*A first-principles primer on glass core IC substrates — the proposed replacement for organic ABF substrate that sits between every advanced silicon die and its PCB. Not the same as PCB-level glass cloth; not the same as Hoya’s photomask glass; not the same as smartphone cover glass. A distinct indu…
A first-principles primer on glass core IC substrates — the proposed replacement for organic ABF substrate that sits between every advanced silicon die and its PCB. Not the same as PCB-level glass cloth; not the same as Hoya’s photomask glass; not the same as smartphone cover glass. A distinct industry, a distinct supply chain, and an emerging investment opportunity that takes 2027-2030 to mature.
In September 2023, Intel held a packaging tech day and announced GlassCore — a research program targeting glass substrate IC packages by 2030. The internal positioning was that glass would replace organic ABF (Ajinomoto Build-up Film) as the substrate material in advanced packages. Most of the industry took this as a long-dated R&D talking point.
Two years later, two things changed: 1. The package areas Intel and Nvidia ship today are at the edge of organic substrate’s physical limits. Per STF Research’s analysis of substrate area progression: Hopper substrate 3,025mm² → Blackwell B200 5,625mm² → Vera Rubin 8,100mm² (+86% / +168% vs H100). Organic ABF warps at large panel sizes; signal loss climbs unacceptably at the 224G SerDes generation. Glass core fixes both. 2. The supplier ecosystem is starting to qualify. Per Pink’s vault: E&R Engineering (8027.TWO) validated TGV laser processing for glass core substrates in 2024 after a 5-year co-development with a North American IDM (implied Intel via a Japanese substrate-maker intermediary). The Japanese customer is “accelerating validation; small production 2026, scaling 2027.” On 24 April 2026, Intel itself “cited rising substrate, T-glass, and memory costs as headwinds to gross margin” — they are paying for the new material today, not in 2030.
The implication: glass substrate is a 2027-2030 industry inflection, but the validated equipment, qualified material, and named customers are already publicly observable. The Intel commit is real; the supply chain is forming; the investable opportunities are upstream of the substrate makers themselves.
Inside a modern AI accelerator package, the silicon die is too small and too pin-dense to bond directly to a PCB. Between them sits an IC substrate: a multilayer rigid substrate that re-routes the chip’s thousands of fine-pitch I/O bumps to the PCB’s coarser solder ball pattern. The substrate is engineered to handle three things simultaneously: huge I/O count (>10,000 contacts on a flagship AI accelerator), massive thermal load (>1kW per package), and increasingly large physical dimensions (90×90mm and growing).
For three decades the dominant substrate material has been organic ABF: Ajinomoto Build-up Film, a high-performance dielectric resin developed by Ajinomoto and laminated with copper to form multilayer organic substrates. Companies like Ibiden, Shinko, Unimicron, Nan Ya PCB, Kinsus, Simmtech, AT&S convert ABF + glass-cloth core + copper foil into the substrate we know today. Hopper, Blackwell, Sapphire Rapids, Granite Rapids — all run on ABF substrates.
ABF was tuned for a world of 30-50W CPUs and small packages. AI changed both numbers by an order of magnitude. The package is now a thermal stress test, a mechanical warpage problem, and a high-frequency signal integrity problem. Organic substrate is approaching its limits on all three axes simultaneously.
The candidate replacement is glass core substrate. A panel of ultra-flat, ultra-thin glass replaces the organic core. Through-glass vias (TGV) are laser-drilled and metallized to connect top and bottom layers. The substrate is then built up the same way as ABF — but on a far more dimensionally stable, thermally inert, and electrically clean foundation.
Let’s walk this from the bottom up.
An IC substrate is a sandwich. A core sheet (typically 0.4-1.0mm thick) provides mechanical structure. Build-up layers (3-5 on each side) carry electrical traces. Through vias connect signals top to bottom. The whole stack is solder-mask-finished and the surface ball-grid-array attached.
The core determines almost everything else. The CTE (coefficient of thermal expansion) of the core has to roughly match the silicon (~3 ppm/°C) so the package doesn’t fracture under heat. The dimensional stability has to allow ±5µm panel-level alignment over 90mm. The dielectric properties have to support tight impedance control. And the panel size has to allow large packages without warping like a potato chip out of the lamination press.
