Why pretreatment decides whether the UPW train survives
A weak pretreatment section is the single most common root cause of premature RO and EDI failure on a US semiconductor ultrapure water train, and the failure usually shows up two pages downstream in the polishing skid. RO membrane life in a well-run fab pretreatment sits at 3–5 years and EDI module life at 5–10 years, and both numbers are controlled by what the feed stream looks like at the inlet of each stage, not by the polishing stage itself (HydropureWater field data, 2026). A single 200 mm wafer needs about 5,600 L of UPW for cleaning, so an unplanned RO or EDI outage costs yield, not just downtime (AXEON, 2026). The spec the train has to defend is fixed: ASTM D5127-13(2018) Table 1 sets online resistivity floors of 18.1 MΩ·cm for Type E-1 (line width 1.0–0.5 µm) and 18.2 MΩ·cm for Types E-1.1, E-1.2 and E-1.3 covering 0.35 µm down to 0.032 µm nodes, with TOC caps of 5, 2 and 1 µg/L respectively (HydropureWater, 2026; ASTM D5127-13(2018)).
The term feed gate refers to the maximum TDS, TOC, hardness, silica, free CO2, SDI and chlorine values that the downstream membrane or module can accept on a continuous basis without losing life or product quality. For an EDI stack fed from an RO+degasser train, the published gate is under 25 ppm TDS, under 1 mg/L TOC, free CO2 under about 5 ppm, hardness under 1 ppm and SDI under 5 (HydropureWater, 2026). Every equipment pick in pretreatment is, in practice, a decision about whether the feed gate can be held on a Monday morning, not just on commissioning day.
Recommended pretreatment stack for RO+EDI+degasifier in US fabs
A 2026 US semiconductor UPW train recommends multimedia filtration, activated carbon, sodium-cycle softening, antiscalant dosing and 5 µm cartridge filtration in front of a two-pass RO, with a membrane degasifier between the RO passes and before the EDI stack. The EDI feed gate is RO permeate under 25 ppm TDS, under 1 mg/L TOC, free CO2 under about 5 ppm, hardness under 1 ppm and SDI under 5, which together protect the 18.2 MΩ·cm resistivity required by SEMI F63 and ASTM D5127 (HydropureWater, 2026; SUEZ Water Handbook).
The eight-stage lineup, in the order it should appear on a P&ID:
- Multimedia filtration using a multi-media filter with anthracite over sand over garnet, target turbidity under 1 NTU and Silt Density Index reduction to 3 or lower. This is the first SDI cut.
- Activated carbon to strip free chlorine, which would otherwise destroy thin-film composite RO membranes by oxidizing the polyamide layer; the fab sequencing matches the SUEZ Water Handbook convention of MMF → carbon → softener before RO.
- Sodium-cycle softening with an industrial water softener, hardness leakage under 1 ppm, sized to limit CaCO3, CaSO4 and BaSO4 scaling in both RO passes and to protect EDI resin from hardness fouling (SUEZ Water Handbook).
- Antiscalant dosing with a phosphate-free formulation through an automatic chemical dosing system, mandatory when silica, sulfate or recovery is high; phosphate-free is required where discharge rules limit phosphorus.
- 5 µm cartridge polishing as the final guard before the high-pressure pump, preventing particulate fouling on the RO element face and protecting the SDI<5 feed gate.
- Two-pass RO with concentrate recycle, with the industrial RO water treatment system typically run at 75% single-pass recovery and 90–95% system recovery with concentrate recycle (HydropureWater, 2026).
- Membrane degasifier, forced-draft or membrane contactor, sized to drop free CO2 from 10–30 ppm down to under 5 ppm so the EDI current removes ions, not bicarbonate (SUEZ Water Handbook; HydropureWater, 2026).
- EDI polisher, continuous electrodeionization stack with the EDI Electrodeionization System, no acid or caustic, resistivity above 18.2 MΩ·cm.
For background on how each RO stage is sized, the RO system design parameters 2026 guide walks through the flux and recovery math used for step 6.
| Stage | Equipment | Primary target | Outlet spec the next stage depends on |
|---|---|---|---|
| 1 | Multi-media filter | Particulate and turbidity cut | Turbidity < 1 NTU; SDI ≤ 3 |
| 2 | Activated carbon | Free chlorine removal | Free Cl2 < 0.1 ppm |
| 3 | Sodium-cycle softener | Hardness removal | Hardness < 1 ppm as CaCO3 |
| 4 | Antiscalant dosing | Scale control on RO | Silica and sulfate held under saturation |
| 5 | 5 µm cartridge | Particulate guard | SDI < 5 at RO feed |
| 6 | Two-pass RO | Bulk ion and organic removal | Permeate TDS < 25 ppm; TOC < 1 mg/L |
| 7 | Membrane degasifier | Free CO2 strip | Free CO2 < 5 ppm |
| 8 | EDI polisher | Final ion polish | Resistivity > 18.2 MΩ·cm |
Feed gates the EDI stack will not forgive
Vendor cut-sheets list limits in different units and different order, so a US fab engineer usually ends up rewriting the gate by hand. The consolidated gate below is the one the EDI module will hold to over a 5–10 year life, not just on commissioning day (HydropureWater, 2026; AXEON, 2026).
