Smooth (< 0.5 nm rms) and subsurface damage-free (010) β-Ga2O3 was achieved with low-pressure chemical mechanical polishing. An applied pressure of 1 kPa along with colloidal silica and poromeric polyurethane polishing pads rotating at 30 rpm was found to be the optimal polishing parameters for (010) β-Ga2O3. Using higher pressures typically employed in the current literature induced subsurface damage in the substrates. Diffuse scatter intensity of triple-axis x-ray rocking curves was used to determine the presence of subsurface lattice damage, which was quantified by measuring peak widths below the half maximum (i.e., FWXM where X < 0.5). The initially rough surfaces of (010) β-Ga2O3 substrates due to wafer slicing and grinding were lapped and polished. A 5 μm Al2O3 slurry followed by a 0.3 μm Al2O3 slurry was used as the primary lap material removal step. The material removal rates were ∼20 and ∼9 μm/h, respectively. Then, chemical mechanical polishing was performed using colloidal alumina followed by colloidal silica. The removal rates were ∼1.3 and ∼0.4 μm/h, respectively. Only colloidal silica showed the complete removal of subsurface damage. The final (020) β-Ga2O3 rocking curve FWHM was ∼13″ and FW(0.001)M was ∼120″, which matches the widths of commercially available pristine (010) β-Ga2O3. A final cleaning step using dilute bleach and dilute citric acid to remove residual silica slurry particles from the surface was demonstrated.
β-Ga2O3 is an ultrawide bandgap semiconductor widely recognized for its potential in next-generation high-power device applications due to its outstanding electrical properties.1–4 Recently, there have been experimental reports of ∼kV breakdown voltage β-Ga2O3 devices.5–7 There have also been major developments in both homo-8,9 and hetero-epitaxial10,11 growth on β-Ga2O3 as well as wafer bonding with β-Ga2O3.12–15 A key step for every semiconductor material, achieving smooth sub-nm β-Ga2O3 surfaces is critical for devices, epitaxy, and wafer bonding.
However, the current literature is deficient in detailed chemical mechanical polishing (CMP) studies for β-Ga2O3. Huang et al.16 utilized colloidal silica in NaOH to CMP (100) β-Ga2O3 while using an applied pressure of 15 kPa. In a follow-up study, Huang et al.17 varied the slurry solvent chemistry while using colloidal silica particles and 13 kPa of applied pressure. While smooth epiready surfaces were achieved, subsurface damage had not been assessed in either study. High pressures are often employed to reduce processing time because the pressure is directly proportional to the material removal rate.18–21 However, as demonstrated in our previous work with polishing various III-V materials, a high applied pad pressure will induce subsurface damage despite achieving smooth surfaces.22 On the other hand, if the applied pressure is insufficient, then little to no polishing occurs.23 Hence, the applied pressure must be optimized such that polishing action occurs while simultaneously not inducing subsurface damage.22,24,25–27 In the most recent study by Huang et al.,28 the applied pressure had been lowered to 1.5 kPa and 800-nm alumina particles were used in the CMP slurry. For abrasive CMP, using particles that are too hard for a given polished material will induce scratches and subsurface damage.22,23 Subsurface damage had not been assessed in that case and, therefore, it was not clear if alumina is suitable for achieving subsurface damage-free β-Ga2O3.
Polishing of a different orientation, (010) β-Ga2O3, had recently been reported in a very comprehensive study by Blevins et al.,29 but the applied pressure was not reported. In that study, subsurface damage was assessed using double-axis x-ray rocking curves by monitoring the changes in the rocking curve full width at half maximum (FWHM). The subsurface damage associated with x-ray diffraction peak broadening may consist of cracks, dislocations, and similar extended defects. Our previous work22,24,25–27 has demonstrated that the peak widths below the half maximum are more sensitive to lattice distortions induced by subsurface damage (i.e., FWXM, where X < 0.5). However, the damage associated with peak broadening below the FWHM may not be visible by other means, such as with transmission microscopy images, e.g., at an intermediate step where a polished substrate is approaching but has not yet achieved a damage-free state. Additionally, our previous work employed triple-axis x-ray rocking curves to assess subsurface lattice damage.22,24,25–27 Performing XRD measurements using triple-axis optics is advantageous over the more widely used double-axis measurements because peak broadening is deconvoluted into its lattice tilt/mosaicity (ω scanning axis) and strain (ω:2θ scanning axis) components. On the other hand, peak widths of rocking curves measured in double-axis combine the contributions of both lattice tilt and strain. While double-axis measurements provide useful information, a more fundamental understanding of the nature of subsurface damage (e.g., tilt versus strain) is essential for studies that range from the elimination of subsurface damage to the impact on epitaxial growth and device performance. Furthermore, the x-ray beam used by Blevins et al.,29 projected ∼20 mm across the wafer surface, spanning nearly the entire diameter of the substrate. By using a wide incident x-ray beam, the rocking curve widths are susceptible to broadening from lattice curvature across a wafer,30 which obscures the effect of polishing, especially toward the final CMP steps. The x-ray rocking curves were also reported to be measured while mounted on the polishing plate, which is a practical measurement that allows for a more rapid assessment of changes. However, as shown in our recent work,31 such sample mounting for x-ray rocking curve measurements can bend substrates by several arcsec of curvature and induce rocking curve broadening that can obscure the contribution of polishing damage to the width.
