Calcium hafnate and calcium zirconate high-permittivity dielectrics for advanced memory packaging

CaHfO3 and CaZrO3 are orthorhombic distorted-perovskite dielectrics that hold an independent, preliminary-FTO-screened device-use position in MIM capacitor applications for HBM4 and DRAM packaging. CaHfO3 carries the widest bandgap in the broader dielectric, ferroelectric and wide-bandgap oxides portfolio at roughly 4.57 eV, and both materials have been confirmed dynamically stable by two independent machine-learning interatomic potentials — a consensus result that the cubic barium analogues BaHfO3 and BaZrO3 failed to achieve. The practical consequence is that device-use coverage in memory dielectrics does not depend on a higher-permittivity Ruddlesden-Popper hafnate lead whose stability adjudication remains open: these materials stand on their own phonon record and their own a preliminary FTO screen review. The strategic timing argument is concrete: the HBM4 high-k integration window opens in 2026 Q3/Q4, and memory foundries are selecting dielectric chemistries now. A preliminary-FTO-screened distorted-perovskite MIM device-use position, anchored to two computationally stable, structurally distinct oxides, is licensable before that window closes. The asset is most valuable as a risk-transfer instrument — its drafted claim would, if granted, keep MIM-capacitor claim coverage in play even if the lead material's four-potential DFT verdict resolves unfavorably, and a preliminary screen (not a freedom-to-operate opinion) identified no blocking genus on alkaline-earth hafnate or zirconate ABO3 capacitor compositions.

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.

CaHfO3 and CaZrO3 crystallize in the orthorhombic Pnma space group, the distorted-perovskite setting produced by cooperative octahedral tilting of the HfO6 or ZrO6 coordination polyhedra. That tilt pattern is the structural origin of both their thermodynamic stability and their wide bandgaps: the reduced B-O-B bond angles that accompany the distortion push the conduction-band minimum upward relative to the undistorted cubic end-members. CaHfO3 reaches approximately 4.57 eV on PBE-level DFT, the widest value in the portfolio family, a gap that translates directly into suppressed Fowler-Nordheim and trap-assisted tunneling leakage in a thin-film MIM stack. The comparison point matters: the cubic-phase analogues BaHfO3 and BaZrO3 were evaluated and returned phonon imaginary modes under both ML potentials — they are soft, dynamically unstable structures that cannot be relied upon for device use. The distorted Ca-based phases are structurally and electronically distinct, and that distinction is the technical foundation for both the claims and the freedom-to-operate position. Computational validation followed Lattice Graph's standard multi-engine consensus protocol. Both CaHfO3 and CaZrO3 were evaluated independently under MACE and CHGNet, the two primary machine-learning interatomic potential engines used in the platform, and both materials cleared the dynamic stability threshold under both potentials — no imaginary phonon frequencies in either calculation for either composition. That two-engine agreement, achieved on May 29, 2026, is the stability verdict the claims rest on. Two independent DFT source calculations provide the structural and electronic baseline, and the CaHfO3/CaZrO3 perovskite system has an established experimental ALD deposition literature, which means the path from computed stability to thin-film fabrication is not speculative. The open validation gates are a measured film-level permittivity and leakage coupon — the experiment that converts computed bandgap and structural stability into the dielectric constant and leakage-current numbers a memory foundry requires for a design-in decision — and an HSE06 hybrid-functional bandgap calculation to refine the PBE-level gap estimate.

The addressable market for high-k MIM and gate dielectrics in HBM4 and advanced DRAM packaging is estimated at $5 billion or more. That figure reflects the aggregate value of dielectric materials and process IP at leading-edge memory nodes, where dielectric performance — permittivity at target equivalent oxide thickness, leakage at operating field, thermal stability through back-end processing — directly governs bit density and therefore wafer yield. The three target licensees in our analysis, Samsung Foundry, SK Hynix, and Micron, are major setters of the high-k roadmap for HBM4 and DRAM; each is likely evaluating dielectric stacks for the HBM4 node (our inference, not sourced), and each could have an interest in an alternative to the incumbent hafnia-based process chemistries. Royalty and licensing logic centers on a per-wafer or per-design-win structure for the MIM capacitor and gate dielectric field of use. A wide-gap dielectric designed for low leakage, which a foundry could design into its HBM4 stack with no blocking patent found on a preliminary screen (not a freedom-to-operate opinion), could command licensing leverage proportional to the node transition cost: foundries sunk billions into HBM4 process development, and a chemistry that resolves a leakage or capacitance constraint without triggering royalty exposure to existing hafnia IP estates could be worth a premium. Non-exclusive licensing across all three major memory makers is the natural structure given that each pursues HBM4 independently; an exclusive field-of-use license is credible only for a single foundry seeking a differentiated dielectric stack. All market-size figures here are estimates and no prices or commitments are stated.

wide-gap, preliminary FTO screen, cross-MLIP-stable high-k arms that carry the device-use claim independent of the split-verdict hafnate lead

distorted-perovskite MIM device-use; PRELIMINARY-FTO-SCREENED full-claim review

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

The natural buyers are the three memory foundries already named as target customers: Samsung Foundry, SK Hynix, and Micron. Each is independently racing the HBM4 high-k integration schedule with a 2026 Q3/Q4 window, and each has organizational incentives to hold a preliminary FTO screen dielectric option that does not depend on the incumbent Hf-Zr IP estates. A field-of-use license scoped to HBM4 and DRAM MIM capacitor and gate dielectric applications fits all three without requiring exclusivity, and non-exclusive licensing across all three maximizes the royalty base in a market where no single foundry can set the standard unilaterally. A single strategic acquirer is also credible if one memory maker wants to hold the distorted-perovskite MIM position exclusively to differentiate its HBM4 stack. The most value-accretive acquisition scenario is a bundle with the Ruddlesden-Popper hafnate lead from the same portfolio: a buyer would then hold both the preliminary-FTO-screened distorted-perovskite arm (MLIP-stable today, claim drafted and not yet granted) and the higher-permittivity hafnate headline (higher upside if its phonon verdict closes favourably), creating a layered memory-dielectric claim position across both structural families. Materials companies and specialty dielectric suppliers serving the memory foundry supply chain are secondary acquisition candidates if the primary foundry approach does not produce a deal before the HBM4 window.

The most direct risk is that the key property in the current proof record is bandgap, not measured permittivity. CaHfO3's 4.57 eV gap supports a low-leakage argument but does not by itself confirm that the material reaches high-k dielectric constants adequate for HBM4 MIM targets. Until a film-level permittivity coupon is measured, the high-k case rests on the family-level dielectric thesis rather than on composition-specific electrical data. A foundry's process integration team will not advance to design-in without that measured dielectric constant and leakage current, so the validation gap is commercially material. A second risk is that the computational stability case, while consensus across two ML potentials, has not yet been confirmed by full DFT phonon calculations with a hybrid functional; the HSE06 bandgap calculation queued as a next step would simultaneously sharpen the electronic picture and provide a higher-fidelity stability cross-check. The roadmap to de-risk both gaps is straightforward in principle: deposit a thin film by ALD (for which literature recipes exist for these compositions), measure permittivity and leakage, and run the HSE06 calculation in parallel. These are standard dielectric characterization steps, not novel experiments. The claim scope risk is manageable provided the negative limitation on cubic BaHfO3/BaZrO3 is maintained and prosecution does not extend coverage into phases the computational record does not support. The window risk is real — the HBM4 high-k selection cycle closes in 2026 Q3/Q4 — which means the film coupon experiment determines whether this asset participates in the current node transition or is carried forward to the next one.