What a semiconductor ZLD train actually contains
A standard semiconductor zero liquid discharge train runs fab wastewater through pretreatment, a primary reverse osmosis stage, a secondary high-recovery membrane step, thermal concentration, and a brine crystallizer, in that order. The pretreatment layer combines ultrafiltration or microfiltration with dissolved air flotation to strip suspended solids, oils and the bulk of the colloidal silica load before the water reaches the first pump train (UF membranes protect the downstream RO). The first-pass RO is sized as the workhorse: on a low-TDS feed at roughly 700 mg/L, the Membranes 2025 pilot reported system recovery of 90–94% with monovalent ion rejection consistently 90–97% (Membranes 2025). The secondary high-recovery stage is where fab-specific chemistry starts to matter: forward osmosis plus multi-stage nanofiltration, or a closed-circuit / semi-batch RO variant, is used to push the concentrate above what a single RO pass can reach without precipitation.
Thermal concentration follows: a multi-effect distillation (MED) or mechanical vapor compression (MVC) unit drops the volume of the membrane reject, and a crystallizer takes the remaining brine to solids. The reuse-and-reclamation frame for this train still traces back to Grant et al. 2012 (Science) and the Tong & Elimelech 2016 (Environ. Sci. Technol.) review, which is the dominant citation in 2026 ZLD chapters (Springer 2025, S1). On the materials side, the Handbook of Environmental Chemistry 2026 chapter on reclaimed wastewater in the semiconductor industry lists ceramic nanofiltration, MXene-coated ceramic NF, and PVA-coated PVDF as 2024–2025 research variants targeting fluoride, silica, VOCs, PFAS and azoles, all still at pilot scale in the cited work (HEC 2026, S2). An industrial RO system remains the only unit operation in this chain with multiple peer-reviewed fab or near-fab pilots in the supplied research; everything downstream is either vendor-characterized or research-scale.
Where the first scaling limit appears: the RO front end
The first membrane in the train sets the recovery ceiling for the rest of the plant, and the Membranes 2025 pilot is the cleanest published 2025 dataset for what that ceiling actually looks like in numbers. First-pass RO at 90–94% system recovery held sodium, calcium and chloride rejection above 90–97% on feed at approximately 700 mg/L TDS, but magnesium and potassium rejection became variable, and the lowest measured magnesium rejection fell to 59% (Membranes 2025). That single number is the first warning sign: a divalent scaling cation is no longer being held back at the recovery limit, so the next concentration step will see an enriched Mg²⁺ stream. The NF stage was operated in two regimes. The first ran at 65–75% recovery on feed at about 3,000 mg/L TDS; the second ran at 75–85% recovery after feed TDS was dropped to about 2,500 mg/L, holding flux at 6.34–6.67 × 10⁻⁴ m/s and permeability at 7.32–7.77 × 10⁻¹⁰ m·s⁻¹·Pa⁻¹. Energy demand for the NF stage was 1.6 kWh/m³ at 76% recovery and 2.0 kWh/m³ at 84% recovery; the RO stage ran at 0.5 kWh/m³ at 90% recovery and 0.8 kWh/m³ at 93% recovery.
On a fab feed, those operating windows tighten immediately. Qiu et al. 2022 (cited in S2) documents fluoride and silica removal via a coagulation-UF process as a pretreatment response, and Lee, Shin, Ryu, Boo & Hong 2025 (cited in S2) describe a forward osmosis plus multi-stage NF hybrid as a research-stage "toward ZLD" pathway for fab wastewater. The energy data above are the only point of comparison: the membrane train is cheap, and the cost lives downstream. The table below fixes the Membranes 2025 operating points in one place.