In this study, polishing parameters for (010) β-Ga2O3 are optimized to achieve both smooth (sub-nm rms) surfaces and subsurface damage-free material simultaneously. These procedures, especially the final polishing steps, are also necessary to achieve high crystalline quality layers produced by light-gas induced exfoliation processes.32 The (010) orientation has the highest thermal conductivity1 compared to the other orientations of β-Ga2O3. Additionally, compared to the other orientations, (010) β-Ga2O3 has been reported to be less susceptible to stacking faults and twinning defects during epitaxial growth.33,34 However, this study is certainly expected to serve as a guide for CMP processes of other orientations.
Lapping and chemical mechanical polishing were performed on a Logitech PM5 CMP tool. The flow rate was 10 ml/min and the pad rotation speed was 30 rpm for both lapping and polishing. A mechanically harder polyurethane impregnated polyester felt polishing pad35 was used exclusively for the lapping steps, while both the felt pad and a softer polyurethane poromeric polishing pad35 were used for the polishing steps. The applied pressure on the substrates was varied from 1 to 10 kPa using DI water on the poromeric pad to optimize the pressure. Because β-Ga2O3 is inert to water,36 there is no chemical component and only mechanical abrasion is supplied by the polishing pad. Based on the results from that study (described below), 1 kPa was then used for all the lapping and polishing steps. The particles used for lapping were either 0.3 or 5 μm alumina particles in de-ionized water. CMP employed colloidal alumina in sodium hypochlorite (Chemlox) or colloidal silica in NaOH (both 70 nm particles). A final 10 min cleaning step on a separate poromeric pad was performed after CMP to remove residual slurry particles: a dilute 1:10 NaOCl solution and a dilute 1:2 citric acid solution were mixed on the polishing pad.23,26,27 1.5 × 1.0 cm2 (010) β-Ga2O3 substrates grown using the edge-defined film-fed growth method were polished.
Subsurface damage was assessed by measuring symmetric (020) β-Ga2O3 rocking curves with triple-axis x-ray diffraction. A high-resolution Bruker-JV D1 diffractometer was employed, whose acceptance angle is ∼10″ in the triple-axis mode. The incident x-ray beam is conditioned by a Göbel mirror37 and a (220) channel-cut silicon crystal, which produces a highly collimated monochromatic beam of Cu Kα1 radiation. The scattered beam optics is a 4-bounce (220) channel-cut silicon analyzer crystal. The incident x-ray beam width was reduced to ∼0.14 mm with mechanical slits in order to reduce the impact of lattice curvature and avoid obscuring the rocking curve data from nonpolishing effects. The projected x-ray footprint across the substrates was ∼0.28 mm for the symmetric (020) β-Ga2O3 reflection. The rms surface roughness after each lapping or CMP step was measured with 40 × 40 μm2 atomic force microscopy (AFM) images using a Quesant QScope 250. A FEI Nova 600 DualBeam focused ion beam (FIB) system was used to prepare transmission electron microscopy (TEM) samples. Thin layers of Pt were used to protect the β-Ga2O3 surfaces during sample preparation. A FEI TITAN S/TEM operating at a 300 keV accelerating voltage was employed to generate scanning transmission electron microscopy (STEM) images aligned to the  zone axis. Both cross-sectional brightfield (BF) and high-angle annular dark-field (HAADF) STEM images were generated. The FIB system was also used to mill fiducial markers on the substrates to measure material removal rates for each polishing step using AFM.
