Water usage at scale requires sophisticated closed-loop systems, digital twins, and multiple filtration strategies, but can water keep up with demand?
Water — lots of it — is a critical enabler for advanced chip architectures, lithography, and back-end packaging. It feeds the ultra-pure water loops that touch every wafer, sluicing heat out of tools that run hotter at each node, and carrying spent chemistries to treatment.
The natural reaction to reports that fabs “use millions of gallons of water” is concern, but the engineering reality is more nuanced. “Use” does not mean “consume.”
Most water withdrawn from a municipal system is treated, recirculated internally through multiple duty cycles, then returned to a plant or a city works for further treatment. The fraction that is truly consumed is the portion that leaves the site primarily as evaporation from cooling and scrubber systems. As the International Roadmap for Devices and Systems (IRDS) states, “An average semiconductor facility uses millions of gallons per day, and the largest single loss pathway is cooling towers.” [1]
“At the scale of the largest greenfield developments, total water demand at full projected production loads can be comparable to a city of 1 million people,” said Slava Libman, CEO of FTD Solutions. “That does not mean all that water is lost. Most of it is treated and re-used internally, and the dominant true loss is evaporation driven by the energy intensity of the facility.”
In hot, arid basins, like the North Phoenix area, evaporative loads dominate the site water balance because thermals and airflow drive tower makeup. In cooler, humid climates, the same scale factory can show much lower evaporative loss for identical production. Fabs can squeeze demand with smarter segregation, reuse, and heat recovery, yet node progression and tool thermal budgets push in the opposite direction. The task for engineers is to quantify the balance, invest in mitigation, and be transparent about the tradeoffs.
“Before construction can occur in central Arizona, designated water providers must demonstrate they already have the water in hand for long-term growth,” said Max Wilson, water resource management advisor for the City of Phoenix. “In the case of TSMC and the surrounding development in North Phoenix, Phoenix is a designated provider and has demonstrated supplies for that growth out to a 100-year build-out.”
For municipal partners, quality is as important as quantity. Industrial reclaim plants and on-site contaminant filtration reduce downstream risk, and they are becoming standard for leading-edge projects.
“Our expectation for the North Phoenix semiconductor facility is an industrial reclaimed water plant that runs reverse osmosis upstream of discharge,” said Berai Kimball, deputy director for environmental and safety for City of Phoenix Water Services. “That step removes constituents of concern before anything reaches our sewer, which helps protect the collection system and the city’s wastewater plants.”
A modern fab’s water system is a set of coupled loops. Raw or reclaimed influent is treated to ultra-pure water (UPW) for on-wafer steps. Spent streams are segregated by chemistry, polished for reuse in lower-criticality duties, and eventually routed to cooling and scrubber systems before discharge. The overall ecosystem is easier to understand if you separate withdrawal, reuse, and consumption.
Withdrawals: Site draw depends on size, node, and climate. Consider Intel’s Ocotillo campus Environmental Assessment. At full build, the site’s three advanced fabs draw about 14 million gallons per day (MGD), with roughly 4 MGD of potable water and 10 MGD of reclaimed supply. Individually, Fab 52 is modeled at 4 MGD, Fab 62 at 5 MGD, and Fab 42 at 3 MGD. [2]
Those are withdrawals for a large multi-fab campus, not net consumption, with a heavy reliance on reclaimed municipal supply that reduces the potable burden by more than two-thirds. Similar multi-fab campuses in hot climates will cluster in this range. Smaller single-module sites will scale down accordingly. These are engineered withdrawals, not losses, and they cover both process and non-process loads.
Reuse: Internally, fabs push water through multiple duty cycles. High-spec UPW sees a wafer once, but the rinse and lightly loaded side streams are often captured and routed to less-sensitive uses such as cooling towers, house scrubbers, and utility water. The practical limit to reuse is cost, chemistry, and risk of cross-contamination at very low defectivity thresholds.
“Reuse is not optional at advanced nodes,” said Joe De Boeck, chief strategy officer and EVP at imec. “There cannot be a waste of water, so our fabs need to be laboratories that show how sustainability becomes part of the manufacturing chain.”
Consumption: True consumption is water that does not return to pipes. Two pathways dominate — evaporation in cooling towers, and humidification or gas scrubbing that vents to the atmosphere. The IRDS highlights cooling towers as the principal loss, and that aligns with site measurements across the industry. Scrubber blow-down and drift exist, but in well-run systems they are secondary. [3]
The heat side explains the scaling behavior. Each new node generation adds patterning steps, chamber cleans, and longer high-temperature cycles. More heat must move from tools into chilled water and then out through evaporative towers. That is why a leading-edge fab in Phoenix will consume a larger share of its withdrawals than the same fab in a coastal climate with higher ambient humidity and lower dry-bulb temperatures.
