Contents lists available at ScienceDirect

Ultramicroscopy

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

3D characterisation using plasma FIB-SEM: A large-area tomographytechnique for complex surfaces like black silicon

Yu Zhanga, Charlie Kongb, Rasmus Schmidt Davidsend, Giuseppe Scarderaa, Leiping Duana,Kean Thong Khooa, David N.R. Paynea,c, Bram Hoexa,⁎, Malcolm Abbotta,⁎

a School of Photovoltaic and Renewable Energy Engineering, University of New South Wales, Sydney, NSW, 2052, Australiab Electron Microscope Unit, University of New South Wales, Sydney, NSW, 2052, Australiac School of Engineering, Macquarie University, Sydney, NSW, 2109, AustraliadNational Centre for Nanofabrication and Characterization, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark

A B S T R A C T

This paper demonstrates an improved method to accurately extract the surface morphology of black silicon (BSi). The method is based on an automated Xe+ plasmafocused ion beam (PFIB) and scanning electron microscope (SEM) tomography technique. A comprehensive new sample preparation method is described and shownto minimize the PFIB artifacts induced by both the top surface sample-PFIB interaction and the non-uniform material density. An optimized post-image processingprocedure is also described that ensures the accuracy of the reconstructed 3D surface model. The application of these new methods is demonstrated by applying themto extract the surface topography of BSi formed by reactive ion etching (RIE) consisting of 2 µm tall needles. An area of 320 µm2 is investigated with a controlled slicethickness of 10 nm. The reconstructed 3D model allows the extraction of critical roughness characteristics, such as height distribution, correlation length, and surfaceenhancement ratio. Furthermore, it is demonstrated that the particular surface studied contains regions in which under-etching has resulted in overhangingstructures, which would not have been identified with other surface topography techniques. Such overhanging structures can be present in a broad range of BSisurfaces, including BSi surfaces formed by RIE and metal catalyst chemical etching (MCCE). Without proper measurement, the un-detected overhangs would result inthe underestimation of many critical surface characteristics, such as absolute surface area, electrochemical reactivity and light-trapping.

1. Introduction and motivation

Black silicon (BSi), also known as surface nano- or micro-texturedsilicon, is a widely used silicon semiconductor feature which inheritsmost of the advantages of bulk silicon material and possesses someunique optical, electronic and electrochemical properties that originatefrom its unique surface geometry. During the last two decades, a diverserange of BSi with different surface morphologies have been created byaltering the fabrication processes to suit the requirements of differentapplications [1]. These morphologies include nano-wires [2–4], nano-porous structures [5,6], needle-like microstructures [7–12], invertedpyramids [13–16], and nano-/micro-hybrid structures [17]. The en-hanced optical absorption and enhanced surface electrochemical re-activity of BSi is correlated with its surface morphology. Understandingthe correlation between BSi surface morphology and its unique prop-erties is critical for optimizing devices incorporating BSi. However, thecomplex nature of BSi surfaces brings extra challenges to characterizeits topographic properties and obtain an accurate surface profile, i.e.,height distribution and roughness characteristics.

Researchers have used various characterization methods to measurethe topography of BSi [18]. The most commonly adopted technique is

atomic force microscopy (AFM) [19–21], which measures the surfaceheight with a scanning probe [22–27]. However, it is incredibly chal-lenging to avoid AFM imaging artifacts when characterizing very roughsurfaces such as BSi. These artifacts can be induced by a non-ideal tipshape, tip bending, flexing, and jumping effects, the adhesion forcesforms between tip and sample, or non-optimal scan settings [19,28–30].The constant tip-sample interaction during the measurement of a highlyroughened surface also makes the AFM tip blunt or contaminates the tipsurface. Moreover, AFM is not able to detect structures that exceed thephysical size of the scanning probe, or micro-nano hybrid structuresthat exceed the oscillation limits of the cantilever. Another option isoptical profilometry, whose working principle is similar to AFM, butuses light as a probe to extract the topographical data from a surface[31,32]. However, the typical BSi surface feature size is near to orbelow the limit of resolution of the optical spot size, which is describedby the Abbe resolution or the Rayleigh criterion [32]. Since all scanningprobe microscopy (SPM) techniques rely on the probe-sample interac-tion from the near-normal or near-vertical direction, none of them canhandle complex overhanging structures.

Focused ion beam – scanning electron microscopy (FIB-SEM) to-mography uses a FIB for flexible micromachining and site-specific

https://doi.org/10.1016/j.ultramic.2020.113084Received 26 March 2020; Received in revised form 20 July 2020; Accepted 27 July 2020

⁎ Corresponding authors.E-mail addresses: b.hoex@unsw.edu.au (B. Hoex), m.abbott@unsw.edu.au (M. Abbott).

Ultramicroscopy 218 (2020) 113084

Available online 29 July 20200304-3991/ © 2020 Elsevier B.V. All rights reserved.

T

material removal in combination with high-resolution SEM imagingafter the removal of each slice [33]. This technique enables the probingof extreme RIE structures that exceed the physical size limitation of thehigh aspect ratio AFM probe; it also has the potential to deal withcomplex features such as overhanging structures. However, FIB-relatedartifacts may occur during ion beam/sample interactions causing biasin the 2D profile, which affects the accuracy of reconstructed three-dimensional (3D) models. Such artifacts include ion damage, ionamorphization, material re-deposition, and the curtain-like surfacestriations formed due to the surface topography induced ion deflectionand spatial variation of the sputter rate [34]. Saab et al. [35] studied thesurface morphology of a high aspect ratio cylindrical drill-hole like BSisample by FIB-SEM tomography. However, the features they extractedfrom the 2D SEM image not only contained the edge of silicon and air,but also picked up the edge that arose from the curtain structures. Thestandard approach to avoid these artifacts is to: (1) Use FIB to createtrenches around the volume of interest (VOI), which not only allows thecross-sectioned surface to be milled and imaged at an oblique angle bythe electron and ion beam, but also avoids the redeposition of sputteredmaterial during tomography; (2) Grow a protective platinum film ofapproximately 1 µm thickness on the sample's top surface by ion beamassisted chemical vapor deposition, enabled by an in situ gas injectionsystem (GIS). Pt-GIS has been used extensively to protect the top surfaceof interest from spurious sputtering during the tomography; (3) Addfiducial markers created by FIB both on the top surface and in front ofthe VOI, these provide independent reference features for registrationwhen the automated tomography progresses [33,36,37]. However, aswe will demonstrate in this work, this procedure can be difficult or evenimpossible to apply to the extreme surface features encountered in BSi.

