How Does SMIC Treat Wastewater at Its Fab Plants? 2026 Engineering Guide

Why SMIC's Wastewater Train Looks the Way It Does

A 300 mm fab can pull millions to tens of millions of gallons of water per day, and the UPW-reject stream is typically the single largest wastewater source on site — even before the process drains are mixed in (Samco, 2025). SMIC's mainland-China fabs at Shanghai, Beijing, and Shenzhen sit inside a regulatory envelope that is fundamentally different from the US 40 CFR 469 baseline: they discharge against GB/T 31962 and local integrated wastewater standards such as DB31/199-2018 for Shanghai, with tighter local caps on ammonia nitrogen in many jurisdictions and fluoride typically held below 10 mg/L (per the relevant GB integrated discharge standards). That regulatory framing forces the unit-process envelope to be tighter on nitrogen and fluoride than the US POTW pass-through case, which is why source segregation, biological nitrification, and fluoride precipitation are non-negotiable legs of any SMIC-class train.

SMIC publishes a net-zero water and ESG framing, but the company has not disclosed fab-level reuse rates at the same granularity as Intel Ocotillo. The honest starting point is that SMIC's specific numbers are inferred from peer fabs and reference installs — Intel's 2019 Circularity in Semiconductor Manufacturing white paper, the EPA Region 9 case study at Ocotillo (2017), the IDE Tech commercial install, and the HydropureWater fab guide (2025). What is not inferred is the source-segregation principle: lithography solvents, metal/CMP, acid/alkali, fluoride, ammonium/TMAH, and UPW reject are kept apart because end-of-pipe mixing produces a combined stream that is simultaneously high-TDS, high-organics, high-fluoride, and high-metals — and therefore very expensive to treat as one. The boundary condition that drives every unit process below is this: the train is shaped by the Chinese regulatory frame, the northern-China water-stress context, and the fact that end-of-pipe blending has been demonstrated to fail at fab scale.

The Six Waste Streams a SMIC-Scale Fab Has to Manage

Before naming any unit process, an engineer needs the working taxonomy. A SMIC-class 300 mm fab generates six distinct streams that are kept segregated at the source header: acid/alkali from wet and RCA cleans, fluoride-bearing HF and BOE etchants, CMP waste from slurry handling, TMAH/ammonium developer and CVD effluent, specialty organics from lithography solvents, and the high-volume UPW reject and cooling-tower blowdown. The first five are the "process drains" that need targeted chemistry; the sixth is the bulk rinse water that drives the reuse economics.

The influent envelope is well-characterized across the fab-equipment literature. HF wastewater runs 100–5,000 ppm F⁻ at pH 2–4, CMP waste carries 100–1,000 mg/L silica and 50–500 mg/L TSS, TMAH developer effluent sits at pH 12–14 with 50–200 ppm TMAH, and UPW reject arrives at 50–200 µS/cm conductivity, 1–5 mg/L TOC, 500–2,000 mg/L TDS, and 10–50 mg/L phosphate, contributing roughly 2–10 GPM per tool and hundreds to thousands of GPM fab-wide (HydropureWater fab equipment guide, 2025).

The practical implication: mixing any two of these streams creates a downstream problem the unit processes cannot fix cheaply. That is why every header on a SMIC-class fab is segregated at the source valve box, not at the central treatment header.

Fluoride Removal: Ca(OH)₂ Precipitation and CaF₂ By-Product Recovery

The fluoride leg is the most distinctive chemistry in a fab train and the one where the 1.2:1 molar ratio of Ca(OH)₂ to fluoride is the single most-cited design parameter. Dosing at that ratio precipitates CaF₂ to roughly 95% removal, with effluent typically below 10 ppm fluoride — matching Taiwan's regulatory limit and the relevant Chinese GB integrated discharge cap (HydropureWater fab equipment guide, 2025). Reactor retention is short: 20–40 minutes is the published envelope, followed by a lamella clarifier for CaF₂ and CMP solids or a DAF for CMP and fluoride sludge with surface loading 20–40 m³/m²/h.

Sludge handling matters as much as the chemistry. A plate-and-frame filter press dewateres the CaF₂ cake to a 5–10% influent-volume sludge yield, which is the same band the HydropureWater field data (2025) reports for the precipitation leg. The recovery question is where the comparison to peer fabs becomes useful: Intel Ireland diverts CaF₂ to cement-product manufacturing, and FAB28 at Qiryat Gat uses it as alternative daily cover at a local landfill. A SMIC-scale fab in northern China generates comparable CaF₂ tonnage and has the same recovery case, particularly given that cement kilns and engineered landfill sites are within economic haul distance of both the Beijing and Shanghai footprints.

Stable chemistry at the head of the leg requires PLC-controlled chemical dosing for pH and Ca(OH)₂ control: a 1.2:1 ratio is tight enough that manual dosing will under- or over-shoot during influent swings, and over-shooting simply wastes lime while under-shooting fails the discharge cap. The parameter envelope below is the working baseline for any SMIC-class fab.

