Rare-earth scandate wide-bandgap high-k dielectric for low-leakage package MIM capacitors

Advanced packaging has arrived at an inflection point where the dielectric quality of passive components embedded within the package — not just on the die — is becoming a hard limiter on system performance. Metal-insulator-metal (MIM) capacitors and through-glass-via (TGV) adjacent structures in glass-core substrates demand dielectrics with simultaneously high permittivity, wide bandgaps, and sub-nanoampere leakage floors. Hafnium oxide (HfO2), the incumbent, has bandgaps around 5–6 eV but is constrained by phase-stability issues at thin-film dimensions and dielectric constants in the 20–25 range that leave significant capacitance density on the table. The rare-earth scandate family — and DyScO3 in particular — offers a largely unexplored path: a bandgap of approximately 4.5 eV combined with higher permittivity than HfO2 and a perovskite crystal structure (Pnma orthorhombic) that is already proven manufacturable at wafer scale as single-crystal substrates for III-V epitaxy research. That existing commercial substrate ecosystem is a decisive advantage: it collapses the materials-availability argument that typically stalls novel dielectric proposals. The timing dimension of this asset is the forced-substitution dynamic in advanced packaging. As leading-edge logic moves to glass-core and panel-scale substrates to escape the dimensional limits of organic laminates, every passive embedded in those substrates must meet new leakage and voltage-hold targets that organic-substrate-era HfO2 films were never designed to hit. This creates a technology transition window — a period during which a licensable high-k dielectric with a wide bandgap and a documented supply chain can displace the incumbent before second-source suppliers and equipment vendors re-optimize around HfO2. The rare-earth scandate family, protected as a composition-plus-device-use patent family covering seven RE-ScO3 members for package-integrated MIM and TGV-adjacent dielectric applications, is positioned to sit inside that window. The asset is a lead filing in the glass-core advanced-packaging substrates portfolio, and it carries honest supply-chain caveats — rare-earth sourcing is a disclosed risk — but those caveats are manageable at the volumes relevant to package-integrated passives, which are far smaller than commodity rare-earth applications.

Each candidate is validated by multiple independent machine-learning interatomic potentials. A material advances only when the engines agree on phonon (dynamic) stability — disagreement is surfaced, not hidden.

DyScO3 crystallizes in the orthorhombic Pnma space group, a distorted perovskite structure related to the GdFeO3-type family. The Dy3+ and Sc3+ cations occupy corner-sharing octahedral environments in a way that suppresses the ferroelectric and antiferroelectric instabilities seen in titanate perovskites, producing a dielectric that is stable across a wide temperature range without the phase transitions that limit applicability of PZT or BST in precision analog circuits. The computed bandgap for DyScO3 is approximately 4.5 eV — wide enough to push the Schottky and direct-tunneling leakage contributions well below the noise floor at package-relevant operating voltages (typically 1–3 V). The co-lead material GdScO3 carries a somewhat narrower bandgap near 3.1 eV, which accepts a trade of modestly higher leakage for potentially higher permittivity or simpler deposition chemistry depending on ALD precursor availability. The full protected family spans seven rare-earth scandates: DyScO3, GdScO3, LaScO3, NdScO3, YScO3, SmScO3, and TbScO3, covering the sweep of lanthanide ionic radii and allowing a licensee to tune the bandgap–permittivity trade-off across a continuous composition space within a single IP family. The dynamic stability of DyScO3 was evaluated using three independent machine-learning interatomic potentials: MACE, CHGNet, and MatterSim. MACE returned a minimum phonon frequency of +0.413 THz and MatterSim returned +0.442 THz — both positive, indicating no imaginary phonon modes and a dynamically stable structure. CHGNet returned -0.703 THz, an apparent imaginary mode suggesting instability. This disagreement triggered a documented methodological review: CHGNet is known to produce false-negative stability predictions on rare-earth oxide compositions, where its training-set coverage of lanthanide-Sc-O ternaries is limited. The MACE result was designated the controlling verdict, and the overall assessment is a majority-stable finding across the three potentials — two independent machine-learning potentials agree the structure is dynamically stable, with the CHGNet outlier attributed to a well-characterized potential artifact rather than a true physical instability. Two independent DFT reference calculations corroborate the majority-stable conclusion, providing additional grounding beyond the machine-learning layer. The computational workflow followed by Lattice Graph runs candidates through progressively more expensive simulations gated by this consensus stability requirement. For DyScO3, the completed simulations include a two-engine run (CHGNet+MACE) and a three-engine run (MatterSim added), from which the stability determination above derives. The dielectric-relevant property of interest — the electronic bandgap — was characterized at 4.5 eV, consistent with prior experimental literature on single-crystal DyScO3 substrates used in III-V MOS research. The existing commercial availability of single-crystal DyScO3 substrates, originally produced for use as lattice-matched growth platforms for InGaAs and related channel materials, provides a critical empirical anchor: this is not a purely hypothetical material. Researchers have grown high-quality thin films on these substrates and characterized their dielectric properties. The asset repurposes that knowledge base and supply chain toward the package-integration context, which is a distinct and currently unprotected application space. One open validation gate remains before the material can be considered ready for licensing discussions at the highest confidence level: a package-integrated thin-film coupon — a demonstration that DyScO3 or GdScO3 can be deposited as a conformal thin film inside a glass-core or silicon-interposer MIM stack and characterized for leakage current density, capacitance density, and breakdown field under packaging-relevant conditions (temperature cycling, humidity, reflow). This coupon-level experiment would close the gap between the single-crystal-substrate literature and the polycrystalline or amorphous ALD-grown thin-film regime that actual MIM capacitor manufacturing requires. Until that gate is opened, the bandgap data and phonon stability are the primary validated claims, and the device-integration story rests on inference from the existing substrate literature.

