772 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 7, NO. 3, MAY 2017

Microscopic Distributions of Defect LuminescenceFrom Subgrain Boundaries in Multicrystalline

Silicon WafersHieu T. Nguyen, Mallory A. Jensen, Li Li, Christian Samundsett, Hang C. Sio, Barry Lai, Tonio Buonassisi,

and Daniel Macdonald

Abstract—We investigate the microscopic distributions ofsub-band-gap luminescence emission (the so-called D-linesD1/D2/D3/D4) and the band-to-band luminescence intensity, nearrecombination-active subgrain boundaries in multicrystallinesilicon wafers for solar cells. We find that the sub-band-gap lu-minescence from decorating defects/impurities (D1/D2) and fromintrinsic dislocations (D3/D4) has distinctly different spatial dis-tributions, and is asymmetric across the subgrain boundaries. Thepresence of D1/D2 is correlated with a strong reduction in the band-to-band luminescence, indicating a higher recombination activity.In contrast, D3/D4 emissions are not strongly correlated with theband-to-band intensity. Based on spatially resolved, synchrotron-based micro-X-ray fluorescence measurements of metal impuri-ties, we confirm that high densities of metal impurities are presentat locations with strong D1/D2 emission but low D3/D4 emission.Finally, we show that the observed asymmetry of the sub-band-gapluminescence across the subgrain boundaries is due to its inclina-tion below the wafer surface. Based on the luminescence asym-metries, the subgrain boundaries are shown to share a commoninclination locally, rather than being orientated randomly.

Index Terms—Crystalline silicon, dislocations, grain bound-aries, photoluminescence (PL), photovoltaic cells.

Manuscript received February 19, 2017; revised March 15, 2017; acceptedMarch 16, 2017. Date of publication March 30, 2017; date of current versionApril 19, 2017. This work was supported in part by the Australian ResearchCouncil and in part by the Australian Renewable Energy Agency under GrantRND009. The work of M. A. Jensen was supported by the National ScienceFoundation Graduate Research Fellowship under Grant 1122374. This workused resources of the Advanced Photon Source, a U.S. Department of Energy(DOE) Office of Science User Facility operated for the DOE Office of Scienceby the Argonne National Laboratory under Contract no. DE-AC02-06CH11357.

H. T. Nguyen, C. Samundsett, H. C. Sio, and D. Macdonald arewith the Research School of Engineering, Australian National Univer-sity, Canberra, A.C.T. 2601, Australia (e-mail: hieu.nguyen@anu.edu.au;chris.samundsett@anu.edu.au; kelvin.sio@anu.edu.au; daniel.macdonald@anu.edu.au).

M. A. Jensen and T. Buonassisi are with the Massachusetts Institute ofTechnology, Cambridge, MA 02139 USA (e-mail: jensenma@mit.edu; buonas-sisi@mit.edu).

L. Li is with the Australian National Fabrication Facility, Department ofElectronic Materials Engineering, Australian National University, Canberra,A.C.T. 2601, Australia (e-mail: lily.li@anu.edu.au).

B. Lai is with the Advanced Photon Source, Argonne National Laboratory,Argonne, IL 60439 USA (e-mail: blai@aps.anl.gov).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JPHOTOV.2017.2684904

I. INTRODUCTION

SUB-BAND-GAP photoluminescence (PL) spectroscopyhas been demonstrated to be a powerful technique to study

defects and impurities in silicon (Si) photovoltaics [1]–[3].Many defects and impurities can emit distinct sub-band-gapPL peaks at lower temperatures, such as dislocations [4]–[13],oxygen precipitates [14]–[16], iron precipitates [17], and Cr–Bpairs [18]. Recently, with the advent of so-called hyperspec-tral imaging techniques, both spatial and spectral informationof PL signals emitted from Si wafers can be captured simulta-neously. Therefore, different radiative recombination centers atvarious locations in Si wafers and solar cells can be investigatedindependently [19]–[22].

In multicrystalline Si (mc-Si) solar cells, dislocation clustersare a major efficiency limiting factor [23]. In mc-Si wafers, largeangle grain boundaries can act as relaxation sites for the highthermal stress and strain during the crystal growth, and thusminimize the formation of dislocations nearby. On the contrary,inside the grains, the material still suffers a high level of stressand strain, causing an elastic bending of the crystal structure.The bending energy can be relaxed by the introduction of dislo-cations. The further the lattice distorted, the more dislocationsneed to be introduced to compensate. These dislocations canaggregate to produce a small angle grain boundary or a subgrainboundary (sub-GB), across which there is a small but permanentmisorientation of the crystal lattice (generally below 2°). Due tothe formation mechanism of these sub-GBs, they always con-tain a high density of dislocations. The residual stress and strainaround these dislocations, in turn, can preferentially trap otherdefects and impurities, and become very recombination active.As such, sub-GBs have detrimental effects on the performanceof mc-Si solar cells [24]. Therefore, an improved understandingof recombination activities of sub-GBs after different process-ing steps is necessary for mitigating their detrimental impactson cell performance.

The sub-GBs are known to emit four distinct sub-band-gappeaks called D1, D2, D3, and D4. The doublet D3/D4 reflectsintrinsic properties of dislocations since it is found to be confinedaround sub-GBs [2], [5], [6], [8], [12]. Meanwhile, the doubletD1/D2 is thought to be emitted from secondary defects andimpurities trapped around the dislocations since it is distributedacross a wider area from the sub-GBs [2], [5], [6], [8], [12].