| Parameter | EDI feed gate | Reason |
|---|---|---|
| RO permeate TDS | < 25 ppm | Resin load and current stay in design band |
| TOC | < 1 mg/L | Organic fouling of anion resin and membrane |
| Free CO2 (post-degasifier) | < 5 ppm | Prevents bicarbonate loading the EDI current |
| Hardness as CaCO3 | < 1 ppm | Stops Ca/Mg scaling on the cell |
| SDI of RO feed | < 5 | Holds RO membrane life at 3–5 years |
| Free chlorine at RO inlet | < 0.1 ppm | Polyamide layer is destroyed by Cl2 |
| Total iron | < 0.05 ppm | Catalytic oxidation of membrane |
| Silica (RO feed as SiO2) | < 150 ppm at standard recovery | Drop recovery or dose antiscalant if higher |
| Temperature | 20–35 °C | Resin life and resistivity stability |
| pH into EDI | 6.5–7.5 | Protects cation and anion resin lifetimes |
If the feed drifts outside any of these envelopes, the EDI product slides below 18.2 MΩ·cm and the fab is suddenly below SEMI F63. Recovery on the EDI module itself runs 90–95% with less than 5% blowdown, so the EDI stage rarely sets the site water balance, but the gate above is what sets the EDI life (HydropureWater, 2026; AXEON, 2026).
Degasification: where CO2 either gets out or quietly kills the EDI current
CO2 rides through RO because it is a dissolved gas, not an ion, and an RO membrane does not reject neutral species. The result is bicarbonate equilibrium shifting into the EDI cell, where the current then spends itself stripping CO2 rather than residual ions (HydropureWater, 2026; SUEZ Water Handbook). A forced-draft or membrane contactor degasifier has a small footprint and drops free CO2 from about 10–30 ppm in the RO permeate to under 5 ppm, which is the right pick when only CO2 is the problem. A vacuum de-aerator drops residual O2 and CO2 to under 10 ppb after second-pass RO, and is the right pick when oxygen and CO2 both matter, for example a power-plant condensate polish or a sub-ppb gas spec (SUEZ Water Handbook).
The cost frame is the part procurement usually misses. A membrane degasifier is a small CAPEX line compared with the EDI modules it protects, and skipping it is the most expensive line-item cut on a US fab UPW P&ID. The EDI replacement budget runs 5–10 years, so a single degasser preserves one and often two module changeouts over a 15-year horizon (HydropureWater, 2026). Vacuum de-aerators cost more in CAPEX and need a sealed skid, but they remove O2 to low ppb, which is required for any high-pressure boiler feed on the same plant. Pick forced-draft or membrane when only CO2 is out of spec; pick vacuum when O2 and CO2 are both out of spec.
US standards that drive the train: SEMI F63, ASTM D5127, and the EPA feedwater rules
SEMI F63 sets fab UPW at >18.2 MΩ·cm, TOC under 1 ppb, particles under 0.3/mL above 0.05 µm, and bacteria under 1 CFU/100 mL; the spec applies at the point of distribution, not only at the skid outlet (AXEON, 2026). ASTM D5127-13(2018) Table 1 sets 18.1 MΩ·cm for Type E-1 (1.0–0.5 µm) and 18.2 MΩ·cm for Types E-1.1 through E-1.3, with TOC caps of 5, 2 and 1 µg/L as the line gets tighter (HydropureWater, 2026). For WFI loops, USP chapter 645 governs conductivity and chapter 643 governs TOC, and the source-water floor is the US EPA National Primary Drinking Water Regulations codified at 40 CFR 141, with no added substance allowed in the bulk water (USP, current to 2025-02). US fabs at sub-7 nm nodes are now quoting TOC under 0.5 ppb, and pretreatment has to overshoot SEMI F63 to keep pace with node shrink (AXEON, 2026).
| Standard | Scope | Key number | What pretreatment must defend |
|---|---|---|---|
| SEMI F63 | Semiconductor UPW spec at point of distribution | >18.2 MΩ·cm; TOC < 1 ppb | EDI product resistivity and TOC at the loop return |
| ASTM D5127-13(2018) Type E-1 | Electronics UPW, 1.0–0.5 µm | 18.1 MΩ·cm; TOC 5 µg/L | EDI gate at minimum |
| ASTM D5127-13(2018) Type E-1.2 | 0.18–0.13 µm nodes | 18.2 MΩ·cm; TOC 2 µg/L | Sub-7 nm fab rinse |
| ASTM D5127-13(2018) Type E-1.3 | 0.065–0.032 µm nodes | 18.2 MΩ·cm; TOC 1 µg/L | Advanced node rinse |
| USP <645> | Bulk water conductivity for WFI/PW | Stage 1/2/3 conductivity at 25 °C | RO permeate conductivity |
| USP <643> | Bulk water TOC for WFI/PW | TOC ≤ 500 ppb (PW) | EDI outlet TOC |
| US EPA NPDWR (40 CFR 141) | Source-water floor for WFI | Drinking-water compliance | City water into pretreatment |
Selecting the pretreatment stack by feedwater type
The eight-stage lineup only works if the first five stages match the feedwater. The decision matrix below maps the feed type to the pretreatment stack we recommend in 2026, with the EDI feed gate held identical across rows. Engineers running a UPW reclaim or high-recovery loop should also review the semiconductor UPW reclaim blueprint for reuse-stream specifics, and the nanofiltration system design guide 2026 where sulfate or hardness pre-softening is required upstream of RO.