Furthermore, an ideal 3D topographical dataset should be statisti-cally representative of the whole surface, with a relatively large area ofinterest and controlled resolution. This can be difficult to achieve if theetch rate of the FIB process is too slow. Kroll et al. [8] employed FIB-SEM tomography to collect a stack of cross-sectional SEM images of 15nm thickness for a BSi sample with over 1.3 µm height needle-likestructures. The reconstructed 3D model was used to simulate the opticalresponse using the finite difference time domain (FDTD) method.However, the material removal rate for the conventional liquid metalion source (LMIS) FIB [38] restrained the dimension of their region ofinterest (ROI) to be only 3 µm × 3 µm. According to [18], a surfaceroughness measurement should be performed over a reasonable sam-pling interval, which is 0.25 to 0.50 times the correlation length1, and ascan size which is equal to or greater than the value at which thestandard deviation or variance (σ) approaches a constant value. Inaddition, it should not only be able to capture the sub-micron or nano-size roughness feature of the texture, but also detect the potentialbackground waviness which is common inMCCE textured surfaces [26]or a micro-nano dual-phase surface [17,39], both of which are typicallyin the range of microns. Generally, it can be concluded that one shouldscan an area of 10 by 10 microns with a scan resolution of 10 nan-ometers to obtain accurate topography for BSi surfaces. The lowthroughput of FIB-SEM means it is currently not a cost-effective can-didate to obtain large-scale statistical data when compared to SPMtechniques. Despite this, FIB-SEM remains an essential tool for sce-narios where SPM techniques are not able to provide sufficiently ac-curate data, but improvements in throughput whilst maintaining re-solution are highly desirable.

In recent years, xenon (Xe+) plasma-based FIB has proven to be apromising substitutional technique to the conventional LMIS FIB as itcan provide faster material removal rates that satisfy applications re-quiring high mill rates [40]. Kwakman et al. [41] and Altmann et al.[42] achieved large area artifact-free deep cross-sections with excellent

surface quality by incorporating a rocking milling technique with thePFIB. A large 3D volumetric dataset for a WC-Co hard metal samplewith a 150 µm × 120 µm × 80 µm volume composed of 100 nm sliceswas demonstrated by Burnett et al. [38], showing that the PFIB tomo-graphy is capable of automating the serial sectioning task and obtaininguniformly spaced large sections [38,40,42]. A comprehensive approachof optimizing the Xe+ plasma FIB-sample interaction toward high-current/high milling rate activities has been developed recently, ac-companied by the implementation of a single-crystal sacrificial mask(SCSM) as a sample protection layer, which improves the image qualitywith minimal artifacts [43]. To date, large area PFIB tomography ac-tivities have been demonstrated with a slice thickness from 50 – 500 nm[44]. Consequently, Xe+ plasma-based FIB-SEM tomography techniqueis an ideal candidate for conducting large-area tomography activitieswith a resolution down to 50 nm, which is still quite inferior to mosttraditional SPM approaches.

In this paper, we employ Xe+ Plasma FIB (PFIB) for the acquisitionof true-2D statistical datasets by using an FEI Helios G4 PFIB dual beammicroscope. We demonstrate an automated PFIB tomography processwith a large volume and a controlled slice thickness of 10 nm, which issignificantly lower than demonstrated previously. This was done byincorporating a rocking milling routine, an SCSM, and optimized mil-ling parameters. Moreover, we present a comparison between twosample preparation techniques. The first is based on the traditional GIS-Pt method, which is described above. The second approach uses anembedding material DurcupanTM ACM resin, which is commonly usedin biology [33,45–47]. For the first method, the paper demonstratesthat the Pt-GIS was not able to provide a uniform, dense layer whenapplied to BSi. The resulting porous Pt layer would result in undesiredimage contrast from both the voids and the void-induced curtain effectduring the automated PFIB-SEM tomography. The second method isshown to be able to completely fill the complex BSi surface withminimized PFIB artifacts and is subsequently employed for 3D topo-graphy characterization. By carefully controlling the experiment con-ditions, we were able to maintain stable PFIB-SEM tomography auto-mated experiment conditions with1,556 SEM images being collected.We then aligned the SEM stack, achieving controlled staircase-like ar-tifacts along the slicing direction. The collection of BSi 3D topo-graphical data using automated PFIB tomography and the post dataprocessing procedures are presented in detail. Finally, we extract thecritical surface roughness statistic characteristics and discuss the po-tential utility of the reconstructed 3D model.

2. Sample information

A polished 6-inch round Cz silicon wafer was textured into BSi bythe maskless RIE process as described in [12]. Saw damage removal wasapplied to the wafer prior to the maskless RIE process by etching in 30%KOH at 75°C for 2 min and subsequent cleaning in 20% HCl at roomtemperature for 5 min and rinsing in deionized water.Texturing of thewafer was achieved by an ICP RIE research level tool from SPTS tech-nologies, with 13.56 MHz radio-frequency platen power of 100 W. Thereactive gases used were O2 and SF6 with a flow rate of 76 sccm and 74sccm, with chamber pressure of 24 mTorr and chunk temperature of0°C. SEM images taken at various angles (plan view, 45° tilted viewand90° cross-sectional view) are shown in Fig. 1 to reveal the surfacemorphology. The red arrow in Fig. 1. (d) indicates an overhang struc-ture that can not be probed by traditional SPM techniques. The mor-phological images of the bare sample were taken by an FEI Nova Na-noSEM 450 field-emission microscope (FE-SEM).

3. Specimen preparation techniques

Two specimen preparation techniques were tested in this paper: theconventional Pt-GIS coating and the modified sample embedding ap-proach by resin. Both approaches will be explained and compared in

1 Correlation length is the length over which the auto correlation functiondrops to a small fraction of its original value [18].

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

2

detail in Section 3.1 and 3.2.

