Growth and properties of semiconducting transition-metal nitride layers

Abstract

In addition to ternary nitride, (Ti0.5Mg0.5)1−xAlxN is used as a model system to explore and demonstrate the tunability of both the bandgap and the strain state in rocksalt structure quaternary nitride semiconductors. 75-135 nm thick (Ti0.5Mg0.5)1−xAlxN layers with x  0.44 deposited on MgO(001) substrates by reactive co-sputtering at 700 °C are epitaxial single crystals with a solid-solution B1 rocksalt structure, as determined by x-ray diffraction ω-2θ scans, ω-rocking curves, and reciprocal space maps. The lattice mismatch with the substrate decreases with increasing x, leading to a transition in the strain-state from partially relaxed (74% and 38% for x = 0 and 0.09) to fully strained for x  0.22. First-principles calculations employing 64-atom Special Quasirandom Structures indicate that the lattice constant for (Ti0.5Mg0.5)1−xAlxN decreases linearly according to a = (4.308 - 0.234x) Å for 0 ≤ x ≤ 1. In contrast, the measured relaxed lattice parameter ao = (4.269 - 0.131x) Å is linear only for x ≤ 0.33, its composition dependence is less pronounced, and x > 0.44 leads to the nucleation of secondary phases. The onset of optical absorption due to interband transitions increases from 2.3 to 2.6 eV for x = 0-0.44. This increase is smaller than that of the expected bandgap, suggesting that the addition of Al in the solid solution relaxes the momentum conservation and causes a shift from direct to indirect gap transitions. The resistivity increases from 9.0 to 708 μm at 77 K and from 6.8 to 89 μm at 295 K with increasing x = 0-0.44, indicating an increasing carrier localization associated with a randomization of cation site occupation and the increasing bandgap which also causes a 33% reduction in the optical carrier concentration.

The potential as refractory infrared plasmonic material of Ti1−xMgxN(001) layers is explored by optical transmission and reflection spectra in combination with ellipsometry and transport measurements. A red-shift in the reflection edge ℏωe (2.0 - 0.8 eV) and the corresponding unscreened plasma energy ℏωpu (7.6 - 4.7 eV) indicate a linear reduction in the free carrier density N with increasing x. However, nitrogen vacancies in Mg rich samples act as donors, resulting in a minimum N = 1.6  1022 cm-3 for x = 0.49. Photoelectron valence-band spectra confirm the diminishing conduction band density of states and indicate a decrease in the Fermi level by 0.9 eV as x increases from 0 to 0.49. The real 1 and imaginary 2 part of the dielectric function are characterized by negative values of 1 and large positive 2 at sub-2.5 eV spectral region due to intraband transition and positive values of both 1 and 2 that are associated with the interband absorption at higher energy. The screened plasma energy Eps that separates these two regions red-shifts between 2.6 - 1.33 eV for x = 0 - 0.39, indicating a tunable plasmonic activity that extends from visible to infrared range (470 - 930 nm). Electron transport exhibits a typical metallic resistivity-temperature dependence for TiN-rich alloys with x  0.26 while a weak carrier localization is observed at temperatures lower than 60 K and 300 K for x = 0.39 and 0.49, respectively, due to Mg-alloying induced disorder. The intrinsic quality factor QiSPP decreases from 2,100 to 30 for TiN and Ti0.51Mg0.49N, indicating an ability to sustain the surface plasmon polaritons that is an order of magnitude larger than for previously reported polycrystalline Ti1−xMgxN layers and is comparably high as for Ti1−xScxN(001).

For the purpose of achieving epitaxial Ti1−xMgxN(001) layers, reactive magnetron cosputtering from titanium and magnesium targets is conducted in 15 mTorr pure N2 at 600 °C onto MgO(001) substrate. X-ray diffraction (XRD) indicates a solid solution rocksalt phase for the composition range x = 0 - 0.55 and a decreasing crystalline quality with increasing Mg content, as quantified by the XRD ω rocking-curve width which increases from 0.25° to 0.80°. XRD φ-scans show that all Ti1−xMgxN layers with x ≤ 0.55 are single crystals with a cube-on-cube epitaxial relationship with the substrate: (001)L║(001)S and [100]L║[100]S. In contrast, a larger Mg concentration (x = 0.85) leads to a polycrystalline, phase-segregated, nitrogen-deficient microstructure.

Overall, the aforementioned synthesis, characterization and analysis provide original discoveries and new knowledge that contribute to a more complete understanding of semiconducting transition-metal nitrides and related alloys.

