Aluminum metallization of III–V semiconductors for the study of proximity superconductivity

Interface abruptness dramatically affects the electronic and optical properties of materials and devices. High quality, reproducible, and well-characterized metal/semiconductor interfaces are critical for the functionality and understanding of the interaction of electrons in a semiconductor with external circuitry since poor crystalline properties at interfaces impede low-resistance charge flow.¹ For example, the formation and height of the Schottky barriers depend on the atomic structure of the metal-semiconductor interface, which was examined in the early days of the molecular beam epitaxy (MBE) development.² Recently, the field has attained renewed interest with the realization that proximity superconductivity effects, using the commonly available element Al, can be studied this way.

However, the close proximity of the superconductor may actually affect the neighboring semiconductor in a negative manner,³ compelling the use of very thin insertion layers. They may be used to control electrostatic potentials and local carrier densities and, when designed properly, to have the additional very desirous effect of preventing or minimizing the interaction of the Al metal layer with the underlying semiconductor,^4,5 which is the focus of this work.

We ultimately want to study optimized complex mixed group V alloys such as InAsSb, but in order to simplify the understanding of the role of the group V species, we start by examining binary materials. In this study, we conduct extensive structural characterization studies with transmission electron microscopy (TEM) and adjust growth parameters to improve the overall crystalline quality of the superconductor metal and the underlying semiconductor, with specific attention paid to the interface.

Table I lists the samples included in this study.

We grew the samples in a solid source Gen II MBE system with liquid nitrogen (LN₂) cooling and an ion pump and cryopump. The host structure consists of a quarter 2-in. (001) InAs (sample A) or GaSb (sample B–K) substrate, a buffer layer of the same material, which is also the semiconductor active layer of interest, capped with a very thin semiconductor insertion layer (AlAs for sample A or AlSb for samples B–K), all grown with standard MBE procedures. We desorbed the oxide on the GaSb substrate in situ. The substrate temperature was increased under Sb overpressure (once the pyrometer temperature was 460 °C). Reflection high-energy electron diffraction (RHEED) indicated that the oxide desorbed ∼580 °C. After holding the substrate for 10 min at 600 °C, it is brought to 490 °C for the buffer layer/insertion layer growth. For sample A, the oxide was desorbed by H-cleaning in the buffer chamber, and then the wafer was transferred into the growth chamber. The V/III beam equivalent pressure (BEP) ratio exceeded 2.5 for GaSb and AlSb and 5 for InAs and AlAs. The growth rate for all layers in all samples, with the exception of the Al layer, was 1 μm/h.

After growth, the sources were cooled to idle temperatures, and we held the sample block at a constant power level, allowing it to cool from the growth temperature to room temperature. For sample A, the As shutter was closed immediately after the growth of the AlAs layer. For samples B–K, the Sb shutter was closed when the substrate fell to 400 °C (the As source was not warmed up before or during the growths of samples B–K).

The residual gases (mainly As) pumped down for a period exceeding 12 h (usually just overnight, but occasionally over a weekend). The background pressure typically reached the 10⁻¹⁰ Torr range in 3–6 h and the ∼10⁻¹¹ range within 10 h.

Before the Al layer deposition, the sample was pointed toward the cryoshroud for 160 min to cool the wafer as much as possible. As discussed previously, although we cannot measure the actual sample temperature at the time of Al deposition, we know it is below 0 °C after ∼96 min.⁴ The Al cell was set to a relatively low temperature (∼950 °C), which corresponds to an Al growth rate (on GaSb) of ∼ 0.09 Å/s to minimize the radiation of heat to the sample during deposition.

We did not do RHEED during the cooling and metal deposition sequence to avoid possible artifacts from the dissipated beam energy.

We examined the crystallography of the interfaces and the local surface morphology with TEM in a JEOL ARM200F equipped with a spherical aberration corrector for probe mode operated at 200 keV. The samples were prepared with cross-sectional tripod polishing to ∼20 μm thickness, and shallow angle argon ion milling with low beam energies (≤3 keV) and LN₂ stage cooling in a PIPS II ion mill. The temperature of the sample in the ion mill with the guns on is in the range of −120 to −95 °C. Care was also taken not to expose the samples to hot plates during epoxy curing, etc. Under these conditions, we do not expect that there is sufficient heating during the sample preparation process to affect the morphology of the structures examined in this paper.