| Unit operation | Feed TDS (mg/L) | Recovery (%) | Key ion rejection | Specific energy (kWh/m³) |
|---|---|---|---|---|
| First-pass RO | ~700 | 90–94 | Na⁺, Ca²⁺, Cl⁻ > 90–97%; Mg²⁺ as low as 59% | 0.5 (90%) – 0.8 (93%) |
| NF (series 1) | ~3,000 | 65–75 | Divalent rejection improved to 78–84% at top of range | 1.6–2.0 |
| NF (series 2) | ~2,500 | 75–85 | Stable permeability 7.32–7.77 × 10⁻¹⁰ m·s⁻¹·Pa⁻¹ | 1.6–2.0 |
| MED (NF permeate feed) | NF permeate | > 85 (CF 8) | Distillate TDS < 12 mg/L | Thermal (driven by TBT 43–50 °C and waste heat) |
The takeaway is straightforward: silica, fluoride, calcium sulfate and TMAH-derived organics all start to control the recovery ceiling at the RO front end on a real fab feed, even when the membranes are running well below their chloride-salt rating (membrane element selection becomes a chemistry decision, not a flux decision).
Pushing past RO: forward osmosis, NF and electrodialysis on fab brine
Three realistic options exist for the secondary concentration step on fab brine in 2026, and they sort by what is actually scaling the membrane rather than by marketing category. Forward osmosis plus multi-stage nanofiltration is the option the recent peer-reviewed fab work has converged on: Lee, Shin, Ryu, Boo & Hong 2025 (cited in HEC 2026, S2) frame this hybrid as a "toward ZLD" pathway, and the same chapter lists ceramic plate pilots (Ang, Teow et al. 2024), MXene-coated ceramic NF (Lee et al. 2024) and bismuth-oxyiodide-coated tubular NF (So & Park 2025) as fab-specific 2024–2025 options, all still at pilot scale in the cited work. Independent peer-reviewed performance numbers for these on production-scale fab feed were not located in the supplied research; treat the published pilots as design-direction evidence, not as sizing data.
High-recovery RO variants are the second option. Gradiant (S3) markets SBRO and CFRO for fab and data-center duty and reports system-level water reuse above 90%, but the supplied research does not contain a peer-reviewed fab-specific case study validating that headline at production scale, so it should be treated as a commercial design target rather than proven performance. Electrodialysis reversal is the third option, and it is the right one when scaling is ionic rather than silica- or organic-driven: Zhao et al. 2019 (Sep. Purif. Technol. 213, 339–347), cited in Springer 2025 (S1), demonstrate EDR on industrial RO brine. The same Springer 2025 chapter also cites Wang & Lin 2022 (ACS EST Eng.) on the thermodynamics of ZLD, which is the cleanest reference for why the second-law cost of concentration rises non-linearly past about 70,000 mg/L. For site context, the fab and data hall process wastewater guide walks through how these chemistries land on a permit line in 2026.
| Secondary stage option | Best-fit scaling problem | Evidence base in supplied research | Recovery claim |
|---|---|---|---|
| FO + multi-stage NF | Silica, fluoride, organics; low-fouling draw | Lee, Shin, Ryu, Boo & Hong 2025 (S2); ceramic/MXene/BiOI pilots 2024–2025 (S2) | Pilot scale; no production-scale fab numbers supplied |
| High-recovery RO (SBRO / CFRO) | Variable feed; brine-volume reduction | Gradiant commercial description (S3); no peer-reviewed fab case in supplied research | > 60% recovery in thermal stages (S3); > 90% system-level reuse (S3) |
| Electrodialysis reversal | Ionic scaling; selective ion separation | Zhao et al. 2019, Sep. Purif. Technol. 213, 339–347 (cited in S1) | Industrial RO brine treatment; specific kWh/m³ not reported in supplied research |
The hard ceiling: thermal concentration and crystallization
Thermal concentration is where ZLD stops being a separation problem and becomes an energy problem. The Membranes 2025 pilot operated a two-effect MED unit on NF permeate at top brine temperatures of only 43–50 °C, driven entirely by low-grade waste heat, and still hit a concentration factor of 8 and recovery above 85% (Membranes 2025). That is the best-case thermal window: a low-TDS feed, vacuum operation at 40 kPa(a) in the first stage and 20 kPa(a) in the second, and distillate quality of TDS below 12 mg/L. The energy story, however, is the one that controls the project. Thermal stages account for more than 80% of ZLD OPEX in the techno-economic analyses cited in that paper, and in the urban pilot they used 86% of the 12 kWh/m³ total, even with the RO stage contributing 71% of the total water recovery.