Organic substrate cores are typically a glass-cloth-and-resin laminate (BT resin or similar). CTE ~12-14 ppm/°C, decent thermal stability, but warpage scales with panel size. Above ~80×80mm the warpage problem is acute. Above ~100×100mm it’s a yield killer.
Glass cores are made from specialty glass (typically a borosilicate or fused-silica formulation). CTE ~3 ppm/°C — matched to silicon. Dimensional stability extraordinary. Surface flatness to <1µm RMS. Dielectric loss far lower than organic at high frequency. The physics of glass solves the three problems organic substrate is hitting simultaneously.
The catch: glass is brittle. You cannot drill mechanical holes through it without micro-cracking. Standard photolithography on glass is hard because the surface is too smooth. Plating chemistry has to be invented. Singulation (cutting individual substrates from a panel) requires laser scoring + breaking. The process flow is, mechanically, harder than organic substrate. The reason it took 30 years to get to glass is that the engineering problems are real.
| Term | Definition |
|---|---|
| IC substrate | The multilayer rigid plate between the silicon die and the PCB |
| ABF | Ajinomoto Build-up Film — the dominant organic dielectric resin in IC substrate |
| BT core | Bismaleimide-triazine resin — common organic core material |
| Glass core | Glass panel used as substrate core, replacing organic core |
| TGV | Through-Glass Via — laser-drilled hole through glass, copper-plated |
| TSV | Through-Silicon Via — used inside silicon stacks, distinct from TGV |
| CTE | Coefficient of thermal expansion |
| Build-up layer | Sequential dielectric + copper layer added on top of core |
| GlassCore | Intel’s branding for its glass substrate program |
| Q-Glass | High-performance glass substrate grade (per STF Research) |
| T-glass | Low-CTE specialty glass for substrate cores |
| Borosilicate glass | Standard substrate glass formulation; CTE ~3 ppm/°C |
| Fused silica | Pure SiO2 glass; even lower CTE; expensive |
| Panel-level packaging (PLP) | Substrate manufacturing on large rectangular panels (e.g., 510×515mm) — vs wafer-level |
[Glass material] → [Panel cutting and surface prep] → [TGV laser drilling] ★ E&R, LPKF
↓
[Via metallization (CVD/PVD copper seed + electroplating)] ← Applied Materials, Lam Research, Tango Systems
↓
[Dielectric build-up layer 1 + copper imaging + cure] ← multiple iterations
↓
[Dielectric build-up layer 2 + copper imaging + cure]
↓
[... build-up layer N ...]
↓
[Solder mask + surface finish + electrical test]
↓
[Singulation — laser scoring + break] ← Disco, E&R
↓
[Final inspection]
↓
[Substrate ready for die attach in OSAT or fab packaging line]| Step | Organic ABF | Glass core |
|---|---|---|
| Core production | Glass-cloth + resin lamination, panel-level | Glass-panel manufacture (different industry) |
| Via formation | Mechanical drill or laser drill | Laser drill only (no mechanical option) |
| Via plating | Mature process | Newer; higher-aspect-ratio constraints |
| Build-up | 25+ years of process maturity | Fresh process development |
| Singulation | Punching or routing; mature | Laser score + break; immature |
| Equipment ecosystem | Widely available | Concentrating to <10 qualified suppliers globally |
The implication: the substrate makers (Ibiden, Shinko, Unimicron, AT&S, Samsung E-M) have to rebuild their process libraries for glass core. They cannot port ABF know-how 1:1. This is what creates the share-shift opportunity — incumbents do not automatically win the next generation.
Organic substrate manufacturing on >500mm panels has been brittle (warpage, yield loss). Glass-core substrates can support larger panel-level packaging because the substrate itself is dimensionally stable. This in turn enables larger packages (Vera Rubin’s 90×90mm and beyond) without introducing new failure modes from substrate warpage.