3.1. BSi and Pt-GIS interaction

The first BSi sample was cleaved into small tokens with a dimensionof 0.5 cm × 0.5 cm for all subsequent processing. The bare BSi samplewas first glued to a standard Ø12.7mm SEM pin stub mount usingconductive silver paste. Then, the sample was placed into the PFIB-SEMdual-beam vacuum chamber and processed through a modified samplepreparation procedure as follows:

(1) A ROI was selected with a dimension of 20 µm by 20 µm re-presenting a contamination-free and damage-free surface.

(2) The ROI was protected by a thin Pt layer of 1 µm thickness, whichwas achieved by subsequent in situ electron-beam-induced deposi-tions (EBID) and ion-beam-induced depositions (IBID). The firstthin layer of Pt with an area of 30 µm × 20 µm were deposited by asequence of EBID deposition at 3 kV beam energy for 3 minutes,with beam current set to 1.6 nA, 3.2 nA then 6.5 nA. Another layerof Pt was deposited on top by the IBID with 30 kV beam energy and7.0 nA beam current. EBID was used as it allows for higher spatialresolution when decomposing the Pt precursor compared to IBIDand thus allows for coating of higher aspect ratio structures.

(3) A 10 µm deep trench was created in front of the ROI by the PFIB sothat material re-deposition during the automated process could beeliminated. Before starting the automated tomography, a cross-sectional surface was exposed by 1.3 nA low current PFIB milling toevaluate the Pt-BSi surface interaction.

Unlike the interaction between Pt and a shallower BSi sample,

which we demonstrated in [48], the much deeper sample structureposed a challenge to the in situ Pt. Due to the proximity effect [49],alsoknown as the structure lateral broadening, with which the Pt built ontop of the structure formed a ‘mushroom’ shape which blocked theelectron-beam from reaching the bottom of the structure. On the otherhand, attenuation of the electron beam resulted in a low decompositionat the bottom of high aspect ratio structures [35]. As shown in Fig. 3(a), the Pt was not able to fully infiltrate the roughened surface andform a uniform, dense layer. The porous structure would result in un-desired striations due to the uneven ion sputtering rate. Such voids andthe void-induced curtain artifacts would result in unwanted imagecontrast and lead to a biased 2D profile extraction.

3.2. Specimen embedding technique with resin

As explained inSection 3.1, Pt-GIS is not a universally suitable op-tion as the top-surface protective layer, especially for samples with avery high surface roughness. Herein, we introduce a new sample pre-paration method that employs the DurcupanTM ACM resin, which sub-stitutes the in situ Pt-GIS for the top-surface protective layer. Durcupanis a colorless epoxy resin that is commonly used as a embedding ma-terial for biological electron microscopy samples, it has a relatively lowviscosity and virtually no shrinkage after curing. The detailed sampleembedding procedure is described below.

For the resin embedding method, the BSi sample was pre-treatedbefore placing it into the dual-beam vacuum chamber. The BSi samplewas sonicated in pure methanol for 2 minutes then immersed into asolution of methanol diluted Durcupan resin with a volume to volumeratio of Durcupan resin: methanol = 20% V/V. The solvent and samplewere then sealed into a Falcon conical tube for 2 minutes sonication and

Fig. 1. SEM images of the BSi sample: (a) plan view, (b) 45° tilted view, and (c) cross-sectional view. (d) Shows the cross-section of the BSi highlighted (blue shade).The presence of an overhang structure is indicated by the red arrow.

Fig. 2. Illustration of BSi sample embedding method used by immersing the sample in a graded Durcupan resin - methanol solution series.

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

3

then put on a tube rotator for 2 hours. Sonicating and rotating wereused to help physically agitate the resin solvent and enhance the resin-sample interaction. As shown in Fig. 2, the solvent was subsequentlyreplaced by a graded Durcupan resin - methanol solution series with aDurcupan resin: methanol volume to volume ratio from 40% V/V, 60%V/V, 80% V/V to 100% V/V. Sonication and rotation of the sample wasrepeated after each solvent replacement. The sample with liquid Dur-cupan resin was then lifted by two plastic rods to drain of excessiveresin. The lifted sample was subsequently placed into the Pelco Bio-wave® Pro±microwave processing system with 20" Hg vacuum treated

for four hours.The use of a diluted solution was found to reduce the trapping of air

bubbles in the inverted rod structure of the BSi. By gradually replacingthe resin solution with higher concentration, the epoxy resin was ableto inherit the better fluidity of the previous solution, and fully infiltratethe sample without introducing new bubbles. Finally, the embeddedsample was put in a compact oven at 60°C for 48 hours for completepolymerizing.

As shown in Fig. 3 (b), in contrast to the conventional Pt-GISmethod, the Durcupan resin can infiltrate the complex BSi surface fully,which not only tailors the ion beam and protects the top surface, butalso provides excellent material contrast. Therefore, the resin methodwas used for the remainder of this work.

4. PFIB-SEM slice-and-view procedure

Tomographical datasets of BSi were obtained using a ThermoFisherHelios G4 PFIB UXe PFIB-SEM DualBeam system. The dual beam systemwas equipped with a 2.5 µA Xe± Plasma FIB ion column, with site-specific fast ion milling capabilities and a 50 × higher throughput

Fig. 3. (a) SEM image of the BSi sample after subsequent in situ EBID and IBID Pt-GIS, with a total thickness of ~ 1 µm; (b) the SEM image of the BSi sampleembedded with Durcupan resin.

Fig. 4. (a) Illustration of sample structure with multiple top-surface protective layers, and (b) overview SEM image of the final sample structure that was slice-and-viewed, the inset in (b) gives an example of a typical 2D SEM image obtained by each slice-and-view activity.

Table 1The pixel size of the slice-and-view experiment and the corresponding voxelsize for the 3D model.