City utilities are building to that reality. They expect industrial customers to do more pre-treatment on-site, and they map out their own plants and aquifer strategies based on the assumption that cooling-linked consumption grows with node heat.
“In our utility, we track emerging contaminants closely and ask high-impact users to remove as much as practical before discharge,” said Kimball. “If the industrial reclaim plant does its job, it protects our downstream assets and keeps rivers and aquifers in better shape when that water eventually cycles back to drinking supplies.”
Reports must identify uncertainties. Real reuse rates are proprietary and vary by site, season, and ramp status. Public enterprise reports can illuminate trend lines, but toolmaker data sets do not map one-to-one onto wafer fabs. For example, Lam Research reports 80.6 million gallons of cumulative savings since 2019 across its own manufacturing and labs, and it documents significant internal reclamation projects. [4]
Those numbers validate the effectiveness of targeted audits and reuse at scale, yet they are not a proxy for a 5nm foundry’s fab-level mass balance. Engineers should read them as evidence that similar approaches inside a wafer fab will find real savings, not as direct benchmarks.
Finally, municipal supply must be planned over long horizons. Phoenix’s designated-provider framework illustrates how cities evaluate industrial demand alongside residential growth and iterative master-plan updates. That governance is part of the water story. It can accelerate fab projects with clarity on reclaimed offsets and pipeline sequencing, or it can stall them if planning signals are weak.
“We revisit our infrastructure master plan every five years,” Wilson said. “That cadence lets us fold in major changes, including semiconductor demand and the halo development around it, and then sequence the new pipelines and plants in the right order.”
The first constraint is not how much water you can bring to the site. It’s how clean you can keep it once it enters the plant. “Advanced-node cleans and CMP rely on UPW at > 18.2 MΩ·cm with sub-ppb organics, according to SEMI F63, and the IRDS identifies sub-10nm particle control as required for modern nodes.” [5]
That standard is achievable only when the entire system behaves like a cleanroom for water. Tanks, liners, valves, gaskets, and distribution piping cannot add extractables back into the loop, or the recycle potential collapses and more streams are forced to blowdown.
“Removing contamination from the raw water in UPW (ultra-pure water) systems is not that complicated. What is more complicated is how to do all this without producing contamination, because the system is not completely inert,” said Libman. “You need all the components to be high purity, and you need to design them to minimize contamination, including materials that will not leach and will not generate bi-product contamination (for example, TOC treating UV lamps produce H2O2 detrimental for production).”
Materials decisions are therefore yield decisions. If a sump coating or tank lining contributes low-molecular-weight organics or trace ions, membranes foul faster, polish steps work harder, and the site must retire partially spent water earlier than planned. That shows up in higher make-up demand and, in hot climates, higher evaporative loss at cooling towers. The way to keep recycling viable is to specify infrastructure by extractables and permeability, not just by generic chemical resistance, and to verify that clean-in-place recipes do not degrade those materials over time.
“Owners are prioritizing infrastructure that supports water reclamation and chemical management as sustainability becomes a central operational pillar,” said Timothy McDonough, construction solutions executive at Sherwin-Williams Protective & Marine. “Low-extractable, low-permeability linings allow higher reclamation without leaching contaminants, which is crucial in maintaining water purity in ultra-pure environments.”
There is also a measurement problem at the edge. As reuse fractions rise, trace neutrals and very small fragments can accumulate in ways that are hard to see with routine metrology. PFAS and other persistent species are the obvious examples, and several sites now design for pre-treatment or destruction at the point where those risks first appear. That is prudent engineering, but it depends on detection limits that often are at or beyond parts-per-trillion. The practical takeaway is to treat purity as a plant-wide constraint.
With purity setting the ceiling, the next lever is how streams are separated and routed, because segregation and fit-for-purpose reuse decide how close a plant can get to that ceiling in production.
Segregation is the difference between a recycle program on paper and a recycle program that survives contact with production. The principle is simple. Keep relatively clean rinses out of high-COD (chemical oxygen demand) or metal-bearing drains, polish them locally or centrally, then send them where the specification fits. In many lines, that means returning a portion to UPW make-up, and routing the rest to services like cooling towers and wet scrubbers that tolerate higher conductivity and Total Organic Carbon (TOC). The more faithfully drains are segregated at the tool, the less work the central plant has to do, and the higher the overall recycle fraction without yield risk.