III. RESULTS AND DISCUSSION
First, the softer poromeric polishing pad was used with DI water to study the effect of applied pressure on β-Ga2O3 CMP. Here, there is no chemical action and only mechanical abrasion is provided by the pad. The subsurface damage was assessed by measuring the diffuse scatter intensity from triple-axis symmetric (020) β-Ga2O3 XRD rocking curves [e.g., peak widths taken at the FW(0.001)M]. The FWHM and FW(0.001)M for commercially available pristine β-Ga2O3 substrates are ∼13 ± 2″ and ∼120 ± 10″, respectively, as shown in Fig. 1. Using pristine substrates is important for studying the effects of various polishing parameters, otherwise poor crystalline quality material would obscure the analysis. With high-quality crystals, any broadening in the rocking curves could unequivocally be attributed to the effect of polishing. After 1 h of polishing at 10 kPa, the rocking curve FWHM and FW(0.001)M increased to 33 ± 4″ and 890 ± 20″, respectively. Even polishing another identical pristine substrate using 3 kPa for 1 h broadened the rocking curve FWHM to 27 ± 3″ and FW(0.001)M to 650 ± 20″. The diffuse scatter intensity is indicative of subsurface damage and shows that even 3 kPa of applied pressure is unsuitable for β-Ga2O3 CMP. These results suggest that subsurface damage was likely introduced in prior studies that used pressures of 13 kPa to 15 kPa.16,17 Using an applied pressure of 1 kPa for 1 h was found to not induce damage; as shown in Fig. 1, the peak widths are unchanged from its as-received pristine state. The rms surface roughness did not change (< 0.5 nm) when using any pressures from 1 kPa to 10 kPa. This indicates that smooth surfaces do not necessarily correspond to pristine, damage-free material.
To determine a compatible polishing pad material, a relatively harder felt pad was compared with a softer poromeric pad. Colloidal silica slurry (70 nm particles) and 1 kPa applied pressure were utilized for both pads. After 1 h on the felt pad, the rocking curves in Fig. 2 show that the harder pad induces subsurface damage. The corresponding FWHM and FW(0.001)M peak widths broadened to 24 ± 3″ and 410 ± 20″, respectively. The polishing damage is due to the mechanical abrasion from the harder felt pad. As shown in Fig. 2, on using the same silica slurry on the softer poromeric pad for ∼135 min, the damage was completely removed. Figure 2 corresponds to the same substrate where subsurface damage is induced and subsequently eliminated. A total of ∼0.9 μm of material was removed to revert the substrate back to its pristine state. We demonstrate that soft poromeric polishing pads, 1 kPa of applied pressure, and colloidal silica slurry are appropriate polishing parameters for achieving smooth surfaces and subsurface-damage-free material simultaneously.
Lapping and polishing are key steps for semiconductor wafers whose surfaces are roughened from wafer slicing and grinding. To demonstrate, the rough sides of as-received single-side polished (010) β-Ga2O3 substrates were lapped and polished. As shown in Fig. 3(a), the (020) rocking curve not only shows a broad main peak at ω = 0″ with a FWHM of 180 ± 20″ and a FW(0.001)M of 8900 ± 100″, but also an even broader peak at ∼5600″ along ω. This second broader peak corresponds to a highly misaligned material, which shows that the grinding process induced so much subsurface damage that the lattice near the surface is tilted ∼1.6° away from the rest of the underlying substrate. The sample was oriented such that the (100) flat was perpendicular to the incident x-ray beam, i.e., this severe lattice tilt distortion is along the direction normal to the (100) cleavage plane (the mechanically weakest plane of β-Ga2O3). AFM shown in Fig. 4(a) also shows surface cracks and voids aligned and elongated preferentially along the (100) plane. The rms roughness of the rough surface was initially 60 nm. Triple-axis (020) symmetric ω:2θ scans were also measured to assess the presence of strain induced by the wafer slicing and grinding process as shown in Fig. 3(c). One ω:2θ scan was measured at the main peak along the rocking curve at ω = 0″, and a second ω:2θ scan was measured at the peak associated with the highly misaligned material at ω = 5600″. The ω:2θ FWHM of the misaligned material is 290 ± 40″ (dotted curve) while the ω:2θ FWHM of the peak measured at ω = 0″ is 70 ± 10″ (dashed curve). The broader ω:2θ peak width is associated with strain, which indicates that the distortion of the highly misaligned material is not only tilted, but is also strained relative to the underlying material deeper in the substrate. There is also some strain (less than in the misaligned material) that extends below the misaligned material because the ω:2θ FWHM of the peak measured at ω = 0″ is broader than commercially available (010) β-Ga2O3 substrates that have triple-axis ω:2θ FWHM of ∼20″.