The cascading benefit: as packages get larger, more die can be co-packaged (3D stacking, chiplet integration, integrated photonics), unlocking new architectural possibilities. Glass substrate isn’t just a material substitution — it’s an enabling layer for the next generation of package architecture.
| Metric | Why it matters | Organic ABF benchmark | Glass core target |
|---|---|---|---|
| Core CTE | Mismatch with silicon = thermal cycling failure | 12-14 ppm/°C | 3-4 ppm/°C ✓ |
| Core surface roughness | Build-up adhesion + signal integrity | 0.3-0.5µm RMS | <0.1µm RMS ✓ |
| Panel-level warpage @100°C | Yield in lamination | 200-400µm over 100mm | <50µm ✓ |
| Maximum panel size | Per-unit cost | ~510×515mm at compromise yield | 600×600mm+ feasible |
| Dielectric loss @ 56GHz | Signal integrity at 224G SerDes | Df ~0.005 | Df <0.001 ✓ |
| TGV aspect ratio | Via density limit | n/a (organic uses other vias) | 10:1 today, 20-30:1 target |
| First-line yield | Commercial viability gate | >85% at scale | 30-50% today; >70% target by 2027 |
The metric to watch most closely is yield. Glass substrate is technically superior on every other axis but yield is what gates commercial deployment. Year-over-year yield improvement at the lead substrate maker (likely a Japanese name working with Intel) is the leading indicator for the industry’s commercial readiness.
| Type | Composition | Pros | Cons | Use |
|---|---|---|---|---|
| Borosilicate | Boron + silicate base | Workable, low CTE (~3-5 ppm/°C), reasonably priced | Some thermal limitations | Mainstream candidate |
| Fused silica | Pure SiO2 | Lowest CTE (~0.5 ppm/°C), highest thermal | Expensive, hard to manufacture in large panels | Premium; limited use |
| T-glass / specialty | Engineered low-CTE blends | Tunable properties | Proprietary recipes | Vendor-specific (Nittobo T-glass, etc.) |
| Q-Glass (per STF) | High-performance grade for AI substrate | Tightest tolerances | Limited supply | Vera Rubin generation |
Each formulation requires a separate qualification cycle at the substrate maker and the end customer. Q-Glass — flagged in STF Research’s Kitagawa Seiki coverage as the next-gen material — is the single most important grade to track for 2027-2028 deployment.
Glass substrate has been an academic and industry research topic since the 1990s. Several factors held it back: - No commercial driver. Organic substrate was good enough for CPUs and GPUs of the time. - No supply ecosystem. Glass cloth, glass panel, and TGV equipment were bespoke. - Yield economics didn’t work. First-prototype yield was 5-15%; commercial substrate yield needed >80%.
Three things changed in 2020-2025: 1. AI accelerator package areas tripled. The substrate-warpage problem became real. 2. High-speed SerDes pushed to 224G PAM4. Organic dielectric loss became unacceptable. 3. Intel made the public commitment. Intel’s GlassCore announcement in September 2023 committed development capital and supplier-ecosystem investment that wasn’t there before.
By April 2026, the chain of validation had built quietly. E&R Engineering’s TGV laser equipment was qualified by a Japanese substrate maker. Corning, AGC, Schott, NEG were all sampling glass to substrate makers. Intel was paying for T-glass and substrate as a reported gross margin headwind. The R&D phase has ended; the engineering and qualification phase is in flight.
PACKAGING GLASS SUBSTRATE VALUE CHAIN
====================================
LAYER 1 — Glass material producers
Premium glass: Corning (GLW), AGC (5201.T), Nippon Electric Glass (5214.T),
Schott (private), Hoya (7741.T) — adjacent (photomask glass)
↓
LAYER 2 — Glass panel + surface preparation
Panel cutting: Various Asian + European specialty firms
Surface prep: DNS Screen, Tokyo Electron, Applied Materials
↓
LAYER 3 — TGV equipment + laser drilling (★ HIGHEST-LEVERAGE BOTTLENECK)
Femto/picosecond laser: LPKF (LPK.DE), E&R Engineering (8027.TWO),
DISCO (6146.T) — adjacent (singulation),
MKS Instruments / Coherent (laser sources),
Trumpf (private)
↓
LAYER 4 — Via metallization + build-up
CVD/PVD/ECD: Applied Materials (AMAT), Lam Research (LRCX), TEL (8035.T),
Tango Systems (private)
↓
LAYER 5 — Substrate fabrication (★ PROCESS KNOW-HOW LAYER)
Glass-substrate development: Ibiden (4062.T), Shinko Electric (6967.T),
Unimicron (3037.TW), Nan Ya PCB (8046.TW),
Samsung Electro-Mechanics (009150.KS),
Daeduck Electronics (008060.KS), AT&S (ATS.VI),
Kinsus (3189.TW), Simmtech (KQ:222800)
↓
LAYER 6 — End customers
Intel, Nvidia, AMD, Google, Amazon, Microsoft, Meta — packaging through OSAT
