Complementary growth technique promises improved LED performance

Epitaxial crystal growth is a critical element in many semiconductor device technologies. And, while successes in fabrication and device design are equally important, epitaxial crystal growth is the foundation upon which everything is built. Or, is it? Inherent in the success of nearly all epitaxial growth is an underlying native substrate, but because bulk crystal growth has reached such a level of refinement in terms of quality and price, this platform for growth is often taken for granted.

In fact, the success of many devices can be traced to the wide availability of high-quality native substrates, with silicon, GaAs and InP obvious examples. There are exceptions, such as HgCdTe for mid-infrared detectors, but by and large without a native substrate one should think twice - and then a few more times - before launching a new material system into the marketplace.

GaN, though, is at least at first glance perhaps a different beast. Its emergence as an electronic and optoelectronic material has all the ingredients of a good myth. It begins with a researcher toiling away in an area of little interest to our community in the confines of a privately owned company with an unrelated core business, and has led to the creation of emergent GaN-based markets such as solid-state lighting, blue and near-ultra-violet lasers and detectors, and high-temperature electronics. GaN is remarkably resilient to defects and is a good optical emitter, despite having dislocation levels of 109 cm-2 that would kill luminescence in GaAs. However, despite this success, reductions in dislocation density can help to make highly reliable GaN products, which brings us back again to the importance of a good substrate.

GaN-based devices suffer from various problems such as poor contacts and difficult processing, but the substrate issue dominates even though it has been with us from the beginning. The reason why it is so difficult to grow a bulk crystal comes down to the incredible bond strength of the nitrogen molecule, which leads to a very strong gallium-nitrogen bond and a high melting point for GaN. For circumstances anywhere close to normal conditions, GaN dissociates into liquid gallium and nitrogen gas before it melts. Only by going to extremely high pressures of 6 GPa at 2220 °C can molten GaN be formed, because at these conditions the melting temperature drops below the dissociation temperature and a nitrogen overpressure allows GaN crystals to form from the melt.

This approach has been used by several groups, beginning with Polish consortium Unipress in the mid-1990s. The crystals are good and have dislocation densities of 102 cm-2, but scaling has been extremely difficult and even today the technique is limited to the production of 10 mm substrates.

Consequently, several companies in Asia, Europe and North America are exploring an alternative approach to GaN substrates production using the well-established hydride vapor-phase epitaxy (HVPE) process. Although this is much more costly than the bulk-crystal substrate process, it is inexpensive for a VPE process and can deliver growth rates of at least 100 μm/h on multi-wafer platforms. It is also better suited to the growth of low-impurity III-V films than MOVPE, but is rarely used for device fabrication. This is primarily because the technique cannot offer the precise thickness control and sharp interfaces demanded by many structures. In addition, the extremely high epitaxial growth rates that are possible with this approach are not in demand, and very low impurity levels are less important in p-n junctions.

However, the HVPE process is well-suited to making relatively large free-standing GaN substrates by depositing a several hundred micron-thick GaN film on a sacrificial substrate such as sapphire and then removing the substrate. Currently, though, this is a difficult and low-yield process. This is partly because of crystal growth and processing issues such as surface roughness and etching. But the main issues are film stresses that result from the thermal and lattice mismatch between the GaN and sapphire that can cause thick layers of GaN to bow or even crack, and make the final separation of GaN from the sapphire substrate difficult. Progress is being made, but substrates produced in this way are expensive, and in the near future will only be suitable for high-cost devices requiring low defect densities, such as GaN lasers. Consequently, an alternative approach is needed for successful commercialization of a range of lower-cost GaN devices.

Our approach addresses the issues associated with substrates and growth as a single entity - as opposed to isolated substrate and epitaxy problems - and focuses away from the laser market that is already best served by expensive GaN substrates. Our aim is to affordably reduce the defect density in a broad spectrum of GaN-related devices grown on foreign substrates, and we believe this can be achieved by combining HVPE and MOVPE in a single process. This allows HVPE to provide an inexpensive high-growth-rate process for the low dislocation-density thick buffer layer, and MOVPE to contribute a very thin, initial starting layer and the final device layers (see figure 1).

MOVPE, not HVPE, is used for buffer-layer growth for several reasons. Firstly, HVPE s high growth rate hinders accurate thickness control of the low-temperature buffer layer (∼25 nm). It also tends to lead to parasitic depositions that can re-enter the gas stream after the reactor is heated for subsequent growth, due to the relatively low substrate temperature compared with the growth chamber and gallium source. Excess chlorine atoms can also cause etching during HVPE growth for temperatures below 800 °C that favor GaCl3 formation.

