Wide-gap crystalline metal-fluoride genus for low-loss superconducting-qubit dielectric layers

Superconducting qubit coherence is throttled by two-level system (TLS) loss in the dielectric layers of shunt capacitors and junction-adjacent regions. The dominant materials in that role today — amorphous SiO2 and AlOx — are structurally disordered, and that disorder is the TLS bath. This patent position covers a computationally selected genus of wide-gap crystalline metal fluorides, M(a)M'(b)F(c), claimed not as bare compositions but as the low-loss dielectric layer of a superconducting qubit device. The selection logic is dual-property and simultaneous: an optical dielectric constant (eps_inf) between 1.7 and 2.15 — low for a crystalline solid — combined with a bandgap at or above 6.0 eV and a lowest JARVIS-DFT zone-center phonon frequency above -5.0 cm⁻¹, a zone-center check consistent with, but not confirming, full-zone dynamic stability. Applied to the JARVIS-DFT database, those filters resolve to exactly 41 members; for 31, a preliminary patent screen returned no hits (not an FTO opinion). The why-now is unambiguous and external. IBM's published 2026-2029 fault-tolerant roadmap names dielectric TLS loss as a specific engineering blocker, not a general concern. DARPA's Quantum Benchmarking Initiative downselect is expected in 2025-2026. A buyer who funds the remaining measured validation gates and takes a license position before this selection rule reaches conference disclosure could, if the claims are granted as drafted, assert the genus against qubit fabricators depositing a qualifying crystalline fluoride as the shunt-capacitor dielectric; scope and validity are set in prosecution.

The loss-reduction mechanism is physical, not incremental. Two independent effects compound. First, we hypothesise that a lower permittivity reduces the participation-weighted field energy in the dielectric at fixed geometry; eps_inf is a proxy for the uncomputed static value. Because TLS loss is thought to scale with loss-weighted field energy in the lossy layer, a material with eps_inf near 1.7-2.0 is hypothesised to store somewhat less such energy than amorphous SiO2 (eps_inf ~ 2.1) or AlOx (eps_inf ~ 2.5+) at identical device dimensions; loss tangent and static permittivity are not computed. Second, crystalline order removes the structural disorder that generates a dense TLS bath. Amorphous oxides present a disordered network with a nearly continuous distribution of two-level defects at millikelvin temperatures; a crystalline material with a confirmed stable phonon spectrum replaces that bath with a well-defined, sparse phonon landscape. The 6.0 eV bandgap floor ensures the layer stays insulating and defect-tolerant at operating temperatures far below any thermal excitation of electronic states. The genus spans five structure types — cryolite, rutile, colquiriite (space group 163), elpasolite, and scheelite — with the cation sites M and M' drawn from the set {Li, Na, K, Rb, Mg, Ca, Sr, Sc, Y, Al, Si, Ge, P}. Named members include Na3AlF6, MgF2, K2SiF6, LiSrAlF6, LiCaAlF6, LiYbAlF6, K3YF6, Na3ScF6, K2GeF6, Rb2SiF6, and LiYF4. Deposition by ALD, MBE, or PVD on Nb, Ta, Al, and TiN substrates — the standard qubit fabrication stack — is expected but has not yet been demonstrated for any genus member. All selection properties were extracted from the JARVIS-DFT database using OptB88vdW DFPT: the optical dielectric constant (eps_inf, averaged) and the lowest zone-center phonon frequency. Both are pre-computed equilibrium columns from a well-validated public DFT framework. The dual-property gate was applied as a compound filter across the full database, resolving the genus to its 41 confirmed members in a single query. Note that the multi-engine consensus phonon stability protocol — where MACE, CHGNet, and additional ML interatomic potentials must agree before a material advances — was not required here, because phonon stability data was drawn directly from JARVIS-DFT rather than from new ML-potential calculations. The stability verdict for this genus rests on the DFT phonon column rather than multi-engine ML consensus, which is an appropriate choice for database-grade materials that already have DFT phonon data, and which a buyer can independently reproduce as a read-only database query.