CMP and Copper-Bearing Streams: pH Balance, IX, and Media Filtration

CMP and copper-bearing waste is segregated at the source valve box and stabilized in the pH window before ion exchange or precipitation can do its work. The control loop is a PLC-controlled chemical dosing for pH and Ca(OH)₂ control skid holding pH in a tight band; a 0.5-unit pH excursion at the head of this leg will change downstream resin capacity by 20–30% and shift the metal speciation in ways the polishing step cannot correct. Intel FAB28 in Qiryat Gat segregates copper-bearing waste into a dedicated pretreatment system, and Intel Oregon added on-site electroplating so that scrap copper is sold rather than hauled — sending more than 600 tons of plating-bath waste per year back to the commodity market (Intel Circularity white paper, November 2019).

Two unit processes dominate the bulk metal cut. Ion exchange delivers >99% Cu/Ni removal at 1–3 eq/L resin capacity, regenerated with 4–6% HCl/NaOH; precipitation with Na₂S achieves >90% metal removal but generates more sludge (HydropureWater fab equipment guide, 2025). Downstream, a multi-media filter takes out residual solids, and the colloidal silica load — 100–1,000 mg/L — is handled with a DAF for CMP and fluoride sludge or lamella clarification before any RO leg, because silica is the dominant foulant on the downstream membrane and the single biggest cause of CIP frequency in fab RO service.

The economic framing is straightforward: copper is a recoverable commodity, not a disposal cost. A SMIC-class fab in Shanghai's Lingang or Beijing's Yizhuang generates the same copper tonnage as a peer Intel site, and the recovery case — selling back to a commodity buyer at LME-linked pricing — is exactly the line that converts a pretreatment cost into a circular-economy credit.

TMAH, Ammonium, and the Organic Load: MBR as the Workhorse

TMAH developer waste arrives at pH 12–14 with 50–200 ppm TMAH, and at that alkalinity the compound is toxic to most conventional biological cultures. The first step is aggressive pH adjustment down to 7–8, then an integrated MBR for fab organic and TMAH streams at 12–24 h HRT removes up to 99% of the TMAH through biodegradation (HydropureWater fab guide, 2025). The performance envelope is well documented: 99% TSS removal, 92–97% COD removal at 10–20 LMH flux, with PVDF membrane life of 12–24 months when the air-scour duty is held in spec. MBR membrane bioreactor modules designed around that envelope are the operating point most Chinese fab MBRs are built to.

The ammonium leg is where by-product recovery shows up. The ammonium sulfate by-product is diverted to fertilizer manufacturing — Intel Oregon and Arizona have been doing this since 2013 and have accumulated more than 35,000 tons; Ireland has since followed (Intel Circularity white paper, November 2019). For a SMIC-scale fab, the local fertilizer market in mainland China absorbs comparable tonnage, and the line is the difference between a compliance cost and a circular-economy credit. The HydropureWater field data (2025) shows PVDF membrane life holding at the upper end of the 12–24 month band when vigorous air scour is maintained; fabs that under-spec the scour duty typically see that number fall toward 12 months or less, which is a 50% swing in membrane OPEX.

The IDE Tech commercial install is the downstream anchor. The MAXH2O Pulse Flow RO system treats ~720 GPM (~4,000 m³/day) of cooling tower blowdown and MBR effluent at 54% recovery, silica-limited, with an 18% gain over a conventional RO that could not run reliably on the same feed (IDE Tech project page, accessed 2026). With a brine-polish step, the same install reaches 88% overall recovery — a real-world data point showing the RO leg alone is silica-limited and that the reuse uplift comes from the upstream MBR doing its job on organics and from the downstream brine stage finishing the recovery curve.

UPW Reject and RO Reuse: Closing the Loop at 70–85% Recovery

UPW reject is the highest-leverage stream on a fab site. At 2–10 GPM per tool, accumulating to hundreds or thousands of GPM fab-wide, with conductivity 50–200 µS/cm and TOC of 1–5 mg/L, it is the largest single wastewater source on site (Samco, 2025) and the prime RO reuse candidate. The standard pretreatment is a hollow-fiber UF as RO guard filtration, holding SDI below the threshold the RO membrane needs to run at high recovery.

The RO duty is well characterized: 95–98% TDS removal, 70–85% recovery, 15–25 LMH flux, 2–5 mg/L anti-scalant dosing, and 3–5 year membrane life when CIP is run on a defined cycle (HydropureWater fab equipment guide, 2025). The industrial RO system for UPW reject reuse is tuned against silica scaling — the IDE Tech PFRO at 54% recovery is silica-limited, and pushing past that requires either a brine-stage polish (88% overall at the IDE install) or a ZLD finish.

