The GaN-on-silicon challenges

The difficulties in GaN-on-silicon growth have been widely discussed during the last twelve years with a focus on lattice and thermal mismatch, but in our experience this focus ignores other difficulties in GaN-on-Si layer growth. In sum they are:

  • 1. high lattice mismatch, low material quality,
  • 2. high thermal mismatch,
  • 3. meltback etching,
  • 4. plastic substrate deformation,
  • 5. vertical conductivity.

In the following we will briefly discuss the state of the art of these topics.

Lattice mismatch

The magnitude of GaN and AlN lattice mismatch on silicon is comparable to lattice mismatch of GaN on sapphire [7]. Using Si as substrate the bond lengths of AlN (1.89 A) and GaN (1.95 A) to Si (2.34 A) differ by -19.2 and -16.7 %, respectively. Thus for AlN on silicon, typically applied as a starting or seeding layer, the situation is comparable to the growth on sapphire. Consequently, the density of misfit dislocations at the substrate/group-III-nitride interface is about the same order of magnitude. The major issue arising from the high misfit is the small coherence length of initial nuclei, leading to small nuclei with no long-range attachment and thus a tendency for high in-plane twist in addition to the usually relatively rough and unstable silicon surface if compared to sapphire. On sapphire the nuclei can be better aligned by applying GaN seeding layers which start as a 2D seed which is then heat-treated to form a few larger 3D islands. Upon further growth they increase in size, coalesce, and again form a 2D layer. By this method, large defect-free islands can be grown. Using AlN seeding layers on sapphire a high material quality probably benefits from the identity of one atom; namely, aluminum. On a microscopic scale most probably an Al2O3-AlN transition is formed. Optimizing this by the addition of oxygen, such seeding layers enable an excellent material quality [8] and reduced in-plane twist [9]. Directly-on-silicon GaN seeding layers do not apply, due to the difficulty in growing GaN directly on silicon, as discussed towards the end of this paragraph. The low surface mobility of Al usually applied for growing AlN seeding layers hinders lateral growth as for GaN on sapphire. In addition, the lower stability of the Si surface and the reactivity of Al and Si, enabling the formation of an alloy, most probably inhibits a good alignment if the seeding layer growth conditions are not well optimized. In sum, the small size of the nuclei leading to a relatively high misalignment of the nuclei, the usually imperfect or microscopically rough Si(111) surface, and its reactivity with group-III and group-V elements, are not beneficial for high material quality. Thus special care has to be taken in seeding layer optimization, starting with surface preparation and substrate heating. The seeding layer determines the overall quality that can be achieved, and a poor seeding layer usually inhibits a subsequent high-quality GaN layer. Due to the six-fold symmetry of the GaN crystal the three-fold surface symmetry of silicon (111) is the first choice for growth. On this surface GaN is oriented with its c-axis vertically to the Si surface plane, thus (0001)AlN У (111)Si or [0001]AlN У [111]Si. The in-plane alignment is [1120]AlN У [110]Si, along the preferred cleavage direction of both materials, or [1100]AlN У Si[112] (Fig. 3.1). Thus cleavage of GaN on Si is more easily applied as on sapphire, where a 30-degree rotation of the GaN layer

Epitaxial relationship of GaN or AlN o

Fig. 3.1. Epitaxial relationship of GaN or AlN on Si(111). In this heteroepitaxial system the preferred cleavage planes of GaN ({1120}) and Si ({110}) are oriented along the same direction. (Reprinted from [7] with permission from Elsevier.)

on the substrate also leads to a 30-degree misalignment of the preferred (1100) cleavage plane of GaN to that of sapphire. As a consequence, cleaved layers on sapphire show rugged cleavage planes.

The alignment of the AlN seeding layer on the substrate depends a lot on the growth conditions and the substrate orientation chosen. There are many approaches to grow AlN on Si, e.g., using different temperatures [10], Al [11, 12], ammonia or even silane pre-flow [13, 14]. It has been also demonstrated that under identical growth conditions the growth on Si(110) can be superior to the growth on Si(111) [15]. In this case, TEM studies revealed the arrangement of AlN on Si and showed that for AlN on Si(110) in the [001]Si У [1l00]AlN direction a nearly perfect lattice matching is achieved with a 2:1 periodicity (Fig. 3.2) [15-18]. In the perpendicular direction the situation is as on Si(111), usually with a «4:5 matching which in sum results in a poor overall matching leading to different bond lengths and a higher initial strain anisotropy in the AlN islands forming on the Si surface. For the growth on Si(110) the better alignment in one direction is already leading to better quality in comparison to Si(111).

In contrast to the growth on Si(111) and Si(110), the growth on Si(001)is more demanding [19-22]. Disregarding the situation of two perpendicular surface reconstructions and by this two perpendicular preferred alignments on this surface, even for one type of surface reconstruction lattice mismatch is large and amounts to 19% and 30% with an alignment of [0001]AlN У [001]Si and in-plane either

Model of lattice matching of an AlN seeding layer on silicon

Fig. 3.2. Model of lattice matching of an AlN seeding layer on silicon (110) (left), and TEM cross-section image showing the 2:1 matching of the lattices and the corresponding model. (Left) Reprinted from [17] with permission from IOP. (Right) Reprinted from [16], © 2008 The Japan Society of Applied Physics.

[01l0]AlN || [110]Si or [2110]AlN || [110]Si [21, 23, 24]. Thus a good alignment is very difficult to achieve, but functional device layers and devices have been demonstrated on up to 150-mm diameter Si substrates [25-27]. Nevertheless, on all substrate orientations a lot depends on the growth conditions of the seeding layer, and its optimization is the key for high-quality layers. Taking particular care on a proper optimization will yield best layers.

 
Source
< Prev   CONTENTS   Source   Next >