2156-3381 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

NGUYEN et al.: MICROSCOPIC DISTRIBUTIONS OF DEFECT LUMINESCENCE FROM SUBGRAIN BOUNDARIES 773

However, D1/D2 and D3/D4 are not always present togetheralong sub-GBs, and the D1/D2 intensity may not be correlatedwith the D3/D4 intensity. In addition, the relative impact ofD1/D2 and D3/D4 on the band-to-band (BB) luminescence fromSi, which reflects the carrier lifetime and therefore the materialquality, remains unclear. If the spatial resolution of PL imagingand PL spectroscopy maps is not high enough, it is likely thatthe spatial distribution of the D3/D4 luminescence is maskedby that of D1/D2, since D1/D2 has a wider spatial distributionaround sub-GBs than D3/D4 [6], [8], [12]. Moreover, recentlywe have reported that the D lines are very often asymmetricallydistributed across sub-GBs in mc-Si wafers [25], [26]. However,there are no conclusive reports on the reasons for this asymmetryto date.

In this work, utilizing the micron-scale spatial resolution of amicro-PL (µ-PL) spectroscopy system, we perform spectral PLmeasurements directly at sub-GBs and the surrounding regionsin order to study their luminescence behaviors. We first reviewsome important characteristics of sub-band-gap luminescencefrom sub-GBs, which are the basis for our later discussions. Wethen examine correlations among intensities of the D lines andthe main BB PL peak after different processing steps, includ-ing the as-cut state, after phosphorus gettering, and after hy-drogenation. In addition, using synchrotron-based micro-X-rayfluorescence spectroscopy (µ-XRF) measurements, we examinethe presence of metal impurities at and around sub-GBs emit-ting strong D-line intensities. Finally, we present and explain theasymmetric distribution of the D lines across sub-GBs, whichprovides insight into the physical arrangements of sub-GBs be-low the wafer surface. These results are validated by secondaryelectron microscope (SEM) and transmission electron micro-scope (TEM) images.

II. EXPERIMENTAL DETAILS

The investigated samples are three standard directionallysolidified industrially grown p-type mc-Si wafers having a back-ground boron doping concentration of 9 × 1015 cm−3. Theywere sister wafers cut consecutively from the same ingot. First,they were chemically etched in an HF/HNO3 solution to re-move saw damage and to achieve planar surfaces. After that,one wafer was kept in the as-cut state. The second wafer wentthrough an extended phosphorous gettering process [27]. Thethird wafer was passivated with a layer of SiNx :H deposited bythe plasma-enhanced chemical vapor deposition technique. Itwas then annealed at 700 °C for 30 min in a N2 gas environmentin a quartz tube furnace to distribute hydrogen throughout thewafer thickness. All three wafers were then chemically etchedagain to reveal bare Si surfaces. Finally, they were immersed ina defect etchant consisting of acetic/HNO3 /HF for 16 h in or-der to reveal sub-GBs [28], [29], which are otherwise invisibleunder an optical microscope.

The µ-PL spectroscopy system employed in this study is aHoriba LabRAM equipped with confocal optics. The incidentlaser light was focused into the sample surface via a 50× objec-tive lens whose numerical aperture is 0.55. The emitted PL lightwas directed into a monochromator whose grating was set at

Fig. 1. Example of a PL spectrum captured at a sub-GB of the as-cut mc-Si wafer, excited with the 810-nm light at 80 K. The peak ∼1130 nm is themain band-to-band (BB) peak emitted from Si, assisted by the emission of atransverse-optical phonon. The spectral region above 1170 nm was decomposedinto five separate Gaussian peaks, corresponding to the D1–D4 lines, and anoptical zone-center phonon replica of the main BB peak [35], denoted as phononreplica of band-to-band (PRBB).

150 grooves/mm, and then was detected by a liquid-nitrogen-cooled InGaAs array detector. The mapping stage had a min-imum step size of 0.1 µm in both X and Y directions. Twoexcitation light sources were employed in this study—a diode-pumped solid-state 532-nm laser, and a supercontinuum NKTlaser whose emission wavelength is tunable between 480 nm and2 µm. A wavelength of 810-nm with a bandwidth of 10 nm wasused in this study. The power was kept constant at 6 mW for allmeasurements. The illuminated spot size was∼1 and∼2 µm forthe 532 and 810-nm excitation light sources, respectively. Thesample temperature was controlled by a liquid-nitrogen-cooledLinkam stage. The spectral response of the entire system wasdetermined by a calibrated halogen–tungsten light source. Themisorientation angles of sub-GBs were determined by electronback scattering diffraction (EBSD) measurements, and foundto be less than 2° for all investigated sub-GBs. The overallconcentration of interstitial oxygen [Oi] in each sample was de-termined using Fourier transform infrared spectroscopy by cal-ibrating the results with a Czochralski Si (Cz-Si) wafer whose[Oi] was known, and was found to be around 2.5 × 1017 cm−3

for the three investigated samples. The µ-XRF scans were per-formed at Argonne National Laboratory’s Advanced PhotonSource Beamline 2-ID-D using an incident X-ray energy of9 keV with a full-width half-maximum beam spot size of∼200 nm [30]. Sub-GBs of interest were located using anoptical microscope aligned with the beam direction. At least20 × 20 µm2 area of each sub-GB was mapped in flyscan mode,which increases the measurement throughput compared to thestandard step-scan mode [31], in 220 nm steps. Further detailsof µ-XRF measurement and analysis procedures can be foundin [32]–[34].

