Why the Polishing Loop Decides Yield at Sub-3 nm
A polishing loop is the continuously recirculating return pipe between the central ultrapure water (UPW) plant and the tools, and at sub-3 nm nodes it is the boundary where yield is won or lost. A 28 nm plant saw a 5% yield loss after silica exceeded 0.5 ppb in the polishing loop, and a separate fab saw a 12% rise in CMP defects when TOC spiked above 2 ppb from biofilm in distribution piping (S2, 2026). SEMI E157-1120 frames those events at $50K–$200K per incident in downtime and scrap, while a 1 ppb TOC rise has been linked to 0.5–1% yield loss on advanced 5 nm nodes (S2, 2026).
The polishing train upstream is the easier half of the problem; the last 50 m of piping between the central skid and the tool is where fabs quietly destroy yield (S4, 2026). The loop has three jobs: hold the spec envelope on the return leg, prevent stagnation-driven biofilm, and reach the point of use without recontamination. That framing is why distribution technology — piping materials, dead-leg control, degasifier placement, point-of-use filtration, and online sensors — has to be specified with the same rigour as the polishers themselves.
SEMI F63 Specs the Distribution Loop Must Hold
SEMI F63 defines the envelope the loop is being designed against, and that envelope tightens with every node shrink. The baseline is resistivity ≥18.2 MΩ·cm at 25 °C, TOC <1 ppb, <0.3 particles/mL at >0.05 µm, and bacteria <1 CFU/100 mL (S1, S4). Sub-5 nm production tightens bacteria to <0.1 CFU/100 mL and TOC to <0.5 ppb; sub-3 nm immersion lithography is the most TOC-sensitive step in the entire flow (S1, S4).
The half-feature rule scales particle filtration to node: particles must be filtered to one-half the smallest feature size, so 3 nm features require roughly 1.5 nm at the tool and ≤200 nm point-of-use filters on the distribution loop (S2, S4). DO is held below 10 µg/L at point of use to prevent oxidation of metal films and low-k dielectrics (S4, 2026). On the return leg, sustained operation at 18.15–18.18 MΩ·cm is the operating target, with anything below 18.15 MΩ·cm treated as an actionable excursion; the online TOC ladder is 0.7 ppb alarm and 1 ppb hard action for sub-3 nm nodes (S1, S4).
| Parameter | SEMI F63 limit | Sub-3 nm operating target | Online measurement |
|---|---|---|---|
| Resistivity at 25 °C | ≥18.2 MΩ·cm | 18.15–18.18 MΩ·cm sustained | Inline conductivity / resistivity meter |
| TOC | <1 ppb | <0.5 ppb; alarm 0.7 ppb, action 1 ppb | UV-persulfate online |
| Particles >0.05 µm | <0.3/mL (S1) / <1/mL (S4 framing) | Half-feature rule, ≤200 nm POU | Laser particle counter |
| Bacteria | <1 CFU/100 mL | <0.1 CFU/100 mL for <5 nm | Membrane filtration / online ATP |
| Dissolved oxygen | — | <10 µg/L at POU (S4) | Optical-fluorescence DO |
| Silica (dissolved + colloidal) | <0.3 ppb (S1/S2); <0.05 ppb for <5 nm (S2) | Sub-ppb; 0.2–1.0 ppb envelope (S1) | Molybdate colorimetric + ICP-MS |
Polishing-Loop Unit Operations and What Each One Must Do
Each distribution-side technology is matched to a specific spec parameter, and the loop is the unit of design. An EDI polisher for chemical-free loop operation sits after RO and sustains >2 MΩ·cm at 90%+ recovery without acid/caustic regeneration; feed hardness must stay below 1 ppm CaCO3 and silica below 1–2 ppm to limit scaling, which is why EDI is a polisher, not a stand-alone (S1, S2, S4). Per S1/S2, EDI tolerates only low-TDS feed and is structurally placed after RO.
A dual-wavelength 185/254 nm UV unit on the return loop does two distinct jobs: 185 nm breaks residual organics down to <1 ppb TOC, and 254 nm handles microbial control; lamp life runs 9,000–12,000 h and the lamp count must be sized to the TOC target, not average flow (S1, S4). For reclaimed feedwater, UV-AOP (UV plus H2O2) is moving from pilot to standard supplementary polishing because small organics such as urea slip past RO/EDI/standard UV (S4, 2026).
