Thermally conductive nitride and carbide liner family for through-glass via thermal management

The glass-core advanced-packaging substrates portfolio addresses one of the most constrained thermal problems in modern semiconductor packaging: the through-glass via (TGV). Glass is an attractive substrate material for advanced packaging — it offers low dielectric loss, dimensional stability, and a coefficient of thermal expansion that can be tuned closer to silicon than organic alternatives — but its inherently low thermal conductivity creates a thermal bottleneck at the via wall. Aluminum nitride has long been the reference liner material for its combination of electrical insulation and thermal conductivity in the range of 150–320 W/m·K, but relying on a single composition creates both supply-chain exposure and a narrow patent position. This asset — the thermally conductive nitride and carbide liner family — answers that problem by establishing a cross-validated set of six compositionally distinct alternatives covering the nitride and carbide chemical space: beta and alpha phase silicon nitride, aluminum oxynitride (AlON), magnesium silicon nitride (MgSiN2), undoped strontium silicon nitride (Sr2Si5N8), and boron carbide (B4C). The strategic value of this family is not that any single member is a direct performance improvement over AlN, but that the family as a whole creates a defensible composition-plus-device-use position that any competitor or customer engineering around AlN must navigate. If a process integrator substitutes Si3N4 liners for cost or deposition-compatibility reasons, if a glass-core supplier selects AlON for its adjustable stoichiometry, or if a thermal engineer reaches for B4C in a high-hardness liner context, the intellectual property in this family speaks to all of those choices. That breadth — backed by independent computational validation — is the asset's primary commercial contribution to the glass-core advanced-packaging substrates portfolio. The timing is also material. The glass-core substrate market is entering what is effectively a forced-substitution window: Intel's glass substrate roadmap, TSMC's advanced packaging push, and the broader industry shift toward chiplet architectures are driving the commercialization of glass-core interposers at scale. Liner material selection is being locked in now, at the design-rule and process-qualification stage, which means a well-positioned IP family filed today sets the landscape for royalty capture over the next full product generation cycle.

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.

The six-member liner family spans two distinct chemical classes. The nitride members — beta-Si3N4, alpha-Si3N4, AlON, MgSiN2, and undoped Sr2Si5N8 — share a common structural logic: they are wide-bandgap insulators with anisotropic covalent bonding that supports phonon-mediated thermal transport well above what conventional oxide liners like Al2O3 can achieve. Boron carbide (B4C) is the lone carbide member, included because its combination of extreme hardness and moderate thermal conductivity (~30–35 W/m·K) opens a different design point where mechanical liner integrity during thermal cycling may be the binding constraint rather than conductivity alone. The alpha and beta polymorphs of Si3N4 are treated as distinct claimed members because they are deposited under different conditions, have meaningfully different microstructures, and their separate inclusion is necessary for claims to cover both CVD and hot-press deposition routes that a process integrator might choose. The computational validation is specific to the two members that required the most scrutiny before inclusion: MgSiN2 and undoped Sr2Si5N8. For MgSiN2, two independent machine-learning interatomic potentials both find the ground-state structure to be dynamically stable, with no imaginary phonon frequencies in either evaluation — MACE reports a minimum phonon frequency of approximately 0.618 THz, and CHGNet independently returns approximately 0.52 THz, both positive and in reasonable agreement. This consensus across two distinct potential families, trained on different data with different architectures, is a meaningful stability signal. The result is supported by two independent DFT source calculations. Sr2Si5N8 in its undoped form received analogous treatment. These two materials are the most computationally novel members of the family; Si3N4, AlON, and B4C have substantially deeper experimental literature bases and did not require the same level of independent cross-validation to justify inclusion. It is important to be precise about what these simulations do and do not address. The multi-engine phonon screening confirms dynamic stability — the structure sits at a local potential-energy minimum and will not spontaneously distort or decompose at the harmonic level. It does not directly measure deposited-film thermal conductivity, adhesion to borosilicate or boroaluminosilicate glass chemistries, or dielectric breakdown behavior under TGV operating conditions. Those are the next validation gates. The open proof gate identified in the workflow is the liner coupon: physical deposition of each candidate on a representative glass substrate, followed by thermal conductivity measurement (laser flash or TDTR), adhesion characterization, and dielectric integrity testing. Until liner coupon data exists for the full family, the thermal conductivity advantage relative to Al2O3 rests on bulk materials literature values rather than deposited-film measurements, which is standard at this stage of materials development but must be disclosed candidly. The negative limitations built into the family deserve explicit technical explanation because they are deliberate and consequential. Beryllium nitrides are excluded on toxicity grounds — the material handling and regulatory burden of Be-containing compounds in a semiconductor fab makes them non-starters regardless of their thermal properties. Hexagonal and cubic boron nitride (h-BN, c-BN) are excluded because their IP landscape is dense and mature; including them would create more freedom-to-operate exposure than competitive value. Europium-activated Sr2Si5N8 — the well-known red phosphor host used in LED applications — is excluded explicitly because the phosphor literature is extensive and well-patented, and including the luminescent form would invite prior-art attacks on the undoped composition. The undoped non-luminescent restriction is therefore both a freedom-to-operate carve-out and a technical differentiation: the property being exploited here is thermal conductivity in an electrically insulating context, not photoluminescence.