III. PL SPECTRA FROM SUB-GBS

In this section, we review some important properties of theso-called D lines emitted from sub-GBs in mc-Si wafers, whichwill serve as a basis for our discussions in later sections. Fig. 1shows an example of a PL spectrum captured at a sub-GB of the

774 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 7, NO. 3, MAY 2017

Fig. 2. Comparison between a single PL spectrum (captured directly at asub-GB) and an average spectrum (from 31 spectra captured from a line scan±15 µm across the sub-GB) of the as-cut mc-Si wafer, excited with the 810-nmlight at 80 K.

as-cut mc-Si wafer at 80 K. The PL peak located ∼1130 nm isthe main BB peak emitted from Si, assisted by the emission of atransverse-optical phonon. Besides that, we can observe anotherfour distinct sub-band-gap peaks denoted as D1–D4 in Fig. 1.The doublet D3/D4 was reported to reflect the intrinsic proper-ties of dislocations by many authors since its spatial distributionwas confined around dislocation cores or sub-GBs [2], [5], [6],[8], [12]. Meanwhile, the doublet D1/D2 was demonstrated tooriginate from secondary defects and impurities trapped by thestrain field around the dislocations, and thus their spatial distri-bution was extended further away from the dislocation cores orsub-GBs [2], [5], [6], [8], [12]. In Fig. 1, we can numerically de-compose the spectrum into five different Gaussian peaks, corre-sponding to the D1–D4 lines, and an optical zone-center phononreplica of the main BB peak [35], denoted as phonon replica ofBB. The main BB line is well-resolved from the other lines,and thus its intensity can be determined precisely without be-ing decomposed from the other lines. Based on the Gaussiandistributions in Fig. 1, integrated intensity mappings taken be-tween 1400 and 1570 nm and between 1220 and 1320 nm can beused to represent the doublets D1/D2 and D3/D4, respectively,without significantly overlapping with adjacent peaks.

Next, Fig. 2 compares a single PL spectrum (captureddirectly at a sub-GB) with an average PL spectrum (averageof 31 spectra captured from a line scan ± 15 µm across thesub-GB) of the mc-Si wafer at 80 K. The single spectrumcaptured at the sub-GB displays high intensities of D3 andD4, whereas the average spectrum displays only a small peakfor D4. Since D3 and D4 are confined around the sub-GBand their intensities are reduced when moving away from thesub-GB, whereas D1 and D2 have broader spatial distributionsaround the sub-GB, the relative intensities of D3 and D4 arereduced significantly in the average spectrum. Therefore, ifthe illuminated spot size is too large, the D1 and D2 linescould overshadow the D3 and D4 lines, and thus affect ourinterpretations on the D line spatial distributions. In this work,the illuminated spot is only 1–2 µm in diameter, allowingmicron-scale spatial mappings of the D lines around sub-GBs.

Another important property of the D lines is the different ther-mal quenching rates between D1/D2 and D3/D4 [4], [12], [36].

Fig. 3. PL spectra at two different sub-GBs from the as-cut mc-Si wafer,excited with the 810-nm light at (a) 80 K and (b) 300 K.

The intensity of the D3/D4 doublet is reduced more quicklyas the temperature rises than that of the D1/D2 doublet. Fig. 3shows PL spectra captured at two different sub-GBs at 80 and300 K. Compared to sub-GB 2, sub-GB 1 emits strong D3/D4 butminimal D1/D2 at 80 K. However, at 300 K, the D3/D4 doubletof sub-GB 1 disappears, whereas the D1/D2 doublet of sub-GB2 is still present with a relatively strong intensity. Therefore, thePL signal around 1300 nm from sub-GB 2 at room temperaturemust be the tail of D1/D2, rather than a residue of D3/D4. Thesignatures in Fig. 3(b) suggest that even though some sub-GBsemit strong D3/D4 (intrinsic properties of dislocations) at lowtemperatures, sub-band-gap PL mappings at room temperaturelikely only contain signatures of D1/D2 (defects and impuritiestrapped around dislocations) [37]–[40]. Therefore, this workwill focus on the luminescence at low temperatures in order toinvestigate properties of sub-GBs in mc-Si wafers.

IV. CORRELATIONS AMONG D LINES AND BB LINE AFTER

DIFFERENT PROCESSING STEPS

In this section, we investigate correlations among the in-tensities of the various D lines and the BB line. Fig. 4 plotsthe intensities of BB versus all D lines [see Fig. 4(a)], BBversus D1 plus D2 [see Fig. 4(b)], BB versus D3 plus D4 [seeFig. 4(c)], D1 versus D2 [see Fig. 4(d)], D3 versus D4 [seeFig. 4(e)], and D1 plus D2 versus D3 plus D4 [see Fig. 4(f)] inlogarithmic scales for the as-cut mc-Si wafer. The intensities ofthe D lines are determined from the areas of their correspondingGaussian fits, whereas that of the BB line is the integrated signalbetween 1070 and 1150 nm, since there is no overlap betweenthe D lines and the BB line in this spectral range. Each datapoint was measured from a separate sub-GB, and the number ineach figure is the correlation coefficient of the pairs.

