Epitaxial Growth of Perovskite Strontium Titanate on Germanium via Atomic Layer Deposition

Perovskite materials are becoming increasingly attractive due to their highly symmetric cubic or pseudocubic structure and myriad of properties. These materials, with general formula ABO₃, consist of A atoms coordinated with 12 oxygen atoms and B atoms coordinated with six oxygen atoms. Owing to their simple structure, yet wide range of potential elements, perovskite materials provide ideal candidates for heterostructure devices. Epitaxial oxide heterostructures boast ferromagnetic,^1–3 anti/ferroelectric,⁴ multiferroic,^5–8 superconductive,^7–12 and magnetoresistive functionalities.^13,14 Many of these desirable electronic properties are interfacial and thus dependent on clean, abrupt transitions between materials. The nearly identical structure and lattice constants shared between members of the perovskite family allow for excellent lattice matching and, therefore, high quality interfaces. Readily lattice-matched to each other as well as some semiconductors, perovskite oxides are now being turned to in next generation metal-oxide-semiconductor electronics.

Monolithic integration of crystalline oxides with silicon, first demonstrated with perovskite strontium titanate, SrTiO₃ (STO), by McKee and colleagues,¹⁵ was a monumental step towards the realization of electronic devices with perovskite-semiconductor incorporation. Molecular beam epitaxy (MBE) is the primary technique for epitaxial growth of oxides on silicon because of the layer-by-layer growth as well as the tunable oxygen partial pressure necessary to control amorphous, interfacial SiO₂ formation.^16–19 Typical MBE growth of STO on Si (001) is achieved by Sr-assisted deoxidation of SiO₂. Under the ultra-high vacuum (UHV) conditions, SrO is volatile and subject to thermal evaporation. Since SrO is thermodynamically preferred over strontium metal and SiO₂, deposition of Sr scavenges oxygen from the SiO₂ layer and the resulting SrO evaporates from the surface. During this process the silicon surface experiences a 2×1 reconstruction at the surface that forms rows of dimerized silicon atoms. Conveniently, ½ monolayer (ML) coverage of Sr atoms on the reconstructed surface fills in the gaps created by these dimer rows.²⁰ The ½ ML coverage provides a protective layer that, with careful control of oxygen pressure, can prevent or control interfacial SiO₂ formation during subsequent oxide growth.^21–23 In the case of STO (and perovskites with similar lattice match), the resulting lattice is rotated 45° in-plane such that (001)_STO‖(001)_Si and (100)_STO‖(110)_Si, allowing registry between the Si (3.84 Å Si-Si distance) and STO (a = 3.905 Å) with only slight compressive strain on the STO. This registry is necessary for high quality interfaces and the desired properties they possess.

Silicon became industrially significant due to the high quality of its interfacial oxide, but SiO₂ use is being phased out for materials capable of equivalent performance at smaller feature sizes. SiO₂ experiences high leakage currents when ultra-thin and this diminishes device performance. The demand for smaller feature sizes could be met by perovskite oxide films with high dielectric constants, k, that provide performance equivalent to SiO₂ and are physically thicker than SiO₂ by the factor k/3.9. Furthermore, alternative semiconductors, like germanium, offer potential for faster device operation due to higher electron and hole mobilities than silicon.^24,25 Germanium also has an interfacial oxide, GeO₂, but in contrast to SiO₂, it is unstable and subject to thermal deoxidation. Thus, 2×1 reconstruction is achievable by simple thermal annealing under UHV, and a protective Sr layer is unnecessary to prevent interfacial oxide growth during perovskite deposition.²⁶

Despite the apparent ease of growth offered by MBE, atomic layer deposition (ALD) provides a more scalable and cost effective method than MBE for the commercial production of oxide materials.^27,28 ALD employs doses of gaseous precursors to the substrate that are self-limiting in their reaction with the substrate surface. Therefore, in an ideal ALD process, up to one atomic layer is deposited for any given precursor dosing cycle and continued dosing of the same precursor will not deposit additional material onto the surface. Reactive functionality is restored with a co-reactant, often an oxidative or reductive precursor (e.g., water or ammonia). Previous work has demonstrated the ALD growth of various perovskite films, such as anatase TiO₂, SrTiO₃, BaTiO₃, and LaAlO₃, on Si (001) that had been buffered with four-unit-cell thick STO grown via MBE. ^29–34 In purely MBE growth of crystalline oxides, ½ monolayer coverage of Sr on clean Si (001) is enough to provide a barrier against SiO₂ formation under the pressures native to the technique (~10^-7 Torr). However, under typical ALD operating pressures of ~1 Torr, previous work has shown that four unit cells of STO is required to avoid oxidizing the Si surface.²⁹

