Letter pubs.acs.org/NanoLett
Three-Dimensional Mapping of Quantum Wells in a GaN/InGaN Core−Shell Nanowire Light-Emitting Diode Array James R. Riley,† Sonal Padalkar,† Qiming Li,‡ Ping Lu,‡ Daniel D. Koleske,‡ Jonathan J. Wierer,‡ George T. Wang,‡ and Lincoln J. Lauhon*,† †
Department of Materials Science and Engineering, Northwestern University, Evanston, Illinois 60208, United States Sandia National Laboratories, P.O. Box 5800, Albuquerque, New Mexico 87185, United States
‡
S Supporting Information *
ABSTRACT: Correlated atom probe tomography, crosssectional scanning transmission electron microscopy, and cathodoluminescence spectroscopy are used to analyze InGaN/GaN multiquantum wells (QWs) in nanowire array light-emitting diodes (LEDs). Tomographic analysis of the In distribution, interface morphology, and dopant clustering reveals material quality comparable to that of planar LED QWs. The position-dependent CL emission wavelength of the nonpolar side-facet QWs and semipolar top QWs is correlated with In composition. KEYWORDS: LED, atom probe tomography, semiconductor, nanowire, GaN, quantum well
G
area. Further, scalable patterning methods to fabricate nanowire arrays have been developed.13−16 Because of the potential benefits of this architecture, there have been several recent reports of core−shell nanowire array LEDs.6,17−20 In all cases, the nanowire axes were parallel to the polar c-axis with InGaN deposition occurring on the nonpolar m-plane side-facets and semipolar apex-facets.6,7,10,17−23 The majority of these studies also reported InGaN deposition on the polar c-plane top-facet.6,7,17,18,21 Variations in surface polarity and growth rates lead to variations in QW thickness and In mole fraction that influence LED optical properties; light emission wavelengths increase from the bottoms to the tops of the nanowires,7,8,10,21 and the peak electroluminescence (EL) wavelength has been observed to shift from red to blue as the injection current is increased.6 Previous studies have suggested that the broad emission spectra of nanowire array LEDs may be useful for the generation of white light via colormixing,8,10 but the viability of this approach requires light emission to be efficient and controllable. Therefore it is crucial to identify and examine the structural factors causing this emission behavior. A direct comparison of the quality of standard planar QWs with that of the unconventional nanowire-QWs is challenging due to their three-dimensional (3D) morphology, but such a comparison could help illustrate the structure−property relationships influencing the broad light
roup III-nitride (III-N) semiconductor alloys produce light with high efficiency, making them appealing materials for solid-state lighting.1 Traditional light-emitting diodes (LEDs) with InxGa1‑xN quantum wells (QWs) grown normal to the polar (0001) c-plane possess large internal electric fields that induce the quantum confined stark effect (QCSE), causing spatial separation of carrier wave functions and reducing radiative recombination.2,3 The QCSE can be avoided by growing QWs on nonpolar surfaces, but nonpolar QWs exhibit high densities of threading dislocations when grown on planar substrates such as sapphire or silicon due to large lattice mismatches, and bulk nonpolar GaN substrates are not readily available. Threading dislocations can act as nonradiative recombination centers and reduce light emission efficiency, so it is of highest importance to develop QW growth strategies that minimize dislocation densities.4 III-N nanowires are promising unconventional substrates for the growth of nonpolar QWs with low dislocation density.5 When nanowire arrays are selectively grown on6−8 or selectively etched from9,10 c-plane GaN on sapphire or silicon, the axis of the resulting nanowires is along the polar c-direction, but the exposed sidefacets are nonpolar m-planes. These GaN surfaces provide better lattice matching than silicon or sapphire for InGaN epitaxy, and radial strain relaxation in the nanowire geometry further reduces the threading dislocation density of epitaxial overlayers.5 In addition, concentric InGaN QW shells grown on nanowire cores provide a larger junction area per unit substrate area, which can increase light output and extraction11−14 and reduce device cost by maximizing the lumens produced per die © 2013 American Chemical Society
Received: June 10, 2013 Revised: July 26, 2013 Published: August 6, 2013 4317
dx.doi.org/10.1021/nl4021045 | Nano Lett. 2013, 13, 4317−4325
Nano Letters
Letter
Figure 1. (a) Side-view SEM image of the nonplanar LED heterostructure with p-InGaN canopy layer and core−shell GaN/InGaN nanowire array. (Inset) Cross-sectional HAADF-STEM image, QWs are visible as parallel lines with bright contrast. (b) Schematic illustrating the 3D geometry and faceting of a nanowire with nonpolar side-facet and semipolar apex-facet QWs. (c) Schematic indicating the approximate regions from which reconstructions were generated.
