Hybrid approach yields the best GaN crystals and wafers

If a perfect process were to exist for forming boules of GaN, it would create material that is completely free from dislocations, impurities and bow. In addition, this process would quickly yield crystals with very large dimensions, so that substrates sliced from them would provide a cheap, ideal foundation for the manufacture of LEDs, lasers and power electronics.

However, as we all know, such a process does not exist. Instead, makers of GaN devices have to choose between a far-from-perfect native platform that is very pricey "“ a 2-inch GaN substrate retails for $2000 or more "“ and several options for larger, foreign substrates that have the drawback of leading to inferior material quality. Makers of laser diodes don't actually have this second choice, because they must work with very high quality material, so GaN is used. However, for the manufacturers of LEDs and transistors, GaN is nearly always deemed to be too expensive, so sapphire, silicon and SiC are widely adopted. These foreign substrates all have a lattice mismatch to GaN that leads to the generation of copious defects in the epistructures, resulting in the loss of device performance and lifetime.

It should be possible to address the lack of an affordable, ultra-high-quality GaN substrate with a process that has been developed by partnership between the Institute of High Pressure Physics of the Polish Academy of Sciences (IHPP PAS), working in close partnership with both Ammono S.A. and scientists at the Wide Bandgap Laboratory at North Carolina State University (NCSU). This team that we are part of employ a technology that involves uniting two existing techniques: HVPE and ammonothermal growth.

Fig. 1: a) A polished 2-inch GaN wafer produced by Ammono S.A. (courtesy of Ammono S.A.); b) An atomic force microscope image of the epi-ready (0001) surface of the Ammono-GaN wafer after chemo-mechanical polishing; the image shows the desired bi-layer steps; root-mean-square roughness is below 0.1 nm (courtesy of G. Kamler and G. Nowak, IHPP PAS).

Virtues of HVPE

Strengths of the former, HVPE, include a relatively high growth rate, which can exceed 100 mm/h, and a possibility to crystallize high-purity material. Thanks to these attributes, HVPE is a well-established method for making GaN substrates "“ this growth technique is employed by the likes of Sumitomo Electric Industry, Hitachi Metals, Furukawa Co, Mitsubishi Chemical, and Saint Gobain (formerly Lumilog).

These leading manufacturers of GaN begin by placing a foreign substrate, typically sapphire or GaAs, into a HVPE reactor, before heating the chamber to around 1300K. Introducing ammonia and gallium chloride at ambient pressure leads to the crystallization of GaN on foreign substrates, and when this thick film is removed via etching or self-separation techniques, a GaN substrate results.

Manufacture of these GaN substrates is well established. For example, market leader Sumitomo already demonstrated 6-inch freestanding material back in 2010. However, the hetero-epitaxial growth causes the substrates to be riddled with a high density of defects, which typically number 10⁶ - 10⁷ cm^-2. What's more, the growth on a foreign substrate causes a large lattice bow "“ for example, when GaN is grown on sapphire the bowing radius of the (0001) crystallographic planes is always below 10 m. This is not good enough for epitaxy and device processing on 2-inch wafers, which require bowing radii exceeding 30 m.

Lattice bow also plays havoc with the growth of thick HVPE-GaN boules, the fabrication of large, freestanding substrates, and the use of freestanding HVPE-GaN crystals as seeds for crystal multiplication. Note that there is no point in crystallizing more material on a seed with a bent lattice, since crystalline quality only deteriorates further with growth.

Fig. 2: HVPE-GaN crystals, thicker than 1 mm, crystallized on 1-inch Ammono-GaN wafers in a few hours (courtesy of T. Sochacki, IHPP PAS).

For the growth of epi-structures and the processing of devices, the wafer surface must be offcut uniformly with an accuracy of one to two tenths of a degree in a specific crystallographic direction, which produces a specific step structure on the wafer's surface. This low offcut promotes bilayer step flow, controls the composition of ternary alloys, and enables uniform incorporation of dopants. Lattice bow precludes all this. If substrates have a significant lattice bow, it is impossible to form homogeneous device layers across the wafer, and low production yield results. For a 2-inch wafer, for example, if the tolerance for the deviation of the offcut across the wafer is below 0.1°, which is a typical value, then the bow radius must be greater than 30 m to be able to meet this tolerance (see table 1 to understand the interplay between the deviation of the surface offcut and the bowing radius for wafers of various sizes).

Fig. 3: A single growth center observed on the as-grown, 1-inch crystal surface (courtesy of T. Sochacki, IHPP PAS).

A better foundation

The problem of lattice bow is not intrinsic to the HVPE process, but is rather a result of hetero-epitaxy. So, what is needed is to begin with a structurally perfect GaN seed "“ a requirement that can be fulfilled if this material is grown by the ammonothermal method, which involves forming GaN from a solution in supercritical ammonia.

The best partner in this approach, which is based on HVPE deposition on ultra-high-quality GaN seeds, is the world's leading grower of ammonothermal GaN, Ammono S.A. This Polish firm uses an approach that is analogous to the hydrothermal crystallization of quartz − but with supercritical ammonia replacing water "“ to produce GaN with many attractive attributes: exceptional lattice flatness, demonstrated by bowing radii of the (0001) crystallographic planes of around 100 m; dislocation densities
of typically just 5 x 10⁴ cm^-2; and a
free carrier, n-type concentration that
may be varied from 5 x 10¹⁷ cm^-3 to 2 x 10¹⁹ cm^-3 (see Figure 1a for a 2-inch GaN substrate).