These rough surfaces were first lapped with 5 μm Al2O3 particles using the harder felt polishing pad and by applying 1 kPa of pressure. After 2 h of lapping, the intensity of the peak due to the highly misaligned material dropped by an order of magnitude, as shown in Fig. 3(a). While the FWHM decreased to 110 ± 10″, the FW(0.001)M broadened to 9580 ± 60″. This indicates that lapping with 5 μm Al2O3 particles reduced the thickness of the initial damage associated with the misoriented layer, but also introduced new subsurface damage (as confirmed with electron microscopy measurements as shown below). Because the peak broadening is now symmetric around the main peak, the new subsurface damage induced by the lapping is isotropic. The rms roughness was reduced to 17 nm, and ∼μm wide and tens of nm deep scratches were observed along arbitrary directions along the surface. Reducing the particle size to 0.3 μm Al2O3 and lapping for 2 h led to a reduced amount of subsurface damage. Note that with each of these steps, a significant amount of material is removed, such that the damage measured corresponds to what was introduced with that step. The rocking curve FWHM remained at 110 ± 10″, while the FW(0.001)M decreased to 3690 (from 9580″) ± 50″. The resulting rms roughness was ∼1 nm and the surface scratches were narrower and shallower: tens of nm wide and ∼1 nm deep. Despite the surface roughness approaching <1 nm smoothness, the diffuse scatter intensity shown in the corresponding rocking curve indicated that subsurface damage remained. After changing the slurry to 70 nm Al2O3 and polishing for 6 h, the rocking curve FWHM and FW(0.001)M reduced to 48 ± 6″ and 1310 ± 50″, respectively. The rms roughness remained unchanged (∼1 nm) due to shallow surface scratches. Even when using colloidal particles, this result indicates that Al2O3 is too hard for β-Ga2O3. As shown in Fig. 3, the rocking curve exhibits diffuse scatter intensity due to subsurface damage indicating that Al2O3 is unsuitable as a final polishing step.
CMP using colloidal SiO2 (∼2.5 × softer than Al2O3) for 10 h was demonstrated to be effective in completely removing the remaining subsurface damage. Figures 3(a) and 3(b) show that both the rocking curve FWHM and FW(0.001)M were reduced to 13 ± 2″ and 120 ± 10″, respectively. This matches the same pristine crystalline quality that is commercially available as shown in Fig. 3(b). The final rms roughness values over 40 × 40 μm2 areas were ≤ 0.5 nm [0.4 nm as shown in Fig. 4(e)]. Furthermore, the (020) ω:2θ FWHM was reduced to 18 ± 2″, which also matches the widths of commercially available pristine substrates and indicates that the strain from subsurface damage was also removed. Figure 3(d) compares the (020) triple-axis ω and ω:2θ scans on the same plot. The widths of the ω:2θ scan are broader than the ω scan [18″ versus 13″ for the FWHM, respectively, and 230″ versus 120″ for the FW(0.001)M, respectively]. The broader widths along the ω:2θ scanning axis are not related to the effect of polishing but rather due to crystal truncation rod scattering. This broadening contribution would be convoluted in double-axis rocking curve measurements and obscure sub-surface damage broadening.
A 10-minute step with diluted 1:10 bleach and 1:2 citric acid on a clean poromeric pad26 was demonstrated to be an effective cleaning method for removing residual colloidal SiO2 particles on the β-Ga2O3 surface. No particles were observed in 40 × 40 μm2 AFM scans. The bleach and citric acid were kept separated and mixed on the polishing pad. The applied pressure was kept at 1 kPa. We find that unlike our previous works for III-V CMP22,24 where bleach and citric acid chemically react by oxidizing the III-V surface, β-Ga2O3 is much less reactive to both bleach and citric acid. No reaction was observed even after 12 h of performing CMP with diluted bleach and citric acid on pristine substrates. The material removal rate was negligible, and no surface roughening was observed while the surface was cleaned.
