Nvidia Advances Glass Substrate Technology, Targeting HBM Integration Bottleneck
nashnova research
Nvidia is working with supply-chain partners of Intel, AMD, and its own to develop glass substrates — aiming to break the size limits of today's silicon interposers so more HBM can fit inside a single package, directly raising the bandwidth ceiling of next-generation AI chips.
Why does Nvidia want to replace the current packaging base?
Today's AI chips use a silicon interposer — a silicon slab that holds the GPU and HBM side by side — but larger silicon slabs are harder to manufacture and more expensive.
This means → how many HBM stacks fit in one package is capped by the interposer's physical size — if HBM can't fit, bandwidth can't grow.
Glass substrates (special glass replacing silicon as the base) are flatter, stronger, deform less at high temperatures, and offer better dielectric properties with lower signal loss.
In plain terms = the silicon base is a small table that warps at the edges as you stack more chips; a glass base is a bigger, flatter, harder table.
Why is HBM the lifeline of AI chips?
HBM — high-bandwidth memory — boosts bandwidth by vertically stacking multiple DRAM layers. It is the core memory component of AI accelerators.
As model sizes and compute demands keep climbing, how much HBM a single package can hold has become a key bottleneck for AI chip performance.
This means → if glass substrates allow a larger package area and more HBM, they directly raise the GPU's data-throughput ceiling.
How far along is the supply chain?
German equipment maker SCHMID disclosed it is co-developing glass-substrate equipment with core companies in the supply chains of Intel, Nvidia, and AMD.
Nvidia CEO Jensen Huang met with SK Group Chairman Chey Tae-won to discuss next-generation semiconductor cooperation including glass substrates, and is pushing TSMC to accelerate related R&D.
This reflects a shift from lab-stage concept to three-way industry mobilization — equipment makers + chip designers + packaging foundries.
What is the biggest barrier to mass production?
SCHMID identified TGV metallization — glass-through-via metallization, creating conductive channels inside non-conductive glass — as the primary technical obstacle.
In plain terms = silicon conducts electricity naturally; drill a hole and signals flow. Glass does not conduct — you have to "tunnel through glass and lay wiring inside," a far harder process.
End-customer qualification is not yet complete, meaning even finished equipment must still pass chipmakers' reliability sign-off.
What determines whether this roadmap actually lands?
Three gates must all clear: manufacturing-process maturity, reliability verification, and customer qualification — any one stalling pushes the timeline back.
This means → the promise of glass substrates is clear, but the pace of execution remains uncertain. Investors should track progress at these three gates, not the concept itself.
SCHMID is simultaneously expanding into AI-server motherboards and next-generation packaging equipment — a sign the equipment maker is betting on the glass-substrate track while hedging its risk.
市场有风险，内容仅供研究参考，不构成投资建议。