shown that low-voltage aberration corrected imaging and spectroscopy can be conducted without inducing significant damage.34 Finally, because HAADF-STEM generates a projection, if the QWs in the cross-sectioned sample are not oriented perfectly parallel to the electron beam, the measured interface width will be broader than the actual width. The 3D HAADF-STEM tomography is one approach to overcome projection artifacts in core−shell nanowires,35 though it can be challenging to measure QW width and composition when the cross-section contains multiple buried interfaces, not all of which are planar. For example, the nanowire apex is composed of a hexagonal pyramid that may terminate in a sharp tip7,9,10 or a flat plane,6 so most cross sections will produce complex projection artifacts, at least in individual images. Surface damage from cross-sectioning and pre-existing defects further complicate interpretation of composition and structure even for STEM tomography.9 Atom probe tomography (APT), which analyzes the structure and composition of a material with part per million sensitivity and subnanometer spatial resolution in three dimensions,36,37 provides capabilities that are highly complementary to STEM-based imaging modalities. APT has been used to analyze In composition in planar InGaN QWs28,33,38−41 and Mg clustering in p-doped GaN/AlGaN superlattices,41−43 and the feasibility of analyzing GaN nanowires has been demonstrated.44,45 While APT is not affected by projection effects or electron beam damage, variations in polarity at surfaces and interfaces45 and differences in evaporation fields between ions can lead to artifacts in APT reconstruction.46−48
emission spectra and point the way toward controlled growth and improved performance. The geometry of nanowire-QW arrays limits the utility of many standard approaches to analyze QW composition and morphology. X-ray diffraction (XRD), secondary ion mass spectroscopy (SIMS), and Rutherford backscattering spectrometry (RBS) for example, are better suited to the analysis of planar samples. Cross-sectional scanning transmission electron microscopy (STEM) based imaging modalities have proven essential to the analysis of planar superlattice structures and are similarly useful in the analysis of nanowire heterostructures. High-angle annular dark field (HAADF-STEM) imaging can be used to measure interface width based on atomic number contrast, energy dispersive X-ray spectroscopy (EDS) can assess composition, and cathodoluminescence spectroscopy (CL) is useful for correlating light emission with composition in direct gap semiconductors.24,25 However, these techniques are not without shortcomings, some of which are particular to the context of InGaN QWs on GaN nanowires. Fluctuations in indium mole fraction are important to identify because In-rich clusters with locally lower band gap within InGaN QWs have been proposed to cause carrier localization,26−28 which may lead to regions of higher-wavelength light emission.29−31 Such fluctuations would lead to contrast variations in TEM and STEM images, but contrast variations can also arise as artifacts of the cross-sectioning process. An additional concern is that In segregation can be induced by exposure to the high-energy electron beam, making it difficult to definitively associate fluctuations with the growth process,32,33 though recently it was 4318
dx.doi.org/10.1021/nl4021045 | Nano Lett. 2013, 13, 4317−4325
Nano Letters
Letter
Figure 2. (a) Cross-sectional HAADF-STEM image of a representative nanowire and associated 3D reconstruction depicting both nonpolar and semipolar QWs. The QW surface is rendered in gray. (b) Frequency distribution analysis for semipolar apex-facet QW and nonpolar side-facet QW showing no significant deviation from a random alloy (binomial distribution, dashed line). (c) The 3.5 atom % In isoconcentration surface shown in (a), rotated to reveal the surface morphology of the outermost nonpolar and semipolar QWs.
GaAs/Cs photomultiplier tube. The samples were analyzed in ultrahigh vacuum at room temperature with a 5 kV electron beam excitation. Further details have been previously discussed.16 FIB lift-out and milling49,50 was used to create specimens for APT analysis targeting specific regions within the nanowireQW LED as shown in Figure 1c: (1) intersection of side- and apex-facet QWs near the tops of the nanowires; (2) side-facet QWs near the bottom of a nanowire; (3) nanowire tips; and (4) p-InGaN canopy layer. Because the volume that can be probed in a single APT specimen tip is smaller than the entire nanowire, distinct specimens were prepared from three different nanowires within a single LED. The sample preparation geometry is outlined in Supporting Information 1. We consider each of these regions sequentially in the remainder of this work, beginning with region (1). The two outermost sets of QWs in the APT reconstruction (Figure 2a) exhibited well-defined side- and apex-facet QWs, enabling frequency distribution analysis (FDA) to compare the In composition in QWs simultaneously grown on nonpolar and semipolar surfaces. In FDA, the APT data is divided into 3D volume units referred to as “voxels” and the concentration of a specific element is measured within each voxel. The frequency of occurrence of voxels with certain composition is then compared. The peak of the FDA indicates the most probable In mole fraction as ∼13.5% in the nonpolar QW and ∼8.0% in the semipolar QW (Figure 2b). The APT analysis is in good agreement with quantitative STEM-EDS (Supporting Information S2) and is consistent with previous reports of core−shell nanowires in which nonpolar QWs accommodate a higher In mole fraction than semipolar QWs.10 Differences in In incorporation efficiency have also been observed in planar QWs grown on GaN substrates of differing polarity,51 but a direct comparison is not made here because the nanowire QW composition is influenced by In diffusion between facets. Within both nonpolar and semipolar QWs, FDA showed that the In distribution is consistent with the distribution of a random alloy. To make this determination, the number of