In Fig. 4(a), the BB intensity is inversely correlated with thetotal intensity of the D lines, i.e., the sub-band-gap lumines-cence globally reflects a reduction of material quality for photo-voltaic applications. The findings are consistent with the results

NGUYEN et al.: MICROSCOPIC DISTRIBUTIONS OF DEFECT LUMINESCENCE FROM SUBGRAIN BOUNDARIES 775

Fig. 4. Correlations among intensities of the D lines and main BB line in the as-cut mc-Si wafer. (a) BB versus all D lines. (b) BB versus (D1+D2). (c) BBversus (D3+D4). (d) D1 versus D2. (e) D3 versus D4. (f) (D1+D2) versus (D3+D4). The intensities of the D lines are the areas of Gaussian fits, whereas theintensity of the BB line is the integration between 1070 and 1150 nm.

reported by Johnston et al. [39], in which the authors saw astrong inverse correlation between the BB luminescence andthe defect luminescence, although the measurements were per-formed at room temperature and the defect luminescence wasa very broad and featureless peak. Moreover, we found that theBB intensity has a significantly stronger correlation with thetotal intensity of D1 plus D2 [see Fig. 4(b)] compared the to-tal intensity of D3 plus D4 [see Fig. 4(c)]. This indicates thatthe D1/D2 emission is a better indicator of high recombinationactivity at subgrain boundaries.

In addition, D1 and D2 are strongly correlated together [seeFig. 4(d)], and so are D3 and D4 [see Fig. 4(e)], consistentwith the fact that D3/D4 are emitted from dislocation coresand D1/D2 are emitted from secondary defects and impurities.However, we found no correlation between the total intensitiesof D1 plus D2 and D3 plus D4, as depicted in Fig. 4(f). Thisresult suggests that the two doublets D1/D2 and D3/D4 do notnecessarily appear together because of their different origins. Infact, Mehl et al. [41] also reported no correlation between thespatial distributions of D1 and D4 using a hyperspectral imagingtechnique, although the spatial resolution was on the order of ahundred of micrometers and D3/D4 could be potentially offsetby D1/D2 at sub-GBs (see, for example, Fig. 2). With the µ-PLtool, we can perform the measurements directly at sub-GBs andthus can find certain sub-GBs which emit only D1/D2 withoutD3/D4 or vice versa. Table I summarizes the correlation coef-ficients among intensities (in logarithmic scales) of the D linesand BB line for all three samples (as-cut, phosphorus gettered,and hydrogenated). These correlations were found to be con-sistent regardless of processing steps, suggesting that getteringand hydrogenation do not significantly alter the PL emissionproperties of subgrain boundaries.

Next, Fig. 5 plots the integrated intensity mappings of D1plus D2 (column 1), D3 plus D4 (column 2), and the BB line(column 3) from four different regions in the as-cut mc-Si wafer.Respectively, the mappings of D1 plus D2, D3 plus D4, and BBare the integrated intensities between 1400 and 1570 nm, 1220and 1320 nm, and 1070 and 1150 nm. These regions of the PLspectra were chosen for mapping since the two doublets D1/D2and D3/D4, and the BB line are not significantly overlapped witheach other within these wavelength ranges, as can be observedin Fig. 1. In Fig. 5, the D3/D4 patterns are more localizedaround sub-GBs than the D1/D2 patterns. The faint, thin darklines within the bright patterns in the mappings of D1/D2 andD3/D4 are artifacts due to etch grooves along the sub-GBs afterthe defect etch. These indicate the positions of the sub-GBs,based on which we can observe an asymmetric distribution ofthe D lines around the sub-GBs. Section VI will discuss thisasymmetry in more detail.

Furthermore, we can observe that dark patterns of the BBmappings follow closely bright patterns of the D1/D2 map-pings consistently for all regions. However, bright patterns ofthe D3/D4 mappings follow neither bright patterns of D1/D2nor dark patterns of BB with consistency. These results confirmthat the doublet D1/D2 has a stronger correlation with the re-combination activity of the sub-GBs than the doublet D3/D4.Moreover, red rectangles in Fig. 5 mark regions of high D3/D4intensities but absent D1/D2, whereas broken yellow rectan-gles mark the regions of high D1/D2 intensities but absentD3/D4. We can see clearly that when D1/D2 is absent, eventhough D3/D4 is high, the contrast of these areas in the BBmappings is minimal. On the other hand, when D1/D2 is highbut D3/D4 is absent, there is a significant reduction in the BBsignal. The results confirm that the sub-GBs emitting a high

776 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 7, NO. 3, MAY 2017

TABLE ICORRELATION COEFFICIENTS AMONG INTENSITIES (LOGARITHMIC SCALE) OF THE D LINES AND BB LINE AFTER DIFFERENT PROCESSING STEPS

AT SUB-GBS OF MC-SI WAFERS

Process steps BB versus all D lines BB versus D1+D2 BB versus D3+D4 D1 versus D2 D3 versus D4 D1+D2 versus D3+D4

As-cut –0.57 –0.54 –0.28 0.73 0.7 –0.14Phosphorus gettering –0.5 –0.43 –0.25 0.69 0.73 0.08Hydrogenation –0.46 –0.52 –0.15 0.77 0.7 –0.18

D1 plus D2 and D3 plus D4 are not correlated, and D3 plus D4 has a very weak correlation with the BB intensity, as indicated by the shaded cells.

Fig. 5. Intensity mappings of D lines and BB line at different regions in the as-cut mc-Si wafer, excited with the 810-nm light source at 80 K. The mapping stepsare 1 µm in both X and Y directions. The brighter color indicates a higher intensity. Broken yellow rectangles indicate regions with D1/D2, but no D3/D4. Redrectangles indicate region with D3/D4, but no D1/D2. Red arrows indicate regions with the strongest D3/D4 intensities. A1, A2, and C1 indicate the regions forwhich µ-XRF scans were performed and shown in Fig. 6. The optical images of the four investigated regions are shown in column 4. Positions of sub-GBs arerevealed by etch grooves after the defect etch.