| Feedwater type | Required pretreatment additions | Notes |
|---|---|---|
| Surface water (lake or river) | Dual MMF + carbon + softener + antiscalant + 5 µm cartridge | Add upstream chlorine residual control; coagulation if TOC > 4 mg/L |
| Well water with high hardness or iron | Softener first, then greensand or aeration for Fe/Mn, then MMF, carbon, RO | Total iron < 0.05 ppm at RO inlet |
| Pre-treated city water (US utility) | Softener and carbon still required; antiscalant optional at conservative recovery; 5 µm cartridge stays | TOC < 2 mg/L at RO feed expected |
| High-TOC or high-silica surface water | Add coagulation stage and tighter antiscalant dose; consider two-pass RO with permeate blending | Drop recovery or blend if silica > 150 ppm as SiO2 |
| Recycled fab UPW reuse stream | Two-stage RO plus UF; intermittent 2nd-stage RO for fouling control | Pilot data shows >75% recovery and 0.5 mgC/L DOC, with RO permeate reaching 18.2 MΩ·cm and <1 ppb DOC at lab scale (Desalination pilot, 2025) |
Commissioning checks and operating costs
Every UPW train we audit fails the same way at startup: the SDI on the RO feed is not held under 5, or the free chlorine slips past the carbon. The commissioning checklist below is the one we walk the operator through, in order:
- RO feed SDI under 5 (15-minute test per ASTM D4189).
- Hardness under 1 ppm as CaCO3 post-softener.
- Free chlorine under 0.1 ppm post-carbon, DPD test.
- RO permeate TDS under 25 ppm on both passes.
- EDI product resistivity above 18 MΩ·cm within 30 minutes of feed-up, with pH 6.5–7.5 at the EDI inlet.
Operating cost splits cleanly between the two stages. RO draws 2–4 kWh/m³ at 200–600 psi; EDI draws 0.5–1.5 kWh/m³ from the stack DC current. RO chemical cost is antiscalant plus periodic cleans on a 3–6 month cadence, with membrane replacement every 3–5 years. EDI chemical cost is zero, and module replacement runs every 5–10 years. A hybrid RO+EDI train typically lands 30–50% below mixed-bed ion exchange on operating cost because acid, caustic and the neutralization tank all disappear (HydropureWater field data, 2026).
Frequently asked questions
What should go in front of RO on a US fab UPW train?
Multimedia filtration, activated carbon, sodium-cycle softening, antiscalant dosing and 5 µm cartridge polishing, in that order, sized to hold SDI under 5, free chlorine under 0.1 ppm and hardness under 1 ppm at the RO inlet. This matches the 2026 US fab convention and protects the 3–5 year RO membrane life (HydropureWater, 2026; SUEZ Water Handbook).
What should go in front of the EDI stack?
Two-pass RO permeate at under 25 ppm TDS and under 1 mg/L TOC, with a membrane degasifier dropping free CO2 to under 5 ppm, and pH held at 6.5–7.5. This combination defends the 18.2 MΩ·cm floor set by SEMI F63 and ASTM D5127 Type E-1.2 (HydropureWater, 2026; AXEON, 2026).
Where does the degasifier sit in the train?
Either between the two RO passes to drop CO2 and load the second pass with better permeate conductivity, or on the second-pass permeate immediately before the EDI stack to keep CO2 off the EDI current. Most US fabs we size for in 2026 place the degasifier on second-pass permeate right before EDI (SUEZ Water Handbook).
What is the EDI feed gate, in one paragraph?
RO permeate under 25 ppm TDS, TOC under 1 mg/L, free CO2 under 5 ppm, hardness under 1 ppm as CaCO3, SDI under 5, free chlorine under 0.1 ppm, total iron under 0.05 ppm, silica under 150 ppm as SiO2 at standard recovery, temperature 20–35 °C and pH 6.5–7.5. Holding this gate every day is the difference between a 5-year and a 10-year EDI module life (HydropureWater, 2026).
How does the stack change for a recycled reuse stream?
A fab UPW reuse stream needs ultrafiltration ahead of two-stage RO, with intermittent second-stage RO operation to control fouling. A 2025 pilot achieved >75% recovery and 0.5 mgC/L DOC on the RO permeate, and the lab-scale UPW train on that permeate held 18.2 MΩ·cm and <1 ppb DOC consistently (Desalination pilot, 2025).