partners (Amkor, ASE, JCET, Powertech)| Layer | Revenue pool ’27E (substrate-only, est.) | Margin profile | Concentration | Investability |
|---|---|---|---|---|
| 1 — Glass material | $0.5-1B | High (specialty glass = 30-50% GM) | Very high (Corning/AGC/NEG/Schott) | Diluted in big-cap names |
| 2 — Surface prep | small | Medium | Medium | Low — incidental capex |
| 3 — TGV equipment | $0.5-1.5B | High (capital equipment, 30-45% GM) | High — <5 qualified suppliers | ★ Best alpha layer (small caps) |
| 4 — Build-up equipment | shared with broader semicap | Medium | Medium-high | Diluted |
| 5 — Substrate fabrication | $5-10B by 2030 | Medium-high (25-40% GM if yield holds) | Medium (top 5 = 70%+) | Direct exposure |
| 6 — End customer | n/a (margin captured at die level) | n/a | High | Diluted |
The most under-priced single layer is Layer 3 — TGV equipment. Total revenue pool is small but margin is structural and the qualified-supplier list is short. E&R Engineering (8027.TWO) is the only public small-cap pure-play with validated TGV process.
The Layer 1 glass material producers (Corning, AGC, NEG, Schott) have meaningful revenue from adjacent business (display glass, optical fiber, photomask blanks, automotive glass) so packaging glass is incremental, not transformative, for these names.
| Company | Ticker | Position | Mkt Cap (USD est) | Pure-play? | Notes |
|---|---|---|---|---|---|
| Corning | GLW | US glass leader; OLED, optical fiber, Gorilla, photomask, packaging glass | ~$32B | No (diversified) | Likely a primary packaging glass supplier; reveals only at high level |
| AGC (Asahi Glass) | 5201.T | Japan glass + chemical conglomerate | ~$10B | No | Vault deep-dive at [[5201]]; flagship in display + photomask glass; emerging packaging glass position |
| Nippon Electric Glass (NEG) | 5214.T | Japan specialty glass | ~$3B | Mostly | Vault deep-dive at [[5214]]; T-glass position |
| Schott | private | German specialty glass | n/a | Mostly | Strong tech; private — no direct exposure |
| Hoya | 7741.T | Japan specialty glass — photomask blanks (EUV/DUV) | ~$30B | No (diversified) | Adjacent (photomask glass), some packaging glass exposure; vault [[7741]] |
| Company | Ticker | Position | Mkt Cap (USD est) | Pure-play? | Notes |
|---|---|---|---|---|---|
| E&R Engineering | 8027.TWO | TGV laser + plasma — validated for glass core 2024 | ~$0.5B | Yes for laser | Vault deep-dive at [[8027]]; Japanese substrate-maker customer; small-cap pure-play; only public name with validated TGV process |
| LPKF Laser & Electronics | LPK.DE | German laser equipment for PCB + glass | ~$0.3B | Yes | Long history in PCB laser; expanding into glass substrate |
| DISCO | 6146.T | World leader in semiconductor singulation + cutting | ~$25B | Yes | Glass singulation; strong adjacency; diversified across semicap |
| MKS Instruments | MKSI | Laser source + photonics; Coherent acquisition | ~$10B | No | Diversified |
| Coherent | COHR | Industrial / fiber laser | ~$10B | No | Diversified; sells to laser equipment makers |
| Trumpf | private | German industrial laser | n/a | Mostly | Premium lasers; private |
| Company | Ticker | Position | Mkt Cap (USD est) | Glass exposure |
|---|---|---|---|---|
| Ibiden | 4062.T | #1 ABF substrate (~30% global share) | ~$10B | Likely lead developer — Intel relationship; needs to migrate or risk displacement |
| Shinko Electric | 6967.T | #2 ABF substrate; FCBGA leader | ~$6B | Less public glass exposure; needs to follow Ibiden |
| Unimicron | 3037.TW | Taiwanese ABF substrate + advanced PCB | ~$10B | Uncertain — primarily ABF |
| Nan Ya PCB | 8046.TW | Taiwanese ABF substrate + PCB | ~$2.5B | Limited |
| Samsung Electro-Mechanics | 009150.KS | Korean substrate + components | ~$8B | Active glass core development; AI accelerator substrate ambitions |
| AT&S | ATS.VI | Austrian/European #5 ABF substrate; Kulim Malaysia ramp | ~$1.5B | AMD anchor; vault deep-dive at [[ATS]]; transition risk if glass displaces ABF |
| Daeduck Electronics | 008060.KS | Korean substrate maker | ~$1B | Some glass core development |
| Kinsus | 3189.TW | Taiwanese substrate | ~$2B | Limited |
Already covered in vault at [[8027]]. Summary points
specific to glass substrate thesis:
Ibiden is the most important name in the substrate maker tier because it is the global #1 ABF substrate maker (~30% share) and therefore has the most to lose from organic-to-glass displacement, the most to gain from owning the transition, and the deepest existing Intel relationship to make the migration commercially.