HFW (µm) 25.9Image Resolution [width × height (px)] 6144 × 4376Pixel size (nm2) 4.21 × 4.21Total number of slices 1,556Slice thickness (nm) 10Voxel size (nm3) 4.21 × 4.21 × 10.00

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

4

milling rate [38,42,50] compared to that of conventional Ga+ basedFIB, and ultra-high-resolution monochromator scanning electroncolumn, with 0.8 nm SEM resolution at dual beam coincidence. ThePFIB-SEM dual beam system was controlled by the Thermo Scientific™Auto Slice and View™ 4 software [51] to automate the slice-and-viewprocedure, which consisted of repetitive PFIB cross-sectioning milling(slice) and the subsequent high-resolution SEM imaging (view). Theepoxy embedded sample was glued to the standard Ø12.7 mm SEM pinstub mounts by conductive silver paste, and then coated with a ~ 60 nmsputtered Pt by Quorum Q300T D Plus dual-target sputter coater. Thesputtered Pt layer was applied to enhance the sample surface con-ductivity and eliminate charging effects associated with the epoxy resinduring the slice-and-view process.

The detailed sample structure is shown in Fig. 4 (a), a SCSM wasplaced site-specifically on top of the ROI in situ by a nano-manipulator.Then, the SCSM was fixed on top of the sample at the rear by Pt-GISdeposition. The SCSM played the role of the ion beam tailor that re-duces the FIB artifacts [42]. Moreover, due to the nature of resin, theROI of the top-down structure was intentionally chosen to be at theedge of the sample, further explanation of this is given in the Supple-mentary Information. Fig. 4 (b) displays the overall SEM image of thesample structure that has been slice-and-viewed. Two ‘X’ patterns werecreated using the in situ site-specific ion patterning. The patterns actedas fiducial markers to help the electron-beam and ion-beam to trackaccurately during the automated slice-and-view process [38], givinginformation about the placement of the milling location of the cross-sectional slices. The incident ion-beam was incoming from the top ofthe SCSM, along the Y direction. And the slice-and-view activity pro-gressed normally to the imaging plane, along the Z direction, from the

outer structure to the inner structure.During the slice-and-view process, the SEM images were collected

using an accelerating voltage of 3 kV, a beam current of 80 pA, and aworking distance of 4.0 mm. The image intensity signal was determinedfrom backscattered electrons collected by the through lens detector(TLD). A comparison of SEM image quality obtained by different signalcollecting mode with different detectors is discussed in theSupplementary Information. The material removal was carried outusing an accelerating voltage of 30 kV, an ion beam current of 15 nA,and a working distance of 16.5 mm. The rocking milling process, whichemploys a compucentric rotation of the sample by +/- 5° offset to theincoming FIB direction, was used to minimize curtaining artifacts [38]during the slice-and-view process.

After the slice-and-view, a stack of 2D cross-sectional SEM images ofthe BSi was collected for further data processing. The pixel size of eachSEM in the stack was calculated by the ratio between the horizontalfield width (HFW) and image resolution. Combining the pixel size andthe slice distance, the voxel size for the reconstructed 3D model wasdetermined and summarized in Table 1.

5. Post SEM stack processing and 3D reconstruction

All of the image post-processing procedures were conducted usingthe open-source image processing package FIJI [52]. A flow chart of thestep-by-step image processing is shown in Fig. 5. The raw stack of SEMswas first smoothed by replacing each pixel with the average of its 3 × 3neighborhood, and then blurred using a Gaussian smoothing algorithmwith a radius factor of 6.0. The line scan profiles of the raw SEM imageand the filtered SEM image (Fig. 5) indicated that this processing

Fig. 5. Flow chart of the image processing procedure of the slice-and-view datasets.

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

5

effectively reduced the high frequency noise of the 2D image signal.However, the edge profile of the silicon/resin interface in the SEM weobtained was not an ideal step edge profile (i.e., an abrupt image

intensity change with sharp discontinuity). In other words, the imageintensity change was not instantaneous but occurred over a finite dis-tance. As highlighted in the line scan of Fig. 5, the silicon/resin edges

Fig. 6. Comparison of (a) SEM image of the sample, (b) slice-and-view reconstructed 3D model in plan view, and (c) 3D reconstructed model. The black arrowindicates the location of the overhang.

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

6

were of ramp-like edge profiles, with an approximate ramp width of200 nm. Such 200 nm ramp width could be associated with a fewfactors: the SE interactive volume across the inclined BSi/resin inter-face, the slightly de-focused SE during automation and the post-imagefiltering. Moreover, the high ion dose received by the sample during theslice-and-view would inevitably result in material amorphisation andfeature smooth-out.

Next, the filtered stack of SEM images was aligned by the Scale-Invariant Feature Transform (SIFT) [53] registration tool2. With over1500 SEM images of 24 MB per image obtained from the automatedslice-and-view procedure, the alignment settings were critical to obtaincontrolled stack drifting while maintaining a highly-efficient calcula-tion speed. All the SIFT alignment setting parameters are listed in theSupplementary Information.

After registration, the outer regions of the stacks were cropped fromthe 3D data set, which removes areas with poor data quality andminimizes the data volume size to reduce the computational resourcesrequired for the subsequent processing. A qualitative visual alignmentcheck was performed after correcting the voxel size and orthographicproject the 3D data set into YZ and XZ directions. From the orthogonalview of Fig. 5, Step 5, we can see that the alignment is not perfect as aslight Z-drifting is observed in the XZ and YZ plane. This Z-driftingcould be eliminated if a fiducial mark was captured together within theSEMs, which can act as a reference point and help improve the post-stack registration.

To extract the 2D profile of the texture from the SEM images, theregistered image data must be analyzed using an edge detection algo-rithm to determine all location points of the silicon/resin interface.However, there is no perfect edge detection solution that is immune tonoise and able to reliably determine the interface location with 100%

accuracy. In this case, edge detection is complicated by the ramp edgeprofiles, which vary with the complicated textured surface feature or-ientation as well as the image blur caused by e-beam interaction vo-lumes between the sample/resin interfaces where the sample's dimen-sion was physically smaller than the minimal detectable edge width[54]. In this work, instead of using the rigorous edge detection algo-rithm or advanced image deconvolution, we simplify the edge detectionby a binary step with identical ‘Isodata’ algorithm3 applied to thresholdall of the SEM images. The impact of different image binarizing algo-rithms in describing the 2D profile is compared in detail in the Sup-plementary Information. The utility of the binary approach, although itmay not be able to accurately identify the features with a size smallerthan the edge width, is sufficient in this work to recover most of thefeatures from the raw SEM images.