The procedure detailed here utilizes the instability of GeO₂ and achieves monolithic integration of STO on germanium via ALD without the need of an MBE-grown buffer layer.²⁶ Furthermore, the Ge-Ge interatomic distance (3.992 Å) on its (100) surface allows for an analogous epitaxial registry with STO that is observed with Si (001). Though the procedure presented here is specific to STO on Ge, slight modifications may allow for the monolithic integration of a variety of perovskite films on germanium. Indeed, direct ALD growth of crystalline SrHfO₃ and BaTiO₃ films have been reported on Ge.^35,36 Additional possibilities include the potential gate oxide, SrZr_xTi_1-xO₃.³⁷ Finally, building on previous studies of ALD perovskite growth on a four-unit cell STO film on Si (001)^29–34 suggests that any film that could be grown on the STO/Si platform could be grown on an ALD-grown STO buffer film on Ge, such as LaAlO₃ and LaCoO₃.^32,38 The multitude of properties available to oxide heterostructures and remarkable similarity between perovskite oxides suggest this procedure could be utilized to study previously difficult or impossible growth combinations with such an industrially viable technique.

Figure 1 depicts the schematic of the vacuum system, which encompasses ALD, MBE, and analytical chambers connected by a 12-foot transfer line. The samples can be transferred in vacuo between each chamber. The baseline pressure of the transfer line is kept at approximately 1.0×10^-9 Torr by three ion pumps. The commercial angle-resolved ultraviolet and X-ray photoelectron spectroscopy (XPS) system is maintained with an ion pump such that the pressure in the analytical chamber is kept at approximately 1.0×10^-9 Torr.

The ALD reactor is a rectangular custom-built stainless steel chamber with a volume of 460 cm³ and length of 20 cm. A schematic of the ALD reactor is shown in Figure 2. The reactor is a hot wall, continuous cross-flow type reactor. Samples placed in the reactor have a clearance of 1.7 cm between the top surface of the substrate and the chamber ceiling and 1.9 cm between the bottom of the substrate and the chamber floor. A heating tape, powered by a dedicated power supply, is wrapped around the chamber from the inlet to approximately 2 cm beyond the exhaust port and provides temperature control of the reactor walls. A temperature controller adjusts the power input to the heating tape according to a temperature measurement taken by a thermal couple located between the heating tape and exterior reactor wall. The reactor is then completely wrapped with three additional heating tapes of constant power provided by a variac, and a final layer of fiberglass wool with aluminum foil covering provides insulation to promote uniform heating. The power output of the variac is adjusted such that the idling temperature (when the dedicated power supply is turned off) of the reactor is approximately 175 °C. The reactor is passively cooled via ambient air. The substrate temperature is calculated using the linear-fit equation (1), where T_s (°C) is the temperature of substrate and T_c (°C) is the temperature of the reactor wall, obtained by directly measuring a substrate fitted with a thermocouple. A temperature profile exists along the flow direction of the chamber due to the cold gate valve that connects the reactor to the transfer line; the temperature profile perpendicular to the flow direction is negligible. The temperature profile causes a richer Sr deposition at the leading edge of the sample, but the composition variation along sample is small (less than a 5% difference between the leading and trailing edges of the sample) according to XPS.³¹ The exhaust of the reactor is connected to a turbomolecular pump and a mechanical pump. During the ALD process, the reactor is pumped by the mechanical pump to maintain the pressure at around 1 Torr. Otherwise, the reactor pressure is maintained below 2.0×10^-6 Torr by the turbomolecular pump.

(1) T_s=0.977T_c + 3.4

The MBE chamber is maintained at a baseline pressure of approximately 2.0×10^-9 Torr or below by a cryogenic pump. The partial pressure of various species in the MBE chamber is monitored by a residual gas analyzer. The background pressure of H₂ is around 1.0×10^-9 Torr, while those of O₂, CO, N₂, CO₂, and H₂O, are less than 1.0×10^-10 Torr. In addition, the MBE chamber is also equipped with six effusion cells, a four-pocket electron beam evaporator, an atomic nitrogen plasma source and an atomic oxygen plasma source with high-precision piezoelectric leak valve, and a reflection high energy electron diffraction (RHEED) system for real-time in situ growth and crystallization observations. The sample manipulator allows the substrate be heated up to 1000 °C using an oxygen-resistant silicon carbide heater.