D1/D2 intensity are more recombination active than the sub-GBs emitting only D3/D4. These findings can be compared withthe results reported by Bauer et al. [42]. These authors foundstrong electron beam induced current contrasts along sub-GBswith the presence of nonsplit Lomer dislocations, but minimalcontrasts along sub-GBs with the presence of only partial dislo-cations and stacking faults. The sub-GBs emitting only D3/D4may, therefore, contain only partial dislocations and stackingfaults. However, a detailed study on the microscopic structures

of dislocations at these sub-GBs is required to confirm thishypothesis.

V. IMPACTS OF METAL IMPURITIES ON D LINES

In this section, we combine both µ-PL mappings and µ-XRFscans in order to provide more insights about the origins of theD lines. We performed µ-XRF scans for regions marked by A1(40 × 40 µm2), A2 (40 × 40 µm2), and C1 (20 × 20 µm2) in

NGUYEN et al.: MICROSCOPIC DISTRIBUTIONS OF DEFECT LUMINESCENCE FROM SUBGRAIN BOUNDARIES 777

Fig. 6. µ-XRF maps of Fe (column 2), Cu (column 3), Ni (column 4), and Ca (column 5) in regions marked by A1 (row 1), A2 (row 2), and C1 (row 3) inFig. 5. High densities of metal impurities are present in regions A1 and C1 (high D1/D2, no D3/D4), whereas those in region C1 (high D3/D4, no D1/D2) are belowthe detection limit. The maps are displayed in logarithmic scales (10ˆ), where the color-bar labels represent the exponent and the unit is ng/cm2. The circles in themap at column 2, row 3 indicate the locations with high Fe counts compared to the background. The detection limit for a given µ-XRF measurement configurationwas quantified by measuring National Institute of Standards and Technology (NIST) standard reference materials 1832 and 1833 in the same configuration as thesample. These limits of Fe, Cu, Ni, and Ca are 19, 9, 12, and 83 ng/cm2, respectively, for region A1; 13, 6, 9, 59 ng/cm2 for region A2; and 9, 4, 5, 37 ng/cm2 forregion C1.

Fig. 5, and show the results for different metals (Fe, Cu, Ni,and Ca) in Fig. 6. The first point to notice in this figure is thatthe densities of metal precipitates in regions A1 and C1 (highD1/D2 but no D3/D4) are high, whereas those in region A2(high D3/D4 but no D1/D2) are below the detection limit. Inaddition, in regions A1 and C1, metal precipitates are presentnot only along the sub-GBs but also in surrounding areas, indi-cating that these regions are highly contaminated with metals,perhaps even before the sub-GB formation during ingot cooling.In other words, these sub-GBs form in a metal rich environmentin which metals are present in various forms (precipitates and asdissolved impurities). If these sub-GBs had been contaminatedwith metals after their formation, the metal precipitates would beexpected to largely occur directly along the sub-GBs, as demon-strated in µ-XRF results by several authors [32]–[34]. Note that,although we could not detect metal precipitates in region A2,this region could also be contaminated with metal impurities orit could have formed in a contaminated environment, as it is only∼50 µm away from region A1 (heavily contaminated). How-ever, the contamination level in A2 is clearly much lower thanthose in A1 and C1.

The D lines (D1–D4) have previously been observed in bothhigh-purity plastically deformed float-zone and Cz-Si wafersby numerous authors [4], [5], [43], [44], indicating that metalimpurities and/or precipitates are not always the direct causes ofthe D lines. However, the D-line intensities could be enhancedwhen the wafers are moderately contaminated with metal im-purities, such as Fe and/or Cu, as reported by Lightowlers andHiggs [6], Kittler et al. [9], and Tajima et al. [11]. Sub-GBs in

mc-Si wafers are the result of dislocation accumulation duringthe ingot growth and cooling, and thus must contain a high den-sity of dislocations. The fact that some sub-GBs emit D1/D2 butnot D3/D4 (for example, A1 and C1 in Fig. 5) demonstrates thata high density of metal impurities may have quenched the lumi-nescence efficiency of the dislocation cores at these sub-GBs,but not the luminescence of the secondary defects/impurities.Meanwhile, the sub-GBs emitting D3/D4 but not D1/D2 (forexample, A2 in Fig. 5) are less contaminated than the sub-GBsalso emitting D1/D2. However, the D3/D4 intensity is not thestrongest at the sub-GBs with absent D1/D2, but at the sub-GBparts with weak D1/D2 as indicated by red arrows in Fig. 5.This behavior is consistent with the fact that a moderate levelof metal contamination could enhance all the D-line intensities[6], [9], [11].

We note that many works have found that oxygen precipitatescan also emit sub-band-gap PL peaks in the spectral range ofD1/D2 in both Cz-Si and mc-Si wafers [12], [14]–[16]. Althoughoxygen cannot be detected by the µ-XRF scans, it could besegregated at sub-GB regions, forming precipitates. Therefore,the observed D1/D2 peaks could partly contain the PL signalemitted from the oxygen precipitates if any. In addition, the D1–D4 lines were demonstrated to be emitted by the recombinationof trapped carriers between a deep level and a shallow level[4], [5], [7], [36]. This two-level model is implied from thestrong temperature dependence of the D lines, suggesting theparticipation of shallow levels; and also from the large energydifference between the D lines and the BB line, suggesting thepresence of deeper levels. That fact that all the D lines are much

778 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 7, NO. 3, MAY 2017

Fig. 7. Comparison of integrated intensity mappings of D1 plus D2 (columns 1 and 2) and D3 plus D4 (columns 3 and 4) between the 810-nm (columns 1 and3) and 532-nm (columns 2 and 4) excitation at different regions in the as-cut mc-Si wafer at 80 K. The mapping steps are 1 µm in both X and Y directions. Redarrows indicate the positions of the sub-GBs. The brighter color indicates a higher intensity.

