Preparing the way for high-quality mass-producible semi-polar GaN 

Careful control of epitaxy, in part through facet engineering, can create a semi-polar GaN template on sapphire that is ideal for the mass-production of ultra-efficient LEDs

BY Jie Song and Jung Han from Yale University

There have been no major breakthroughs in LED technology in recent years. While progress is still being made, gains in the efficacy of lighting products are now steady, yet sluggish. And although there is still a tangible performance gap between "tier 1" and "tier 2" vendors, it is narrowing.

Further expansion and penetration of solid-state lighting hinges on a step change in performance. This could come from solving either of two biggest outstanding technology issues, which are known as the efficiency droop and wavelength droop. The former is the rapid decrease in an LED's external quantum efficiency above a current density of typically 1 A cm-2, while the latter refers to the systematic decrease in InGaN LED efficiency from blue to green, yellow, and amber.

Figure 1. Cross-sectional g = <1100> two-beam bright-field transmission electron microscopy images near the [1120] zone axis of two samples. The defects running through the layers are identified as basal plane stacking faults. Reprinted with permission of AIP Publishing.

Both these issues, which have been the subject of much investigation, are beyond the scope of this article. However, they are included as a backdrop to the semi-polar technology developed by our team at Yale University and Saphlux that has the potential to take LED performance to a new level.

While details of various mechanisms that contribute to the droop continue to court debate, there is a consensus of opinion that the polar nature of the wurtzite GaN crystal configuration, in particular the strong polarization-induced electric field along the c-axis (0001), plays a central role in complicating and compromising the performance of GaN light emitters.

The presence of high electrostatic fields in III-nitride heterostructures is well known, with reports of its existence appearing as far back as the late 1990s. Since then, researchers have been hunting for more favourable GaN orientations.

A tough start

Early success came at Paul-Drude Institute, Berlin. These researchers revealed that non-polar [1010] m-plane GaN can be grown on foreign substrates. Their technology, referred to as the beginning of the Gen 1 approach for preparing non-polar and semi-polar GaN on planar heteroepitaxial substrates, produced films with a rather primitive material quality. Stacking faults and dislocations riddled the epilayers (see Figure 1).

It did not take long for researchers in the III-nitride community to turn to their toolkit of approaches for eliminating defects. That included epitaxial lateral overgrowth (ELO), a technique that can suppress structural defects from heteroepitaxy. The ELO technique, which we describe as the Gen 2 approach, is known to deliver success on c-plane GaN, where it leads to the formation of dislocation-free layers. But when it is applied to non-polar and semi-polar GaN planes, new stacking faults form in the overgrown region along the -c-axis [0001]. The tenacity and recurrence nature of these stacking faults has cast serious doubt on the ultimate viability of non-polar and semi-polar devices.

Figure 2. The growth evolution for (1100) GaN formed by epitaxial layer overgrowth. Reprinted with permission of Japan Society of Applied Physics.

One of the critical advances within the nitride community in the years spanning the late 1990s to the middle of the following decade was the use of hydride vapour phase epitaxy (HVPE) to prepare c-plane GaN with a thickness in the millimetre or even centimetre range. This technology, perfected by several Japanese semiconductor manufacturers, is capable of growing GaN at hundreds of microns per hour. It enables the preparation of GaN crystals that are shaped like hockey pucks and have a thickness of 5 mm or more. Manufacturers take the upper portion of these pucks, which have a very low density of dislocations, and slice them into free standing GaN wafers. This yields high-quality wafers that support the commercialization of performance demanding applications, such as blue laser diodes.

These GaN pucks can also be sliced at different angles to produce samples with different crystallographic orientations. This approach becomes a convenient shortcut for researchers to quickly access high-quality, non- and semi-polar GaN to validate the anticipated benefits.

Further success came in 2005, when researchers at UCSB, working with Mitsubishi Chemical Corporation, built the first non-polar, m-plane GaN LEDs on a free-standing GaN substrate. These devices delivered an instantaneous, compelling improvement in light output, asserting the promises of non-polar and semi-polar orientations while emphasizing the importance of employing a low-defect template.