Where the reuse header demands UPW-grade water, a polishing ion-exchange or EDI step goes downstream of the RO. Where it does not — scrubbers, cooling towers, landscape — the RO permeate can go directly to the reuse header, and at Ocotillo it goes to drinking-quality aquifer recharge (US EPA Region 9 case study, 2017 snapshot). The 2026 industry context is moving in the same direction: Samsung's 330,000 t/day fab reuse announcement at KIWW 2026 and SK Hynix's 2026 water recycling performance confirm the direction of travel.

SMIC-Class Fab vs. Intel Ocotillo: A Reuse and Throughput Benchmark

The defensible comparison point for any SMIC-class fab is the EPA-documented Ocotillo campus in Chandler, Arizona: ~75% reuse, ~5.2 MGD conserved, and 3.5+ billion gallons of drinking-quality water recharged to the underground aquifer since the Chandler RO Recharge Facility began operation (US EPA Region 9 case study, 2017 snapshot). Ocotillo is the reference design not because it is the highest reuse number in the industry but because the EPA has independently verified both the rate and the recharge volume, and because the site sits in a water-stressed watershed where reuse is a permit and supply necessity rather than a CSR choice.

Samco (2025) puts the industry baseline at ~60% recovery, achieved by re-routing spent streams to less exacting uses (cooling, scrubbers). SMIC-class fabs in northern China should be measured against the Ocotillo envelope, not the Samco average, because the regulatory and water-stress context is closer to Arizona than to the Samco baseline. The IDE Tech commercial install at ~720 GPM and 54% RO recovery (88% with brine polish) is the real-world anchor showing what a well-designed RO leg can do on fab feedwater, even with silica-limited feed.

The honest gap: SMIC has not published fab-level reuse numbers at the same granularity as Intel Ocotillo, so the comparison has to be drawn from peer fabs, vendor reference installs, and the equipment-performance literature — and the article should say so explicitly to preserve credibility. For a deeper read on the Intel side of the comparison, see Intel's 2026 fab wastewater process train, and for the equipment side, the chip fab wastewater treatment equipment specs for 2026.

Equipment Selection, CAPEX, and Lead Time for a SMIC-Scale Build

Two procurement paths cover the SMIC-class envelope. The modular skid path runs $1.2M–$3M CapEx for 100–300 GPM, with OPEX at $0.20–$0.40 per 1,000 gallons treated and 12–18 month lead time — the right choice for phased fab ramp or pilot-scale validation (HydropureWater field data, 2025). The centralized path runs $3M–$8M CapEx for 500–2,000 GPM, with OPEX at $0.10–$0.25 per 1,000 gallons and 24–36 month lead time — the right choice once the fab is in volume production and water is a permit-binding constraint. Payback on either path is 2–5 years at 70–90% reuse; a $2M CapEx system saving $500K/year in water and hauling clears in roughly 4 years.

For mainland-China projects, the sourcing and after-sales question is decisive. Lead times and equipment sourcing should be benchmarked against domestic MBR and RO vendors as well as European membrane suppliers, because import logistics and after-sales membrane replacement — 3–5 year RO life, 12–24 month MBR life — materially affect lifecycle cost. The hybrid ZLD path is now the design baseline for new fabs in stressed watersheds, and the engineering write-ups linked below give the design and cost breakdown.

For the ZLD and hybrid design reference, see the microelectronics wastewater ZLD engineering blueprint and the wafer fab developer wastewater treatment hybrid system design for the 99.8% TSS removal envelope and cost breakdown.

Frequently Asked Questions

What is the most distinctive chemistry in a SMIC-class fab wastewater train?

Fluoride from HF and BOE etchants, at 100–5,000 ppm F⁻ and pH 2–4, is the most distinctive chemistry. It is treated with Ca(OH)₂ at a 1.2:1 molar ratio to precipitate CaF₂, achieving ~95% removal and effluent below 10 ppm, with sludge dewatered to 5–10% of influent volume (HydropureWater fab guide, 2025).

What reuse rate should a SMIC-class fab in northern China target?

70–85% RO recovery on UPW reject and cooling-tower blowdown is the published envelope, in line with the Samco (2025) ~60% industry baseline and Intel Ocotillo's ~75% EPA-documented benchmark (US EPA Region 9, 2017 snapshot). A brine-polish stage can push overall recovery to 88%, as demonstrated at the IDE Tech commercial install.

Can an MBR handle TMAH developer waste at pH 12–14?

Yes, but only after aggressive pH adjustment down to 7–8. An integrated MBR at 12–24 h HRT then achieves up to 99% TMAH removal through biodegradation, with 92–97% COD removal and 99% TSS removal at 10–20 LMH flux. PVDF membrane life runs 12–24 months when air-scour duty is held in spec (HydropureWater field data, 2025).

How long do RO membranes last in fab UPW-reject service?

3–5 years with effective UF pretreatment, anti-scalant dosing at 2–5 mg/L, and a defined CIP cycle. The dominant foulant is colloidal silica from the CMP leg, which is why the hollow-fiber UF as RO guard filtration is standard on this stream.