Skip to content
primarysourced Photonics sector Tower
TSEM
~9 min read · 2,005 words ·updated 2026-04-29 · confidence 60%

Tower-ST Agrate R3 — the 300mm specialty-analog joint

Updated: 2026-04-29 Status: ✓ Verified via Tower IR press release (June 24, 2021), STMicroelectronics newsroom mirror, eeNews Europe, and Tower’s annual Form 20-F filings. Capex / wafer-output-by-quarter granularity flagged ⚠ where Tower / ST do not break it out separately. Cross-references: Technology overview · PH18 process · Newport Beach + Maxim history · overview

Executive summary

On June 24, 2021 (Tower IR ✓; STMicroelectronics newsroom ✓; Tower IR mirror ✓), STMicroelectronics and Tower Semiconductor announced an agreement under which Tower would join ST inside its Agrate R3 300mm wafer fab then under construction at the Agrate Brianza site in Lombardy, Italy — about 25 km northeast of Milan.

The structure is not a joint venture in the equity sense — it is a shared-cleanroom, capacity-extension arrangement:

  • ST owns and operates the R3 facility ✓
  • Tower installs its own equipment in approximately one third of the total cleanroom space ✓
  • ST manages facility operations; selected Tower personnel are seconded to ST roles supporting qualification, ramp-up, engineering, and process functions ✓
  • Tower establishes a wholly-owned Italian subsidiary to operate its share of the Agrate R3 capacity ✓

The headline strategic outcome at the time of announcement was that Tower would more than triple its 300mm foundry capacity through this arrangement (Tower CEO statement, 2021-06-24) ✓. The original ramp window was R3 equipment install in late 2021, production start in the second half of 2022 (Tower IR release) ✓. Public reporting since suggests the actual ramp slipped relative to the H2 2022 target, with full process qualification working through 2024-2026; ⚠ Tower’s annual 20-F filings are the cleanest source for the realized ramp cadence and should be cross-checked for each fiscal year.

The Agrate R3 site is positioned for smart-power, analog mixed-signal, and RF process flows at 130 nm / 90 nm / 65 nm nodes (Tower IR) ✓ — automotive, industrial, and personal-electronics end markets. This is not a silicon-photonics asset; the photonics-relevant TSEM platform remains PH18 at Newport Beach (and, since November 2024, the new 300mm SiPh process hosted at Migdal Haemek). For the AI-photonics scope of this KB, Agrate R3 matters as balance-sheet capex context (the 300mm capex is mostly going into Agrate, not into photonics) and as diversification context (Tower’s 65 nm BCD platform serves the broader specialty-analog cycle).

1. Deal structure

1.1 Counterparty arrangement

ElementDetailSource
Announcement dateJune 24, 2021Tower IR press release ✓
SiteAgrate R3 (the third Agrate fab at the Agrate Brianza, Lombardy, Italy site)STMicroelectronics newsroom ✓
Wafer size300 mmTower IR ✓
Cleanroom shareApproximately one third Tower; balance STTower IR; eeNews Europe ✓
Operating partyST operates the facility; Tower personnel seconded to specific ramp / engineering rolesTower IR ✓
Tower legal vehicleWholly-owned Italian subsidiaryTower IR ✓
Equity structureNot a JV in the equity sense — capacity-share / co-investment arrangementeeNews Europe ✓

⚠ The economics of the capacity-share arrangement — whether Tower is paying ST a fab-services fee, whether there is a buy-out option on the cleanroom share, and how shared infrastructure costs (gases, water, FAB-wide utilities) are allocated — are not publicly disclosed at counterparty granularity. Tower’s annual 20-F discloses aggregate capex contributions and investment commitments at the Agrate vehicle level.

1.2 Tower’s investment

Tower’s investment is CapEx-led rather than via cash payment to ST. Tower committed to fund and install its share of the equipment in the R3 cleanroom. ⚠ The exact dollar amount of the Tower commitment was not specified in the June 2021 press releases; it has been estimated at the multiple-hundreds-of-millions-of-USD scale based on standard 300mm fab tool-cost economics. Tower’s Form 20-F annual filings disclose annualized capex figures by site that allow back-out of the Agrate cumulative spend.

