Epitaxial lateral overgrowth (ELO) is a classic approach for reducing defects in nitride epilayers. This study, conducted entirely on r-plane sapphire substrates, develops and compares two patterning strategies: the conventional one-dimensional stripe template and a novel two-dimensional chessboard-like microrod array.

Fabrication Process Flow

The complete process consists of five steps:

1. Growth of the native GaN template: A 1 μm-thick (11–20) non-polar GaN layer is grown on r-plane sapphire by low-pressure MOVPE using a high-temperature AlN buffer layer. This is designated the "as-grown" template.

2. SiO₂ mask deposition: A 150 nm-thick SiO₂ film is deposited by plasma-enhanced chemical vapor deposition (PECVD).

3. Lithography and dry etching: Standard photolithography followed by reactive ion etching (RIE) patterns the SiO₂ into a regular microrod array. The SiO₂ pattern then serves as a hard mask for inductively coupled plasma (ICP) etching of the underlying GaN, etching all the way down to the sapphire surface. Each SiO₂ micro-mask remains on top of its GaN microrod.

4. KOH photo-electrochemical (PEC) etching: The patterned template is immersed in a 10% KOH solution for photo-electrochemical etching. This etch is highly selective — it removes GaN from the [000–1] N-polar direction while leaving the [0001] Ga-polar facets essentially untouched, and does not etch the SiO₂ mask. The result is a mushroom-shaped microrod cross-section.

5. MOVPE overgrowth: The patterned template is returned to the MOVPE reactor for GaN overgrowth at controlled temperatures.

Why the Mushroom Shape Matters

It is well established that growth along the [0001] c-direction produces relatively defect-free material, whereas growth along the opposite –c direction allows defects to propagate vertically through the overgrown layer. The overhanging mushroom geometry physically blocks defect propagation from the –c direction, further improving crystal quality.

Template Specifications

Parameter	Microrod Array (Samples A/B/C)	Stripe Template (Sample D)
Geometry	Chessboard-patterned circular microrods	Parallel GaN stripes
Feature size	2.5 μm diameter	1.5 μm stripe width
Spacing	Regular 2D array pitch	1.5 μm gap between stripes
Orientation	Two-dimensional periodicity in-plane	Strips run along [1–100] (perpendicular to c-direction)
Etch depth	Etched through to sapphire	Etched through to sapphire
Special structure	KOH PEC etch creates mushroom cross-section	SiO₂ stripe mask retained on top

 
 
[Figure 1] Top-view SEM image of the regular microrod array template. Scale bar: 4 μm.
 
 
[Figure 2 ] Cross-sectional SEM image of the mushroom-shaped microrod structure. Scale bar: 2 μm. The SiO₂ mask and GaN overhang are visible.

Overgrowth Morphology Evolution on Microrods

The study tracks surface evolution by varying overgrowth time from 1000 s to 8000 s and imaging with SEM.
 
 
[Figure 3 ] Plan-view SEM images of non-polar GaN overgrown on the microrod array at (a) 1000 s, (b) 2000 s, (c) 3000 s, (d) 4000 s, (e) 5000 s, and (f) 8000 s. Scale bar: 4 μm.
 
The growth proceeds through four stages:

1. 1000 s (nucleation): Growth initiates preferentially along [0001] c-direction, with partial contribution from [1–100] and [–1100], while [000–1] –c growth is effectively suppressed by the mushroom structure. Truncated prisms form around each microrod; the first coalescence begins before 1000 s.

2. 2000 s (first coalescence): The gaps between microrods are filled by lateral growth, completing the first coalescence. Vertical growth then becomes dominant.

3. 3000–4000 s (second coalescence): Lateral growth extends over the SiO₂ mask, generating multiple semipolar facets as adjacent overgrown regions merge.

4. 5000–8000 s (full coalescence): Complete coalescence is achieved, yielding a smooth surface. Voids remain between microrods and above the SiO₂ mask.

Geometric Limitation of the Stripe Template

By contrast, the one-dimensional stripe template imposes a geometric constraint: growth begins at the stripe sidewalls and expands primarily along [0001], while [1–100] lateral growth is blocked by the stripes themselves. Voids form beneath the overgrown stripes, but the [1–100] direction remains constrained throughout the process.

Crystal Quality: XRD Rocking Curve FWHM

Because non-polar GaN is intrinsically anisotropic, rocking curve FWHM must be measured as a function of azimuthal angle. Azimuth 0° corresponds to the incident X-ray projection parallel to the c-direction; 90° corresponds to projection along the [1–100] m-direction.
 
[Figure 4 ] Azimuth-dependent (11–20) on-axis XRD rocking curve FWHM for the as-grown sample, the microrod-array overgrown sample, and the stripe-template overgrown sample.
 

Template	FWHM along [0001]	FWHM along [1–100]
Planar as-grown on sapphire	1180 arcsec	1880 arcsec
Stripe template (Sample D)	380 arcsec	420 arcsec
Microrod array (A/B/C)	270 arcsec	380 arcsec

 
The microrod template achieves the best reported values to date for (11–20) non-polar GaN on r-plane sapphire. Cross-sectional transmission electron microscopy (TEM) reveals that the dislocation density drops from ~10¹¹/cm² in the as-grown template to ~6×10⁸/cm² — a reduction of more than two orders of magnitude.

Why it works: A one-dimensional stripe provides lateral defect filtering in only one direction. A two-dimensional chessboard microrod array permits lateral growth — and thus dislocation bending and annihilation — along both orthogonal in-plane directions, which is why the rocking curve widths are superior in both directions.

Crystal quality is only half the story. Strain distribution is the other make-or-break factor for non-polar GaN devices. The next post examines how sapphire substrate patterning rewrites the strain state of the epilayer.