Multi-engine phonon-consensus screening method for rare-earth ternary silicide superconductor candidates

The rare-earth silicide superconductor candidates portfolio's most defensible asset is not a composition — it is a method. Rare-earth ternary silicides in the approximate 1:1:1 stoichiometry class are well-documented in the heavy-fermion literature going back to the 1980s, which forecloses composition-of-matter claims on the known genus. The inventive contribution here is the verification and selection engine itself: a four-engine phonon-consensus screen, paired with a fail-closed freedom-to-operate prescreen over roughly 306,000 patent claims, that converts a broad, partly-characterized genus of approximately 1,788 candidates into a curated, evidence-anchored shortlist with traceable IP clearance. No single prior actor has assembled that combination at genus scale. The timing is material. Superconducting-device and quantum-hardware programs are scaling rapidly, and their bottleneck is shifting from discovery to credible down-selection. A buyer who holds this triage layer — the engine that decides which candidates advance to synthesis — would hold drafted method claims on this screening workflow, if granted, upstream of the device families it selects for. Because the method is composition-agnostic, it can be redeployed across other intermetallic and superconductor screening campaigns, extending its value well beyond the silicide class. The claim strategy explicitly avoids asserting any composition-of-matter right to literature-known members, which is precisely the design choice that keeps this asset alive where a composition claim would not survive.

Candidates are screened for imaginary phonon modes with the machine-learning interatomic potentials named above. This is a computed harmonic check, not a measurement; where engines disagree, that is shown.

The method targets rare-earth transition-metal silicon compositions near 1:1:1 stoichiometry, drawn from the ThCr2Si2-derived, PbFCl-CeFeSi, and related layered intermetallic structure families. Dynamic stability — the question of whether a crystal structure will hold together against vibrational perturbations — is assessed by four independent machine-learning interatomic potential engines: MACE, CHGNet, MatterSim, and ORB. Each engine executes a finite-displacement phonon calculation independently, and a candidate is classified as consensus-stable only when at least three of the four engines agree the structure carries no imaginary phonon modes. Dissenting votes are recorded rather than suppressed. Across 99 phonon runs conducted across the candidate set, 77 cleared the 3-of-4 majority threshold, a consensus-stable rate of 78%. The multi-engine design is a deliberate response to a known limitation: machine-learning potential reliability is reduced for f-electron rare-earth intermetallics, where relativistic and correlation effects are difficult to capture in general-purpose potentials. Requiring majority agreement across four architecturally distinct potentials is intended to reduce the false-stable rate relative to any single-engine call, though no DFPT benchmark is reported and the potentials share much training data, and the recorded-dissent protocol gives the verdict a rough indication of disagreement. Each candidate is also anchored to a density functional theory convex hull reference, providing a thermodynamic stability check alongside the dynamic one. Proxy superconducting transition temperature is computed and used strictly as a screening signal to rank and filter candidates within the stable subset. It operates in a sub-30% accuracy regime and is treated as a directional indicator only — it is never a selection criterion in the formal sense and is not claimed as a property. This design choice is intentional: it prevents an examiner or litigant from challenging the method on the basis that a predicted Tc was wrong, while still allowing the screening pipeline to prioritize the most promising members for downstream characterization. The output of the full pipeline is the intersection of consensus-stable and freedom-to-operate-prescreened candidates — a curated shortlist that is computationally screened and has passed a preliminary patent prescreen (not a freedom-to-operate opinion).

The addressable market spans superconducting electronics, quantum computing hardware, cryogenic sensing, and materials informatics platforms. We estimate the total addressable range at $1 to $5 billion, with the capturable slice being the selection and IP-diligence layer that gates device programs, not the device hardware itself. That distinction matters: a method license prices against the cost and risk it eliminates at the candidate-selection stage, not against end-product revenue, and those avoided costs — failed synthesis runs, FTO surprises at manufacturing scale — are large. Royalty and licensing logic favors a hybrid structure: a running royalty or per-member fee tied to validated candidates that advance to fabrication, combined with a platform or method subscription for repeated screening campaigns. The key commercial insight is recurrence. Superconducting-device developers, quantum-hardware programs, and materials-discovery licensees each run multiple screening campaigns per year as their material scope expands; a per-campaign or seat-based fee captures that recurrence directly. The 78% consensus-stable rate on this run — 77 of 99 phonon runs meeting the majority-consensus rule, on a subset of the genus — shows the pipeline can run in batch, though it does not by itself establish commercial-scale volume. The value narrative is straightforward to translate into procurement language: the method reduces wet-lab spend by eliminating dynamically unstable candidates before synthesis, reduces IP risk by running a full-claim freedom-to-operate prescreen before any candidate advances, and reduces diligence friction by producing a traceable, documented selection record. Each of those benefits has a direct cost analog inside a device program's R&D budget, making the licensing fee easy to justify against avoided expenditure rather than hypothetical future revenue.

converts a partly-known genus into an evidence-anchored, FTO-screened selection no single-engine screen or title-only flag can deliver

method/selection claim not foreclosed by any composition prior art; the verification engine is the independently-defensible asset

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

The natural acquirers are quantum-hardware programs and superconducting-device developers who need a credible triage layer upstream of their synthesis pipelines, materials-discovery licensees who run repeated screening campaigns across multiple material classes, and materials-informatics platforms seeking to differentiate their offerings from single-engine competitors. Materials-informatics vendors are strong acquisition candidates. Owning the method outright allows a vendor to bundle the fail-closed full-claim prescreen into diligence offerings in a way competitors cannot replicate with a keyword flag, and the composition-agnostic design means the same engine can be sold across material classes beyond silicides. The license-versus-acquire decision turns on exclusivity and competitive positioning: a strategic that wants to deny the screening engine to competitors should acquire; a platform monetizing breadth across many licensees should take a non-exclusive license and deploy the method across material classes. The recurrent campaign structure of the target customer base supports either model, since the fee recurs naturally with each new screening campaign a licensee runs.

The principal risk is that the stability evidence is MLIP consensus, not first-party density functional perturbation theory, and machine-learning potential reliability is known to be reduced for f-electron rare-earth intermetallics. The 3-of-4 majority threshold mitigates but does not eliminate this risk: a minority of selected candidates may carry split votes, meaning some stability calls remain provisional until DFPT confirms them. A buyer must accept that the current evidence is screening-grade, not synthesis-ready, for the flagship members. The roadmap to de-risk this is the first proof gate described above: first-party DFPT adjudication of the leading members, which converts the consensus call to a first-party computational result. Predicted transition temperature is screening-only in a sub-30% accuracy regime and carries no performance guarantee; the method is designed around this limitation rather than despite it, so it does not represent a claim defect, but a buyer expecting measured Tc data at closing should build that expectation into the diligence scope and post-closing validation budget.