Why Glass? — The Limits of ABF Substrates
Why Glass?
The Limits of ABF Substrates and the Rise of Glass Core Technology
The semiconductor packaging paradigm is shifting — and it's driven by three physical limits that organic materials simply can't overcome anymore.
TSMC is mass-producing 2nm chips. Samsung has pushed transistors vertical with its GAA architecture.
The semiconductor industry ships a faster, smaller chip every single year.
But what about the substrate those chips sit on?
Surprisingly, the core material underneath today's most advanced AI accelerators
hasn't changed much since the 1990s.
It's still an organic stack of epoxy resin and glass fiber — the ABF substrate.
While front-end fabrication battles at the atomic scale,
packaging has been quietly getting away with gluing chips onto what is essentially sophisticated cardboard.
That free ride is ending. And the material stepping in to replace it is glass.
What Is an ABF Substrate?
ABF stands for Ajinomoto Build-up Film. Ajinomoto — yes, the Japanese seasoning company best known for MSG and bouillon cubes. In the 1990s, Ajinomoto applied chemical expertise from its food business to semiconductor materials, and ABF film was born. Today, more than half of all high-performance semiconductor substrates worldwide are made with material supplied exclusively by Ajinomoto.
The chips inside Intel CPUs, AMD processors, and NVIDIA GPUs all depend on a single seasoning company's material. That's the quiet irony buried inside every high-end PC.
The substrate's job is straightforward: it acts as a bridge between the chip and the motherboard, routing hundreds to thousands of signals with precision while distributing power and spreading heat. ABF has been the standard for good reasons.
- ①Processability — Epoxy-based materials are easy to laser-drill and plate with copper
- ②Cost — Decades of mass production have driven unit costs down dramatically
- ③Ecosystem — Equipment, processes, and design tools are all optimized around ABF
The problem is that all of these advantages describe the past.
Three Physical Limits of ABF
There are three fundamental physical constraints baked into ABF substrates. They've always been there — but AI chips have finally made them impossible to ignore.
① CTE Mismatch — The Chip and the Substrate Expand at Different Speeds
Every material expands when heated. The problem is that silicon chips (CTE ≈ 2.3 ppm/°C) and ABF substrates (CTE ≈ 40 ppm/°C) expand at wildly different rates. Every time a chip powers on or off, the joint between chip and substrate is pulled and pushed in slightly different directions — thousands of times over its lifetime. Eventually, it cracks. With smaller dies this was manageable, but large AI accelerator dies have turned this into a serious reliability crisis.
② Flatness — Organic Materials Warp
Epoxy-based organic materials warp slightly under heat. That might sound like a minor cosmetic issue — it isn't. Warpage directly limits how tightly bump interconnects can be placed. Once bump pitch drops below 100μm, ABF simply can't maintain stable, reliable joints across a large die. Glass doesn't warp. Its surface flatness holds at the micrometer level, enabling far denser interconnects and finer routing.
③ High-Frequency Signal Loss — The Hidden Bottleneck of the AI Era
As signals travel through a substrate, some energy bleeds off as heat. This is called dielectric loss (Df). The higher the frequency, the more energy is lost. In AI accelerators and HBM interfaces that demand tens of gigahertz of bandwidth, ABF's high Df becomes a serious performance wall. Glass substrates exhibit far lower dielectric loss than organic alternatives — keeping signals intact where it matters most.
Why Glass Solves All Three
Glass doesn't just chip away at these problems — it addresses all three simultaneously. That's not a coincidence. The material properties are simply different at a fundamental level.
- ✓CTE close to silicon — Thermal stress drops dramatically, and solder joint reliability takes a major leap forward
- ✓Exceptional surface flatness — Tens of times flatter than organic alternatives, enabling finer traces and tighter bump pitch
- ✓Low dielectric loss (Df) — High-frequency signals travel further and cleaner, meeting the bandwidth demands of AI accelerators
- ✓Optical transparency — A hidden bonus: glass is inherently compatible with photonic interconnects, making it future-ready for the silicon photonics era
"The performance ceiling is no longer the transistor — it's the substrate.
Without a materials change, AI accelerator scaling stops here."
So Why Isn't It Being Used Yet?
Fair question. If glass is this obviously better, why aren't chipmakers shipping it today? The honest answer: glass breaks.
⚠️ Brittleness — Glass is notoriously difficult to handle at scale. Processing 300mm or 500mm glass panels through a semiconductor fab without cracking them is a genuine engineering challenge — not a minor inconvenience.
- !TGV process immaturity — Drilling thousands of through-glass vias (TGVs) at sub-50μm diameter with acceptable yield is still an unsolved problem at volume
- !Ecosystem lock-in — Equipment, inspection systems, and process flows have been optimized for ABF over decades. You can't swap substrates overnight
- !Yield economics — Until TGV yields improve meaningfully, glass can't compete with ABF on cost per unit at high volume
These are precisely the reasons Intel's glass substrate production timeline keeps slipping. The technical superiority of glass is already proven in the lab. What remains is the hard work of process maturity and ecosystem transition.
Why Right Now Is the Inflection Point
AI is forcing the issue. NVIDIA's H100 and B200 accelerators consume hundreds of watts and move terabytes of data in real time. These demands are pushing substrate materials past their physical limits — right now, not in theory.
Intel is investing billions targeting glass substrate production around 2030. SKC, Samsung Electro-Mechanics, Corning, and AGC have all entered the race. ABF isn't dead yet — but the direction of its replacement is already fixed. The only open question is timing.
ABF substrates face three physical walls: CTE mismatch, warpage, and dielectric signal loss. Glass solves all three simultaneously — and AI accelerators are making these walls impossible to ignore. The question isn't whether glass substrates will replace ABF. It's when.
In the next post, we trace the full 60-year history of packaging substrates — from metal lead frames to fiberglass PCBs to ABF — and what each transition taught the industry about what comes next. That history makes the scale of the coming glass transition unmistakably clear.
Comments
Post a Comment