The 18‑Megohm Race: Inside How Fabs Build City‑Scale Ultra‑Pure Water Plants

Semiconductor makers are pushing water systems to extremes: double‑pass RO, CEDI, UV, UF, and point‑of‑use filters working in concert to deliver >18.0 MΩ·cm water at industrial volumes. A design brief for senior UPW engineers shows how to get there while boosting recovery and reuse.

Industry: Semiconductor | Process: Wafer_Fabrication

Water is now a front‑row constraint in chipmaking. IDTechEx forecasts semiconductor water use will double by 2035, with TSMC alone withdrawing about 101×10^6 m³ in 2023 (www.idtechex.com). One recent report calls a “typical fab” a 3.78×10^7 L/day ultra‑pure water (UPW) operation (www.businesstoday.com.my).

Most of that water doesn’t leave the site for long. As SemiEngineering puts it, “almost all freshwater entering a fab will leave as wastewater… [with] the dominant true loss [via] evaporation” (semiengineering.com) (semiengineering.com). In practice, >90–95% of process water is reused in loops; only cooling tower/metabolic losses consume water.

Against that backdrop, the UPW plant is both a yield engine and a sustainability lever: it must supply extreme purity at high reliability while enabling aggressive recycle/reuse. SK Hynix, for instance, increased its reused water by 51% from 2020–2023 (www.idtechex.com), and firms like NXP/Onsemi routinely repurpose wastewater (e.g., for cooling).

Fab water demand and reuse baselines

Design capacity must match multi‑kiloliter‑per‑day realities—up to 10^4–10^5 m³/day for large fabs—while facilitating >90% recoveries. The engineering implications are clear: supply reliability at very high purity and embedded recycle schemes that minimize true losses to evaporation (semiengineering.com).

These targets are not theoretical. IDTechEx’s doubling projection to 2035 and TSMC’s 2023 withdrawal (~101×10^6 m³) (www.idtechex.com) frame the scale; a “typical fab” tapping 3.78×10^7 L/day shows the day‑to‑day magnitude (www.businesstoday.com.my).

Multi‑stage pretreatment design

UPW starts by taming raw feed (ground, surface, or municipal) with a pretreatment train that protects membranes and stabilizes recovery. Clarification/sedimentation removes grit and most TSS (total suspended solids) and is effective at >70–80% TSS removal, typically leaving turbidity in the low NTU range; a compact clarifier sets detention time to 0.5–4 hours while controlling footprint. Indonesian potable‑water regulations specify turbidity ≤5 NTU (grinviro-global.com), and high‑end pretreatment aims ≲0.2 NTU entering RO.

Coagulation sets the stage for clarification, with dosage handled by an accurate dosing pump and commonly supported by high‑purity PAC (polyaluminum chloride), as in PAC programs that improve particle aggregation by reducing turbidity. Floc growth and settling are strengthened with polymers, and engineered blends are available under flocculants that routinely improve clarifier efficiency by 30–50%.

Downstream, multimedia filters polish remaining particles. Dual‑media beds using sand/silica often reduce turbidity to <0.5 NTU (frequently <0.2 NTU), and top layers of anthracite extend run times and depth filtration.

Activated carbon (GAC) is next to strip chlorine and organics that can attack membranes; commercial beds remove ~80–90% of residual chlorine and volatile organics, making activated carbon a standard line of defense before RO. Because chlorine destroys polyamide RO membranes, designs may also include chemical quenching via a dechlorination agent to drive free chlorine to ~0 mg/L.

Hardness removal prevents scale. Ion‑exchange softeners reduce Ca/Mg to ≪1 mg/L, and a common rule of thumb is to limit Ca hardness to ≈1 grain/gal (17.1 mg/L as CaCO₃) to avoid RO scaling (www.researchgate.net); pretreatment aims to bring hardness to <0.1–1 mg/L, typically via a softener paired with appropriate resin.

For more robust particle barriers, ultrafiltration (UF) ahead of RO can remove ≈98–99% of turbidity—evidence from one study underscores how a combined sand + UF train can virtually eliminate particles (www.researchgate.net); packaged ultrafiltration units are common pretreatment to RO.

Double‑pass RO and CEDI

Primary purification hinges on reverse osmosis (RO), typically arranged in two passes to maximize salt and contaminant removal. A first‑pass RO runs at high recovery (~80–90%), and its permeate feeds a second pass that operates at lower recovery; overall RO recovery of ~70–80% is common (www.researchgate.net). Industrial installations often center on brackish‑water RO for feeds around a few hundred mg/L TDS (total dissolved solids).

Real‑world performance data show UF+RO can drive permeate conductivity to ~6 μS/cm (~0.17 MΩ), which is already orders of magnitude cleaner than feed (www.researchgate.net); double‑pass RO would lower conductivity further, into single‑μS or sub‑μS territory. For membrane selection, utilities often specify recognized elements such as FilmTec RO membranes inside a broader packaged membrane system.

RO permeate alone is not UPW. Continuous electrodeionization (CEDI)—electrically driven ion removal that does not require chemical regeneration—serves as the final primary purifier. CEDI consistently delivers ~15–17 MΩ·cm water (felitecn.com) and is widely deployed via EDI modules. By contrast, mixed‑bed DI can reach ~18.2 MΩ·cm at its peak but requires periodic chemical regeneration and downtime (felitecn.com), a trade‑off some plants still make with mixed‑bed polishers.