Fig. 4: a) A polished freestranding HVPE-GaN wafer sliced from the Ammono-GaN seed (courtesy of T. Sochacki, IHPP PAS); b) Wafer surface after the defect selective etching; the etch pit density was 5x10⁴ cm^-2 (courtesy of J.L. Weyher, IHPP PAS).

Making great use of these seeds, the part of our team based at the IHPP PAS has developed a combination of mechanical and chemo-mechanical processes for transforming the surfaces of these wafers into an epi-ready state. Using this technique, one can obtain surfaces with uniform bilayer steps and a root-mean-square (RMS) surface roughness below 0.1 nm (see Figure 1b).

To open the door to a new era of GaN bulk crystal and substrate production, the hybrid HVPE-ammonothermal growth technique has to succeed on two fronts: the perfection associated with the ammonothermal GaN seeds must be maintained in the HVPE-grown material; and it must be possible to grow thick HVPE boules to multiply the ammonothermal GaN crystals. IHPP PAS and Ammono S.A. have triumphed in both areas while working on a two-year project that finishes this July and has been backed by $1 million by the Polish National Center for Research and Development.

One of the highlights of this effort has been the formation of 1 mm-thick HVPE-GaN crystals, which were crystallized on 1-inch Ammono-GaN wafers in a few hours (see Figure 2). This additional GaN grown by HVPE is macroscopically flat and free from cracks or pits. HVPE conditions governed the growth rates, which ranged from 120 mm/h to

240 mm/h. As desired, only one growth center on the entire 1-inch crystal surface after the growth process was observed (see Figure 3).

By slicing the freestanding wafers of HVPE-grown GaN from the ammonothermally grown seeds, members of our team were able to form about 300 mm-thick GaN substrates that replicated the crystalline quality of the original Ammono seed, and were free from cracks and pits (see Figure 4a). Note that the Ammono-GaN seeds that were separated could be used for further HVPE crystallization without any limitations.

IHPP PAS has scrutinized the quality of the freestanding, HVPE-grown GaN substrates with a variety of techniques. Using a molten KOH-NaOH eutectic revealed an etch pit density, and thus a threading dislocation density, of 5 x 10⁴ cm^-2, the same as the original seed. This effort also revealed three types of etch pits: large etch pits, which were correlated to the screw dislocations; small pits that were associated with threading edge dislocations; and medium-sized pits that originated from threading mixed dislocations (see Figure 4b). These insights showed that the structural properties of the freestanding HVPE-GaN do not differ from the structural properties of the Ammono-GaN seeds. However, IHPP PAS still needs to conduct further work to directly correlate the defects in Ammono-GaN with those found in HVPE-GaN.

Team members at NCSU have determined the structural and optical properties of the freestanding, HVPE-grown GaN. X-ray rocking curves of the symmetric (0 0 2) and skew-symmetric (3 0 2) and (1 0 2) reflections produced very intense, narrow Bragg peaks, indicative of excellent crystallinity and low dislocation density (see Figure 5). The full-width at half-maxima values of the (0 0 2), (1 0 2), and (3 0 2) reflections were 22, 17 and 35 arcsec, respectively; these values were close to the theoretical values for a perfect GaN crystal.

Meanwhile, low-temperature photoluminescence spectra acquired from the (0001) surface exhibited several well-defined bound and free exciton peaks. Dominating the spectra were two sharp donor-bound exciton emission lines at 3.471 eV and 3.472 eV that have values for the full-width at half maximum of 127 meV and 167 meV, respectively (see Fig. 6a). This width for the bound exciton peaks is the narrowest ever reported for GaN, confirming not only the high crystalline quality of this material, but also its very high purity.

Fig. 5: X-ray rocking curves of the symmetric (0 0 2), and skew-symmetric (3 0 2) and (1 0 2) reflections for the freestranding HVPE-GaN (courtesy of M. Bobea, NCSU).

Additional measurements by the NCSU team revealed that the room-temperature transmission spectrum was relatively featureless up to the band edge (see Figure 6b). What was even more impressive was that, according to secondary ion mass spectrometry, oxygen and carbon contents were below 10¹⁶ cm^-3, while the silicon impurity level was 3 x 10¹⁶ cm^-3. In addition, the substrates had a high surface quality, according to low-temperature photoluminescence and high-resolution X-ray studies of the diffuse scatter and crystal truncation rods.

These measurements confirmed that the work has laid the foundation for the manufacture of high-quality, larger GaN crystals that can be cut into wafers with the well-defined and uniform offcut that device makers are looking for. However, that's not to say that the research in this area is over. Next, the scientists at IHPP PAS want to examine doping via HVPE, to see if it is possible to create semi-insulating and n-type substrates with controlled doping levels. It is also needed to transfer the process to 2-inch material. As IHPP PAS do all this, Ammono will be optimizing their ammonothermal GaN growth on these HVPE crystals, to increase the production and availability of ultra-high-quality GaN substrates.

Fig. 6: Photoluminescence spectra obtained at 3K from the (0001) surface of the freestranding HVPE-GaN (linear scale); b) RT transmission spectrum of the freestanding HVPE-GaN (courtesy of Z. Bryan and I. Bryan, NCSU).

Further reading

T. Sochacki et. al. Appl. Phys. Express 6 075504 (2013)

T. Sochacki et. al. Journal of Crystal Growth 394 55 (2014)