A 3D reconstructed surface with dimensions of 21.62 × 15.21 µm2

is shown in Fig. 6 together with the plan-view SEM image of the sample.The insets of Fig. 6. (a) and (b) shows the direct comparison of the plan-view SEM image and the reconstructed 3D model for the BSi sample.Note that the 3D model is not reconstructed from the same location ofthe sample where the SEM image was taken as the tomography char-acterization is destructive. Although the SEM micrograph [Fig. 6(a)]provides better sensitivity for shallow holes, the signal decreases dra-matically when the depth of deep holes exceeds a certain level where allexcited electrons were re-absorbed by the sample. As a comparison, thereconstructed 3D model [Fig. 6(b)] provides better insights into thedepth variation of deep holes.

The binarised stack was then rendered by the Volume Viewer4 toreveal the presence of the overhang structures [Fig. 6(c)] that could notbe detected using an SPM technique. Unfortunately, although the Z-resolution is determined by the thickness of each slice (10 nm), theaccuracy of the 3D model is mainly limited by a combination of factors,including the quality of the 2D SEM image, which is affected by therelative large interaction volume of back scattered electrons, the de-focused SE image, the post-image filtering, and un-optimized edge de-tection. The nano-sized holes visible on top of the RIE in the SEM image[Fig. 6(a)] are ‘smoothed-out’ in the reconstructed 3D model [Fig. 6(b)]because the ramp edge profile's width exceeds the physical size of thefeature. The ‘smoothed-out’ model would inevitably lead to biased re-sults for the absolute surface area, roughness characteristics, and theestimated volume that was etched when fabricating the BSi sample.Since there is currently no convincing solution to evaluate the quality ofedge detection or to quantitatively measure the error [55], we need totake into account the bias when calculations and simulations are carriedout based on this tomographically reconstructed 3D model.

In order to create a 3D reconstruction that is compatible with es-tablished SPM analysis software (e.g. Gwyddion, MountainsSPIP, etc.),the binarised images were outlined by using a one-pixel wide blackoutline (value of ‘0’ in the TIFF format) between the foreground/background interface, and all other pixels were replaced by white(value of ‘255’ in the TIFF format). A custom MATLAB script was de-veloped to generate the 3D surface data. In this script, only the highestZ value for each XY data point will be extracted (e.g., only the topsurface of an overhang will be registered) so that a conventional surfacefunction can be utilized to describe the surface. Some preliminarysurface roughness analysis was conducted using the Gwyddion [56]software using the surface function as shown in Table 2 and Fig. 7.

The surface height profile of the BSi sample, as well as the criticalroughness characteristics, help us to better understand the surface. Forexample, the height distribution of the BSi can be used to determine thesilicon-to-air ratio along the Z-direction, which can be used as the

Table 2Statistical roughness characteristics of the BSi sample calculated from thehighest Z indexed dataset.

Max Height (µm) 2.10Average Height (µm) 0.95RMS roughness (nm) 487Correlation Length (nm) 148Surface Enhancement Ratio (Absolute surface area/Projected area) 5.65

0.0 0.5 1.0 1.5 2.00.0E+00

2.0E+05

4.0E+05

6.0E+05

8.0E+05

1.0E+06

Cou

nts

(a.u

.)

Height (μm)

Height Distribution

Fig. 7. Height distribution of the 3D model from the highest Z indexed dataset.

2 the Scale Invariant Feature Transform (SIFT) [53] registration tool isavailable at the FIJI Plugins list.

3 An iterative thresholding procedure based on the isodata algorithm [59],which is available at the FIJI threshold list.

4 Volume Viewer is a 3D reslicing and threshold-enabled 3D visualization toolwhich is available in the FIJI Plugins list.

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

7

optical simulation input file for effective medium approximation [23]or the surface profile can be directly imported to more rigorous simu-lation techniques such as Finite Difference Time Domain [35,57,58].With insights about critical surface roughness or feature shape, simu-lating the behavior of BSi can provide design criteria for BSi devicefabrication and help improve BSi device performance to suit a widerrange of applications.

6. Conclusion

In this work, we were able to reconstruct the 3D surface morphologyof a BSi with 2 µm tall needles and accurately measure its criticalroughness characteristics. We have adapted the Xe+ plasma FIB toconduct the precise tomographical task, which consists of using Xe+

plasma-based FIB to remove the material rapidly with concurrent op-timized SEM imaging. Through this approach, we were able to char-acterize the surface profile of a BSi sample with high aspect ratio fea-tures that would be incompatible with AFM probes. By incorporating asingle-crystal sacrificial mask,rocking milling and optimizing the ex-periment procedure, the automated PFIB slice thickness was reduceddown to 10 nm. Thanks to the fast material removal rate of Xe+ PlasmaFIB, our dataset is statistically representative for the surface char-acteristics of BSi. The data covered lateral dimensions of 21.62 × 15.21µm2 and a voxel size of 4.21 × 4.21 × 10.00 nm3 was obtained. Ourresults have also proven that Plasma Xe+ FIB is capable of conductingautomated tomography tasks with precise milling requirements. Inorder to obtain the true-2D cross-sectional profile with controlled ar-tifacts, the top surface of a BSi sample needs to be fully infiltrated witha second phase material. Here, we have verified that the more con-ventional Pt-GIS deposition approach can not be successfully used onsamples with a very high surface roughness. In contrast, an alternativeapproach of incorporating the Durcupan resin, which is usually used inembedding biological samples for FIB processing, was shown to fullyinfiltrate the BSi without residual voids. The resin also served as a top-surface protective layer and protected the extreme rough top-surface ofthe BSi sample from potential PFIB artifacts. The sample preparationmethod described in this paper is not limited to the characterization ofBSi, it also gives guidance on FIB processing of high-roughness surfacesfor other purposes, such as TEM lamellae and site-specific defect ana-lysis. Finally, we used the collected SEM stack to reconstruct the 3Dmodel of the BSi surface, which could help us to better understand theunique properties of BSi. In the future, the model can be used as aninput file to conduct various optical simulations and process-relatedsimulations, which provide insights of BSi surface structure correlatedproperties and design guidance for maximizing BSi device performance.

Declaration of Competing Interest

The authors declare that they have no known competing financialinterests or personal relationships that could have appeared to influ-ence the work reported in this paper.