UF at 0.01 µm (10,000 MWCO) strips colloidal silica and bacterial fragments (S1, S4). The non-regenerable mixed-bed / scavenger DI polisher lands the loop on 18.2 MΩ·cm and is the redundancy stage after EDI; per S2, EDI removes 90–95% of silica and the scavenger DI takes the remaining 5–10% to sub-ppb. The finishing block is a 0.2 µm (≤200 nm at point of use per S4) polish filter and a loop-side membrane or vacuum degasifier — the degasifier is the answer to the ~70% CO2 ingress root cause called out in S2 and is the reason a distribution-side degasifier must be sized separately from any central unit. Distribution materials are PVDF, PFA, PP or SS-316L pipe with orbital-welded joints; solvent-welded plastic is excluded because leachables add measurable TOC (S4).
| Unit operation | Spec it controls | Sizing rule | Failure mode to watch |
|---|---|---|---|
| EDI (post-RO) | Resistivity, silica to ~5–10% of feed | Feed hardness <1 ppm CaCO3, silica <1–2 ppm; 90%+ recovery | Scaling → resistivity drop, module failure |
| Dual-wavelength UV (185/254 nm) | TOC, microbial control | Lamp count sized to TOC target, 9,000–12,000 h life | 185 nm intensity <80% of new lamp → TOC climb |
| UF 0.01 µm (10,000 MWCO) | Colloidal silica, bacterial fragments | Peak demand, not average | Fouling, integrity loss |
| Scavenger / mixed-bed DI | Final ionic polish to 18.2 MΩ·cm | Redundancy after EDI for sub-ppb silica | Sodium breakthrough, silica slip |
| Final 0.2 µm (≤200 nm POU) filter | Particles, half-feature rule | ≤200 nm at the tool, 3 nm → ~1.5 nm target | Bypass, microbial floc breakthrough |
| Loop-side membrane / vacuum degasifier | CO2, DO <10 µg/L | Sized to the last 100 m, not the central skid | Air ingress, pump cavitation |
Sensors and Setpoints: Monitoring the Return Loop in Real Time
Online monitoring is what makes the loop enforceable. Per S4, three instrument locations catch drift before wafer lots are exposed: post-UV, on the return loop, and at point of use. Inline resistivity on the return loop is the first line of defence because CO2 ingress through fittings is the most common cause of a slow drift (S4). A 0.1 ppb NaCl excursion drops resistivity from 18.18 to 18.11 MΩ·cm — a tiny but detectable shift that flags ionic breakthrough before particle bridging shows up at sub-7 nm (S4, 2026).
Online TOC analysers (UV-persulfate) sit post-UV and at point of use, with an alarm at 0.7 ppb and a hard action at 1 ppb for sub-3 nm nodes (S1, S4). Online sodium ion-selective electrodes catch cation-bed exhaustion early — sodium is the first ion to break through a depleted cation exchanger and conductivity alone cannot see it (S4). Silica is tracked by online molybdate colorimetric for trend plus periodic ICP-MS for confirmation, with dissolved and colloidal fractions reported separately (S1, S4). DO is measured by optical-fluorescence sensors on the return loop and at point of use, alarming at 10 µg/L; optical-fluorescence drifts less than membrane-electrochemical cells over multi-month campaigns (S4). Laser particle counters >0.05 µm at point of use are trended against specific tool uptime so slow drift is caught before a scrap event (S1, S4).
