1 INTRODUCTION

Owing to a unique combination of features, AlN has attracted a great deal of attention in recent decades. Its wide band gap (6.2 eV) lends this semiconductor a high breakdown voltage, fine thermal conductivity, and a large piezoelectric constant; it also has strong potential for application in high-power electronic [1], ultraviolet photonic [2], and acoustic [3] semiconductor devices. Since bulk substrates are limited in size and costly, AlN layers are normally grown on foreign substrates, such as sapphire and silicon (Si) [4]. Their large size and the potential for integration of gallium-nitride and silicon electronics [5] are the key advantages of silicon substrates.

However, it is rather hard to grow high-quality AlN on Si substrates. First, a large lattice mismatch (~19%) between AlN and Si(111) normally results in high densities of threading dislocations and an initial tension stress. Second, a huge mismatch between the coefficients of thermal expansion (~43%) of AlN and Si leads to cracking in the process of cooling the structure from the epitaxy temperature to room temperature [6]. These are the reasons why the thickness of AlN layers grown on Si is normally below 1 μm and is insufficient to suppress dislocations from the heterointerface [7]. It has been proposed in recent years to perform AlN heteroepitaxy on Si(100) substrates, which are the ones used most widely in silicon electronics, with their surface patterned in the form of rectangular or triangular ridges of micrometer or nanometer dimensions [8]. When forming a structured surface, one may expose planes Si(111) and Si(\(\overline {111} \)) after the deposition of a mask and chemical etching and grow layers at a certain angle (e.g., 54°). Semipolar (\(10\bar {1}1\)) layers on a silicon substrate may be fabricated this way [8].

Thin buffer SiNx layers are formed on flat silicon or sapphire substrates in advance to improve the quality of AlN layers. Specifically, the use of a buffer SiNx-layer helped reduce the density of basal-plane stacking faults in epitaxy of semipolar AlGaN(\(11\bar {2}2\)) on a flat m-Al2O3 substrate [9, 10].

It was established that SiNx may exist in several crystallographic variations. Modification β-Si3N4 is known to be the most stable [11] and is characterized by a hexagonal unit cell with lattice parameters a = b = 7.63 Å, α = 60°, and c = 2.91 Å [12], which is used efficiently in selective epitaxy of GaN layers on a sapphire substrate [13].

Fig 1.
figure 1

SEM images: (a) NP-Si(100) substrate, (d) Si3N4/NP-Si(100) template; AFM images and surface profiles of structures: (b, c) AlN/NP-Si(100), (e, f) AlN/Si3N4/NP-Si(100).

Full size image

In the present study, we compare the surface morphology of AlN layers grown by metalorganic chemical vapor deposition (MOCVD) on flat Si(111) substrates and V-shaped nanostructured Si(100) substrates with and without a deposited nanometer buffer SiNx-layer.

2 PROCEDURE

Nanostructured substrates were fabricated as in [14]. Embryo waves with a small (nanometer) amplitude formed under bombardment of the surface of a silicon substrate by nitrogen ions [14]. Substrates with a nanostructured NP-Si(100) surface (see Fig. 1a) then formed in the process of chemical caustic etching, and templates SiNx/NP-Si emerged after bombardment of NP-Si(100) by a flux of N+. It is evident that bright SiNx strips with different thicknesses of SiNx at the apex and on the slopes are positioned at the tops of nanoridges of templates (see Fig. 1d).

AlN layers were grown on NP-Si(100), SiNx/NP-Si(100) templates, and flat Si(111) substrates with a thickness of ~200 nm by MOCVD as in [15]. Sample surfaces were examined with a SolverNEXT scanning probe microscope. AlN layers were grown on a flat Si(111) substrate and on NP-Si(100) substrates with a V-shaped profile with ridges ~40 nm in height and Si(111) and Si(\(1\bar {1}\bar {1}\)) faces of two types: uncoated (see Fig. 1a) and coated with SiNx (Fig. 1d).

3 RESULTS AND DISCUSSION

Panels a–f of Fig. 1 show the topographic features of AlN surface morphology in AlN/NP-Si(100) and AlN/Si3N4/NP-Si(100) heterostructures. These features indicate that the surface of AlN grown on the template is smoother than the surface of AlN on NP‑Si(100). The root-mean-square (RMS) and average (RA) roughness values of AlN layers grown on the Si(111) substrate, the nanostructured NP-Si(100) substrate, and the SiNx/NP-Si(100) template are lusted in Table 1.

Table 1. Root-mean-square (RMS) and average (RA) roughness values of AlN layers grown on different substrates
Full size table

It is evident that the RMS and RA values of the layer grown on Si(111) are close to the corresponding values of layers on the template (see the table) and differ from the RMS and RA levels of the layer fabricated on NP-Si(100).

As is known, the initial V/III ratio specifies the polarity and morphology of the surface in MOCVD epitaxy of AlN layers. AlN layers grown at high initial V/III ratios have N-polarity and a rough surface, while those grown at low initial V/III ratios feature Al polarity and a smooth surface [16]. The V/III ratio in our experiments remained constant in the process of growth of layers of substrates and the template.

We attribute the variation of surface morphology to differences in the surface diffusion length of atoms on the nanostructured substrate and the template. Indeed, surface diffusion length λs may be expressed in terms of adsorption energy barrier Ead and diffusion potential energy barrier Ed [17]:

$${{\lambda }_{s}} = {{\lambda }_{e}}{{e}^{{\frac{{{{E}_{{ad}}} - {{E}_{d}}}}{{2{{k}_{{\text{B}}}}T}}}}},$$
(1)

where kB is the Boltzmann coefficient and λe is the effective jump distance on the surface or the distance between two neighboring lattice sites.

The (Ead–Ed) value is 0.55 eV/atom for a Si(111) surface and 0.62 eV/atom for a Si(111) surface coated with Si3N4 [18]. It follows from these data that λs on SiNx/NP-Si(100) should be greater than on NP‑Si(100) in the course of formation of an AlN seed on substrate and template faces. This should facilitate growth in the region of template nanogrooves and result in a reduction in the average relief height, which is what is observed experimentally. Note that the surface roughness of the AlN layer for AlN/3C-SiC/Si(111) heterostructures grown by molecular beam epitaxy decreases with decreasing growth rate [19].

4 CONCLUSIONS

Thus, the roughness of an AlN layer grown on a SiNx/NP-Si(100) template is 3.8 times lower than that of a layer grown on NP-Si(100) and is close to the level of roughness of AlN grown on a flat Si(111) substrate.