Acknowledgment

This work was supported by the Australian Government through theAustralian Renewable Energy Agency (ARENA: 2017/RND009).Responsibility for the views, information, or advice expressed herein isnot accepted by the Australian Government. The authors acknowledgethe facilities and the scientific and technical assistance of MicroscopyAustralia at the Electron Microscope Unit (EMU) within the MarkWainwright Analytical Centre (MWAC) at UNSW Sydney. The BSifabrication work was supported by the Velux Foundation (project nr.13891) in Denmark. The authors acknowledge the use of the compu-tational cluster Katana supported by Research Technology Services atUNSW Sydney. Yu Zhang appreciates the financial support from the

Australian Government Research Training Program Scholarship.

Supplementary materials

Supplementary material associated with this article can be found, inthe online version, at doi:10.1016/j.ultramic.2020.113084.

References

[1] J. Lv, T. Zhang, P. Zhang, Y. Zhao, S. Li, Review application of nanostructured blacksilicon, Nanoscale Res. Lett. 13 (2018), https://doi.org/10.1186/s11671-018-2523-4.

[2] I. Oh, J. Kye, S. Hwang, Enhanced photoelectrochemical hydrogen production fromsilicon nanowire array photocathode, Nano Lett. 12 (2012) 298–302 https://doi.org/10.1021/nl203564s.

[3] Y. Wang, T. Wang, P. Da, M. Xu, H. Wu, G. Zheng, Silicon nanowires for biosensing,energy storage, and conversion, Adv. Mater. 25 (2013) 5177–5195 https://doi.org/10.1002/adma.201301943.

[4] H.Y. Chen, H.L. Lu, Q.H. Ren, Y. Zhang, X.F. Yang, S.J. Ding, D.W. Zhang, Enhancedphotovoltaic performance of inverted pyramid-based nanostructured black-siliconsolar cells passivated by an atomic-layer-deposited Al<inf>2</inf>O<inf>3</inf>layer, Nanoscale 7 (2015) 15142–15148 https://doi.org/10.1039/c5nr03353e.

[5] D. Dimova-Malinovska, M. Sendova-Vassileva, N. Tzenov, M. Kamenova,Preparation of thin porous silicon layers by stain etching, Thin Solid Films 297(1997) 9–12 https://doi.org/10.1016/S0040-6090(96)09434-5.

[6] M. Saadoun, N. Mliki, H. Kaabi, K. Daoudi, B. Bessaïs, H. Ezzaouia, R. Bennaceur,Vapour-etching-based porous silicon: a new approach, Thin Solid Films 405 (2002)29–34 https://doi.org/10.1016/S0040-6090(01)01757-6.

[7] M. Steglich, D. Lehr, S. Ratzsch, T. Käsebier, F. Schrempel, E.B. Kley,A. Tünnermann, An ultra-black silicon absorber, Laser Photonics Rev. 8 (2014)13–17 https://doi.org/10.1002/lpor.201300142.

[8] M. Kroll, T. Käsebier, M. Otto, R. Salzer, R.B. Wehrspohn, E.-B. Kley,A. Tünnermann, T. Pertsch, Optical modeling of needle like silicon surfaces pro-duced by an ICP-RIE process, SPIE 7725, photonics sol, Energy Syst. III. 7725(2010) 10 https://doi.org/10.1117/12.854596.

[9] Y. Xia, B. Liu, J. Liu, Z. Shen, C. Li, A novel method to produce black silicon for solarcells, Sol. Energy. 85 (2011) 1574–1578 https://doi.org/10.1016/j.solener.2011.03.012.

[10] K.N. Nguyen, D. Abi-Saab, P. Basset, E. Richalot, M. Malak, N. Pavy, F. Flourens,F. Marty, D. Angelescu, Y. Leprince-Wang, T. Bourouina, Study of black siliconobtained by cryogenic plasma etching: approach to achieve the hot spot of athermoelectric energy harvester, Microsyst. Technol. 18 (2012) 1807–1814 https://doi.org/10.1007/s00542-012-1486-0.

[11] H. Savin, P. Repo, G. Von Gastrow, P. Ortega, E. Calle, M. Garín, R. Alcubilla, Blacksilicon solar cells with interdigitated back-contacts achieve 22.1% efficiency, Nat.Nanotechnol. 10 (2015) 624–628 https://doi.org/10.1038/nnano.2015.89.

[12] R.S. Davidsen, H. Li, A. To, X. Wang, A. Han, J. An, J. Colwell, C. Chan, A. Wenham,M.S. Schmidt, A. Boisen, O. Hansen, S. Wenham, A. Barnett, Black silicon laser-doped selective emitter solar cell with 18.1% efficiency, Sol, Energy Mater. Sol.Cells. 144 (2016) 740–747 https://doi.org/10.1016/j.solmat.2015.10.018.

[13] Z.G. Huang, K. Gao, X.G. Wang, C. Xu, X.M. Song, L.X. Shi, Y. Zhang, B. Hoex,W.Z. Shen, Large-area MACE Si nano-inverted-pyramids for PERC solar cell appli-cation, Sol. Energy. 188 (2019) 300–304 https://doi.org/10.1016/j.solener.2019.06.015.

[14] W. Chen, Y. Liu, L. Yang, J. Wu, Q. Chen, Y. Zhao, Y. Wang, X. Du, Difference inanisotropic etching characteristics of alkaline and copper based acid solutions forsingle-crystalline Si, Sci. Rep. 8 (2018) 2–9 https://doi.org/10.1038/s41598-018-21877-x.

[15] P. Li, Y. Wei, X. Tan, X. Li, Y. Wang, Z. Zhao, Z. Yuan, A. Liu, Effective optimizationof emitters and surface passivation for nanostructured silicon solar cells, RSC Adv. 6(2016) 104073–104081 https://doi.org/10.1039/c6ra20945a.

[16] A. Kumagai, Texturization using metal catalyst wet chemical etching for multi-crystalline diamond wire sawn wafer, Sol. Energy Mater. Sol. Cells. 133 (2015)216–222 https://doi.org/10.1016/j.solmat.2014.11.008.