Cross-sectional STEM images were used to image defects associated with the subsurface damage that contributed to the diffuse scatter intensity measured in the XRD rocking curves. A cross-sectional HAADF STEM image of the as-received rough face is shown in Fig. 5(a). Note that AFM of the surface in Fig. 4(a) shows surface voids and cracks, which corresponds to the surface morphology observed on the top surface of the HAADF STEM image in Fig. 5(a). The HAADF STEM image shows that these cracks and voids propagate downward toward the bulk of the substrate. The long vertical cracks and voids are speculated to be responsible for the second broad peak observed in the rocking curve shown in Fig. 3(a). These cracks and voids induce severe lattice tilt, and the resulting tilt distortion is observed to be preferentially along the (100) cleavage plane. The depth of these vertical cracks and voids extends beyond the thickness of the TEM sample (∼8 μm). After lapping with 5 μm Al2O3 for 2 h, the vertical cracks and voids are no longer observed as shown in Fig. 5(b), and the surface is flattened. Because ∼40 μm of material was removed, the aforementioned vertical cracks and voids do not extend farther than 40 μm below the original damaged rough surface. Magnified BF images are shown in Fig. 6 to enhance the contrast from dislocations. These high-density dislocations also make up for the subsurface damage induced by wafer slicing and grinding [Figs. 6(a)–6(c)] and lapping [Figs. 6(d)–6(f)]. Each image is taken at a different depth from the surface. These dislocations contribute to the diffuse scatter intensity measured in the XRD rocking curves in Fig. 3(a). In both cases for grinding or lapping, there is a relatively dark contrast region within the first 200 nm to 300 nm from the surface where the extent of damage is so large that dislocations are not resolvable. The near-surface damage region is then followed by a region consisting of mosaic light contrast surrounded by dislocations. Qualitatively, the density of dislocations appears to decrease with depth, i.e., comparing the contrast going from Figs. 6(a)–6(c) and from Figs. 6(d)–6(f). After the last CMP step using the softer colloidal SiO2 slurry, the subsurface damage was removed, and uniform contrast is observed in the cross-sectional BF STEM image as shown in Fig. 7. No dark contrast from dislocations is observed, which indicates that the cracks, voids, and dislocations were induced by the wafer slicing, grinding, lapping, and aggressive polishing steps and not due to the inherent crystalline quality of β-Ga2O3.
The SiO2 slurry used to obtain subsurface-damage-free and smooth β-Ga2O3 surfaces was suspended in a NaOH-based solvent and the material removal rate was ∼0.4 μm/h. Work by Huang et al.,16,17 proposed that NaOH reacts with β-Ga2O3 to form a passivating gallium hydroxide salt layer. NaOH has been shown to exhibit extremely slow free etch rates at room temperature, on the order of a few nm/h.38 Hence, we propose that the SiO2 particles provide the mechanical abrasion to remove the passivated hydroxide surface layer, causing the material removal rate to increase by 2 orders of magnitude compared to the free etch rate of NaOH alone. This is analogous to the polishing mechanism in our previous CMP studies for various III-V materials where bleach and citric acid were used to oxidize the substrate surfaces and the oxide was removed mechanically (the polishing pad was sufficient for mechanical abrasion in those cases).22,24,25
The measured material removal rates for the various abrasives used for lapping and CMP are shown in Fig. 8. The analysis presented in this current work can be used to optimize lapping and CMP parameters to achieve both smooth surfaces (< 0.5 nm rms roughness) and subsurface-damage-free material simultaneously. Work by Blevins et al.29 estimated that wafer slicing induced a 75 μm of subsurface damage. They then optimized their polishing parameters to remove 100 μm and reduced the total processing time from ∼250 h to ∼28 h. Using the results presented in this current work, the processing time to achieve smooth, damage-free surfaces would be further reduced to ∼17 h: (1) 4 h of 5 μm Al2O3 to remove ∼80 μm of material, (2) 1.5 h of 0.3 μm Al2O3 to remove ∼14 μm of material, (3) 1.5 h of colloidal Al2O3 to remove ∼2 μm of material, and finally, (4) 10 h of colloidal SiO2 to remove ∼4 μm of material. The removal rate using colloidal SiO2 was ∼0.4 μm/h, which is suitable for the CMP of thin layers exfoliated using ion implantation. For example, our previous work32 used He ion implantation to exfoliate ∼0.7 μm thick (010) β-Ga2O3 layers, which had surface roughnesses of ∼4 nm post-exfoliation. Polishing for at least 1 min with colloidal SiO2 would smoothen the exfoliated surfaces and remove ∼7 nm of material. Even finer control over the removal rate could be achieved by reducing the colloidal SiO2 concentration. Demonstrated with other materials systems over a weight percent range of 0% to 15%, reducing the abrasive particle concentration reduces the material removal rate.39–41 The weight percent of colloidal SiO2 used in this current work was ∼40 wt%, so reducing the colloidal SiO2 concentration to achieve even finer removal rates is expected to be feasible.
