Crystal growth & ingot (Czochralski/Float-Zone) (process step)

Step 16 of 56 in the semiconductor flow (Wafer). Prev: Polysilicon (electronic-grade silicon) (process step) · Next: Wafer slicing, lapping & polishing (process step)

What this step does

This step turns purified polysilicon feedstock (the 9N+ chunks from the prior step) into a single, defect-free silicon crystal: a cylindrical ingot that is one continuous crystal lattice. Almost all device-grade silicon, and effectively all 300mm wafers, is grown by the Czochralski (CZ) method; Float-Zone (FZ) is a smaller, premium route for high-resistivity power and detector wafers. Sub-processes inside the step:

• Melt the polysilicon in a quartz crucible (CZ) or melt a moving zone of a polysilicon rod with an RF coil, crucible-free (FZ).
• Dip a seed crystal, then pull and rotate it slowly so the melt solidifies onto the seed in a single orientation.
• Control diameter, dopant level (resistivity) and oxygen/defect content; crop and grind the boule to the target diameter.

Where it sits and why it matters

Crystal growth sets the substrate quality that every downstream transistor inherits: orientation, resistivity, oxygen content, and crystal-defect density. It is the gate between commodity polysilicon and a value-added wafer. Crucially, the conversion of polysilicon to finished wafer costs roughly $22 to $26 per square metre, split about evenly between crystal growth and the later wafering/polishing steps [1]. So crystal growth is on the order of half the value added between feedstock and bare wafer, even though the bare-wafer market itself (~$12-13bn of shipments) is small relative to the $600bn+ chip industry it feeds.

Equipment market

The honest read: published “crystal furnace” TAMs are unreliable because they fold in solar PV pullers, which dwarf the semiconductor pool. The widely cited “monocrystalline silicon furnace” market of ~$3.4bn in 2024 (12.9% CAGR to ~$7.0bn by 2030) is overwhelmingly solar [2]; the semiconductor-grade CZ/FZ puller pool is a fraction of that, plausibly low-hundreds of millions of dollars a year of new-tool revenue (rough estimate, no clean third-party number isolates it). PVA TePla’s whole Semiconductor Systems segment, of which crystal-growing systems are one product line, was EUR 187.6m in 2024 [3], which frames the order of magnitude.

Concentration note: for semiconductor-grade silicon, value is concentrated in a handful of Western/Japanese specialists (PVA TePla, Ferrotec, Linton, Cyberstar, Mitsubishi); for solar and SiC, Chinese vendors (JSG, NAURA) dominate by volume. The two markets share furnace concepts but are commercially separate.

Materials & consumables

Each CZ pull consumes the polysilicon charge plus a hot zone of consumables that wear out and recur:

• Fused-quartz crucibles (the melt vessel) - single-use or few-use, 14 to 36 inch diameters. Lead supplier Momentive Technologies (US); also Ferrotec and several Asian makers [5].
• Graphite hot-zone parts: heaters, susceptors, heat shields, insulation. Lead suppliers SGL Carbon (Germany), Toyo Tanso, Tokai Carbon [6]. Recurring-revenue note: consumables are the steadier annuity in this step. Tool sales are cyclical and capex-gated, but every wafer maker burns crucibles and graphite continuously, so the consumables layer (Momentive, SGL) has more durable, less cyclical revenue than the puller OEMs.

Volumes, revenue, profitability

Global silicon wafer shipments were ~12,178 MSI (million square inches) in 2024, recovering toward ~13,328 MSI in 2025 [7] - these are the wafers that crystal growth feeds. The bare-wafer market is ~$12-13bn of revenue, and crystal growth is roughly half the polysilicon-to-wafer conversion cost [1]. Margins by layer: the value-capturing layer here is mostly the wafer makers themselves (they grow in-house), not the tool vendors. As a proxy for the equipment layer, PVA TePla ran a 21.4% group EBITDA margin in 2024 (EUR 40.2m EBITDA) [3] - healthy for capital equipment but well below front-end litho/etch OEMs. Wafer makers (Shin-Etsu, SUMCO) earn the substrate margin; their economics swing hard with the silicon cycle.

Competitive landscape & value capture

The defining structural fact: crystal growth is mostly captive. The big wafer makers - Shin-Etsu (~32-33% of 300mm) and SUMCO (~26-27%) [7], plus GlobalWafers, Siltronic, SK Siltron - run their own crystal pullers as a core trade secret. They buy some furnaces but also build and tune their own hot zones, so the merchant puller-OEM revenue pool is thinner than the wafer market implies. Value accrues to (a) the wafer makers, who hold the process know-how and the consolidated, oligopolistic market, and (b) the consumables suppliers who feed every pull. FZ is a defensible premium niche: FZ wafers cost ~30% more than CZ and serve high-voltage power devices (IGBTs, EV inverters), growing ~8% a year [8]. EU/specialty angle: PVA TePla (DE, listed) is the clear European pure-play and the go-to for SiC and specialty-crystal pullers, which matters for the power-electronics and compound-semi build-out.

Market drivers, constraints & trends

Net read: the served market is cyclical with a structurally higher floor. AI has pulled 300mm wafers from glut to tightness, but the value still accrues to captive wafer makers, and a 2027 correction is the consensus base case.

• Drivers

• AI/datacentre demand has flipped the cycle: Q1 2026 silicon wafer shipments hit 3,275 MSI, up 13.1% year-on-year, led by 300mm for AI logic and HBM [9][10]. SemiAnalysis frames the industry as now in a genuine “silicon shortage” phase, with hyperscaler capex gated by fab/wafer supply, not appetite [11].