[17] T. Rahman, R.S. Bonilla, A. Nawabjan, P.R. Wilshaw, S.A. Boden, Passivation of all-angle black surfaces for silicon solar cells, Sol. Energy Mater. Sol. Cells. 160 (2017)444–453 https://doi.org/10.1016/j.solmat.2016.10.044.

[18] B. Bhushan, Chapter 2: surface roughness analysis and measurement techniques,Mod. Tribol. Handb. Vol. One Princ. Tribol. CRC Press, 2000, pp. 79–150 https://doi.org/10.1201/9780849377877.

[19] F.J. Giessibl, Advances in atomic force microscopy, Rev. Mod. Phys. 75 (2003)949–975 https://doi.org/10.1103/RevModPhys.75.949.

[20] N. Jalili, K. Laxminarayana, A review of atomic force microscopy imaging systems:Application to molecular metrology and biological sciences, Mechatronics 14(2004) 907–945 https://doi.org/10.1016/j.mechatronics.2004.04.005.

[21] D. Sarid, Review of scanning force microscopy, J. Vac. Sci. Technol. BMicroelectron. Nanom. Struct. 9 (1991) 431 https://doi.org/10.1116/1.585585.

[22] D. Payne, M. Abbott, A.C. Lopez, Y. Zeng, T.H. Fung, J. Cruz-campa, R. Davidson,M. Plakhotnyuk, D. Bagnall, Rapid optical modelling of plasma textured silicon,(2017) 1996.

[23] T.H. Fung, M.U. Khan, Y. Zhang, N.J. Western, D.N.R. Payne, K.R. McIntosh,M.D. Abbott, Improved ray tracing on random pyramid texture via application of

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

8

phong scattering, IEEE J. Photovoltaics. 9 (2019) 591–600 https://doi.org/10.1109/JPHOTOV.2019.2894688.

[24] L. Duan, H. Yi, Y. Zhang, F. Haque, C. Xu, A. Uddin, Comparative study of light- andthermal-induced degradation for both fullerene and non-fullerene-based organicsolar cells, Sustain. Energy Fuels. 3 (2019) 723–735 https://doi.org/10.1039/c8se00567b.

[25] L. Duan, Y. Zhang, H. Yi, F. Haque, R. Deng, H. Guan, Y. Zou, A. Uddin, Trade-offbetween exciton dissociation and carrier recombination and dielectric properties iny6-sensitized nonfullerene ternary organic solar cells, Energy Technol. 1900924(2019) 1–12 https://doi.org/10.1002/ente.201900924.

[26] D. Payne, T.H. Fung, M.U. Khan, J. Cruz-Campa, K. McIntosh, M. Abbott,Understanding the optics of industrial black silicon, AIP Conf. Proc. 2018 (1999),https://doi.org/10.1063/1.5049297.

[27] K. Jäger, On the scalar scattering theory for thin-film solar cells, 2012. http://repository.tudelft.nl/view/ir/uuid:4220e3ee-bdcb-4a46-ade1-470d3c2ad6da/.

[28] E. Ukraintsev, A. Kromka, H. Kozak, Z. Reme, B. Rezek, Artifacts in atomic forcemicroscopy of biological samples, Force Microsc. Investig. Biol. Cell Protein (2012),https://doi.org/10.5772/36203.

[29] J.I. Paredes, A. Martinez-Alonso, J.M.D. Tascon, Adhesion artefacts in atomic forcemicroscopy imaging, J. Microsc. 200 (2000) 109–113 https://doi.org/10.1046/j.1365-2818.2000.00755.x.

[30] U.D. SCHWARZ, H. HAEFKE, P. REIMANN, H. ‐J GÜNTHERODT, Tip artefacts inscanning force microscopy, J. Microsc. 173 (1994) 183–197 https://doi.org/10.1111/j.1365-2818.1994.tb03441.x.

[31] E.P. Ivanova, J. Hasan, H.K. Webb, G. Gervinskas, S. Juodkazis, V.K. Truong,A.H.F. Wu, R.N. Lamb, V.A. Baulin, G.S. Watson, J.A. Watson, D.E. Mainwaring,R.J. Crawford, Bactericidal activity of black silicon, Nat. Commun. 4 (2013) 1–7https://doi.org/10.1038/ncomms3838.

[32] J.W. Haus, Nanocharacterization, in: Fundam. Appl. Nanophotonics, 1st ed., 2016:pp. 185–210. https://doi.org/10.1016/B978-1-78242-464-2.00006-3.

[33] L. Holzer, M. Cantoni, Review of FIB Tomography, in: and P.R. Ivo Utke, StanislavMoshkalev (Ed, Nanofabrication Using Focus. Ion Electron Beams (2012) 410–435.

[34] T. Ishitani, K. Umemura, T. Ohnishi, T. Yaguchi, T. Kamino, Improvements inperformance of focused ion beam cross-sectioning: aspects of ion-sample interac-tion, J. Electron Microsc. (Tokyo). 53 (2004) 443–449 https://doi.org/10.1093/jmicro/dfh078.

[35] D.A. Saab, P. Basset, F. Marty, D.E. Angelescu, M. Trawick, Accurate 3D re-construction of silicon micro/nanostructures, based on high resolution FIB-SEMtomography: application to black silicon, DTIP 2014 - Symp. Des. Test, Integr.Packag. MEMS/MOEMS, 2014, pp. 4–7 https://doi.org/10.1109/DTIP.2014.7056695.

[36] L.A. Giannuzzi, F.A. Stevie, A review of focused ion beam milling techniques forTEM specimen preparation 10_1016-S0968-4328(99)00005-0 MicronScienceDirect_com, Micron 30 (1999) 197–204 http://zenobi.ethz.ch/content/dam/ethz/special-interest/chab/organic-chemistry/zenobi-group-dam/documents/Education/LecturesExercises/Analytical Strategy 2014/HS2015/WepfAdditional Material 1.pdf.

[37] M.D. Uchic, L. Holzer, B.J. Inkson, E.L. Principe, P. Munroe, Three-dimensionalmicrostructural characterization using focused ion beam tomography, MRS Bull. 32(2007) 408–416 https://doi.org/10.1557/mrs2007.64.

