Progress in Near-Equilibrium Ammonothermal (NEAT) Growth of GaN Substrates for GaN-on-GaN Semiconductor Devices

Abstract

This paper reviews the near-equilibrium ammonothermal (NEAT) growth of bulk gallium nitride (GaN) crystals and reports the evaluation of 2″ GaN substrates and 100 mmbulk GaN crystal grown in our pilot production reactor. Recent progress in oxygen reduction enabled growing NEAT GaN substrates with lower residual oxygen, coloration, and optical absorption. The oxygen concentration was approximately 2 × 10¹⁸ cm⁻², and the optical absorption coefficient was 1.3 cm⁻¹ at 450 nm. Maps of full-width half maximum (FWHM) of X-ray diffraction rocking curveswere generated for grown crystals and finished wafers. The X-ray rocking curve maps confirmed high-quality and uniform microstructure across the entire surface of the bulk crystals and substrates. The average FWHM of the 50 best bulk crystals from the recent batch was 28 ± 4 arcsec for the 002 diffraction and 34 ± 5 arcsec for the 201 diffraction, with an average radius of curvature of 20 m. X-ray topography measured on both sides of the bulk crystals implied that the density of dislocations wasreduced by one order of magnitude during the NEAT growth. A typical NEAT GaN substrate shows dislocation density of about 2 × 10⁵ cm⁻².

1. Introduction

Gallium nitride (GaN) is one of the wide bandgap semiconductors. Successful commercialization of GaN-based light-emitting diodes (LEDs) has promoted further research and development of GaN for other applications, including laser diodes (LDs), high electron mobility transistors (HEMTs), and power transistors. However, except for LDs, the majority of GaN devices are developed using dissimilar substrates, such as sapphire, silicon carbide (SiC), and silicon (Si). Usage of low-cost substrates, such as sapphire and silicon, is mandatory for applications in consumer electronics, where cost-reduction pressure is extremely high. Nevertheless, new applications are currently emerging in the energy, defense, and industrial sectors. Devices in these sectors require reliable high-power operation, which is often more important than cost. Since heteroepitaxial growth generates threading dislocations that degrade device performance and lifetime [1], homoepitaxial GaN substrates with lower dislocation density are of interest despite their higher price.

Although hydride vapor phase epitaxy (HVPE) is currently used to produce the majority of GaN substrates in the marketplace, the method shows limitations in achieving low dislocation density. Several non-vapor phase methods including high-pressure solution growth [2,3,4], flux growth [5,6,7,8,9,10], and ammonothermal growth [11,12,13,14,15,16,17,18,19,20,21,22] have been researched. We have developed the near-equilibrium ammonothermal (NEAT) method to produce low-dislocation GaN substrates. The NEAT method is one type of ammonothermal growth with growth conditions set near equilibrium. Utilizing a weak driving force enables continuous growth for months through a reduction in spontaneous nucleation. The NEAT condition achieves improvement in crystal quality over the growth period. The drawback of a lower growth rate is offset by an increase in the number of bulk GaN crystals grown per batch.

In 2018, we successfully demonstrated 2″ crack-free GaN substrates [23]. However, GaN crystals grown by the ammonothermal method typically contained high levels of oxygen in the order of 10¹⁹ cm⁻³. Previously reported 2″ substrates also contained 10¹⁹ cm⁻³ of oxygen and the color was light brown [23]. In 2019, we successfully reduced the oxygen concentration in small-sized NEAT bulk GaN [24]. Scaling up the same oxygen reduction strategy to a large production reactor, we successfully grew reduced-color 2″ bulk GaN crystals. This paper reports the characterization of reduced-oxygen 2″ NEAT GaN substrates. In addition, by using the same pilot-production reactor, we attempted growth of 4″ bulk GaN. We confirmed that the dislocation density in the 4″ bulk GaN was similar to our 2″ bulk GaN.

2. Experiment

GaN bulk crystals were grown by the NEAT method using our pilot-production high-pressure reactor. Single crystalline GaN seed crystals, polycrystalline GaN and/or metallic Ga, and an alkali-based mineralizer were loaded into the reactor. After sealing the reactor, the air inside the reactor was pumped out and replaced with anhydrous ammonia. The ammonia fill factor was 40~60%. Then, the reactor was heated with external heaters to 450–600 °C. The resulting ammonia pressure was 100~300 MPa. The driving force of crystal growth was controlled with growth parameters including temperature, pressure, seed location, and nutrient location. In the NEAT method, we controlled the driving force to achieve consistent growth over a few months.

Seed crystals were prepared by HVPE. As reported previously, the microstructure of seed crystals has a significant impact on the quality of grown bulk crystals [25]. Low-quality seed crystals were filtered out from usage by evaluating a mapping of X-ray full-width half maximums (FWHMs) over the wafer, as outlined below. Selected seeds typically showed an average FWHM of 002 rocking curves less than 100 arcsec with a radius of curvature of about 5–10 m. The lateral sizes of the seed crystals were 2″, 2.3″, or 4″.

