Silicon and sapphire substrates for superconducting qubits

Short answer: Silicon and sapphire are the two common substrates for superconducting qubits, and the choice affects how much energy the substrate absorbs. A 2025 arXiv preprint from a Princeton group reports tantalum transmons on high-resistivity silicon with lifetimes up to 1.68 milliseconds, while a 2017 preprint from Caltech reported aluminum transmons on silicon with lifetimes of tens of microseconds. The difference reflects materials work in the intervening years as much as the substrate.

The 2025 tantalum-on-silicon result

The preprint (arXiv 2503.14798, "2D transmons with lifetimes and coherence times exceeding 1 millisecond") reports that replacing the substrate with high-resistivity silicon dramatically decreases bulk substrate loss. The authors report two-dimensional transmons with time-averaged quality factors above 1.5 times ten to the seventh, reaching a maximum of 2.5 times ten to the seventh. This corresponds to an energy relaxation time T1 of up to 1.68 ms.

The low loss lets them observe decoherence effects related to the Josephson junction. With improved, low-contamination junction deposition, they report Hahn echo coherence times T2E exceeding T1. They say these improvements come without changes to the qubit architecture, so standard control gates can be used, and they demonstrate single-qubit gates with 99.994% fidelity. They also say the tantalum-on-silicon stack is simple and could potentially be fabricated at wafer scale and translated to large-scale processors. That is their projection, and it is not a demonstrated result.

The paper's body text adds that surface and bulk dielectrics contribute comparably to two-level-system loss in these devices. That point comes from reading beyond the abstract and is the authors' own analysis.

The 2017 aluminum-on-silicon result

Keller and coauthors (arXiv 1703.10195, 2017) describe transmons with aluminum Josephson junctions on high-resistivity silicon and on silicon-on-insulator. Their process uses an anhydrous hydrofluoric vapor step that removes silicon surface oxides without attacking aluminum. For qubits at about 5 GHz they report lifetimes and coherence times comparable to those attainable on sapphire, with T1 of 27 microseconds and T2 of 6.6 microseconds on silicon, and T1 of 3.5 microseconds and T2 of 2.2 microseconds on silicon-on-insulator. They say the process in principle allows co-fabrication of silicon photonic and mechanical elements for chip-scale transducers. This is their stated aim.

Tantalum resonators on both substrates

A conference abstract from the University of Glasgow describes growth of tantalum on heated sapphire and on silicon using a niobium seed layer, along with dry etch recipes and the internal quality factors extracted from resonators. The authors say tantalum resonators on silicon at room temperature could allow highly coherent circuits on systems without heating capabilities. The abstract is a presentation summary, and it gives no qubit results.

Limits

The three sources differ in material, year and measurement, so the numbers should not be lined up as a controlled comparison. The 2025 result is an arXiv preprint and has not been confirmed by this site as peer reviewed.

FAQ

Does silicon beat sapphire for qubits?

The 2025 preprint reports very low bulk loss on silicon with tantalum films, and the 2017 preprint reports silicon comparable to sapphire.

What lifetime was reported?

T1 up to 1.68 ms, per the 2025 preprint.

What limits these devices now?

The 2025 authors point to two-level-system loss from surfaces and bulk dielectrics.

Is tantalum required?

The 2025 result uses tantalum. The 2017 result uses aluminum.

Is it ready for large processors?

The authors say it could be fabricated at wafer scale, which is a projection.