[38] T.L. Burnett, R. Kelley, B. Winiarski, L. Contreras, M. Daly, A. Gholinia, M.G. Burke,P.J. Withers, Large volume serial section tomography by Xe Plasma FIB dual beammicroscopy, Ultramicroscopy 161 (2016) 119–129 https://doi.org/10.1016/j.ultramic.2015.11.001.

[39] X. Tan, Z. Tao, M. Yu, H. Wu, H. Li, Anti-reflectance investigation of a micro-nanohybrid structure fabricated by dry/wet etching methods, Sci. Rep. 8 (2018) 1–11https://doi.org/10.1038/s41598-018-26234-6.

[40] N.S. Smith, W.P. Skoczylas, S.M. Kellogg, D.E. Kinion, P.P. Tesch, O. Sutherland,A. Aanesland, R.W. Boswell, High brightness inductively coupled plasma source forhigh current focused ion beam applications, J. Vac. Sci. Technol. B Microelectron.

Nanom. Struct. 24 (2006) 2902–2906 https://doi.org/10.1116/1.2366617.[41] L. Kwakman, G. Franz, M.M.V. Taklo, A. Klumpp, P. Ramm, Characterization and

failure analysis of 3D integrated systems using a novel plasma-FIB system, AIP Conf.Proc. 1395 (2011) 269–273 https://doi.org/10.1063/1.3657902.

[42] F. Altmann, S. Klengel, J. Schischka, M. Petzold, Defect analysis using highthroughput plasma FIB in packaging reliability investigations, Proc. - Electron.Components Technol. Conf. 2013, pp. 1940–1945 https://doi.org/10.1109/ECTC.2013.6575843.

[43] S. Subramaniam, J. Huening, J. Richards, K. Johnson, A comprehensive approachtowards optimizing the xenon plasma focused ion beam instrument for semi-conductor failure analysis applications, Microsc. Microanal. 23 (2017) 769–781https://doi.org/10.1017/S1431927617000563.

[44] T.L. Burnett, B. Winiarski, R. Kelley, X.L. Zhong, I.N. Boona, D.W. McComb,K. Mani, M.G. Burke, P.J. Withers, Xe+ Plasma FIB: 3D microstructures fromnanometers to hundreds of micrometers, Micros. Today. 24 (2016) 32–39 https://doi.org/10.1017/s1551929516000316.

[45] K.C. B. A.G., D.J. H. B.M., FIB-SEM Tomography in Biology, Kuo J, Humana Press,Totowa, NJ, 2014https://doi.org/10.1007/978-1-62703-776-1_24.

[46] A.J. Bushby, K.M.Y. P'Ng, R.D. Young, C. Pinali, C. Knupp, A.J. Quantock, Imagingthree-dimensional tissue architectures by focused ion beam scanning electron mi-croscopy, Nat. Protoc. 6 (2011) 845–858 https://doi.org/10.1038/nprot.2011.332.

[47] G. Knott, S. Rosset, M. Cantoni, Focussed ion beam milling and scanning electronmicroscopy of brain tissue, J. Vis. Exp. (2011) 3–6 https://doi.org/10.3791/2588.

[48] Y. Zhang, D.N.R. Payne, C. Kong, M.U. Khan, T.H. Fung, R.S. Davidsen, O. Hansen,G. Scardera, M.D. Abbott, B. Hoex, Advanced characterisation of black siliconsurface topography with 3D PFIB-SEM, in: 2020: pp. 0825–0828. https://doi.org/10.1109/pvsc40753.2019.8980830.

[49] T.H.P. Chang, Proximity effect in electron-beam lithography, J, Vac. Sci. Technol.12 (1975) 1271–1275 https://doi.org/10.1116/1.568515.

[50] S. Kellogg, R. Schampers, S. Zhang, A. Graupera, T. Miller, W. Laur, A. Dirriwachter,High throughput sample preparation and analysis using an inductively coupledplasma (ICP) focused ion beam source, Microsc. Microanal. 16 (2010) 222–223https://doi.org/10.1017/S1431927610058733.

[51] Auto Slice and View 4 Software, (n.d.) 1–2. https://www.fei.com/documents/auto-slice-view-4-datasheet/.

[52] J. Schindelin, I. Arganda-Carrera, E. Frise, K. Verena, L. Mark, P. Tobias, P. Stephan,R. Curtis, S. Stephan, S. Benjamin, T. Jean-Yves, J.W. Daniel, H. Volker, E. Kevin,T. Pavel, C. Albert, Fiji - an Open platform for biological image analysis, Nat.Methods. 9 (2009) 241 https://doi.org/10.1038/nmeth.2019.Fiji.

[53] D.G. Lowe, Distinctive image features from scale-invariant keypoints, Int. J.Comput. Vis. 60 (2004) 91–110 https://doi.org/10.1023/B:VISI.0000029664.99615.94.

[54] R. Jain, R. Kasturi, B.G. Schunck, Edge Detection, in: Mach. Vis. McGraw-Hill, Inc.,1995, pp. 140–185 https://doi.org/10.1016/b978-012206093-9/50008-3.

[55] C. Lopez-Molina, B. De Baets, H. Bustince, Quantitative error measures for edgedetection, Pattern Recognit. 46 (2013) 1125–1139 https://doi.org/10.1016/j.patcog.2012.10.027.

[56] D. Nečas, P. Klapetek, Gwyddion: An open-source software for SPM data analysis,Cent. Eur. J. Phys. 10 (2012) 181–188 https://doi.org/10.2478/s11534-011-0096-2.

[57] T.H. Fung, T. Veeken, D.N.R. Payne, B. Veettil, A. Polman, M.D. Abbott, Applicationand validity of the effective medium approximation to the optical properties ofnano-textured silicon coated with a dielectric layer, 27 (2019) 38645–38660.

[58] S.K. Kim, R.W. Day, J.F. Cahoon, T.J. Kempa, K.D. Song, H.G. Park, C.M. Lieber,Tuning light absorption in core/shell silicon nanowire photovoltaic devices throughmorphological design, Nano Lett. 12 (2012) 4971–4976 https://doi.org/10.1021/nl302578z.

[59] T.W. Ridler, S. Calvard, Picture thresholding using an iterative selection method,IEEE Trans. Syst. Man Cybern. smc 8 (1978) 630–632.

Y. Zhang, et al. Ultramicroscopy 218 (2020) 113084

9