After the NEAT growth, photographs of grown crystals were taken with a digital camera (Canon, Powershot SX510RX). The bulk crystals were rounded with a core drill and a cylindrical grinder. After making a primary orientation flat on the m-plane and a secondary orientation flat on the a-plane, the bulk GaN crystals were sliced into wafers with a multiple wire saw. Sliced GaN blank wafers were shaped with a surface grinder, polished, and finished with chemical mechanical polishing (CMP). Additional details regarding the wafering process are outlined in a previous paper [26]. Removal of the subsurface damaged layer after CMP was confirmed by measuring the intensity of the shoulders of the 114 rocking curve in a grazing configuration [27]. The incident beam angle was 10.8662° to the Ga surface of the wafer. The intensity of shoulders was compared with an empirically obtained standard peak from an unprocessed bulk GaN crystal with no processing-related surface or subsurface damage. Surface morphology of samples was observed with a Nomarski microscope and a tapping-mode atomic force microscope (AFM, Asylum MFP-3D).

After the double-side CMP, optical absorption of the GaN substrates was measured using a spectrophotometer (Shimazu UV-2401PC). Secondary ion mass spectroscopy (SIMS) was measured at EAG Inc. The depth profile of oxygen was measured for smaller-sized GaN crystals co-loaded in the production growth. The oxygen detection limit was about 8 × 10¹⁵ cm⁻³.

The qualities of seed crystals, grown bulk crystals, and processed wafers were characterized with an X-ray diffractometer (PANalytical X’Pert) using a Cu Kα X-ray source and a four-crystal Ge (002) monochromator. The X-ray source was operated at 40 mA and 40 kV. Seeds, grown crystals, and processed wafers were routinely characterized by mapping the FWHMs of X-ray rocking curves from the 002 and 201 diffractions over the entire wafer. The spatial separation of each measuring point was 2 mm. To obtain sufficient spatial separation of the incident X-ray beam, 0.3 mm × 0.3 mm slits were used after the monochromator. The estimated X-ray beam sizes on the sample were 0.3 mm × 1.0 mm and 1.17 mm × 0.52 mm for the 002 and 201 measurements, respectively.

The X-ray topography was measured at the BL16B2 of SPring-8 synchrotron radiation facility in Japan. The X-ray beam was monochromated with a Si (111) double crystal monochromator. The X-ray beam with 9.16 keV of energy was irradiated onto the sample with an incident angle of about 3°, and the topography image from the (114) reflection at 2θ = 84.6° was recorded on an X-ray film.

3. Results and Discussion

3.1. Appearance, Coloration, and Optical Absorption

After the growth, photographs of the crystals were taken with a digital camera. Figure 1 shows bulk GaN crystals grown in one batch. As demonstrated here, we can grow many crystals simultaneously by using a large-sized pressure reactor. Simultaneous growth of many crystals offsets the low growth rate (approximately 30 μm per day). This is the advantage of the ammonothermal method over the other growth methods.

Figure 2 shows a comparison of two crystals from different runs, before (left) and after (right) implementation of our improved oxygen removal procedure in the pilot production reactor. The picture shows that the crystal coloration was greatly reduced with better oxygen removal in the growth environment. The residual oxygen in co-loaded smaller crystals was 7.5 × 10¹⁸ cm⁻³ for the previous run and 1.5 × 10¹⁸ cm⁻³ for the latest run.

A photograph of a 2″ GaN substrate after single-side CMP is presented in Figure 3. The substrate looks almost colorless relative to the crystal it was sliced from, because the thickness of the substrate is much thinner, about 400 μm. We confirmed removal of the subsurface damaged layer in the finished wafer with grazing angle X-ray diffraction [27]. AFM images were not recorded for all substrates, because it potentially damages the finished surface. However, the surface typically shows an atomic step structure after CMP, as reported previously [23].

Optical absorption was measured for double-side polished substrates. The absorption coefficient of GaN substrates in the previous high-oxygen growth was 5.6 cm⁻¹ at 450 nm, whereas that from the reduced-oxygen growth was 1.3 cm⁻¹ at 450 nm. These results are in line with our previous results. Figure 4 shows the relationship between the optical absorption coefficient at 450 nm and the residual oxygen concentration in the NEAT substrates. As shown in the figure, the optical absorption shows direct correlation with the residual oxygen concentrations for the oxygen concentration from 1.5 × 10¹⁸ cm⁻³ to 2.5 × 10²⁰ cm⁻³.

