Novel Oxide Buffer for Scalable GaN-on-Silicon

The economic argument for moving compound semiconductors from their expensive native substrates to large format silicon is well established [1]. To support the economic argument “on-silicon” versions have to be able to run on depreciated fab lines – this places limits on, for example wafer bow and silicon thickness from the perspective of being able to be run on standard fab equipment

For GaN on silicon the current standard approach is to use a mixture of (Al)GaN interlayers grown at lower temperatures to induce the necessary compressive stress to offset the tensile stress that is inherent to GaN on silicon [2]. Some issues with this approach are the Al induced Si diffusion from the substrate into the GaN which can lead to device breakdown, gallium meltback, and the unwanted electrical transport in the interlayers themselves [3].

In this paper we will present an alternative approach using a hybrid epitaxial process, in which the buffer is first optimized with the templatethen used in a standard upstream epitaxial process. For GaN-on-silicon clearly the AlGaN/GaN HEMT has to be grown by MOCVD. For this approach to be successful there has to be a thorough understanding of any possible interaction between the buffer material and the subsequent thermal duty cycles and gas environments typical of MOCVD. This adds an additional design constraint to the engineering of buffer material. In choosing to separate the buffer epitaxy from the III-N epitaxy there has to be some added value in the final epi wafer. In the case of using a rare earth oxide as the intermediate layer the most obvious benefits is that the GaN is now spatially separated from the silicon which removes one of the key breakdown paths [4]

The epi process used is robust, repeatable using high vacuum solid source epitaxy reliant on O₂ and pure metal sources. The process has been successfully scaled to 200mm. The various oxides available can be combined into ternary alloys to produced graded and stacked layers [5] starting from a lattice coincident material at the silicon surface to an upper surface that is now only 9% mismatched from GaN, as opposed to 17% for GaN on silicon. In order to demonstrate the suitability of the oxide within an MOCVD process, bulk layers of GaN were grown using a standard sapphire like process with the typical nitridation, recrystallization, and recovery steps preceding the growth of the GaN. The completed layers were characterized by X-ray diffraction and GaN films with a (002) FWHM of 750arcsecs have been achieved. The surface morphology of the films determined by AFM showed a step flow growth mode typical of MOCVD with <1nm RMS roughness over a 20umx20um area. Photoluminescence from the GaN excited by a 266nm laser showed a strong band-edge emission indicating a high quality GaN film. The oxide-silicon templates were also used as a starting wafer in a full MOCVD HEMT growth which exhibited a fully strained 25% AlGaN barrier. C-V measurements indicated a high mobility channel at the interface between the GaN buffer and AlGaN channel with a sharp pinchoff and a high carrier density of 5x10¹² /cm².

The aforementioned additional value that this hybrid silicon-oxide-GaN process can deliver comes from two directions. Firstly the high quality crystalline oxide itself has a breakdown voltage of >4MV/cm with a dielectric constant of ~15 meaning that the buffer itself can take on a portion of the breakdown voltage thereby reducing the total thickness of GaN required beneath the AlGaN channel. Thinner GaN will require less bow engineering to meet the necessary device fabrication specifications. Secondly results will be presented that show the correct engineering of the oxide growth process can be used to pre-strain the silicon wafer in such a way as to offset the tensile strain imparted by GaN. This offers the potential for a simplified III-N epi process eliminating a large portion of the AlN interlayers currently used for GaN-on-silicon

In conclusion the use of the rare earth oxide represents a novel approach to growing GaN-on-silicon in which the buffer itself adds to the overall functionality of the epiwafer, allows a greater degree of freedom in the growth of GaN on silicon and opens up the possibilities for new ways to engineer power devices.

[1] www.i-micronews.com January 2012 issue 4 p6

[2] D Zhu et al Rep. Prog. Phys. Vol76 p106501 (2013)

[3] Yi-Qiang et al Chin. Phys. B vol22 p088104 (2013)

[4] D. Visalli et al Applied PhysLett, vol97, p113501 (2010)

[5] Compound Semiconductor July 2012 p37