Single layers of ∼500 nm thick homoepitaxial β-Ga2O3 films were grown on the substrates polished using the above described technique and also on the substrates with commercial polish. The AFM image of the surface after growth is shown in Fig. 9, which shows that the rms roughness is ∼0.3 nm on the substrate polished in this work and ∼0.5 nm on the substrate with the commercial polish, which indicates that smooth surfaces were maintained throughout the homoepitaxial growth process. The triple-axis ω:2θ scan of the symmetric (020) reflection after growth is shown in Fig. 10. A single peak with a FWHM of ∼20″ (which matches commercially available bare substrates) is observed with no thickness fringes. Work by Okumura et al.,42 observed thickness fringes after homoepitaxial growth on (010) β-Ga2O3 substrates and attributed these fringes to non-stoichiometry and/or impurities at the growth interface. The rocking curve after growth was very similar to the rocking curve from the epilayer grown on the standard sample, with the rocking curve FWHM of 30″ for the commercial sample and 17″ for the epitaxial material with our CMP process. There was a slight increase in the diffuse scatter [≥150″ at the FW(0.001)M] compared to 120″ for the pre-epitaxial case in both cases. These preliminary results are promising, and a further detailed analysis of the epitaxial layer and growth interface is underway.
IV. SUMMARY AND CONCLUSIONS
CMP parameters of (010) β-Ga2O3 were optimized to achieve both smooth (< 0.5 nm rms) surfaces and subsurface-damage-free material simultaneously. A 1 kPa of applied pressure with colloidal silica slurry on poromeric polyurethane polishing pads rotating at 30 RPM was found to be the optimal parameter for (010) β-Ga2O3. The corresponding (020) β-Ga2O3 rocking curve FWHM and FW(0.001)M peak widths were ∼13″ and ∼120″, respectively, which matches the widths of commercially available pristine β-Ga2O3. Pressures ≥3 kPa were found to induce subsurface damage despite maintaining smooth surfaces. X-ray rocking curves were invaluable in characterizing the effects of various lapping and polishing steps and the diffuse scatter intensity measured by the peak widths below the half max was the most sensitive to subsurface damage. Cross-sectional STEM images showed that wafer slicing and grinding can induce long vertical cracks and voids that extend ∼40 μm from the surface. These cracks and voids appear to cause severe distortion that tilted the lattice ∼1.6° preferentially toward the (100) cleavage plane. STEM images show that after the final CMP step, all of the subsurface damage was removed and the wafer was free of voids, cracks, and dislocations. Material removal rates were measured for various lapping and polishing steps. 5 μm Al2O3 and 0.3 μm Al2O3 lappings resulted in the removal rates of ∼20 μm/h and ∼9 μm/h, respectively. Colloidal Al2O3 and colloidal SiO2 CMP resulted in the removal rates of ∼1.3 μm/h and ∼0.4 μm/h, respectively. We provide a systematic approach for CMP optimization to determine appropriate parameters that can be applied to other materials.
The authors would like to acknowledge the support from the Office of Naval Research through a MURI program, Grant No. N00014-18-1-2429. The authors would like to thank Arkka Bhattacharyya and Professor Sriram Krishnamoorthy for growing homoepitaxial β-Ga2O3 films on the substrates polished in this work.
Conflict of Interest
The authors have no conflicts to disclose.
Michael E. Liao: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Project administration (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Kenny Huynh: Data curation (equal); Formal analysis (equal); Investigation (equal). Lezli Matto: Investigation (equal). Dorian P. Luccioni: Investigation (equal). Mark S. Goorsky: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal).
The data that support the findings of this study are available from the corresponding author upon reasonable request.