Between 2007 and 2014, academic groups and industrial labs, including those at Sumitomo, Sony, Nichia, and Osram, have reported performance enhancements. Measurements made in these facilities show that devices formed from various semi-polar orientations, prepared by cross-slicing of GaN crystal pucks, are less prone to current and wavelength droop. This validates the potential for next-generation GaN devices formed on orientations with reduced polarization fields.

Figure 3. Semi-polar GaN bulk substrates can be produced via HVPE growth of thick, c-plane GaN bulk substrates and subsequent cross-slicing.

The main drawback of HVPE-grown, free-standing GaN substrates is that they are difficult to scale-up and manufacture. Due to the geometry of the GaN puck, cross-slicing at a high angle to the c-plane (to move away from the polar-axis) yields tiny strips with irregular, inconsistent dimensions (see Figure 3). This form of substrate is a nonstarter for any serious LED-based research and development activity in growth and processing. In spite of the compelling results from a handful research labs specialized in coping with these defect-free, free-standing non- or semi-polar GaN pieces, the majority of the III-nitride community has yet find a way to cross the chasm, and to commercialize non- and semi-polar GaN technology.

This situation could change if high-quality GaN can be made through hetero-epitaxy on large-area, mainstream substrates, such as sapphire or silicon. Beginning from 2008, several groups began to investigate a new approach to accomplishing this, which we refer to as orientation-controlled epitaxy (OCE) on patterned sapphire substrates.

Revisiting, improving

The idea of OCE on sapphire originated from an earlier work by Nobuhiko Sawaki at Nagoya University in Japan. That team demonstrated selective growth of GaN on inclined {111} sidewalls of stripe patterned silicon substrates. The GaN growth proceeds in the c-direction, inclined to the surface normal, which can coalesce to produce a flat semi-polar or non-polar surface depending on the inclination angle of the sidewalls. This method allows a new definition and design of epitaxial relations between the layer and the substrate. Success stemmed from integrating several established techniques, including high-precision patterning of sapphire substrates, selective area growth, and the preparation of large-area, arbitrarily orientated sapphire substrates (see Figure 4).

Figure 4. Growth of semi-polar (1122) GaN layer on a maskless, patterned r-plane sapphire substrate. Reprinted with permission of AIP Publishing.

With this approach, to realise the required final GaN surface orientation, one must simply calculate the required direction for the GaN c-axis (as well as the corresponding a- or m-axis). Together, these GaN axes will uniquely specify the orientation of the c-plane sapphire that is inclined from the surface normal. By applying simple geometric algebra, it is possible to determine: the crystallographic plane for the sapphire surface; the direction of trenches that should be opened up; and the angle of inclination of the trench sidewalls, such that one of the sidewall facets will have the c-plane sapphire orientation that supports the desired semi-polar or non-polar GaN surface.

This orientation-controlled approach, which we refer to as Gen 3, has several advantages over other approaches. First, the lattice and crystallographic mismatches that plague planar semi-polar and non-polar heteroepitaxy are no longer an issue. Instead, formation of semi-polar and non-polar material becomes an exercise in c-plane growth of GaN on c-plane sapphire "“ a topic that the III-nitride community has mastered for a few decades. Second, orientation-controlled epitaxy employs selective-area overgrowth, a technique proven to be effective in reducing the densities of threading dislocations. And third, this approach is capable of producing, at least in principle, every possible non- and semi-polar orientation by design.

Figure 5 Cross-sectional scanning electron microscopy images of (a) (1122), (b) (2021), and (c) (2021) GaN films grown on sapphire substrates. Left and right columns of scanning electron microscopy images correspond to a short growth time before coalescence and a long growth time after coalescence, respectively.

Using r-plane and (2243) plane sapphire, with a trench angled at 58° and 75°, we have prepared (1122) and (2021) GaN, respectively (see Figure 5). What's more, by combining orientation-controlled epitaxy with the nitridation of c-plane sapphire, we have been able to change the growth of GaN on (0001) sapphire sidewalls from the (0001) plane to the (0001) plane, making it possible to produce the (2021) plane on GaN-on-sapphire. Together these three demonstrations show that it is possible to produce any semi-polar GaN orientation of GaN-on-sapphire.