The addressable market is the superconducting qubit dielectric layer, within the broader quantum hardware and fault-tolerant quantum computing segment. We estimate that addressable market at $1-2 billion, consistent with the scale of capital flowing into superconducting qubit platforms and their supply chains. The dielectric layer is a small-mass, high-leverage component: it controls qubit coherence time (T1 and quality factor Qi) yet constitutes a negligible fraction of system bill of materials. That asymmetry supports royalty-on-value or per-node licensing logic rather than commodity-volume pricing. Named potential licensees span the full field of active superconducting qubit builders: IBM Quantum, Google Quantum AI, Rigetti Computing, AWS Center for Quantum Computing, Intel Quantum, PsiQuantum, and performers in the DARPA Quantum Benchmarking Initiative. The genus is broad enough that any builder who deposits a qualifying crystalline fluoride as the shunt-capacitor dielectric practices it, which means parallel non-exclusive licenses to multiple parties are structurally viable. Royalty attachment would most naturally be to qubit nodes or wafers shipped, not to the mass of dielectric material deposited. The near-term funding logic is straightforward: qubit builders have public roadmaps naming dielectric TLS as a performance bottleneck, which converts a materials patent position into something acquirers can tie to a specific engineering program. A buyer who moves during the 60-90 day option window can fund the remaining measured validation experiments, confirm the loss-tangent advantage in a real device, and hold a blocking genus position before the selection rule becomes public knowledge at a technical conference.

lower participation-ratio-weighted field energy via low eps_inf + crystalline dynamical-stability TLS suppression vs amorphous SiO2/AlOx of same geometry

use-limited qubit-dielectric device+method genus, not bare composition; novel dual-property selection rule; 31/41 members computationally-predicted patent_count=0; 7 COD members carved out by express proviso

Preliminary screen only. This is not a freedom-to-operate opinion.

The natural licensees are the named superconducting qubit platform builders, with IBM Quantum and Google Quantum AI carrying the strongest near-term urgency given IBM's published 2026-2029 fault-tolerant roadmap, where dielectric TLS loss is a named engineering blocker. DARPA Quantum Benchmarking Initiative performers face a 2025-2026 downselect, creating a parallel pressure point for government-program buyers. Rigetti, AWS Center for Quantum Computing, Intel Quantum, and PsiQuantum round out the field of active superconducting platform builders who would need a license to practice the genus in the qubit-dielectric use. Because the genus reads across all qubit builders who deposit a qualifying crystalline fluoride, non-exclusive field-of-use licenses to multiple parties in parallel are structurally viable and value-maximizing. A single strategic buyer on a fault-tolerant program who wants durable coherence advantage might instead pursue an exclusive license or outright acquisition to deny rivals access to the member set. The 60-90 day option structure is appropriate for an initial transaction: a buyer takes an option, funds the Na3AlF6 DFPT confirmation and an initial loss-tangent measurement, then converts to a full license or acquisition once device-level data exists.

The most significant risk is that all selection properties are database-computed rather than measured, meaning the core claim that these materials reduce TLS loss in real qubits is physics-supported but experimentally unconfirmed. Crystalline fluorides should outperform amorphous oxides on both eps_inf and structural disorder arguments, but the magnitude of T1 improvement in an actual device — at millikelvin temperatures, single-photon power, and with real interface states — is unknown until measured. A buyer acquiring this position before the validation experiments are complete takes a technical risk that the measured improvement is smaller than the physics argument projects. The validity risk is real and should not be minimized: the strongest genus members are historically studied UV-optics materials, and a patent examiner or challenger will focus on whether the use-limitation and dual-property selection rule are sufficient to distinguish from prior art on the crystal structures and their known optical properties. The selection rule is genuinely novel as applied to qubit dielectrics, but that argument must survive examination. There is also a cross-portfolio overlap to track: Na3AlF6 and K2SiF6 appear in sibling filings for other functional uses, and the independence of the qubit-dielectric use claim from those filings requires ongoing management as prosecution proceeds. The decisive de-risking path is sequential: DFPT confirm Na3AlF6 eps_inf, fabricate a coupon qubit with one or two genus members as the shunt-capacitor dielectric, and measure T1 and Qi against an amorphous-oxide control at identical geometry.