Database-driven method for selecting low-loss crystalline fluoride dielectrics for superconducting qubits

The superconducting qubit community has spent the better part of a decade discovering, often expensively and slowly, that the dielectric materials surrounding a qubit's tunnel junction and capacitor pads are the dominant source of microwave loss and decoherence. The standard approach to finding a better dielectric has been empirical: deposit candidate films, fabricate test devices, measure loss tangents, iterate. That cycle can take months per material and requires significant fab access. This patent covers a fundamentally different approach — a computational selection rule that queries pre-existing DFPT database entries for three simultaneously satisfied criteria (high-frequency dielectric constant in the range 1.7–2.15, optical bandgap at or above 6.0 eV, and a lowest phonon frequency no lower than −5 cm⁻¹) and identifies viable crystalline fluoride candidates without running a single new simulation. The method converts a multi-dimensional search problem into a reproducible database query, and every relevant result is verifiable from data already in hand. The timing is material. Large-scale quantum processors require hundreds to thousands of qubit-dielectric interfaces, and even modest reductions in loss tangent translate to substantially longer coherence times and lower error rates at scale. As the leading hardware teams (IBM, Google, AWS, DARPA-funded programs) push toward fault-tolerant operation, the pressure to move beyond silicon dioxide and silicon nitride — materials whose loss characteristics are well-understood but not particularly good — is intensifying. A defensible, patented selection methodology that collapses the candidate-search phase from months of fab work into a database query is commercially useful well before any specific composition reaches production. It licenses naturally as a method claim, and it positions the broader portfolio of specific fluoride compositions as outputs of a validated, IP-protected pipeline rather than lucky guesses.

The selection rule operates on DFPT (density functional perturbation theory) equilibrium-property databases covering crystalline fluoride compounds. Three properties must be simultaneously satisfied for a candidate to pass. First, the high-frequency dielectric constant (ε_inf) must fall between 1.7 and 2.15. This range is not arbitrary: a lower ε_inf correlates with weaker ionic polarizability and fewer low-energy phonon modes that can couple to microwave-frequency electromagnetic fields, which is the physical mechanism of two-level-system (TLS) loss. An upper bound is included because very low ε_inf materials often lack the structural cohesion needed for robust thin-film deposition. Second, the optical bandgap must be at or above 6.0 eV. Wide-gap fluorides suppress electronic contributions to loss and ensure that the material is a true insulator at millikelvin operating temperatures; it also substantially reduces the probability of quasiparticle-poisoning pathways. Third, the lowest phonon mode must be at or above −5 cm⁻¹ — a threshold that accommodates mild numerical noise in DFPT phonon calculations while filtering out genuinely dynamically unstable structures that would be unrealizable as deposited thin films. The gate defined by these three criteria was applied to a 41-member candidate set drawn from computed fluoride entries (designated in the computational record). Of those 41, the FTO split revealed 31 that are "computed-only" — meaning they have DFPT property data in the database but are not in the Crystallography Open Database with prior publicly deposited crystal structures, preserving maximum freedom-to-operate space. Seven additional members are excluded because their crystal structures appear in the open crystallographic literature (already-published), and all compositions containing beryllium, lead, thallium, cadmium, mercury, or hydrogen are explicitly excluded from the candidate set by negative limitation, on toxicological and electrochemical grounds. A ranking exhibit scores the surviving members by a composite figure of merit — the product of (1/ε_inf), the phonon margin above the −5 cm⁻¹ threshold, and the bandgap — providing an explicit, reproducible ordering that a practitioner can act on without subjective judgment. The key computational feature of this method is that it requires no new simulation at decision time. The dielectric constant and bandgap entries are sourced from DFPT calculations already present in materials databases (such as the Materials Project or JARVIS-DFT), and phonon stability is similarly available from prior high-throughput calculations. The patent's novel contribution is not the underlying DFPT calculations themselves but the simultaneous, multi-property functional specification applied specifically to the qubit-dielectric use case — a combination that the computational record indicates has not been previously codified as a selection rule for this application. The method sits within the broader metal-fluoride qubit dielectric materials portfolio, which includes specific compositions validated by full consensus-based multi-potential analysis (requiring agreement across at minimum two independent machine-learning interatomic potentials before advancing to DFT-level confirmation). For those specific compositions, MACE, CHGNet, MatterSim, and ORB potentials are used as independent validators, with consensus on the absence of imaginary phonon modes required. The method claim covered here is upstream of that validation layer: it identifies which compositions enter that pipeline in the first place. This hierarchical architecture — cheap database query first, expensive simulation only for survivors — is itself a defensible engineering advance in computational materials discovery workflow.