The June 2021 release framed Tower’s strategic interest as:

“Tower’s 300mm foundry capacity will be more than tripled by this activity in Agrate.” — Tower IR press release, June 24, 2021 (towersemi.com) ✓

This is the most concrete capacity disclosure Tower has made. ⚠ The pre-Agrate 300mm baseline that this triples is presumably the TPSCo Arai 300mm capacity (Niigata, Japan, JV with Nuvoton, ex-Panasonic Semi); a tripling implies the Agrate share alone is roughly 2× the TPSCo Arai contribution at full ramp.

2. Node and process focus

The R3 fab is positioned for the specialty-analog process portfolio at the 130 nm / 90 nm / 65 nm nodes — not for advanced logic. Process flows confirmed for qualification on the line (Tower IR press release ✓):

NodeProcess focusEnd markets
130 nmSmart-power, BCD (bipolar-CMOS-DMOS)Automotive power management, industrial, personal-electronics
90 nmMixed-signal CMOS, analogAutomotive, industrial, personal-electronics
65 nmRF / mmWave, advanced BCD, analog mixed-signalRF / 5G fronthaul, automotive radar, advanced power management

The 65 nm node is the load-bearing one for Tower’s strategic positioning. In December 2024, Tower released its 300mm 65 nm 3.3 V BCD power-management platform (Tower IR, 2024-12-23) ✓ — a productized process flow for 300mm 65 nm power-management ICs. This is the first major productization milestone of the Agrate share, indicating that the line was qualified and producing in customer-ready form by late 2024. The realized ramp window vs the original H2 2022 target is therefore approximately 18-24 months delayed ◐ — consistent with industry-wide 300mm fab build-out / equipment-delivery slippage during 2021-2024.

⚠ The R3 cleanroom space and equipment assembly was first equipped during 2021-2022; first-wafer-out / pilot production likely in 2022-2023 with full process-flow qualification in 2024. The exact “first commercial wafer” date is not separately disclosed — Tower’s 20-F filings phrase progress in qualitative ramp language.

3. Capacity at full ramp

⚠ The full-ramp capacity of Tower’s Agrate share has not been disclosed at wafer-per-month granularity in primary sources. The “more than tripled 300mm capacity” framing is the only public quantitative reference.

A reverse-engineering pass:

  • TPSCo Arai 300mm pre-Agrate: ⚠ ~30,000-40,000 wafers per month is the public-aggregator estimate (not Tower-disclosed)
  • Tripling of 300mm capacity implies Tower’s total 300mm reach at Agrate full ramp is in the range of ~90,000-120,000 WPM, of which Agrate contributes the majority
  • Tower’s share of total R3 cleanroom (one third) at typical 300mm fab cleanroom capacity (~30,000-50,000 WPM at full equipment fill for the entire R3 building) implies Tower’s Agrate slot is plausibly ~10,000-17,000 WPM at full ramp

⚠ All numerical capacity estimates above are aggregator-level and unverified against Tower / ST primary disclosure. The Tower 20-F segment reporting is the cleanest authority but does not break out Agrate-specific WPM.

4. Business-model differentiation

The Agrate arrangement is structurally different from typical foundry capacity acquisitions in three ways:

4.1 Capacity-share vs full ownership

Tower does not own R3. Tower owns its equipment inside R3 and contracts with ST for the facility services (cleanroom utilities, tool maintenance, gas / chemical supply). This is a lower-capital-intensity path to 300mm scale than building a full new fab — Tower avoids the multi-billion-dollar fab-shell construction cost and rides on ST’s existing facility infrastructure.