Information from correlated measurements including STEMbased imaging can therefore improve the reliability of APT reconstruction and analysis. Here we report correlated APT, HAADF-STEM, and STEMEDS characterization of a nanowire LED array. The growth of QWs on multifaceted nanowires leads to interfacet as well as intrafacet variations in In content. The 3D structural and compositional analysis is correlated with spatially resolved optical properties using cathodoluminescence (CL) spectroscopy. The light emission wavelength increases from the bottoms to the tops of the nanowires due to variations in In mole fraction along the nonpolar side-facet QWs, and lower intensity longer wavelength emission at the nanowire tips is linked to polar top-facet QWs with higher In mole fraction. We examine the nature and magnitude of these composition variations to assess whether the broad emission spectrum may be useful for white light LEDs. Nanowire-QWs were fabricated using a top-down approach employing a mask of silica microspheres and a multistep etching process,9,15 producing a regular array of hexagonal ntype GaN nanowires ∼100 nm in diameter and ∼900 nm in height with nonpolar {11̅00} m-plane side-facets and a Ga-polar (0001) c-plane top-facet. InGaN QWs were grown radially on the nanowire cores via metal−organic chemical vapor deposition (MOCVD), producing nonpolar {11̅00} side-facet QWs and semipolar {101̅1} apex-facet QWs. A p-InGaN canopy layer was then grown to complete the LED heterostructure (Figure 1). Additional growth details have been discussed previously.9 HAADF-STEM images of cross sections made by focused-ion beam (FIB) milling reveal flat InGaN QWs with side-facet QWs wider than the apex-facet QWs (Figure 2a).9 Both types of QWs exhibited contrast variations in HAADF-STEM images that are similar to prior analyses of nanowire-QWs.6 APT was performed with a LEAP 4000× Si, using 0.01 pJ laser energy, 200 kHz pulse frequency, 0.5−1.0% target detection rate, and 27 K background temperature. CL spectroscopy was conducted with a Zeiss Gemini SEM and a Gatan MonoCL3 system with water-cooled 4319
dx.doi.org/10.1021/nl4021045 | Nano Lett. 2013, 13, 4317−4325
Nano Letters
Letter
Figure 3. (a) Cross-sectional SEM image of cleaved nanowire array with shaded regions indicating the nanowire core (purple), nonpolar side-facet QWs (blue), semipolar apex-facet QWs (cyan), and nanowire tip (green). Labels i−vi indicate the electron beam position for the CL spectra shown in (b). Color-coding is schematic and does not directly correlate to light emission color. (c) The 3D reconstruction of nonpolar side-facet QWs for a specimen analyzed in the vertical nanowire orientation. The apparent angle of the QWs is likely due to a slight inclination in the liftout during mounting as well as variations in QW spacing and morphology near the nanowire base (shown in Figure 1a(inset) and Supporting Information 2c). (d) Frequency distribution analysis for four different regions within the QW indicated by the black arrow. (e) Monochromatic CL spectroscopy signals (false color) overlaid on SEM images obtained from a cleaved region of the nonplanar LED showing band-edge emission (365 nm) at a GaN nanowire core and higher wavelength emission at the nanowire side-facets. The wavelength of this emission increases from the bottoms to the tops of all nanowires examined.
STEM images, no In-rich clusters are present, eliminating one potential source of carrier localization in region (1). Fluctuations in QW width have also been hypothesized to induce carrier confinement,28−30,53,54 and subnanometer fluctuations have been identified in APT measurements of InGaN QWs on planar substrates.33 Surfaces of constant concentration (“isoconcentration surfaces”) were used to characterize individual QW interfaces in 3D (Figure 2c) without projection artifacts. The isoconcentration surfaces are generated with 1 nm3 voxels to create surfaces that are continuous and border each QW on both sides without any connection between the individual surfaces. The grid parameters used to generate the surfaces did not significantly impact the measured roughness. The two outermost sets of QWs exhibited subnanometer fluctuations in QW width with RMS roughness of approximately 0.5 nm. The same 3D grid parameters used for analysis of the nanowire QWs were used to generate isoconcentration surfaces for planar m-plane QWs extracted from an LED (not shown).55 These surfaces also exhibited subnanometer width fluctuations of approximately 0.5 nm.56 Therefore, we do not expect additional carrier confine-
voxels with a certain composition was compared to the expected number of voxels of that composition if all atoms were randomly distributed throughout the material.37,52 The random distribution is represented by the binomial distribution when treating InxGa1‑xN as a pseudobinary alloy with In and Ga occupying the group-III sites in the III-N lattice. The magnitude of a deviation from randomness was quantified using a χ2 test, in which the p-value illustrates the probability of obtaining a χ2 value at least as extreme as the observed χ2 value for a certain confidence level. The data were binned into voxels containing 100 ions each in order to balance resolution and counting errors and achieve a 99% confidence level; this binning created voxels with average volume of ∼1.6 nm3. The null hypothesis (that the measured distribution exhibits fluctuations consistent with a random alloy) was rejected if the p-value was