Fig. 8. (a) Fig. 7(a) was rotated 90° counter clockwise (D1 plus D2 intensitymapping in region A with the 810-nm excitation light at 80 K). (b) SEM imageof the area marked in (a). The very dark pattern around the sub-GB is due todamage caused by the focused ion beam during the milling process. (c) Brightfield TEM image of the vertical cross-section of the sub-GB indicated in (b).The wavy curves are bending contours due to the elastic bending of the thinTEM specimen.

broader than the main BB peak suggests that these energy levelsare not well-defined, but rather form continuous bands insidethe band gap of silicon.

VI. ASYMMETRIC DISTRIBUTION OF D LINES

In the D line mappings in Fig. 5, we can notice dark cores lyingwithin the bright patterns of D1/D2 and D3/D4 at numerous sub-GBs. These are optical artifacts from the etch grooves along thesub-GBs, and which indicate the exact locations of the sub-GBs. In this section, we utilize these artifacts to investigatemicroscopic spatial distributions of the D lines around sub-GBsin mc-Si wafers. Fig. 7 compares the intensity mappings of theD lines between the 810 and 532-nm excitation lights at 80 K.

In all mappings in Fig. 7, within the bright patterns along thesub-GBs, the thin dark lines (for mappings with 810-nm light)or thin bright lines (for mappings with the 532-nm light) arelocations of sub-GBs, and indicated by red arrows in Fig. 7(a),for example. In Fig. 7(a), the mappings of D1 plus D2 with the810-nm excitation light are highly asymmetric across all sub-GBs, whereas those with the 532-nm excitation light are lessasymmetric. The same trend is observed for the D3 plus D4mappings, although the asymmetry is less pronounced in thiscase. We repeated the measurements for numerous sub-GBs[see Fig. 7(b) and (c)], and found the same behaviors. Thesesignatures suggest that these sub-GBs are inclined underneaththe wafer surface rather than perpendicular to the wafer surface.The higher asymmetry observed with the 810-nm excitation isthen due to the deeper generation profile achieved with thissource. The reason we observe a less asymmetric distributionof D3/D4 compared to D1/D2 is because D1/D2 has a broaderspatial distribution around the sub-GBs than D3/D4.

Now, we verify the above inclination hypothesis with SEMand TEM results. Fig. 8(a) shows Fig. 7(a) rotated 90° counterclockwise. Fig. 8(b) shows an SEM image of the sub-GB dis-playing a highly asymmetric distribution of D1/D2. In Fig. 8(b),the etch groove is the location of the sub-GB, and its shape is alsoasymmetric. The left-hand side slope of the groove is shallow,whereas the right-hand side slope is steep. This signature couldindicate the inclination direction of this sub-GB underneath thesurface. In fact, a TEM vertical cross-section foil (∼100 nm

NGUYEN et al.: MICROSCOPIC DISTRIBUTIONS OF DEFECT LUMINESCENCE FROM SUBGRAIN BOUNDARIES 779

Fig. 9. Optical microscope images of the four investigated regions (A, B, C, and D). Positions of sub-GBs are revealed by etch grooves after the defect etch.Directions of arrows indicate the sides of wider D-line emissions across the sub-GBs. The images have the same X and Y directions relative to one another.

thin) across this sub-GB was prepared by a focused ion beamtool, and its bright field TEM image is displayed in Fig. 8(c).The TEM micrograph clearly shows an inclined sub-GB un-derneath the surface, angled toward the broader side of the Dline mappings, or the steeper side of the etch groove. Therefore,the observed asymmetric patterns of the D lines are due to theinclined sub-GBs underneath the wafer surface.

Finally, we examine if the asymmetry of the D lines, i.e.,the inclination of the sub-GBs, has a common direction.Fig. 9 shows optical images of the four investigated regions(A, B, C, and D). They have the same X and Y directions relativeto one another, and they came from the same dislocation clusterin a wafer. The directions of the arrows indicate the sides withwider D-line emissions across the sub-GBs. We can see clearlythat the D lines are consistently skewed to the left (red arrows)and downward (yellow arrows). Since sub-GBs are the result ofdislocation accumulation, these results suggest that dislocationswere formed and evolved in a common direction locally, ratherthan in random directions, during the crystal growth process.This conclusion is difficult to make using common microscopictools such as TEM or EBSD due to impractical aspects of samplepreparations.

VII. CONCLUSION

Utilizing micron-scale spatial resolution from a µ-PL spec-troscopy system, we performed a microscopic investigation onluminescence behaviors of subgrain boundaries in mc-Si wafers.We demonstrated that when the spatial resolution is low or themeasurement temperature is high, the spatial distribution ofluminescence signals of subgrain boundaries can become in-distinct. In addition, we showed that at subgrain boundaries,there is no spatial and intensity correlation between D1/D2 andD3/D4. Although both of them negatively affect the BB lumi-nescence from silicon, D1/D2 are significantly more detrimentalthan D3/D4. Moreover, based on the synchrotron-based µ-XRFmeasurements, we confirmed that high densities of metal impu-rities are present at the sub-GBs with strong D1/D2 emission,and their presence can quench the intrinsic luminescence fromdislocation cores. Finally, we concluded that the asymmetricdistribution of the D lines across subgrain boundaries is due tothe inclined subgrain boundaries underneath the wafer surface.The subgrain boundaries were formed and evolved in a com-mon direction locally, rather than in a random manner, duringthe crystal growth process.