4.2 Counter-party operational risk

Because ST operates the facility, Tower’s customers tape out on a fab that is majority-operated by a competitor (ST also runs analog / power businesses). The mitigation is firewall protocols on customer IP and process recipes, plus the segregated equipment-ownership structure that Tower seconds engineering personnel to oversee. ⚠ How the firewall is enforced operationally, and whether there has been customer pushback on the arrangement, is not publicly disclosed.

4.3 Strategic alignment to the specialty-analog cycle

Agrate’s node focus (65 nm BCD + RF) is explicitly specialty-analog, not advanced logic. This aligns Tower’s 300mm capex to the same end-markets (automotive, industrial, RF) that Tower’s existing 200mm portfolio serves, rather than chasing the logic node-shrink curve. The strategic differentiation vs TSMC / Samsung / Intel is therefore node-mature 300mm scale at specialty-analog process flows — which is exactly the white space that GlobalFoundries’ 22FDX / 12LP+ / 45RFSOI portfolio also targets.

5. Timeline of milestones

DateMilestoneSource
2021-06-24Announcement of the Tower-ST Agrate R3 capacity-share arrangementTower IR ✓
H2 2021 → 2022R3 cleanroom built out; equipment installation beginsTower IR ✓
H2 2022 (originally targeted)Production startTower IR ✓
2022-2023 (realized)First-wafer-out / pilot production; ramp slippage relative to original target⚠ Inferred from 20-F qualitative disclosure
2024-12-23Tower 300mm 65 nm 3.3 V BCD platform released as standard offering — first productized Agrate-share process flowTower IR ✓
2024-2025Continued 65 nm ramp; full process-flow qualification⚠ 20-F disclosure granularity
2025-2026Full-ramp progression; Tower’s 300mm contribution to total revenue rising from low-single-digit % toward mid-teens % share⚠ 20-F segment reporting needed for confirmation

6. Strategic context — what Agrate means for the AI-photonics thesis

For the photonics-research scope of this KB, Agrate matters in three indirect ways:

  1. Balance-sheet absorption — Tower’s biggest 300mm capex commitment is going into Agrate, not into photonics. This affects how much capex headroom Tower has to expand the new 300mm SiPh process at Migdal Haemek (announced 2024-11-26) ◐ — the more Agrate consumes, the less is available for photonics expansion.

  2. Diversification cushion — the Agrate 65 nm BCD ramp gives Tower a non-photonics revenue lever that grows independently of the AI-photonics cycle. For an investor sizing TSEM’s photonics exposure, the Agrate ramp is a counter-cyclical diversifier — the BCD power-management business serves auto/industrial cycles that are decorrelated from datacenter optics.

  3. Capacity flexibility — Tower’s 300mm footprint at Agrate is dominantly for analog/RF/BCD, but the same 300mm site capability gives Tower the option to migrate photonic-process flows onto 300mm in Italy at a future date. ⚠ No public Tower disclosure suggests this is on the roadmap; the new 300mm SiPh process is hosted at Migdal Haemek, not Agrate.

7. Open audit items

  1. ⚠ Exact Tower CapEx commitment to Agrate — multi-year cumulative dollar figure. 20-F annual filings have aggregate capex but not site-specific breakouts at full granularity.
  2. ⚠ Realized first-wafer-out date at Agrate — the H2 2022 target was original; actual ramp dates need to be reconstructed from 20-F language across multiple fiscal years.
  3. ⚠ Wafer-per-month capacity at full ramp for Tower’s Agrate share — the “more than tripled 300mm capacity” framing is the only quantitative anchor.
  4. ⚠ Customer roster specifically for the Agrate-served products — not separately disclosed; bundled with Tower’s broader power-management / RF customer base.
  5. ⚠ Operational firewall protocols between Tower and ST on shared cleanroom IP — not publicly disclosed.
  6. ⚠ Tower’s option / exit terms for the Agrate arrangement — not disclosed at counterparty-agreement level.
  7. ⚠ 65 nm BCD process competitive positioning vs the GF 22FDX / 12LP+ / 45RFSOI portfolio at equivalent node — needs side-by-side device-spec comparison.

Cross-references

Sources