InAs is known to interface relatively well with metals and superconductors, forming ohmic contacts rather than a Schottky barrier.⁶ Two-dimensional electron systems confined to surface InAs layers are of interest because of their potential for realizing topological states of matter.^7,8 As we discussed in a prior report⁴, depositing Al directly onto InAs surfaces free of excess As tends to induce a strong interaction, causing surface pitting and spherical Al balls to form rather than a flat film. An unintended interaction layer may result from Al deposition onto InAs in the presence of excess As, particularly if the growth chamber is not LN₂ cooled and has a high As background. High-resolution imaging with energy dispersive x-ray spectroscopy (EDS) data on later samples indicates that the interaction layer consists of Al, In, and As.

Figure 1 shows an aberration corrected scanning transmission electron microscope (STEM) bright field (BF) image of sample A taken along a $[1¯10]$ substrate zone axis. We had chosen the 3.3 Å AlAs thickness because our prior studies⁴ indicated that this could be thick enough to mitigate the Al-InAs interaction. As shown in Fig. 1 and in a corresponding background filtered⁹ high-resolution (HRTEM) image of the Al film (Fig. 2), the Al layers are locally of high crystalline quality through the thickness, undistorted, and atomically flat within the bulk. They are oriented along the [110] growth direction. The Al surface is still somewhat undulated and HRTEM images collected at lower magnifications (Fig. 3) show the presence of Moiré fringes, indicating that there are large overlapping crystal grains across the width of the Al film.

Figure 4(a) shows EDS maps with the intensities representing non-normalized atomic percentages of Al, In, and As taken from a region at and near the interface. The normalized line profile [Fig. 4(b)] is an average over the width of the STEM image of the composition. The red dashed line is a guide to the eye and indicates the end of the As-containing layer. We can see that Al signal becomes much stronger once As terminates and that a band consisting of Al, In, and As exists below the Al layer. There is no way to definitely know whether Al is migrating into InAs, or if residual As is incorporating into the growing Al film, but the As band in the EDS corresponds to the intended thickness of the InAs layer. This suggests that residual As is incorporated within a few monolayers, but Al interacts into the InAs layer. Because Al is not as flat as hoped, and Al still seems to interact with the InAs layer, we decided to attempt Sb-based insertion layers.

Although the Al interaction with InAs improved with intentionally deposited As-based insertion layers, our final goal was to form abrupt semiconductor-superconductor interfaces onto InAs_(1−x)Sb_x. Therefore, we moved to Sb-based insertion layers. Ideally, we would have worked with InSb substrates and buffer layers. The InSb surface quality is highly dependent on nontrivial substrate preparation to desorb the oxide, and even with that, they are not as high quality as the starting GaSb surface. Therefore, as a baseline, we deposited Al directly onto a 1000 Å GaSb buffer layer/GaSb substrate with no intentional insertion layer (sample B).

Figure 5(a) shows the interface between GaSb and Al for sample B. Al has large grains twinned relative to one another and Moiré fringes due to overlapping, differently oriented grains. The Al film is continuous [Fig. 5(b)] but has an undulated surface, although considerably flatter than those previously seen in Al films directly deposited onto As-containing layers such as InAs and GaAs.⁴ Figure 5(c) shows the EDS maps for Al, Ga, and Sb along with the corresponding STEM BF image of the interface. Al and Ga do not intermix like Al and In do when Al is deposited onto InAs. The white guidelines in the STEM image [Fig. 5(c)] are highlighting a crystallographic feature that is similar to that seen near the interface in Fig. 5(a). It consists only of Al, and the contrast is due to diffraction effects, not chemical composition variation.

We grew a series of nine films consisting of Al deposited onto AlSb insertion layers with thickness ranging from 0.8 to 200 Å onto (001) GaSb, as listed in Table I. The growth temperature of the GaSb buffer and AlSb insertion layer was 490° C. Figure 6 is a schematic of the structures.

