixel of a map, andthe resulting DF images were given the name of EBSD-DF images. Theoffline process allows improving the quality of the final image bychoosing the virtual beam collection area (number of pixels) to beused in the EBSP for the image reconstruction in order to enhancecontrast and reduce noise.

With only one EBSD scan, one can generate as many EBSD-DFimages as the number of pixels in the EBSP image, i.e., tens ofthousands images. By selecting specific reflections on a referenceEBSP, the deformation of micro-hardness indents was assessedmore precisely. Moreover, a high contrast was achieved when a

high angular resolution EBSD camera at long detector distance wasused making the deformation area more visible.

Apart from the fact that even with the latest generation ofcommercially available EBSD cameras the angular resolution ofEBSPs is lower than that of ECPs, some limitations need to bementioned. The more limiting parameter of the method is thechoice of the reference EBSP on which are selected the reflectionsrendering the interpretation of EBSD-DF images more difficult ifthe reference area is also deformed. Secondly, this technique iseven more valuable if high resolution EBSPs are recorded duringthe EBSD scan. This increases considerably the acquisition timewhich is still an issue when EBSD is performed at high magnifica-tion due to drift and carbon contamination. However, this latterissue may be solved when faster and more efficient EBSD camerawill be commercialized.

In the future, the technique developed here might be helpful ifapplied to improve our understanding of materials deformationand of the ECCI contrast mechanism through the use of EBSD-DFimages with known conditions. In fact, because it permits to goback and forth from the EBSD-DF image to the EBSPs images, itwill provide a simple and efficient way to relate the intensity ofspecific regions of the image to a virtual beam on the EBSP. Thiswill lead to interpret more easily and rapidly the contrast observedon the ECCI and DF images in the SEM.

Acknowledgments

We are thankful to Philippe Bocher and Florent Bridier from theÉcole de Technologie Supérieure de Montréal for giving us accessto the SU-70 SEM and to YaoYao Ding, Matthew Gallaugher, andProfessor Richard Chromik from McGill University, Montréal, forproviding the steel and iron samples.

References

[1] A. Morikawa, et al., Low-voltage dark-field stem imaging with optimumdetection angle, Microsc. Microanal. 12 (S02) (2006) 1368–1369.

[2] D.C. Joy, D.E. Newbury, D.L. Davidson, Electron channeling patterns in thescanning electron microscope, J. Appl. Phys. 53 (8) (1982) R81–R122.

[3] P.B. Hirsch, et al., Electron Microscopy of Thin Crystals, vol. 320, Butterworths,London, 1965.

[4] D. Joy, Electron channeling patterns in the SEM, in: D.B. Holt, et al., (Eds.),Quantitative Scanning Electron Microscopy, Academic Press, London, NewYork, San Francisco, 1974, pp. 131–182.

[5] E.J. Payton, G. Nolze, The backscatter electron signal as an additional tool forphase segmentation in electron backscatter diffraction, Microsc. Microanal. 19(04) (2013) 929–941.

[6] D.J. Prior, et al., Orientation contrast imaging of microstructures in rocks usingforescatter detectors in the scanning electron microscope, Mineral. Mag. 60(1996) 859–869.

[7] E. Payton, L. Agudo Jacome, G. Nolze, Phase Identification by image processingof EBSD patterns, Microsc. Microanal. 19 (S2) (2013) S842–S843.

[8] Schwarzer, R.A., J. Sukkau, Electron Back Scattered Diffraction: Current State,Prospects and Comparison with X-Ray Diffraction Texture Measurement.

[9] M.M. Nowell, I. Wight Stuart, T. Rampton, R. De Kloe, A new microstructuralimaging approach through EBSD pattern region of interest analysis, Microsc.Microanal. 20 (Suppl. 3) (2014) S1116–S1117.

[10] T. Maitland, S. Sitzman, Electron backscatter diffraction (EBSD) techique andmaterial characterization examples, in: W. Zhou, Z.L. Wang (Eds.), ScanningMicroscopy for Nanotechnology, Springer, New York, 2006.

[11] B.-C. Ng, B. Simkin, M. Crimp, Electron channeling contrast imaging ofdislocation structures in deformed stoichiometric NiAl, Mater. Sci. Eng.: A239 (1997) 150–156.

[12] B. Simkin, M. Crimp, An experimentally convenient configuration for electronchanneling contrast imaging, Ultramicroscopy 77 (1) (1999) 65–75.

[13] M.A. Crimp, Scanning electron microscopy imaging of dislocations in bulkmaterials, using electron channeling contrast, Microsc. Res. Tech. 69 (5) (2006)374–381.

[14] I. Gutierrez-Urrutia, S. Zaefferer, D. Raabe, Electron channeling contrastimaging of twins and dislocations in twinning-induced plasticity steels undercontrolled diffraction conditions in a scanning electron microscope, Scr. Mater.61 (7) (2009) 737–740.

[15] S.I. Wright, B.L. Adams, Automatic analysis of electron backscatter diffractionpatterns, Metall. Mater. Trans. A 23 (3) (1992) 759–767.

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131130

[16] J. Wu, et al., Image quality analysis: a new method of characterizing micro-structures, ISIJ Int. 45 (2) (2005) 254–262.

[17] S.I. Wright, M.M. Nowell, EBSD image quality mapping, Microsc. Microanal. 12(1) (2006) 72–84.

[18] P.T. Pinard, et al., An open-source engine for the processing of electronbackscatter patterns: EBSD-Image, Microsc. Microanal. 17 (03) (2011) 374–385.

[19] A. Wilkinson, Methods for determining elastic strains from electron back-scatter diffraction and electron channeling patterns, Mater. Sci. Technol. 13 (1)(1997) 79–84.

[20] B. Simkin, et al., Orientation determination and defect analysis in the near-cubic intermetallic $$-TiAl using SACP, ECCI, and EBSD, Intermetallics 11 (3)(2003) 215–223.

[21] J. Venables, C. Harland, Electron back-scattering patterns—a new technique forobtaining crystallographic information in the scanning electron microscope,Philos. Mag. 27 (5) (1973) 1193–1200.

[22] L. Reimer, Scanning Electron Microscopy: Physics of Image Formation andMicroanalysis, Springer, New York, 1998.

[23] O.C. Wells, Comparison of different models for the generation of electronbackscattering patterns in the scanning electron microscope, Scanning 21 (6)(1999) 368–371.

[24] A.J. Wilkinson, et al., Direct detection of electron backscatter diffractionpatterns, Phys. Rev. Lett. 111 (6) (2013) 065506.

[25] Rasband, W. ImageJ U.S. National Institutes of Health, Bethesda, Maryland,USA, 1997–2012, ⟨http://imagej.nih.gov/ij/⟩.

[26] K. Marthinsen, R. Hoier, Determination of crystal symmetry from electronchanneling patterns, Acta Crystallogr. Sect. A: Found. Crystallogr. 44 (5) (1988)693–700.

[27] M.T. Welsch, et al., Measuring the plastic zone size by orientation gradientmapping (OGM) and electron channeling contrast imaging (ECCI), Adv. Eng.Mater. 9 (1-2) (2007) 31–37.

[28] A.J. Wilkinson, D. Randman, Determination of elastic strain fields andgeometrically necessary dislocation distributions near nanoindents usingelectron back scatter diffraction, Philos. Mag. 90 (9) (2010) 1159–1177.

[29] T. Britton, et al., Factors affecting the accuracy of high resolution electronbackscatter diffraction when using simulated patterns, Ultramicroscopy 110(12) (2010) 1443–1453.

N. Brodusch et al. / Ultramicroscopy 148 (2015) 123–131 131