Although combining HVPE and MOVPE growth in one reactor has many advantages, it has previously been avoided because the two growth processes have different requirements. The MOVPE chemical reaction is strongly exothermic and produces a positive net entropy change with temperature. Higher temperatures increase the driving force and deposition is most favorable in the hottest region of the growth system, which should be the substrate area. In contrast, the HVPE reaction is weakly exothermic, with a negative change in entropy - as temperature increases the driving force decreases. Deposition is greatest in the coolest part of the growth system, within decomposition limits. The HVPE reaction is also a near-equilibrium deposition method and at ∼1000 °C, a typical growth temperature, the driving force for HVPE deposition is just 3% of that for MOVPE. This means that the HVPE process produces fewer point defects than MOVPE, but also creates interfaces that are less sharp, which is why the latter technique is preferred for heterostructure growth.

We have unified the two methods by combining a cold-wall approach for MOVPE growth with a hot-wall approach for the HVPE process. During the MOVPE process a resistive heater under the substrate raises the wafer s temperature above that of the reactor walls, which are unheated. For HVPE growth the situation is reversed, and the multi-zone external tube furnace heats the surroundings more than the substrate.

Using this combined process we have grown a 12 μm thick GaN layer onto a 2 inch diameter (0001) sapphire substrate. After the substrate was cleaned, etched and loaded into the reactor, it was heated to 1000 °C for thermal cleaning. The reactor s temperature was then lowered to 500 °C using the cold-walled resistive heating configuration for MOVPE growth of the 25 nm thick buffer layer. The external heater then raised the temperature of the wafer to 1025 °C for recrystallization and the growth of 12 μm of GaN by HVPE.

AFM measurements over 10 × 10 μm areas show a root-mean-square surface roughness of less than 1 nm, while cross-sectional TEM reveals the dislocation networks (see figure 2). These images identify a dense network of threading dislocations that appear as dark lines concentrated at the GaN-sapphire interface. Most dislocations are curved and some are even bent into the basal plane, suggesting that they can interact, annihilate, or tie themselves up and limit propagation. The upshot is that this epitaxial process can efficiently reduce dislocations during the growth. Most of the action occurs within a few microns of the GaN-sapphire interface, where the dense tangle of dislocations diminishes dramatically, and after 12 μm of growth the surface s dislocation density drops to the 107 cm2 range.

Although the reduction in GaN dislocation density by two orders of magnitude to 107 cm2 is significant, it is still too high for the GaN substrate market. However, for a major reduction in dislocation density it is neither time-consuming nor expensive, and could benefit lower-cost commodity markets where GaN substrates would be too costly, such as LED-based general lighting.

If so much is possible with just 12 μm of growth, why shouldn t we turn to even thicker buffer layers to cut the dislocation density even more? Firstly, as can be seen in figure 2, the greatest reductions in dislocation density occur close to the GaN-sapphire interface, and additional growth brings diminishing returns. More importantly, further growth has a downside, as greater thicknesses increase the accumulated thermal and lattice mismatch strain, which can reduce device performance and reliability and can cause the material to bow.

The improvements in crystal-growth quality produced by growing very thick layers are seen by the continued reductions in the full-width at half-maximum (FWHM) values from our high-resolution X-ray diffraction rocking-curve measurements. Measurements on the 12 and 50 μm thick films show that the FWHM of the X-ray peaks from the symmetric (0002) and asymmetric (202Bar1) reflections reduce from 460 and 450 arcsec to 300 and 280 arcsec, respectively. This shows that the thicker films do produce an improvement in crystal quality, but bowing becomes significant in these films, which prevents them making good large-area substrates.

Free-standing GaN substrates will have an important role in the continued improvement of GaN-based devices, especially in situations where performance can justify the significant substrate costs. However, lower-cost products are currently better served with a simple hybrid MOVPE-HVPE crystal growth process. This approach allows the most suitable process to be used for the low-temperature starting layer, the thick buffer layer and the heterostructure, and can form high-quality devices in a single reactor using one continuous growth process.

David Miller at CBL Technologies has played a significant role in all phases of this work. Thanks to Oliver Brandt, Klaus Ploog, Manfred Ramsteiner and Achim Trampert of the Paul Drude Institute (PDI) in Berlin, Germany, for film characterization and discussion, and the PDI for their support.