Heteroepitaxy of \(La_2O_3\) and \(La_{2-x}Y_xO_3\) on GaAs (111)A by Atomic Layer Deposition: Achieving Low Interface Trap Density

: GaAs metal-oxide-semiconductor devices historically suffer from Fermi-level pinning, which is mainly due to the high trap density of states at the oxide/GaAs interface. In this work, we present a new way of passivating the interface trap states by growing an epitaxial layer of high- k dielectric oxide, La 2-x Y x O 3 , on GaAs(111)A. High-quality epitaxial La 2-x Y x O 3 thin films are achieved by an ex-situ atomic layer deposition (ALD) process, and GaAs MOS capacitors made from this epitaxial structure show very good interface quality with small frequency dispersion and low interface trap densities ( D it ). In particular, the La 2 O 3 /GaAs interface, which has a lattice mismatch of only 0.04%, shows very low D it in the GaAs bandgap, below 3×10 11 cm -2 eV -1 near the conduction band edge. The La 2 O 3 /GaAs capacitors also show the lowest frequency dispersion of any dielectric on GaAs. This is the first achievement of such low trap densities for oxides on GaAs.

High-mobility InGaAs metal-oxide-semiconductor field-effect transistors (MOSFETs) have shown promising performance compared to Si-based devices for high-speed complementary MOS (CMOS) logic applications. However, GaAs MOS devices suffer from Fermi-level pinning, which is mainly due to the high trap density of states at the oxide/GaAs interface. 1,2 In this work, we present a new way of passivating the interface trap states by growing an epitaxial layer of high-k dielectric oxide, La 2-x Y x O 3 , on GaAs(111)A, which effectively reduces the trap density and minimizes the frequency dispersion of capacitance.
For GaAs MOS structures, usually the dielectric oxide is either amorphous or polycrystalline, and therefore a high density of dangling bonds exists at the oxide/GaAs interface. These dangling bonds form interface states in the midgap, 3 which trap carriers and produce a large frequency dispersion of capacitance and Fermi-level pinning. GaAs MOS devices with epitaxial dielectric layers should have a low interface trap density of states (D it ), since a perfect epitaxial interface is supposed to have no dangling bonds. Also, contrary to polycrystalline oxides, a perfect epitaxial oxide should contain no grain boundaries, 4 which preserves the desired features of the low leakage current and uniformity. However, growing epitaxial oxides on GaAs is rather challenging, since GaAs is neither chemically stable nor thermally stable: GaAs can be oxidized easily to form low quality surface oxides that compromise the interface quality; 5 and GaAs starts to lose As over 400 ˚C. 6 Hong et al. 7,8 have demonstrated a method of using in-situ electron beam evaporation to grow epitaxial (Ga,Gd) 2  directions of GaAs, respectively). Getting MOSFETs even of this quality also requires that there is no air-break between growth of the GaAs and the oxide, so that complex multichamber MBE systems are necessary. Several follow-up structural analyses of Gd 2 O 3 /GaAs(100) [11][12][13] revealed that perfect strained epitaxy only occurs in the first few layers.
When the oxide film thickness exceeds ~3 nm, the Gd 2 O 3 film starts to relax by generating misfit dislocations, so that the film is no longer perfectly epitaxial. 8 Unfortunately, simply substituing Gd 2 O 3 with other lanthanide sesquioxides cannot accommodate the mismatch simultaneously in two orthogonal in-plane directions, since the in-plane lattice spacing of Gd 2 O 3 is greater than GaAs in one direction but smaller in the other. Very recently, epitaxial growth of cubic high-k oxide, LaLuO 3 , on GaAs(111)A has been achieved by an ex-situ atomic layer deposition (ALD) process in our group. 14 The heteroepitaxy relationship was found to be (111) LaLuO3 //(111) GaAs (a LaLuO3 ≈ 2a GaAs ) with a relaxed interface. 14 Since the (111) plane has a three-fold symmetry, the in-plane mismatch between oxide and GaAs can be simultaneously engineered with lanthanide sesquioxides that have appropriate cation sizes.