Similarly, epitaxial (Ti1−xMgx)0.25Al0.75N(0001)/Al2O3(0001) layers with 0.0  x  1.0 are used as a model system to explore how Fermi-level engineering facilitates structural stabilization of a host matrix despite the intentional introduction of local bonding instabilities that enhance the piezoelectric response in wurtzite structure quaternary nitride. The crystalline quality of the AlN matrix deteriorates for both Ti-rich (x < 0.2) and Mg-rich (x ≥ 0.9) alloys, which is attributed to the destabilizing octahedral bonding preference of Ti dopants and the preferred 0.67 nitrogen-to-Mg ratio for Mg dopants. Conversely, x = 0.5 leads to a stability peak with a minimum in the lattice constant ratio c/a that is attributed to a trend towards non-directional ionic bonding and leads to a maximum in the expected piezoelectric stress constant e33. The optical band gap increases from 5.15 to 5.24 to 5.93 eV for x = 0.0, 0.5 and 1.0, the refractive index at a wavelength λ = 400 nm decreases linearly according to n = 2.78-0.45x for x ≤ 0.79, and a sub-gap absorption at 3-4 eV for x < 0.5 indicates Ti-induced mid-gap states. The Fermi-level decreases from within the conduction band for x = 0 to the band gap at x = 0.5, but is stabilized near the valence band maximum by an increasing N vacancy concentration for x > 0.5.

Correspondingly, the maximum stability at x = 0.5 is attributed to filled bonding and empty antibonding states. The measured elastic constant along the hexagonal axis C33 = 270 ± 14 GPa is nearly composition independent, leading to a maximum in the expected piezoelectric constant d33 = 6.4 pC/N at x = 0.5 which is 50% larger than for the pure AlN matrix. Thus, contrary to the typical anti-correlation between stability and electromechanical coupling, the overall results show that the (Ti1−xMgx)0.25Al0.75N system exhibits simultaneous maxima in the structural stability and the piezoelectric response at x = 0.5.

Semiconducting transition-metal nitrides emerge as a new category of semiconductors that are stable in hostile environments and are compatible with conventional semiconductor device processing, which broadens their potential application in industrial and technology fields. These compounds are widely used as hard protective coating layers, diffusion barriers in microelectronics and photovoltaics, and optical or decorative surfaces, as they have prominent physical properties with respect to demanding requirements including high hardness, wear and corrosion resistance, and high temperature stability. In addition, enhanced mechanical, optoelectrical, and piezoelectrical performance can be achieved by alloying, to form rocksalt or wurtzite-derived ternaries and/or quaternaries. While these alloys share the same cubic or hexagonal symmetry with their matrixes, mechanically and/or electronically they can exhibit higher degree of tunability and stronger sensitivity to external variables. Specifically, the formation of TiN-based compounds, like Ti1-xMgxN and (Ti0.5Mg0.5)1−xAlxN, has the potential to extend the opportunities for manipulating the strains in the epitaxial films, engineering bandgap of corresponding ternaries/quaternaries and controlling the carrier density in rocksalt structure nitrides. In parallel, the co-addition of Ti and Mg into AlN matrix to form wurtzite structure nitrides like (Ti1−xMgx)0.25Al0.75N, softens and destabilizes the hexagonal AlN which causes a reduction in the elastic constant and a simultaneous increase in the internal strain sensitivity and therefore an overall enhancement of the piezoelectric coefficient.

In order to investigate strain relaxation in epitaxial Ti1−xMgxN(001)/MgO(001) system, structural information is extracted from Ti1−xMgxN(001) layers with different thickness d. Layers with thickness d = 35-58 nm are fully strained, with an in-plane lattice parameter a|| = 4.212  0.001 Å matching that of the MgO substrate, while the out-of-plane lattice parameter a increases with x from 4.251 Å for TiN(001) to 4.289 Å for Ti0.51Mg0.49N(001). This yields a relaxed lattice parameter for Ti1−xMgxN(001) of ao = (1-x)aTiN + xaMgN – bx(1-x), where aTiN = 4.239 Å, aMgN = 4.345 Å, and the bowing parameter b = 0.113 Å. In contrast, thicker Ti1−xMgxN(001) layers with d = 110-275 nm are partially (pure TiN) or fully (x = 0.37 and 0.49) relaxed, indicating a critical thickness for relaxation of 50-100 nm. The in-plane x-ray coherence length is large (100-400 nm) for fully strained layers with 0.00  x  0.45 but drops by an order of magnitude for x = 0.49 as the composition approaches the phase stability limit. It is also an order of magnitude smaller for thicker (d ≥ 110 nm) layers, which is attributed to strain relaxation through the nucleation and growth of misfit dislocations facilitated by glide of threading dislocations.