Push the train one stage further and the numbers change shape. The same paper reports that a brine crystallizer can reach 99% water recovery, but only at an energy cost of 52–70 kWh/m³ of feed for brines up to 300,000 mg/L TDS. A full ZLD train combining RO, a brine concentrator, and a crystallizer draws about 500 kWh/m³ of inflow; adding an NF unit and a third brine concentrator raises this to about 700 kWh/m³ of inflow (Panagopoulos techno-economic cited in the Membranes 2025 paper). The distillate quality from the thermal step was TDS below 12 mg/L, so the water itself is good — the energy penalty is the binding constraint, not the separation. For the equipment side, a well-specified industrial RO system front-end makes the thermal numbers look better, but it cannot move the crystallizer off its kWh/m³ curve. The 52–70 kWh/m³ number is the single most important figure in this article: it is the marginal cost of the last 9% of recovery on a fab brine, and it sets the upper bound on what true ZLD is worth to a fab.
MLD vs ZLD: a 2026 decision rule for fab engineers
Minimal liquid discharge and zero liquid discharge are not binary choices — they sit on the same recovery spectrum, and the operating point is set by feed chemistry, available waste heat, and the fab's marginal water cost. Panagopoulos & Haralambous 2020 (J. Environ. Chem. Eng. 8(5), 104418), cited in Springer 2025 (S1), is the working reference for that framing in 2026. The decision rule that follows is grounded in the operating points actually published in the supplied research, not in vendor targets.
If the membrane train can be capped at 75–85% recovery on feed at or below about 2,500 mg/L TDS — the second-series NF operating point in the Membranes 2025 pilot — and the site has access to low-grade waste heat to drive an MED polisher, MLD with thermal polishing is the lower-risk path: the energy is dominated by the membrane stage, the thermal stage runs at 43–50 °C TBT, and the crystallizer is avoided entirely. If the influent TDS or the permit pressure forces greater than 90% recovery at the membrane front end, the marginal water from 90% to 99% will be produced by the crystallizer at 52–70 kWh/m³ of feed, and economics must be checked against the fab's marginal water cost and the local power tariff. Gradiant (S3) claims greater than 90% system-level reuse, but the supplied research does not contain a fab-specific case study validating that figure at production scale, so MLD is the conservative default unless a full mass and energy balance is available. For fabs already planning a polishing loop, the UPW scale-up guide shows where the membrane front end tends to break on tool-count growth.
| Decision input | MLD path | True ZLD path |
|---|---|---|
| Feed TDS at NF inlet | ≤ ~2,500 mg/L (S5 series 2) | > ~2,500 mg/L, or scaling ions force higher recovery |
| Membrane recovery target | 75–85% on NF; 90–94% on RO if low TDS | Push past 90% on the membrane front end |
| Energy at the last step | MED at 43–50 °C TBT, waste-heat driven (S5) | Crystallizer at 52–70 kWh/m³ of feed (S5) |
| Full-train energy | ~12 kWh/m³ in the urban pilot (S5), thermal share ~86% | ~500 kWh/m³ (RO + BC + BCr); ~700 kWh/m³ with NF + 3 BC (Panagopoulos in S5) |
| Evidence base in supplied research | Pilot-validated operating points (S5) | Commercial claims at system level (S3); no fab-specific production-scale validation supplied |
What to demand from a ZLD or MLD supplier in 2026
The single biggest procurement failure on fab ZLD is accepting a system-level headline number instead of per-stage data tied to feed chemistry. The Membranes 2025 pilot and the Panagopoulos techno-economics cited there give a buyer a defensible set of reference points to push back with. Recovery, rejection, and specific energy should be requested per unit operation — not as a single "system reuse above 90%" claim — and the feed TDS and target recovery at which each number was measured should be on the page. The supplied research sets the reference values at 0.5–0.8 kWh/m³ for RO, 1.6–2.0 kWh/m³ for NF, and 52–70 kWh/m³ for the crystallizer, with a full ZLD train at about 500 kWh/m³ of inflow and roughly 700 kWh/m³ with the NF + three brine-concentrator configuration (Panagopoulos cited in Membranes 2025).