3.2. X-ray Mapping

X-ray FWHM maps were measured for 2″ NEAT GaN substrates after CMP. Figure 5a shows an example of a FWHM map for 002 rocking curves, and Figure 5b shows a map for 201 rocking curves. For this particular sample, average FWHMs were 28 arcsec for the 002 diffraction and 36 arcsec for the 201 diffraction. The higher 201 FWHM was due to higher values in the right-edge region of the substrate. The average FWHM of the 50 best bulk crystals from the batch shown in Figure 1 was 28 ± 4 arcsec for the 002 diffraction and 34 ± 5 arcsec for the 201 diffraction, with an average radius of curvature of 20 m. This indicates the crystals are of excellent quality. We expect additional wafers from these bulk crystals to have similar X-ray FWHM numbers to the bulk crystals they are sliced from.

The X-ray maps for HVPE-grown substrates highlight the superior microstructure uniformity of NEAT GaN substrates. Figure 6a shows an sample X-ray FWHM map for 002 rocking curves, and Figure 6b shows a map for 201 rocking curves. The HVPE substrate shows more yellow and even red regions for both the 002 and 201 maps. Additionally, some spots show double peaks and triple peaks in rocking curves, represented by a single diagonal line and double diagonal lines in the box, respectively.

From these results, it is clear that measuring X-ray FWHM at one or several points can yield a misleading evaluation of the GaN substrates. There are many green spots in the HVPE GaN substrates. If the selected measurement spot falls into such green zones, one will obtain narrow X-ray rocking curves and may expect that the substrate has high quality everywhere. The X-ray FWHM mapping we employ is a powerful tool to evaluate crystal quality and uniformity of GaN substrates. These X-ray FWHM maps demonstrate the superior microstructure of 2″ NEAT GaN substrates.

3.3. Growth and Characterization of 4″ Bulk GaN Crystal

A photo of the 100 mm diameter bulk GaN crystal after rounding and flat formation is presented in Figure 7. The dark stain around the edge is a roughened surface caused by precipitation of the alkali mineralizer during cool-down at the end of the growth. We surmise that the concentrated alkali metal-based compound etched the surface of the crystal. The oxygen concentration was 1.32 × 10¹⁸ cm⁻³ on the top surface of the crystal, i.e., the final oxygen concentration at the end of the growth. The dots did not extend through the thickness of the growth. They were holes in the HVPE seed that closed early in the NEAT growth. The ability of the NEAT growth to close holes is a clear advantage over HVPE, where holes grow larger with time, limiting the run time and growth thickness [28]. Holes in the substrate cause problems during device fabrication, and substrates with holes are sold at a lower price tier.

X-ray topography was measured on both sides of the boule (Figure 8). Since the Ga polar surface of the seed crystal was masked during the growth, c-plane growth only occurred on the N-polar face. Therefore, the Ga-polar side of the boule exposed the seed. Figure 8a shows an X-ray topography image of the as-grown N-polar face of the boule. Due to the lattice bow of the sample, the image area with constructive reflection is sandwiched between off-Bragg regions covered with many white islands (top and bottom regions in the image). The white dots in the dark region of the image (one example is indicated with a white arrow), are contrasts of threading dislocations. Although detailed analysis is needed, we think that the majority of the dislocations are either edge or mixed dislocations. The dislocation density counted from the two areas indicated with white 200 × 200 µm squares was about 2 × 10⁵ cm⁻² and 5 × 10⁵ cm⁻².

Figure 8b shows an X-ray topography image from the Ga-polar surface of the boule, which was the seed crystal. As shown in the image, the defect density was so high that it was practically impossible to adequately distinguish one dislocation contrast from another; hence, the estimated dislocation density was greater than 10⁶ cm⁻². In addition, the seed image shows a macro defect, as indicated with a white arrow. It was confirmed that the dislocation density was improved from >10⁶ cm⁻² level to 2~5 × 10⁵ cm⁻² during the NEAT growth of 100 mm bulk GaN.

4. Conclusions

Bulk GaN crystals of 2″ and 4″ were grown by the NEAT method. The 2″ GaN substrates produced by the NEAT method were characterized with optical observation methods, X-ray rocking curves, and X-ray topography. The latest NEAT GaN substrate had a reduced oxygen concentration of 1.5 × 10¹⁸ cm⁻³. The optical absorption coefficient at 450 nm was 1.3 cm⁻¹, which followed the trend line of the previously reported relationship between the absorption coefficient and oxygen concentration. FWHM mapping of X-ray rocking curves and X-ray topography both demonstrated high degrees of crystalline perfection in the 2″ substrates. The dislocation density was estimated to be low-10⁵ cm⁻² from the X-ray topography. A 100 mm bulk GaN crystal grown by the NEAT method also showed reduced dislocation density (2~5 × 10⁵ cm⁻²) from that of the seed crystal.