UMC publishes enough hard numbers to show what disciplined segregation and reuse look like at foundry scale. Companywide, UMC reports a process-water recycling rate of 84.3%, with “new fab areas” routing wastewater into as many as 27 categories to avoid co-mingling and to simplify downstream treatment. In 2023 alone, the group logged ~825,000 tons of incremental water savings and ~5.47 million tons cumulative savings from multi-year projects. Routing and fit-for-purpose polish do the heavy lifting once purity is protected. [6]
Reclaimed sources are also a UMC priority for operations rather than a contingency. Fab 12i (Singapore) used about 4.0 million tons of reclaimed water in 2024, covering 97.6% of that fab’s total withdrawal. Fab 12A (Taiwan) added ~0.58 million tons after adding reclaimed supply in late 2022, bringing combined reclaimed use to 4.58 million tons, which is up 16.9% year-over-year. UMC’s forward targets formalize this mix, with reclaimed plus desalinated water planned to reach 18% in 2025, and 32% by 2030 across the company. [7]
The architecture of these reclaimed systems matters too. Point-of-use reclaim can return lightly loaded rinses quickly with short residence time, which reduces biological growth and limits re-contamination. It asks for sub-fab space, local control, and robust failure isolation. Central reclaim simplifies maintenance and operations, but it increases transport and mixing risk if segregation is weak. Many facilities now run hybrid schemes that polish at the edge and finish at the center, with automated valves and analyzers that switch routing when quality moves out of range. The control objective is not heroic recovery at all costs. It is predictable recovery that never compromises the wafer loop.
At OSAT scale, ASE shows what disciplined segregation and fit-for-purpose reuse look like when they are treated as production constraints rather than side projects. The company reports 2023 total withdrawals of 21.47 million tons, 6.08 million tons consumed, and site-level recycle programs that cut effluent by 12% year over year. Reclamation rates are about 70% at Kaohsiung and Chungli, 50% in Malaysia, and 37% in Singapore, with ISO 46001 water-efficiency management now embedded at multiple sites. [8]
Those numbers are not wafer-fab benchmarks, but they are proof that segregation at the drain, polish matched to the target spec, and aggressive repurposing can scale in complex assembly and test operations without destabilizing yield.
The infrastructure underlying those rates is the real lesson. Kaohsiung’s dedicated recycling plant was built in phases to process up to 30,000 tons per day at about 75% recovery, returning roughly 22,500 tons per day to service. Chungli runs about 7,000 tons per day at about 70% recovery. (8)
ASE’s operators route relatively clean rinses to local polish, push reclaimed water to utilities that tolerate higher conductivity and TOC, and keep the UPW loop protected from back-contamination.
Segregation also sets up the thermal side. Streams that no longer meet front-end specs still have value as cooling tower makeup, especially when pretreatment allows higher cycles of concentration. That reduces blowdown volumes and cuts the fraction of withdrawals that end up as consumption. The tradeoff is site-specific. Pushing cycles too high raises scaling risk and inhibitor demand. Under-segregating forces clean streams into dirty drains and throws away the UPW plant’s work. The middle path is to instrument drains with automate routing, and treat the reuse network like a production tool with SPC limits, alarms, and maintenance windows.
“We are still, for the most part, a single-pass industry. Fresh water becomes ultra-pure water and after being used in the fab becomes wastewater,” said Rushikesh Matkar, global manager for sustainability at Ovivo. “Generation on generation we see reuse increasing, but process and non-process users are actually competing against each other for the same fresh water, so segregation and routing decide how much you can truly recycle.”
Routing and reuse get you partway there. The harder step is running the entire water–and–energy loop as one coupled system. That is where plant-level digital twins and supervisory controls matter. A useful twin ingests live data from UPW skids, reclaim units, cooling circuits, and scrubbers, and ties those signals to KPIs engineers actually use — cycles of concentration at the towers, UPW recovery ratio, COD and TOC excursions in segregated drains, and drift against tool setpoints. With that model in place, you can do what ordinary PLC logic cannot. You can test “what if” scenarios before you change recipes, schedule adiabatic assist only when it saves water per megawatt rejected, and push recovery without crossing purity trip lines. The payoff is fewer gallons per unit heat and fewer surprises when the tool mix or the weather shifts.
A second benefit is failure foresight. Twins let you combine hard sensors with “soft sensors” that infer fouling or leaching from subtle shifts in pressure, temperature, and conductivity. That matters when you are running high-fraction recycle, because the risk is not a single catastrophic event. It is a slow drift in organics or extractables that your routine metrology does not catch until it touches yield. Model-based alarms and playbooks shorten that loop.
