Atomically Thin ${\mathrm{Al}}_2{\mathrm{O}}_3$ Films for Tunnel Junctions

Metal-insulator-metal tunnel junctions are common throughout the microelectronics industry. The industry standard ${\mathrm{AlO}}_x$ tunnel barrier, formed through oxygen diffusion into an Al wetting layer, is plagued by internal defects and pinholes which prevent the realization of atomically thin barriers demanded for enhanced quantum coherence. In this work, we employ in situ scanning tunneling spectroscopy along with molecular-dynamics simulations to understand and control the growth of atomically thin ${\mathrm{Al}}_2{\mathrm{O}}_3$ tunnel barriers using atomic-layer deposition. We find that a carefully tuned initial ${\mathrm{H}}_2\mathrm{O}$ pulse hydroxylated the Al surface and enabled the creation of an atomically thin ${\mathrm{Al}}_2{\mathrm{O}}_3$ tunnel barrier with a high-quality $M\text{−}I$ interface and a significantly enhanced barrier height compared to thermal ${\mathrm{AlO}}_x$. These properties, corroborated by fabricated Josephson junctions, show that atomic-layer deposition ${\mathrm{Al}}_2{\mathrm{O}}_3$ is a dense, leak-free tunnel barrier with a low defect density which can be a key component for the next generation of metal-insulator-metal tunnel junctions.

Figure 1

Illustration which shows the structural differences between the (a) thermal ${\mathrm{AlO}}_x$ tunnel barrier, formed through oxygen diffusion into the Al wetting layer, and (b) the ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ tunnel barrier, formed through layer-by-layer atomic-layer deposition of ${\mathrm{Al}}_2{\mathrm{O}}_3$.

Figure 2

AIMD simulation and STS study of the ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ growth on an Al wetting layer from the pre-ALD sample heating to the first ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ cycle ($0.12\mathrm{nm}/\mathrm{cycle}$). (a) Exemplary STS $dI/dV$ spectra are plotted for an Al sample after (I) 75 min of heating in the ALD chamber, (II) after 15 min of heating, and (III) after one ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ cycle. The arrows (blue) depict the tunnel-barrier height, calculated as the intersection of the fit lines (red). Diagrams (top) illustrate the expected surface as seen by the STM tip. The inset in (II) is the $dI/dV$ spectrum of a sample that is directly transferred to the STM chamber after Al sputtering and the inset in (III) is the corresponding $I\text{−}V$ curve for the one-cycle ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ $dI/dV$ spectra. (b) AIMD simulations are shown for ${\mathrm{H}}_2\mathrm{O}$ adsorption onto an Al (111) surface. When only one ${\mathrm{H}}_2\mathrm{O}$ molecule is present on the Al surface, dissociation is thermodynamically unfavorable (I, II). However, when ${\mathrm{H}}_2\mathrm{O}$ molecules are in close proximity, dissociation into ${\mathrm{OH}}^{-}$ and ${\mathrm{H}}^{+}$ is nearly instantaneous (III, IV). (c) The percentage of the Al surface which had a barrier height consistent with ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ after one ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ cycle with a variable initial ${\mathrm{H}}_2\mathrm{O}$ pulse duration.

Figure 3

A comparative STS study of ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ vs thermal ${\mathrm{AlO}}_x$ tunnel barriers. (a) Exemplary constant height $dI/dV$ spectra are taken on a 1.3-nm thermal ${\mathrm{AlO}}_x$ tunnel barrier (top) and a ten-cycle (1.2 nm) ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ tunnel barrier (bottom) with 15-min heating. The arrows (blue) depict the tunnel-barrier height calculated as the intersection of the fit lines (red). (b) The average tunnel-barrier height (dashed lines) for thermal ${\mathrm{AlO}}_x$ (red) and the ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ [15 min (blue) and 75 min (black) heating] tunnel barriers plotted as function of tunnel-barrier thickness, respectively.

Figure 4

$\mathrm{Nb}\text{−}\mathrm{Al}/{\mathrm{Al}}_2{\mathrm{O}}_3/\mathrm{Nb}$ Josephson junctions with an ALD ${\mathrm{Al}}_2{\mathrm{O}}_3$ tunnel barrier are measured. (a) The $I\text{−}V$ characteristics of a five-ALD cycle $10\mu \mathrm{m}\times 10\mu \mathrm{m}$ Josephson junction at $T=4.2\mathrm{K}$ are shown which display a very low leakage current. The bias current waveform is triangular at 5 Hz and is ramped up linearly from zero to 0.6 mA, then from 0.6 mA to $-0.6\mathrm{mA}$, and finally from $-0.6\mathrm{mA}$ to zero. (b) The critical current density $J_c$ as a function of the ALD cycle, or equivalently thickness, which follows the expected exponential dependence (solid line). The inset shows a chip with 12 JJs with areas ranging from $5\mu \mathrm{m}\times 5\mu \mathrm{m}$ to $10\mu \mathrm{m}\times 10\mu \mathrm{m}$. (c) The magnetic field dependence of the average switching current is shown for a similar five-cycle JJ processed from the same batch. The magnetic field and switching current have been normalized to the field at the first minimum (12 Oe) and the switching current at the central maximum ($76\mu \mathrm{A}$). (d) The measured switching current distributions of a ten-cycle junction at $T=0.76\mathrm{K}$ and 1.17 K. The lines are calculated switching current distributions based on thermal activation theory.