• Node transitions lift wafer spec and price: Shin-Etsu and SUMCO put in JPY 150bn (~$1bn) in 2025 for 200,000 wpm of ultra-flat 2nm/3nm-grade 300mm capacity, and SUMCO is killing 200mm at Miyazaki by late 2026 to shift to AI-grade 300mm [12].

• Reshoring/sovereignty adds greenfield demand: GlobalWafers opened Sherman, Texas (now a $7.5bn programme) and started Italy’s first 300mm plant in Oct 2025 on multi-year STMicro/Infineon deals [12].

• Constraints

• Hard cyclicality: SEMI’s base case is +5.1%/+5.4% in 2025/26, a -6.2% correction in 2027, recovery in 2028 [13] — the served market is not a smooth compounder.

• ASP erosion: all five incumbents have secured greenfield capacity and, under depreciation pressure, will run it hard; Chinese suppliers pivoting from 200mm to 300mm add structural price pressure [13].

• Captive value capture: the big-5 hold ~85% of 300mm capacity and grow crystals in-house as a trade secret, so the merchant puller-OEM pool stays thin even as wafer demand climbs [12].

• Trends & inflections to watch

• Silicon-shortage re-rating: substrate scarcity could re-price the wafer layer above its historic cyclical band. Tripwire: SEMI quarterly shipments holding double-digit YoY growth into the back-half of 2026 [9][10], rather than rolling over toward the forecast 2027 dip [13].

• SiC to larger diameters: the 150mm to 200mm shift is underway and Wolfspeed demoed a single-crystal 300mm SiC wafer [14], lifting demand for specialty pullers (PVA TePla). Tripwire: first volume 200mm SiC supply contract or a foundry committing to 300mm SiC.

• China 300mm entry: Chinese wafer capacity moving into 300mm is the clearest downside catalyst. Tripwire: a Chinese maker qualifying 300mm at a non-Chinese logic/memory fab [13].

Connections

Materials & Process · thesis: Specialty Silicon Non Leading Edge

Sources

1. Cost to convert polysilicon to wafer ~$22-26/m2, split ~50/50 between crystal growth and wafering - cited via wafer-cost discussion: https://www.zmsh-semitech.com/the-300mm-silicon-wafer-manufacturing-process-crystal-growth-slicing-and-polishing/ (rough/old industry figure)
2. Monocrystalline silicon furnace market ~$3.36bn 2024, 12.9% CAGR to ~$6.97bn 2030 (solar-heavy); vendor list incl. JSG, NAURA, PVA TePla, Ferrotec, Linton: https://www.marketresearchreports.com/lpi/global-czochralski-monocrystalline-silicon-furnace-market-growth-2024-2030 (report-mill, treat as directional)
3. PVA TePla FY2024: Semiconductor Systems segment EUR 187.6m; group EBITDA EUR 40.2m, 21.4% margin: https://www.pvatepla-cgs.com/en/news/news-detail/news/pva-tepla-with-slight-sales-growth-and-robust-earnings-increase-in-2024/
4. Linton Crystal Technologies, CZ growing equipment: https://www.lintoncrystal.com/products/cz-growing-equipment
5. Momentive Technologies quartz crucibles for silicon crystal growth (14-36 inch): https://www.momentivetech.com/products/crucibles
6. SGL Carbon specialty graphites for semiconductor crystal growth (heaters, susceptors, heat shields): https://www.sglcarbon.com/en/markets-solutions/applications/semiconductor-crystal-growth/
7. Silicon wafer shipments ~12,178 MSI 2024, ~13,328 MSI 2025; Shin-Etsu ~32-33%, SUMCO ~26-27% of 300mm: https://marklapedus.substack.com/p/silicon-wafer-market-upturn-higher
8. Float-Zone silicon wafers ~30% costlier than CZ, ~7.9-8.5% CAGR, power-device driven: https://www.indexbox.io/blog/float-zone-silicon-wafers-market-forecast-points-higher-toward-2035-driven-by-power-semiconductor-demand/
9. SEMI: Q1 2026 silicon wafer shipments 3,275 MSI, up 13.1% YoY on AI datacentre demand: https://ninescrolls.com/news/semi-q1-2026-silicon-wafer-shipments-hit-3-275-msi-up-13-1-year-on-year-on-ai
10. AI and 300mm demand drive 2025 silicon wafer growth (5.4% to ~13,076 MSI, 300mm +7%): https://www.semiconductor-digest.com/ai-and-300mm-demand-drive-2025-silicon-wafer-growth/
11. SemiAnalysis “The Great AI Silicon Shortage” — hyperscaler capex gated by silicon supply: https://newsletter.semianalysis.com/p/the-great-ai-silicon-shortage
12. Wafer-maker capacity moves: Shin-Etsu/SUMCO JPY 150bn for 2nm/3nm ultra-flat 300mm; SUMCO exits Miyazaki 200mm by late 2026; GlobalWafers Sherman ($7.5bn) + Italy 300mm; big-5 ~85% of 300mm: https://www.mordorintelligence.com/industry-reports/semiconductor-silicon-wafer-market
13. SEMI 2025 silicon wafer market — cyclical limits vs structural change; +5.1%/+5.4%/-6.2%/+9.8% 2025-28; ASP erosion, China 300mm pivot: https://www.semi.org/en/blogs/2025-silicon-wafer-market-at-the-threshold-between-cyclical-limits-and-structural-change
14. Wolfspeed FY2026 8-K — 150mm to 200mm SiC transition; single-crystal 300mm SiC wafer demonstrated: https://www.sec.gov/Archives/edgar/data/0000895419/000089541926000012/ex991q2-26.htm