“UMC has implemented digital-twin and real-time monitoring systems to optimize water and chemical flows in our facilities,” said a UMC spokesperson. “For example, UMC utilizes intelligent control in wastewater treatment to reduce chemical consumption, and applies smart management to chiller systems to increase energy efficiency.”
On the tooling side, virtualization pushes the same idea upstream into R&D. Lam describes “virtual twins” of processes and assets to cut physical experiments. Their results show that replacing portions of bench work with validated simulation can conserve resources like water and chemicals in addition to energy and materials. In parallel, Lam reports deploying ECO sensors and dashboards to monitor process-cooling water and other utilities in real-time, which is the telemetry feed that makes twins useful on the factory floor.
The implementation details still matter. Twins are only as good as the data and the actuators behind them. If drain segregation is leaky, or if tower chemistry and fan setpoints are not addressable from the supervisory layer, the model becomes a nice visualization with no authority. Likewise, if analytics cannot see parts-per-trillion trends in PFAS-bearing streams or small neutrals from materials of construction, the twin will not prevent the slow accumulation that stresses UPW polish and narrow-margin cleans. The practical test is simple. Can the model predict and prevent the next excursion, and can you see the water saved per unit heat on a meter across seasons?
“Our version of the digital twin is a model representation of the existing (or future) facility, interconnected water systems, and their performance,” said Libman. “You do not need artificial intelligence to build the model, but some questions require AI once you start feeding continuous information into it, including automated reports, recommendations, and risk management.”
Even in a fab with strong segregation and internal recycling, the largest share of true consumption is physical rather than chemical — evaporation in heat rejection. Tool power density, duty cycle, and local weather set the tower makeup rate. Blowdown depends on cycles of concentration and pretreatment. Drift is small with modern eliminators but never zero, which is why geometry, eliminator condition, and fan setpoints still matter at campus scale. The engineering problem is to convert a variable thermal load into the smallest practical evaporative demand without trading away reliability.
The levers are familiar but coupled. Raising cycles reduces blowdown, which cuts withdrawals, but it tightens the chemistry window for scaling and corrosion. Increasing pretreatment quality for tower makeup helps cycles, yet it adds capital and operating cost. Hybrid and dry coolers chip away at evaporation, although space, noise, and approach temperature limits keep them from displacing towers in many climates. Heat recovery lowers the load that reaches the towers, but it needs a steady grade of waste heat. The upside is that tenants that can use it, which is not always a given in advanced nodes.
None of these choices is a one-time fix. Tower chemistry, fan algorithms, and setpoints drift with seasons and with tool mix. A plant that does not instrument and tune these loops will see consumption creep up even if the flowsheet is correct on paper.
“The entire industry is banking on cooling towers using the latent heat of water for heat rejection,” said Ovivo’s Matkar. “Efficiency is getting better, but, at the same time, the fab is getting bigger, hotter, and faster. Generation on generation, we see reuse increasing, but process and non-process users are actually competing against each other for the same fresh water.”
For engineers, the practical path is to connect the water model to the energy model. Target the thermal sources that most strongly drive tower evaporation, then verify the gain with instrumentation. Recipe changes that cut idle heat in a few high-load tools can be worth more than heroic tower chemistry. Point-of-use reclaim that routes lightly used rinses to tower makeup raises cycles at the same time it avoids sending clean water into dirty drains. Where climate permits, adiabatic mode scheduling around peak dry-bulb hours reduces makeup without hurting approach.
On the controls side, Cohu gives a clear example of how non-process water can be reduced with better system design and analytics. Company-wide water withdrawal fell to 48.7 million liters in 2024, a 16% decrease from 2023, driven in part by a rainwater collection and recycling system at the new Laguna, Philippines, facility, which saved about 9% of that site’s annual withdrawal. The same reports document 57.9 million liters in 2023 and note additional projects, including DI water recycling in Poway, California. [9]
While these are not wafer-fab volumes, they illustrate the same levers that matter inside fabs — separate clean streams, harvest non-process sources, and instrument savings.
“Technologies for individual water and wastewater treatment units exist, but keeping the whole site’s circularity optimized, with climate-dependent evaporation and location-specific constraints, is the core challenge,” said Libman. “The hard part is doing all this with minimum cost and risks to production or compliance.”
The chemistry at issue is not abstract. PFAS (per- and polyfluoroalkyl substances) can enter from photoresist chemistries and from polymer infrastructure, while TMAH (tetramethylammonium hydroxide) sits in the acute toxicity spotlight for legacy reasons that are still relevant. Capture is only half the job. Without credible destruction or verified sequestration, captured PFAS becomes deferred liability.