The addressable market for through-glass via liner materials is best understood at two levels. At the direct material-supply level, the market for specialty liner materials and precursors in glass-core packaging is estimated in the range of $200–500 million, acknowledging that this is a developing segment where shipment volumes are currently small but growing at rates consistent with the broader advanced packaging market, which analysts broadly project to compound in the low-to-mid double digits through the late 2020s. The liner material itself is a small-area, high-value process step: the per-unit material cost is modest, but the qualifying supplier base is narrow and switching costs are high once a material is embedded in a qualified process flow. This creates durable pricing power for IP holders with composition-and-device-use coverage. At the licensing level, the relevant buyers are not primarily liner material suppliers — they are the glass-core substrate manufacturers, the advanced packaging foundries, and the major IDMs and fabless semiconductor companies that will specify liner materials in their design rules. A royalty-on-substrate or royalty-on-packaged-unit model is more natural than a royalty-on-material model at this level of the value chain, because the value being captured is the thermal performance of the packaged device, not the raw liner material. Standard royalty logic for enabling process patents in packaging tends to run in the range of 1–3% of substrate value, though negotiated rates depend heavily on exclusivity and the depth of coverage. The customers most directly affected by liner material selection are the glass-core substrate integrators currently qualifying TGV processes — companies in Japan, South Korea, Taiwan, and increasingly the US that are building out glass substrate capacity to serve AI accelerator and high-bandwidth memory packaging demand. These integrators are making liner material decisions now, in the 2025–2027 window, which means the licensing or acquisition window for this family is near-term rather than speculative.

thermal-liner breadth

undoped non-luminescent Sr2Si5N8 only

The most natural acquirers or licensees for this family are glass-core substrate manufacturers who are actively qualifying TGV processes and need to secure IP freedom and potential exclusivity in liner material selection before their process flows are locked. Companies such as AGC, Corning, and the major Korean glass manufacturers fall into this category, as do the advanced packaging foundries — ASE, Amkor, JCET — that are building glass-core interposer capability. For these buyers, a composition-plus-device-use family covering six validated alternatives to AlN is a straightforward defensive and offensive IP acquisition: it protects their own process choices while creating licensing leverage over competitors who select any of the same materials. A second buyer category is the semiconductor IDMs and fabless companies that are specifying glass-core packaging for next-generation AI accelerators and high-bandwidth memory stacks — companies that want to ensure their packaging supply chain is not restricted by third-party IP in liner materials. For these buyers, a license rather than an acquisition is the more natural structure, and the clean FTO status makes a straightforward field-of-use license the path of least resistance. The family's breadth across six distinct compositions is a particular advantage in licensing negotiations because it means a licensee's process engineers retain flexibility to select the material that best fits their deposition equipment without exiting the licensed scope.

The primary technical risk is the gap between computed bulk stability and measured thin-film performance. Thermal conductivity in deposited films — particularly nitrides deposited at the via-aspect-ratio thicknesses relevant to TGV liners — can be substantially lower than bulk single-crystal values due to grain boundary scattering, film stress, and interfacial phonon resistance. If liner coupon measurements reveal that one or more family members do not achieve meaningful thermal conductivity improvement over Al2O3 at practical deposition thicknesses, the commercial rationale for those members weakens. The mitigation is to prioritize the coupon campaign early, beginning with MgSiN2 and undoped Sr2Si5N8 as the computationally novel members, and accepting that some family members may be narrowed or abandoned based on data. The secondary risk is claim durability during prosecution. The device-use claim structure is stronger than composition alone, but patent offices examining glass-core packaging applications are increasingly sophisticated, and the specific functional recitation of "thermally conductive insulating liner on a through-glass via wall" must be supported by adequate specification disclosure. The inclusion of six diverse compositions in a single family also creates prosecution risk that an examiner will require restriction, splitting the family into multiple applications and increasing prosecution cost. The mitigation here is ensuring that the specification provides sufficient enabling disclosure for each member, particularly MgSiN2 and Sr2Si5N8, supported by the computational validation data. A well-resourced prosecution strategy that anticipates restriction requirements and plans continuation filings accordingly is the standard path to maintaining family breadth.