Initial electrical characterizations showed quite promising results, as the MOS capacitors made from epitaxial ALD LaLuO 3 /GaAs showed an order of magnitude reduction in interface trap density (D it ~7×10 11 cm -2 eV -1 ) compared with amorphous ALD Al 2 O 3 /GaAs (~8×10 12 cm -2 eV -1 ). 14 But, still LaLuO 3 has a fairly large lattice mismatch with respect to GaAs (-3.8%), and another concern is that Lu is one of the rarest elements on earth, which would be problematic for large scale fabrication.
In this work, we report an ALD process for depositing another high-k oxide, La 2-x Y x O 3 , epitaxially on GaAs(111)A. The k value of La 2-x Y x O 3 was reported as high as 27, 15  is ~6% smaller than that of La 2 O 3 . Therefore, we can adjust the lattice constant of the ternary compound, La 2-x Y x O 3 , to study the effect of mismatch by varying the ratio of La and Y. As an ALD process grows films in a layer-by-layer manner, the compositional ratio of these two cations can be tuned by varying the ratio of La 2 O 3 and Y 2 O 3 cycles. A high-resolution X-ray structural analysis indicates a high-quality heteroepitaxy of Electrical measurements on MOS capacitors show a promisingly small frequency dispersion of capacitance and a low D it ~2×10 11 cm -2 eV -1 in the GaAs bandgap close to the conduction band edge. In addition, our process tolerates an air-break between growth of the GaAs and ALD of the epitaxial oxide. ALD is known to produce uniform films over large areas with good reproducibility, 16 so we believe that this process is very promising for large scale manufacturing. However, TEM does not have enough resolution to determine precisely the small difference between their lattice constants. Therefore, high-resolution X-ray diffraction (HRXRD) was used to investigate the detailed epitaxial structures.
Coupled 2-ω HRXRD scans were performed for the oxide/GaAs(111)A samples. The peaks from the GaAs substrate were used as the internal references, and the oxide/GaAs lattice mismatch, which is defined as (a oxide -2a GaAs )/2a GaAs , was calculated from the relative shift of the oxide peak with respect to the GaAs peak, assuming a fully relaxed heteroepitaxy relation at the interface. 14 For the La 1.1 Y 0.9 O 3 /GaAs sample, the coupled 2-ω scan clearly shows both peaks of the GaAs(111) and La 1.1 Y 0.9 O 3 (222) reflections, as shown in Figure 2a.
The corresponding ω rocking curves of GaAs(111) and La 1.1 Y 0.9 O 3 (222) reflections have a similar shape with the same full width at half maximum of ~32" (Supporting Information Figure S3). This indicates a high quality heteroepitaxy of La 1.1 Y 0.9 O 3 /GaAs over a large area (several mm 2 ). The 2 angle of the La 1.1 Y 0.9 O 3 (222) peak was found to be 0.958˚ greater than that of the GaAs(111) peak, which corresponds to a lattice mismatch of -3.32% for  Figure S4). Therefore, we performed another 2-ω scan around the GaAs(333) reflection to determine the mismatch with greater sensitivity. As shown in Figure 2c, the 2 angle of the La 2 O 3 (666) peak was only ~0.046˚ smaller than that of the GaAs(333) peak, suggesting a much smaller lattice mismatch of only +0.04% for La 2 O 3 with respect to GaAs.
In summary of the above structural analysis, both of the TEM and HRXRD results suggested a high-quality heteroepitaxy relation of La 2-x Y x O 3 /GaAs(111)A (x = 0, 0.2 and 0.9) with smaller lattice mismatch for higher La-content oxide. The measured lattice mismatch approximately follows Vegard's law (Supporting Information Figure S6). In addition, we also found that pure Y 2 O 3 on GaAs(111)A is also epitaxial (Supporting Information Figure S5). 15.6%, and 9.9%, respectively. We also measured the D it by the conductance-voltage method (Supporting Information Figure S7). 17 The distribution of D it within the GaAs band gap is plotted in Figure 4. Consistent with the C-V results, the interface of the amorphous This material is available free of charge via the Internet at http://pubs.acs.org

Author Information
Corresponding Author *E-mail: gordon@chemistry.harvard.edu

Notes
The authors declare no competing financial interest.