Engineers know the toolkit. On the capture side, that means reverse osmosis (OR) and nanofiltration (NF) where appropriate, with ion exchange (IX) and granular activated carbon (GAC) where appropriate. On the destruction side, thermal and electrochemical routes are advancing, but proof of mineralization at practical energy budgets is still uneven. The engineering judgment call is where to set boundaries between what gets measured and treated at the tool or sub-fab, what moves to central treatment, and what never leaves the facility at all.
“We have been focusing on the PFAS MCL that came out for drinking water, and we are currently working toward bringing on the wastewater method so that we can start gathering data and information,” said Baria Kimball, deputy director for environmental and safety for the City of Phoenix Water Services. “Our strategy right now is to get data, understand where our levels are, and then do what we need to do to comply with the limits.”
Measurement is the first limiting reagent. At parts-per-trillion levels, the metrology budget can exceed the treatment budget. That is why some operators design for zero liquid discharge (ZLD) as a compliance tool rather than a water-conservation tool. ZLD removes discharge uncertainty, but it does not remove physics. It concentrates brines that must be handled, it adds thermal load that shows up elsewhere in the balance, and it requires reliability equal to any front-end utility.
“The EPA guidance for PFAS is at 4 parts per trillion in drinking water, which is very, very, very low. At that trace level, monitoring in wastewater becomes difficult,” said FTD’s Libman. “Efficient PFAS treatment solutions are still under development. If you cannot guarantee that you can efficiently monitor and control, quite often it’s easier to just build as a zero liquid discharge. While ZLD maximizes water reuse opportunity, it has its own side effects, adding carbon footprint and risks as it necessitates reuse of a portion of the treated effluent to UPW system.”
PFAS control is improving, but practice still lags goals at leading-edge nodes where reuse fractions are high and metrology must set the guardrails. ZLD can be the right risk hedge in bounded cases, yet it is not a universal answer once energy, brine crystallization, and maintenance are included in the cost function. The priority for both fabs and cities is to segregate PFAS-bearing streams early, avoid co-mingling that multiplies treatment volumes, and set monitoring plans that withstand third-party scrutiny.
Three fault lines run through every advanced-node program. The first is measurement at parts-per-trillion for persistent chemistries. The second is whether high-fraction recirculation can stay clean enough to protect yield without hidden side effects. The third is shrinking evaporative loss without trading it for unmanageable energy and maintenance burdens. Each is solvable in principle, and each still needs production-grade evidence, not just pilot results.
When it comes to measurement, science is ahead of many plants. Capture trains can remove a wide spectrum of PFAS species and toxic bases like TMAH, but credibility rests on what can be shown at the meter. At parts-per-trillion, sampling artifacts and analytical noise can rival the signal. This is why some facilities evaluate zero liquid discharge as a compliance hedge, not a water-savings program. It resolves the discharge question, but only by moving the burden to brine management and energy.
Recirculation raises a different kind of uncertainty. Every recycle loop increases residence time and the chance that low-molecular-weight neutrals or short-chain fluorinated fragments will accumulate where routine metrology struggles. The engineering answer is stronger segregation at the tool, earlier pretreatment for risky streams, and continuous analytics that can see the problem before it reaches the wafer. Until those controls are routine, claims of near-closed UPW loops should be treated as site-specific rather than universal.
The story is not that fabs do or do not use a lot of water. They do. The story that matters is where that water goes, how much is recycled, and how much is truly consumed. Consumption tracks heat more than it tracks wafer count, which is why evaporation sits at the center of the balance. Engineers have real levers. Segregation and routing raise recycling without impacting yield. Material choices protect purity so water can live a second or third life. Instrumentation closes the loop so changes hold through seasons and tool mix. And municipal partners can turn reclaimed sources into a reliable supply when pretreatment at the fence keeps specialty chemistries out of the sewer.
Skepticism persists. Parts-per-trillion monitoring must match the confidence of the claims. High-fraction recirculation needs production-grade evidence that it does not accumulate trace neutrals or polymer-derived extractables in ways that slip past routine metrology. ZLD is a tool for compliance, not a virtue by itself, and it brings energy and brine obligations that need long-term plans.
Engineers will recognize the pattern. The next gains will come from orchestration, not from a single breakthrough. Drain segregation and fit-for-purpose reuse keep the cleanest liters in circulation. Heat-aware operations reduce the gallons that must leave as vapor. And fence-line pretreatment and transparent reporting build trust with cities and the public.
When those pieces operate together and the savings show up on the meter, water at scale becomes a solvable part of advanced-node manufacturing rather than a showstopper.
Sources
 |
 |
 |
 |
 |
 |
 |  |
 |
 |
 |
Leave a Reply