Second, request a mass balance for the scaling ions that actually limit fab recovery — silica, fluoride, calcium, sulfate — and for the fab organics that travel with them (IPA, TMAH, VOCs, azoles, PFAS), as catalogued in the Handbook of Environmental Chemistry 2026 chapter (S2). The Membranes 2025 numbers all sit below typical fab influent TDS, so a supplier who can only show urban-wastewater performance is not yet at the fab problem. Third, ask for an MLD option with a defined brine-out specification as a contractual fallback, following the Panagopoulos & Haralambous 2020 framing in S1; this protects the fab if permitting or economics change before the crystallizer is justified. Finally, request OEM-supported membrane and chemical-cleaning protocols so the recovery and rejection targets on the data sheet can actually be held on fab feed, and reference the polishing-loop engineering specs for the resistivity side and the evaporation crystallization engineering specs for the thermal-side numbers. A supplier who refuses per-stage kWh/m³ is a supplier who cannot defend the crystallizer line item.
Frequently Asked Questions
What is the maximum practical recovery for RO on semiconductor wastewater?
In the Membranes 2025 pilot, first-pass RO reached 90–94% system recovery on feed at about 700 mg/L TDS, with Na⁺, Ca²⁺, and Cl⁻ rejection consistently above 90–97%; the lowest measured Mg²⁺ rejection was 59% at the same operating point, which is the first warning that divalent scaling ions are no longer being held back. Request a per-stage recovery and rejection table at your actual feed TDS rather than relying on a system-level reuse headline.
Where does scaling actually limit ZLD in fabs?
At the membrane and thermal interface, not at the RO pump. The Handbook of Environmental Chemistry 2026 chapter catalogues silica, fluoride, calcium sulfate and TMAH-derived organics as the species that govern the recovery ceiling on fab feed, with fluoride-and-silica removal via coagulation-UF (Qiu et al. 2022) and a forward osmosis plus multi-stage NF hybrid (Lee, Shin, Ryu, Boo & Hong 2025) as the research responses. Ask the supplier for a mass balance on those species across the train.
How much energy does a crystallizer use, and what does that mean for fab water cost?
The Membranes 2025 paper reports a brine crystallizer at 52–70 kWh/m³ of feed for brines up to 300,000 mg/L TDS, and a Panagopoulos techno-economic (cited in the same paper) at about 500 kWh/m³ of inflow for a RO + brine concentrator + crystallizer train, and about 700 kWh/m³ with the NF + three-brine-concentrator configuration. Multiply those by your local power tariff to size the marginal cost of the last 9% of recovery; this is the figure a supplier should put in writing.
When should a fab stop at MLD instead of pushing to true ZLD?
When the membrane train can be held at 75–85% recovery on feed at or below about 2,500 mg/L TDS — the second-series NF operating point in the Membranes 2025 pilot — and waste heat is available to drive an MED polisher at 43–50 °C top brine temperature, MLD with thermal polishing is the lower-risk path. Permit pressure forcing greater than 90% membrane recovery, or a feed that cannot be diluted below that point, is what tips the project into crystallizer energy of 52–70 kWh/m³, and that is where the MLD-vs-ZLD decision actually lives. A defensible MLD line with a defined brine-out specification should be a contractual option regardless of which path is chosen.