The addressable market is the superconducting quantum computing hardware supply chain, including fab-level dielectric material decisions made by IBM Quantum, Google Quantum AI, Amazon Web Services' Center for Quantum Computing, and the constellation of DARPA Quantum Benchmarking Initiative performers and university-affiliated hardware groups. Estimates of the superconducting quantum computing hardware and adjacent materials/services market range from $1 billion to $2 billion annually in the near term, growing as processor qubit counts scale and as foundry-style quantum chip manufacturing services come online. These are rough estimates reflecting a market that is expanding rapidly but still in its pre-fault-tolerance phase; actual licensing revenue will depend on which compositions eventually reach production and how broadly the method claim reads on competing selection methodologies. The natural licensing mechanism for a method-of-screening claim is a paid option or technology access agreement tied to the specific database query and ranking methodology, potentially structured as an upfront access fee plus milestone payments if a selected candidate enters device qualification. Unlike a composition patent that requires tracking a specific material through a supply chain, a method claim licenses at the level of the engineering decision — the selection step — making it well-suited to licensing arrangements with process engineering teams that are actively evaluating alternatives to their current dielectric stack. The "data-room-verifiable" character of the claim is a practical commercial advantage: a prospective licensee can confirm that the pre-computed property data supports the selection rule without running independent experiments, compressing due-diligence timelines. The method also monetizes the negative space: even compositions that ultimately underperform in device testing generate data that validates which sub-regions of the ε_inf / bandgap / phonon space are worth targeting, and that validated computational methodology retains value for successive rounds of material search.

zero-new-simulation, data-room-verifiable de-risking (both properties pre-computed); monetizes immediately via paid option

novel dual-computed-equilibrium-property selection rule applied to qubit-dielectric selection; FTO leg 2; no prior art recites it for dielectric-layer selection

The most natural licensees are the quantum hardware groups that are actively making dielectric stack decisions at scale: IBM Quantum (which has publicly disclosed that TLS loss from dielectrics is among its primary qubit coherence bottlenecks), Google Quantum AI (similarly focused on loss reduction as it pushes qubit count and gate fidelity), and Amazon Web Services' quantum computing group. A paid option structure — granting access to the method and the ranked candidate list in exchange for an upfront fee, with milestone payments tied to fab qualification of a selected member — fits the procurement model of a hardware team that wants to explore the computational shortlist without committing to a full joint-development agreement. DARPA Quantum Benchmarking Initiative performers, particularly those at university labs and smaller hardware startups that lack internal computational materials capacity, represent a second tier of buyers who would value the pre-screened candidate list precisely because they cannot replicate the underlying DFPT calculations themselves. A non-obvious strategic buyer is a quantum foundry or process-development organization that is building a standardized qubit fabrication process for multiple customers — the quantum equivalent of a semiconductor fab. Such an organization has strong incentive to lock in a well-validated dielectric material selection methodology early, because the choice propagates across every customer's device. Licensing this method at the foundry level would be more commercially efficient than licensing it to individual hardware teams, and would create a broader-reaching royalty stream tied to process adoption rather than to any single end product.

The primary technical risk is the validation gap between computed equilibrium properties and measured thin-film device performance. DFPT calculations of phonon stability and dielectric constants are ground-state, bulk-crystal results; actual qubit dielectrics are amorphous or nanocrystalline thin films deposited under non-equilibrium conditions, and their loss characteristics are dominated by surface and interface chemistry that bulk calculations do not address. If the first experimental measurements of selected candidates show loss tangents no better than current silicon-based dielectrics, the commercial case weakens substantially, though the method claim retains validity as long as the selection rule itself is novel and non-obvious. The de-risking roadmap is to prioritize one or two of the highest-ranked candidates from for deposition and resonator-level loss tangent measurement by a fabrication partner, with results closing the open proof gate. The primary IP risk is prior-art timing: a conference presentation or preprint describing a similar multi-property computational screening rule for qubit dielectrics — drawing on JARVIS-DFT or Materials Project data — before the filing date would create a significant validity challenge. Filing should therefore be prioritized relative to the publication calendar of computational quantum materials groups. A secondary risk is claim scope: method-of-use claims are sometimes difficult to detect infringement on when the method is applied internally by a hardware team without public disclosure. This is a standard enforcement challenge for process method claims and argues for pairing the method claim with at least one specific composition claim that covers the most commercially attractive output, creating multiple assertion points across the portfolio.