Same logic as Corning. Diversified, premium businesses, packaging substrate is a small slice of forward growth optionality. Best held as part of the broader specialty-glass + photomask + substrate complex (e.g., Hoya at [[7741]]) rather than as a glass substrate pure-play.
Vault deep-dive at [[ATS]]. Smaller substrate maker
(Europe’s only scaled ABF player), AMD anchor, Kulim ramp.
The glass core IC substrate TAM is small today (~$0-200M, mostly R&D/qualification volume) but projected to grow to $2-5B by 2030 if Intel’s commitment holds and competitors follow. The largest single-name beneficiary is unclear — it depends on whether glass substrate displaces ABF (substrate makers shift) or supplements it (parallel supply chain).
The TGV equipment TAM is even smaller (~$50-200M in 2027-2028) but projected to grow to $0.5-1.5B by 2030 with very high concentration (top 3 equipment makers >70% share).
Currently secular — the technology is in early adoption. Cyclical risk reappears once primary deployment is reached (likely 2028-2030).
Once a substrate maker qualifies on glass core for a specific customer (Intel, hyperscaler), switching costs are extremely high. The upfront qualification cost (12-24 months) creates structural lock-in for the early movers.
| Year | Status | Key milestone |
|---|---|---|
| 2024 | E&R TGV equipment validated by Japanese substrate maker | Equipment-side qualification |
| 2025 | Intel reports T-glass + substrate cost as gross margin headwind | Intel paying for material today |
| 2026 | Small production at Japanese substrate maker; first AI accelerator engineering samples | Sub-commercial deployment |
| 2027 | Commercial scaling at first substrate maker; 2-3 hyperscaler programs qualifying | Early commercial deployment |
| 2028 | Multiple substrate makers qualified; Vera Rubin Ultra possible glass package | Mainstream entry |
| 2029-2030 | Intel GlassCore production target; ABF/glass dual-track standard | Mass deployment |
The single most important tracking signal is first-production-line yield improvement at the Japanese substrate maker. Year-over-year yield % is a leading indicator for the entire industry’s commercial readiness.
| Layer | Smallest pure-play | Mkt cap | Concentration | Bypassable? | Market priced? |
|---|---|---|---|---|---|
| Glass material (specialty packaging-grade) | Hoya / NEG / AGC | $3-30B | High (<5 firms) | No | Partly priced (small as % of parent revenue) |
| TGV equipment | E&R Engineering (8027.TWO) | ~$0.5B | Very high (<5 qualified) | No | Under-priced for the option value |
| Build-up equipment | shared semicap | n/a | Medium | n/a | Priced |
| Glass-substrate fabrication | (no public small-cap pure-play) | various | Medium | No | Mixed |
“What is the $100M-$1B Mkt cap pure-play with unique exposure to the glass core packaging substrate transition that the market hasn’t priced?” → E&R Engineering (8027.TWO). Already in vault as [[8027]] with full deep-dive + management DD. Re-confirm thesis against this primer.