ACKNOWLEDGMENT

The authors acknowledge E. E. Looney, A. Youssef,S. Wieghold, and J. Poindexter for their assistance withsynchrotron-based measurements, Dr. F. Kremer from CAMfor his assistance with TEM measurements, and C. Sun for hisassistance with FTIR measurements. The Australian NationalFabrication Facility and Center for Advanced Microscopy aregreatly acknowledged for providing access to some of the facil-ities used in this work.

REFERENCES

[1] S. Binetti, A. Le Donne, and M. Acciarri, “Processing step-relatedupgrading of silicon-based solar cells detected by photo-luminescencespectroscopy,” Sol. Energy Mater. Sol. Cells, vol. 86, pp. 11–18, 2005.

[2] S. Binetti, A. Le Donne, and A. Sassella, “Photoluminescence andinfrared spectroscopy for the study of defects in silicon for photovoltaicapplications,” Sol. Energy Mater. Sol. Cells, vol. 130, pp. 696–703, 2014.

[3] M. Tajima, “Spectroscopy and topography of deep-level luminescence inphotovoltaic silicon,” IEEE J. Photovolt., vol. 4, no. 6, pp. 1452–1458,Nov. 2014.

[4] M. Suezawa, Y. Sasaki, and K. Sumino, “Dependence of photolumines-cence on temperature in dislocated silicon crystals,” Phys. Status Solidi A,vol. 79, pp. 173–181, 1983.

[5] R. Sauer et al., “Dislocation-related photoluminescence in silicon,” Appl.Phys. A, vol. 36, pp. 1–13, 1985.

[6] E. C. Lightowlers and V. Higgs, “Luminescence associated with thepresence of dislocations in silicon,” Phys. Status Solidi A, vol. 138,pp. 665–672, 1993.

[7] V. V. Kveder, E. A. Steinman, S. A. Shevchenko, and H. G. Grimmeiss,“Dislocation-related electroluminescence at room temperature in plasti-cally deformed silicon,” Phys. Rev. B, vol. 51, pp. 10520–10526, 1995.

[8] V. Higgs, F. Chin, X. Wang, J. Mosalski, and R. Beanland, “Photolumi-nescence characterization of defects in Si and SiGe structures,” J. Phys.:Condens. Matter, vol. 12, pp. 10105–10121, 2000.

[9] M. Kittler, W. Seifert, T. Arguirov, I. Tarasov, and S. Ostapenko, “Room-temperature luminescence and electron beam-induced current (EBIC) re-combination behaviour of crystal defects in multicrystalline silicon,” Sol.Energy Mater. Sol. Cells, vol. 72, pp. 465–472, 2002.

[10] M. Tajima and H. Sugimoto, “Defect analysis in solar cell silicon byphotoluminescence spectroscopy and topography,” ECS Trans., vol. 25,pp. 3–10, 2009.

[11] M. Tajima, M. Ikebe, Y. Ohshita, and A. Ogura, “Photoluminescenceanalysis of iron contamination effect in multicrystalline silicon wafers forsolar cells,” J. Electron. Mater., vol. 39, pp. 747–750, 2010.

[12] M. Tajima et al., “Deep-level photoluminescence due to dislocationsand oxygen precipitates in multicrystalline Si,” J. Appl. Phys., vol. 111,pp. 113523-1–113523-6, 2012.

[13] C. Krause, D. Mankovics, H.-M. Krause, T. Arguirov, and M. Kittler,“On the origin of intense luminescence at 0.93 eV from multi-crystallinesilicon,” J. Appl. Phys., vol. 114, pp. 034902-1–034902-6, 2013.

[14] S. Pizzini, M. Guzzi, E. Grilli, and G. Borionetti, “The photolumines-cence emission in the 0.7-0.9 eV range from oxygen precipitates, thermaldonors and dislocations in silicon,” J. Phys.: Condens. Matter, vol. 12,pp. 10131–10143, 2000.

780 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 7, NO. 3, MAY 2017

[15] S. Binetti et al., “Optical properties of oxygen precipitates and dislocationsin silicon,” J. Appl. Phys., vol. 92, pp. 2437–2445, 2002.

[16] K. Bothe, R. J. Falster, and J. D. Murphy, “Room temperature sub-bandgap photoluminescence from silicon containing oxide precipitates,”Appl. Phys. Lett., vol. 101, 2012, Art. no. 032107.

[17] P. Gundel et al., “Micro-photoluminescence spectroscopy on metal precip-itates in silicon,” Phys. Status Solidi: Rapid Res. Lett., vol. 3, pp. 230–232,2009.

[18] H. Conzelmann and J. Weber, “Photoluminescence from chromium-boronpair in silicon,” Physica B+C, vol. 116, pp. 291–296, 1983.

[19] F. Dreckschmidt and H.-J. Moller, “Defect luminescence at grainboundaries in multicrystalline silicon,” Phys. Status Solidi C, vol. 8,pp. 1356–1360, 2011

[20] E. Olsen and A. S. Flø, “Spectral and spatially resolved imaging ofphotoluminescence in multicrystalline silicon wafers,” Appl. Phys. Lett.,vol. 99, pp. 011903-1–011903-3, 2011.