Unfortunately, orientation-controlled epitaxy is not a perfect solution, as nasty stacking faults remain present in these Gen 3 layers (Figure 6). Again, it is the tenacity of stacking faults that presents an impasse for mass-producible, semi-polar GaN-on-sapphire.

Figure 6. Cross-sectional transmission electron microscopy images of (2021) GaN grown on patterned sapphire by orientation controlled epitaxy (Gen 3) under two-beam conditions with g vectors of (a) <0002> and (b) <1100>, respectively.

After reviewing a large volume of experimental work, we conclude that basal plane stacking faults are typically generated in non- or semi-polar GaN growth when the N-polar basal plane (0001) front is brought into contact with (or over) a foreign surface, such as the surface of SiO2, during overgrowth. This led us to develop a refined version of our process, which we call facet-engineered, orientation-controlled epitaxy, or the Gen 4 process.

Eliminating the faults

According to the classical theory of crystal growth, the Wulff principle states that rapid-growing facets tend to become extinct, and that a crystal would be bound by slow-growing facets under equilibrium. In our Gen 4 process, our proprietary method accelerates the typically slow-growing N-polar plane, to the extent that the N-polar GaN basal plane disappears very quickly. This eliminates the formation of basal plane stacking faults.

To demonstrate the capability of our technology, we have incorporated this technique in the growth of (2021) GaN-on-sapphire. X-ray diffraction shows two peaks corresponding to GaN (2021) and sapphire (2243) diffractions, indicating that a single (2021) orientation of GaN has been achieved with direction parallel to the sapphire (2243) orientation (see Figure 7 (a)).

With this approach, we have eliminated the basal plane stacking faults in (2021) GaN. Work from Kazuyuki Tadamoto's group from Yamaguchi University, Japan, shows that in plan-view cathodoluminescence images, the non-radiative centres associated with stacking faults cause them to be exhibited as dark bands/straight lines. That feature is not present in our images "“ all we see are dark spots, due to the extension of threading dislocations from GaN-on-sapphire heteroepitaxy (see Figure 7(b)).

Figure 7. (a) X-ray diffraction 2 θ/ ω scan about (2021) GaN grown on sapphire. (b) Plan-view panchromatic cathodoluminescence image of (2021) GaN grown on sapphire cannot uncover any stacking faults. (c) Cross-sectional transmission electron microscopy image under two-beam condition taken along a diffraction vector of g = <1010>. (d) A photo of a 2-inch, stacking fault-free (2021) GaN grown on sapphire.

To take a closer look at the microstructure of our material, we have turned to transmission electron microscopy (TEM, see Figure 7(c)). Numerous TEM studies reveal no detectable stacking faults, confirming again that our Gen 4 technique is effective at suppressing basal plane stacking faults through the elimination of N-polar planes.

We have also inspected the defects over a 2-inch wafer, using plane-view cathodoluminescence (see Figure 7(d)). Inspecting different positions indicates that we can achieve large-area, stacking-fault-free (2021) on a 2-inch sapphire substrate. This is a very promising result, and more should follow, as our technology is applied to the growth of other semi-polar and non-polar orientation stacking-fault-free, GaN-on-sapphire substrates.

Figure 8. Successive generations of growth technologies have lowered and then eliminated the stacking faults density in semi-polar and non-polar GaN grown on sapphire substrates.

There's no doubt that it's a long road to reach defect-free, semi-polar and non-polar GaN layers on large-area, mass-producible wafers. This is highlighted by our plot of the density of stacking faults in semi-polar and non-polar GaN on various heteroepitaxial substrates (see Figure 8). Successive generations of growth technologies have initially reduced stacking fault densities from generally above 105 cm-2 to 103-104cm-2, before we eliminated them. This eradication is a significant milestone toward mass-producible truly high-performance semi-polar LEDs, as even one stacking fault in these devices is one too many.

Having addressed the fundamental issue of the control and elimination of stacking faults for semi-polar GaN layers on sapphire, we are cautiously optimistic that semi-polar and non-polar LEDs on sapphire will impact, if not displace, mainstream c-plane LEDs in the future. With improvements in the incumbents nearing saturation, the "˜end of Moore's law' could be in sight for c-plane LEDs, forcing the community to look for and adopt new directions, both figuratively and (in our case) literally.