Primary‑purification metrics are straightforward: high‑performance RO removes ~99% NaCl‑equivalent salts; if feed TDS is ~500 mg/L, double‑pass RO typically reduces that to <10 mg/L (≪0.1 mS/cm). CEDI then eliminates essentially all residual ionic species. Achieving ≥18.0 MΩ·cm (measured at 25°C per industry roadmaps) requires final gas removal—dissolved CO₂ must be stripped (via degassers) along with any residual ions (www.mks.com). Where scaling risks persist, plants dose antiscalants in proportion to residual hardness using membrane antiscalants.

UV, ultrafiltration, and distribution polishing

19f89424-0ebe-4fd7-8596-96e41472bb28 (1)

After RO/CEDI, a polishing loop maintains sterility and keeps TOC (total organic carbon) and particles in check. UV oxidation at 185/254 nm destroys trace organics and sterilizes bacteria; low‑OPEX mercury‑lamp units deliver a 99.99% pathogen kill rate without chemicals, as in ultraviolet systems. Degasification (often integrated or via separate membrane degassers) strips CO₂ and oxygen to push resistivity beyond 18.0 MΩ·cm.

Loop ultrafiltration removes sub‑micron colloids and bacteria, and final point‑of‑use (POU) barriers at each tool interface are typically 0.1–0.2 μm. Housings matter in UPW; 316L stainless is standard for hygienic service, and stainless cartridge housings pair with high‑purity cartridge filters to control particles at the point of delivery. A closed stainless tank and a continuously circulating distribution ring maintain low bioburden, with microbial counts targeted at <1 CFU/100 mL (CFU: colony‑forming units).

In practice, mercury lamp UV + UF can achieve negligible turbidity (~<0.01 NTU) and TOC limited only by water/carbonates; one study reported RO+UF output turbidity ≈0.01 NTU (www.researchgate.net). Industry reports note that combined RO/ion‑exchange/UV/degass methods are “employed to reduce the TOC to acceptable levels in UPW” (www.mks.com).

Plant performance metrics and recovery

A correctly sized UPW plant should deliver: final resistivity ≥18.0 MΩ·cm, TOC <10 ppb (typically), turbidity around 0.01 NTU, and microbial counts <1 CFU/100 mL. Overall recovery of ~75–80% (including RO and CEDI) means only ~20–25% of feedwater leaves as brine. For a 10,000 m³/day output, waste is ~2.5–3.3×10^3 m³/day.

Compared to mixed‑bed DI, continuous EDI’s chemical‑free operation avoids regeneration downtime; in large UPW systems (>150 L/min), its OPEX advantage typically outweighs higher CAPEX, with industry practice showing breakeven around 40–50 GPM flows (felitecn.com). High‑availability is maintained by advanced process control and real‑time monitors (digital Twin/SCADA), with online resistivity, UV254, TOC analyzers, and flow meters as standard ancillaries.

Internal recycle and reuse strategies

Net consumption can drop sharply by recycling rinse and scrub effluents. For example, reprocessing spent CMP rinse by UF+RO can yield ~90% reuse of that water in non‑critical loops (www.researchgate.net). Many companies aim to recycle >50% of total usage in aggregate (www.idtechex.com), a target aligned with the observation that “most water withdrawn… will leave as wastewater,” with dominant true loss via evaporation (semiengineering.com).

Regulatory context and effluent management

UPW far exceeds typical potable standards, but plants still operate within national rules. Indonesian drinking‑water standards (Permenkes 492/2010 and updates) specify turbidity ≤5 NTU (grinviro-global.com) and zero coliforms/E. coli; the pretreatment and final filtration described above routinely deliver ~0.1 NTU and effectively sterile water.

Effluent—e.g., ED brine or RO concentrate—must meet discharge limits such as those under PermenLH 68/2016, which set BOD, COD, and heavy‑metal thresholds (e.g., Cu, Pb, Ni, Zn) (www.intilab.com). Although UPW systems produce negligible residual concentrations, the plant should include a waste‑treatment step (neutralization, polishing) to ensure any drain disposal meets local “bakumutu” requirements.

Summary of outcomes and standards

When executed well, the train—clarification, media filtration, activated carbon, softening, double‑pass RO, CEDI, UV, UF, degasification, and POU filters—produces 18+ MΩ·cm water, <50 ppb TOC (with polishing typically driving TOC into the single‑digit ppb range), ~0.01 NTU turbidity, and <1 CFU/100 mL. These targets align with industry roadmaps that call for >18.0 MΩ·cm at 25°C (www.mks.com) and are supported by pretreatment that reduces turbidity well below the 5 NTU potable limit (grinviro-global.com), plus high‑recovery RO/CEDI operations documented in performance studies (www.researchgate.net).

Strategically, this setup answers two imperatives—reliability at extreme purity and embedded reuse—against a backdrop of surging fab demand and tightening local constraints, as tracked by IDTechEx (www.idtechex.com) and operational insights from SemiEngineering (semiengineering.com). The engineering playbook is fixed; the execution window is now.

Chat on WhatsApp