The addressable market for high-k dielectrics in advanced packaging MIM capacitors and TGV-adjacent structures is estimated at $1–2 billion annually, growing as glass-core and fan-out panel-level packaging displace organic substrates in high-performance compute, memory, and RF applications. This estimate reflects both the direct materials supply market — dielectric films consumed in package fabrication — and the embedded licensing value that accrues to whoever controls the composition IP covering novel high-k materials in this configuration. The distinction matters because a buyer of this asset is more likely to extract value through licensing to substrate manufacturers and OSATs than through direct materials supply, although the two models are not mutually exclusive. The customer set for a licensee of this IP falls into two primary buckets. First, MIM capacitor vendors and substrate manufacturers — companies producing embedded passive components for advanced packaging on glass or silicon interposer platforms. Second, foundries and OSATs that have vertically integrated passive fabrication into their panel or wafer processing lines and need dielectric IP to underpin their process differentiation claims. Both groups are actively evaluating non-HfO2 high-k options as they push MIM capacitance density above what legacy HfO2 films can deliver at sub-50 nm equivalent oxide thickness. A royalty or licensing structure could reasonably be indexed to wafer starts using the dielectric, to total capacitance area licensed, or as a one-time IP acquisition price bundled into a broader high-k dielectrics portfolio. The strategic value of the broad RE-ScO3 family — seven members in a single composition claim — is that it forecloses design-arounds within the rare-earth scandate chemical space, giving a licensee a durable position rather than a single-compound holding that can be stepped around by choosing an adjacent lanthanide.

wide-gap low-leakage, bench-validated availability

package-integrated config; single-crystal-substrate provenance is background

The most direct strategic fit is with advanced packaging substrate manufacturers and their dielectric supply chains — companies such as AGC, Corning, and Toppan who are commercializing glass-core substrates, and the OSAT and IDM players (Intel Foundry, Samsung Electro-Mechanics, Ibiden, Shinko) who embed passive components into high-density interposers. For these buyers, acquiring the RE-ScO3 composition-plus-device-use family provides a defensible IP position in next-generation high-k MIM dielectrics that can be enforced against competitors adopting the same materials, and it gives them a potential process-differentiation story for customers who need lower leakage in embedded capacitors than HfO2 can deliver. The asset would also be of interest to specialty dielectric materials suppliers and ALD precursor companies seeking to attach IP to a product line targeting the advanced packaging segment. A second buyer category is fabless semiconductor companies and system integrators who are building custom chiplet-based systems on glass or silicon interposer substrates and want to secure freedom-to-use across the full RE-ScO3 family for their own packaging technology platforms. For these buyers, the asset is less about licensing-out and more about freedom-to-operate assurance combined with the optionality to license or cross-license. The breadth of the seven-member composition family and the clean FTO status make it straightforward to integrate into a portfolio acquisition without requiring significant claim surgery. The supply-chain risk around dysprosium is the variable that would most commonly arise in buyer diligence, and it is worth addressing directly: at the thin-film deposition volumes relevant to package-integrated passives (grams per tool per year rather than kilograms), rare-earth supply constraints that affect permanent-magnet or phosphor markets are largely irrelevant, and ALD precursor suppliers already support Dy and Gd oxide precursor lines for gate-dielectric research.

The most material risk is rare-earth supply chain concentration. Dysprosium and several of the other lanthanides in the protected family are subject to export controls and price volatility driven by Chinese mining and processing dominance. While thin-film dielectric volumes are small in absolute mass terms, any technology dependent on critical rare earths faces scrutiny in supply-chain risk assessments, particularly from defense-adjacent or export-controlled customers. This risk is partially mitigated by the breadth of the composition family: LaScO3 and NdScO3 use lighter lanthanides with more diversified supply, and a licensee can anchor their commercial product to those members while retaining freedom to use Dy and Gd as the supply environment allows. The second material risk is the open validation gate on package-integrated thin-film behavior. The gap between bulk crystal stability (computationally established) and thin-film dielectric performance in a real packaging stack is non-trivial, and without a coupon-level demonstration, a buyer is acquiring a well-grounded computational and composition claim rather than a process-ready technology. De-risking this gate requires a targeted thin-film deposition and characterization campaign — estimated to be months rather than years of laboratory work — which could be structured as a condition in a licensing agreement, a joint development arrangement, or a milestone-gated acquisition. The CHGNet false-negative on rare-earth oxides, while explained and addressed by the majority-stable determination, will also appear in buyer diligence and should be disclosed proactively with the documented rationale rather than buried; the explanation is technically sound and strengthens rather than undermines confidence in the computational workflow.