| Rank | Company | Ticker | Layer | Thesis | Risk | Timeframe |
|---|---|---|---|---|---|---|
| 1 | E&R Engineering | 8027.TWO | TGV equipment | Validated TGV → Japanese substrate maker → Intel; only public small-cap pure-play; option value | Loss-making; cash declining; single analyst | 18-36 months |
| 2 | Ibiden | 4062.T | Substrate transition leader | ABF re-acceleration + glass core option | Substrate cycle risk; transition execution | 18-36 months |
| 3 | Hoya | 7741.T | Specialty glass diversified | Photomask + emerging packaging glass exposure | Diversified; glass substrate is small slice | 24-48 months |
| 4 | NEG | 5214.T | T-glass + specialty glass | Lower-conviction Tier 2 glass material | Diversified; long-cycle | 24-48 months |
| 5 | Samsung Electro-Mechanics | 009150.KS | Substrate + glass core developer | Korean alternative to Japanese substrate makers | Korean conglomerate exposure | 24-48 months |
| 6 | LPKF | LPK.DE | Laser equipment | Smaller TGV exposure than E&R | Mobile/auto cyclicality drags | 18-36 months |
| 7 | AT&S | ATS.VI | ABF substrate; transition risk | AMD anchor + ABF re-acceleration | Glass migration could compress | 18-36 months |
Tier 1 — Core (asymmetric, real-option): - E&R Engineering (8027.TWO) — the cleanest single-name play on the TGV equipment layer; vault has full deep-dive + mgmt DD. Size as a real option, not a fundamental compounder.
Tier 2 — Tactical (diversified exposure): - Ibiden (4062.T) — substrate cycle re-acceleration with glass core optionality. - Samsung Electro-Mechanics (009150.KS) — Korean substrate alternative.
Tier 3 — Watchlist: - Hoya, NEG, AGC — only meaningful as part of broader specialty glass holdings; vault has [[5201]], [[5214]], [[7741]]. - LPKF (LPK.DE) — smaller relative play; less pure than E&R. - AT&S (ATS.VI) — substrate transition risk; better watched as ABF cycle play.
Avoid: - Pure ABF substrate names with no glass roadmap — long-term displacement risk if Intel commit accelerates. - Small Chinese substrate makers at any scale-up valuation — they cannot qualify on US-origin tooling and US-listed end customers in the current export control regime.
| Company | TGV equipment exposure | Glass material exposure | Substrate fabrication exposure | Total purity score |
|---|---|---|---|---|
| E&R Engineering | 9/10 | n/a | n/a | High purity, equipment layer |
| LPKF | 5/10 | n/a | n/a | Medium purity, equipment layer |
| Ibiden | n/a | n/a | 7/10 (transition leader) | Medium purity, substrate layer |
| Samsung E-M | n/a | n/a | 5/10 | Medium purity |
| Hoya | n/a | 3/10 | n/a | Low purity (diversified) |
stfbutnou.substack.com)
— Kitagawa Seiki coverage flags Q-Glass as Vera Rubin material; expect
packaging substrate coverage.collyerbridge.com) —
Intel supply chain ideas mirrored in vault; tracks substrate / glass
headwinds.KB/wiki/8027/8027.md — E&R
Engineering deep-dive (TGV laser equipment, Japanese substrate-maker
customer, Intel indirect glass core exposure)KB/wiki/8027/8027-mgmt-dd.md —
December 2025 earnings call corrections (TGV customer = Japanese, not
Intel directly)KB/wiki/ai-server-pcb-primer.md
— companion primer; PCB layer below substrateKB/wiki/ATS/ats-deep-dive.md —
AT&S ABF substrate maker (transition risk perspective)KB/wiki/5201/,
KB/wiki/5214/, KB/wiki/7741/ —
adjacent specialty glass deep-divesKB/raw/substack-archive/stf-research/2026-03-16-kitagawa-seiki-hidden-champion-pressing-ai-era.mdKB/raw/substack-archive/illyquid/2026-04-26-intel-supply-chain-ideas.md
(jukan05 quote on Intel substrate / T-glass cost headwind)Pre-delivery checklist: redundancy sweep ✓ (cut two repeated yield-economics paragraphs); word justification ✓ (every value-chain table earns its space); guide pass ✓ (Register D — investment writeup; em-dashes used per Register D as needed; concentrated technical content but accessible to the patient reader). Length is shorter than the AI Server PCB primer (40-45% length) because (a) the industry is earlier-stage and there is less to map, and (b) the bottleneck identification (E&R + glass material parents) emerges quickly. Pink instructed “no length limit; thoroughness beats brevity” — calibrated thoroughness for a smaller industry.