| Instrument | Location | Setpoint / alarm | What it catches |
|---|---|---|---|
| Inline resistivity / conductivity | Return loop, POU | Sustained 18.15–18.18 MΩ·cm; <18.15 = action | Ionic breakthrough, CO2 drift (S4) |
| Online TOC (UV-persulfate) | Post-UV, POU | 0.7 ppb alarm, 1 ppb action (sub-3 nm) | UV decay, biofilm, organic ingress (S1, S4) |
| Sodium ISE | Post-DI | Trend on first ion slip | Cation-bed exhaustion (S4) |
| Silica (molybdate + ICP-MS) | Return loop / grab | <0.3 ppb, <0.05 ppb for <5 nm (S2) | DI exhaustion, EDI scaling (S1, S2) |
| Optical-fluorescence DO | Return loop, POU | 10 µg/L alarm | Degasifier underperformance, air ingress (S4) |
| Laser particle counter >0.05 µm | POU | Trend vs tool uptime | Filter bypass, biofilm flocs (S1, S4) |
Three Loop Architectures: RO+DI, RO+EDI, and Hybrid RO+EDI+DI
Architecture choice is the single biggest CAPEX/OPEX lever on the loop, and the three options are not interchangeable. RO+DI usually carries about 20% lower CAPEX than RO+EDI, but about 30% higher OPEX from resin regeneration at ~$0.20/m³ versus ~$0.05/m³ for EDI; RO+DI OPEX typically sits near 1.1 kWh/m³ because regeneration cycles add energy (S2, 2026). RO+EDI runs continuous chemical-free operation near 0.8 kWh/m³, suits 14 nm and larger nodes with stable feed, and is the standard answer for steady quality without regeneration downtime (S2).
Hybrid RO+EDI + scavenger DI is the highest-CAPEX option and the strongest contamination buffer, standard for 3 nm and smaller nodes or variable feed (S2). The architecture takes 90–95% of silica in EDI and the remaining 5–10% in the scavenger DI to sub-ppb (S2). For 2026 procurement, two sustainability numbers are buyer-evaluation criteria rather than nice-to-haves: 85–92% closed-loop recycling and 3–7 kWh per 1,000 gallons of UPW produced (S1, S4). Fabs that cannot show both are off the shortlist for water-stressed US sites.
| Architecture | CAPEX vs RO+EDI | OPEX (per S2) | Silica handling | Best-fit nodes |
|---|---|---|---|---|
| RO+DI | ~20% lower CAPEX | ~$0.20/m³ regeneration; ~1.1 kWh/m³ | DI polisher alone | Older / smaller fabs, non-critical (S2) |
| RO+EDI | Baseline | ~$0.05/m³; ~0.8 kWh/m³ | EDI takes 90–95% of silica | 14 nm+ with stable feed (S2) |
| RO+EDI + scavenger DI | Highest | EDI + DI management | EDI 90–95%, scavenger 5–10% to sub-ppb | <5 nm, variable feed, low-risk requirement (S2) |
The industrial RO system feeding the polishing loop is the common front end for all three; the architectural choice is what sits downstream of it.
CAPEX, OPEX and TCO: Numbers for the Bid Review
New-plant CAPEX ranges from $2M for a 50 m³/h system to $5M for a 200 m³/h system, and about 60% of that spend sits in pretreatment and primary RO (S2, 2026). OPEX stacks energy at 0.5–1.2 kWh/m³, chemicals at $0.10–$0.30/m³, labor at $0.05–$0.15/m³, and membrane or resin replacement at $0.03–$0.10/m³; RO+EDI often operates near 0.8 kWh/m³ while RO+DI sits nearer 1.1 kWh/m³ because regeneration cycles add energy (S2).
Worked example: a 100 m³/h RO+EDI plant at $3.5M CAPEX and $0.25/m³ average OPEX produces ~4,380,000 m³ over 5 years, giving ~$1.095M OPEX and ~$4.595M TCO (S2, 2026). Contamination events priced at $50K–$200K per incident (SEMI E157-1120 framing in S2) make the monitoring and redundancy line items easy to defend at bid review, not nice-to-haves. Buyers evaluating replacement intervals should note RO membranes typically last 3–5 years ($10K–$50K for a 50 m³/h train) and EDI modules 5–7 years ($50K–$200K to replace at that capacity), with site history rather than vendor averages driving the real interval (S2).