[21] I. Burud, A. S. Flø, and E. Olsen, “On the origin of inter band gap radiativeemission in crystalline silicon,” AIP Adv., vol. 2, pp. 042135-1–042135-7,2012.

[22] D. Lausch et al., “Classification of crystal defects in multicrystalline sili-con solar cells and wafer using spectrally and spatially resolved photolu-minescence,” J. Appl. Phys., vol. 119, pp. 054501-1–054501-6, 2016.

[23] B. Sopori et al., “Performance limitations of mc-si solar cells caused bydefect clusters,” ECS Trans., vol. 18, pp. 1049–1058, 2009.

[24] J. Chen and T. Sekiguchi, “Carrier recombination activity and structuralproperties of small-angle grain boundaries in multicrystalline silicon,”Jpn. J. Appl. Phys., vol. 46, pp. 6489–6497, 2007.

[25] H. T. Nguyen, F. E. Rougieux, F. Wang, H. Tan, and D. Macdonald,“Micrometer-scale deep-level spectral photoluminescence from dislo-cations in multicrystalline silicon,” IEEE J. Photovolt., vol. 5, no. 3,pp. 799–804, May 2015.

[26] H. T. Nguyen, F. E. Rougieux, F. Wang, and D. Macdonald, “Effects ofsolar cell processing steps on dislocation luminescence in multicrystallinesilicon,” Energy Procedia, vol. 77, pp. 619–625, 2015.

[27] S. P. Phang and D. Macdonald, “Direct comparison of boron, phosphorus,and aluminum gettering of iron in crystalline silicon,” J. Appl. Phys.,vol. 109, pp. 073521-1–073521-6, 2011.

[28] W.C. Dash, “Copper precipitation on dislocations in silicon,” J. Appl.Phys., vol. 27, pp. 1193–1195, 1956.

[29] Y. Kashigawa, R. Shimokawa, and M. Yamanaka, “Highly sensitiveetchants for delineation of defects in single- and polycrystalline siliconmaterials,” J. Electrochem. Soc., vol. 143, pp. 4079–4087, 1996.

[30] Z. Cai et al., “Performance of a high-resolution x-ray microprobe at theadvanced photon source,” in Proc. 11th US Nat. Conf. AIP CP 521, Syn-chrotron Radiat. Instrum., 1999, pp. 31–34.

[31] A. E. Morishige et al., “Accelerating synchrotron-based characterizationof solar materials: Development of flyscan capability,” in Proc. 43rd IEEEPhotovolt. Spec. Conf., 2016, pp. 2006–2010.

[32] D. P. Fenning et al., “Precipitated iron: A limit on gettering efficacy inmulticrystalline silicon,” J. Appl. Phys., vol. 113, 2013, Art. no. 044521.

[33] M. A. Jensen et al., “Synchrotron-based analysis of chromium distribu-tions in multicrystalline silicon for solar cells,” Appl. Phys. Lett., vol. 106,2015, Art. no. 202104.

[34] A. E. Morishige et al., “Synchrotron-based investigation of transition-metal getterability in n-type multicrystalline silicon,” Appl. Phys. Lett.,vol. 108, 2016, Art. no. 202104.

[35] H. T. Nguyen and D. Macdonald, “On the composition of luminescencespectra from heavily doped p-type silicon under low and high excitation,”J. Lumin., vol. 181, pp. 223–229, 2017.

[36] M. C. Schubert et al., “Spatially resolved luminescence spectroscopy onmulticrystalline silicon,” in Proc. 23rd Eur. Photovolt. Sol. Energy Conf.Exhib., 2008, pp. 17–23.

[37] F. Yan et al., “Defect-band photoluminescence imaging on multi-crystalline silicon wafers,” Phys. Status Solidi RRL, vol. 6, pp. 190–192,2012.

[38] D. Mankovics, R. P. Schmid, T. Arguirov, and M. Kittler, “Dislocation-related photoluminescence imaging of mc-Si wafers at room temperature,”Cryst. Res. Technol., vol. 47, pp. 1148–1152, 2012.

[39] S. Johnston et al., “Correlating multicrystalline silicon defect types us-ing photoluminescence, defect-band emission, and lock-in thermographyimaging techniques,” IEEE J. Photovolt., vol. 4, no. 1, pp. 348–354,Jan. 2014.

[40] D. Lausch and C. Hagendorf, “Influence of different types of recombina-tion active defects on the integral electrical properties of multicrystallinesilicon solar cells,” J. Sol. Energy, vol. 2015, pp. 1–9, 2015.

[41] T. Mehl, M. D. Sabatino, K. Adamczyk, I. Buruda, and E. Olsen, “Defectsin multicrystalline Si wafers studied by spectral photoluminescence imag-ing, combined with EBSD and dislocation mapping,” Energy Procedia,vol. 92, pp. 130–137, 2016.

[42] J. Bauer et al., “Recombination at Lomer dislocations in multicrystallinesilicon for solar cell,” IEEE J. Photovolt., vol. 6, no. 1, pp. 100–110,Jan. 2016.

[43] M. Tajima and Y. Matsushita, “Photoluminescence related to dislocationin annealed Czochralski-grown Si crystals,” Jpn J. Appl. Phys., vol. 22,pp. L589–L591, 1983.

[44] T. Sekiguchi and K. Sumino, “Cathodoluminescence study on dislocationsin silicon,” J. Appl. Phys., vol. 79, pp. 3253–3260, 1996.

Authors’ photographs and biographies not available at the time of publication.