Abstract
State-of-the-art transmon qubits rely on large capacitors which systematically improves their coherence due to reduced surface loss participation. However, this approach increases both the footprint and the parasitic cross-coupling and is ultimately limited by radiation losses - a potential roadblock for scaling up quantum processors to millions of qubits. In this work we present transmon qubits with sizes as low as 3639$ \mu^2\gtrsim$100 nm wide vacuum gap capacitors that are micro-machined from commercial silicon-on-insulator wafers and shadow evaporated with aluminum. After the release in HF vapor we achieve a vacuum participation ratio up to 99.6\% in an in-plane design that is compatible with standard coplanar circuits. Qubit relaxation time measurements for small gaps with high vacuum electric fields of up to 22 V/m reveal a double exponential decay indicating comparably strong coupling to long-lived two-level-systems (TLS). The exceptionally high selectivity of 20 dB to the superconductor-vacuum surface allows to precisely back out the sub-single-photon dielectric loss tangent of aluminum oxide exposed to ambient conditions. In terms of future scaling potential we achieve a qubit quality factor by footprint area of , which is on par with the highest devices relying on larger geometries and expected to improve substantially for lower loss superconductors like NbTiN, TiN or Ta.