| Cost line | Range / figure | Source |
|---|---|---|
| New-plant CAPEX | $2M (50 m³/h) – $5M (200 m³/h); ~60% in pretreatment + primary RO | S2 (2026) |
| Energy | 0.5–1.2 kWh/m³ ($0.05–$0.12) | S2 |
| Chemicals | $0.10–$0.30/m³ | S2 |
| Labor | $0.05–$0.15/m³ | S2 |
| Membrane / resin replacement | $0.03–$0.10/m³ | S2 |
| RO+EDI OPEX example | ~0.8 kWh/m³ | S2 |
| RO+DI OPEX example | ~1.1 kWh/m³ | S2 |
| Worked 5-yr TCO (100 m³/h RO+EDI) | $3.5M CAPEX + ~$1.095M OPEX ≈ $4.595M TCO | S2 |
| Contamination event cost | $50K–$200K per incident | S2 (SEMI E157-1120) |
Common Loop Excursions and How to Diagnose Them
Most loop excursions are diagnosed the same way regardless of fab or node. When resistivity falls below 18 MΩ·cm, S2 attributes about 70% of events to CO2 ingress, 20% to EDI failure, and 10% to exhausted DI resin — the fix is degasifier optimisation, EDI cleaning/replacement, or DI replacement, not central-skid rework. Bypass testing of RO permeate versus final UPW isolates whether the polishing train or an upstream stage failed (S2, 2026).
TOC spikes usually track 185 nm UV intensity falling below 80% of a new lamp, biofilm in piping, or exhausted activated carbon (S2). An on-site ClO2 generator for loop sanitisation restores disinfectant residual in the loop without dragging chloride into the polishers. Silica breakthrough >0.3 ppb points to DI resin exhaustion, EDI module scaling, or poor RO rejection; a documented 7 nm fab case was fixed with an emergency resin change followed by a second DI polisher in series for redundancy (S2, 2026). Particle counts >1/mL or microbial counts >1 CFU/100 mL point to final-filter bypass, upstream media breakthrough, microbial flocculation, insufficient UV dose, or stagnant zones; remediation is filter change, loop sanitisation, and dead-leg elimination (S2).
For broader fab water-system pretreatment reliability, the 2026 membrane-fouling prevention guide for fab water systems pairs with this loop design; the 2026 SEMI F63 plant-design guide and the 2026 UPW spec and reliability guide are the upstream reference documents for any bid review.
Frequently Asked Questions
What CAPEX and OPEX should we put on the table for a polishing loop at 100 m³/h?
Per S2 (2026), a 100 m³/h RO+EDI plant is the worked example at $3.5M CAPEX and ~$0.25/m³ OPEX, giving a 5-year TCO around $4.6M; new-plant CAPEX spans $2M (50 m³/h) to $5M (200 m³/h), with ~60% in pretreatment and primary RO. OPEX stacks from energy 0.5–1.2 kWh/m³, chemicals $0.10–$0.30/m³, labor $0.05–$0.15/m³, and membrane/resin replacement $0.03–$0.10/m³, so a buyer should request a line-itemised OPEX build at the bid stage rather than a blended number.
How do we choose a polishing-loop supplier without locking into a single architecture?
Tie supplier selection to the 2026 buyer-evaluation criteria from S1/S4: 85–92% closed-loop recycling, 3–7 kWh per 1,000 gallons of UPW produced, EDI feed-hardness <1 ppm CaCO3 and silica <1–2 ppm capability, and a documented distribution-side degasifier sized for the last 100 m (not just a central skid unit). A supplier who cannot show all four should not be on the shortlist for a US fab targeting sub-3 nm production.
Where does the loop fail first when resistivity drifts below 18.15 MΩ·cm?
Per S2 (2026) field data, ~70% of resistivity drops on operating loops trace to CO2 ingress through fittings, ~20% to EDI failure, and ~10% to exhausted DI resin. The first action is to verify the loop-side membrane or vacuum degasifier performance and inspect fittings for air ingress before opening the EDI modules or scheduling a DI change-out (S2, S4).
What lead time and consumable budgets should procurement bake into the contract?
Per S2, RO membranes typically last 3–5 years with replacement near $10K–$50K for a 50 m³/h train, and EDI modules 5–7 years at $50K–$200K to replace at the same scale; UV lamps run 9,000–12,000 h (S1, S4). Buyers should request a planned-replacement schedule with site-specific intervals, a guaranteed membrane and EDI spare-stock commitment, and a sanitisation consumables line (including ClO2 generation if the loop is reclaimed) before signing — these are the line items that swing TCO across a 5-year window.