We find that even very thin, submonolayer thicknesses of AlSb make a substantial difference in the Al morphology. Figure 7 shows an HRTEM image of the sample grown with the thinnest AlSb insertion layer (0.8 Å). The surface roughness (not shown here) is about one to two monolayers. The Al film is single crystalline with no observed Moiré fringes and is oriented along the (111) growth direction.

We did not observe any substantial change in the Al crystallography or interface quality for AlSb film thicknesses of 1.6, 3.2, 6.3, and 12.5 Å. Figure 8 shows an HRTEM image of the sample grown with a 25 Å insertion layer. The AlSb/Al interface is atomically flat, and the Al film is single crystalline with the (111) growth direction. As shown in Fig. 8(b), the surface is not undulated and has only about one monolayer of roughness (average roughness was 2.5 Å in TEM).

Further increases in the AlSb insertion layer thickness did not improve the quality of the Al surface layer. Samples with 50, 100, and 200 Å AlSb layers had Al regions of different crystallographic orientations beginning at the AlSb interface. For example, Figs. 9(a) and 9(b) show HRTEM of the sample grown with a 50 Å insertion layer. The Al film has various growth orientations, twinned regions, and Moiré fringes. The AlSb/Al interface is not as flat as it was for the thinner AlSb layers. A lower magnification HRTEM image [Fig. 9(c)] shows that the Al film’s surface is also not flat. EDS results (not shown here) are similar to those shown in Fig. 5(c) for the thinner samples. It may be that the thicker AlSb layer is strained, causing a growth front that results in the Al layer being less flat and forming a polycrystalline microstructure. AlSb has a Matthews–Blakeslee critical thickness of approximately 200 Å on GaSb. Despite this, it remains highly strained for the film thicknesses above 1000 Å and even retains significant strain (∼18%) for the thicknesses of 2 μm.¹⁰ Our insertion layers are all significantly thinner than where we would expect to observe threading dislocations. The strain causes perturbations at the Al/AlSb interface, leading to the crystallographic features seen in our Al films grown on insertion layers that are at or greater than 50 Å thickness. The Al/AlSb interface looks nearly identical in TEM samples for I–K (AlSb thicknesses of 50, 100, and 200 Å). The TEM samples of samples J and K did not have the surfaces intact due to the ion milling, so we cannot comment on if the surface became increasingly rough beyond what is seen in Fig. 9(c).

We find that thin (≤∼25 Å) AlSb insertion layers promote atomically flat superconductor/semiconductor interfaces with smooth surfaces and little chemical intermixing. This and our prior work shows that Sb insertion layers are preferable to As insertion layers in terms of abruptness. Al and Ga seem to have less intermixing than In and Al or In and Ga.

The sample in our previous report showing proximity superconductivity in InAs_(1−x)Sb_x was grown prior to this systematic study of antimonide insertion layers. That particular sample had a two-monolayer GaAs insertion layer. Although the interface abruptness was significantly better than seen for Al grown directly onto InAs, it was not as sharp as seen for our AlSb insertion layers or even when Al was deposited directly onto GaSb. The Al surface was also not as flat. We can only speculate about the reasons at this point without more thorough study. GaAs does have a more substantial lattice mismatch with InAs_(1−x)Sb_x than AlSb, although it is not clear how much this matters for only a few monolayers. If we end the growth with a III-Sb layer, we then cool the wafer under Sb overpressure, which may lead to a more stable surface for the next day’s Al deposition compared to an As-terminated surface. Despite the nonideal interface, we were able to fabricate Josephson junctions in our InAs_(1−x)Sb_x structures. In a future study, we will grow lattice matched InAsSb onto metamorphic buffers and deposit a very thin AlSb insertion layer and then metalize with Al. The flatter Al surface and sharp InAs_(1−x)Sb_x enabled with the AlSb insertion layer will likely produce higher quality devices than previously demonstrated.

The work at NYU was supported by the Army Research Office Agreement No. W911NF1810067.