I live in Albany, which is quietly becoming a semiconductor town. The federal NSTC EUV Accelerator opened up the road in July 2025, and the Albany NanoTech Complex has been running 300mm cleanroom R&D for years. I spent a decade diagnosing machines for a living — closed-loop coolant systems, hydraulics, HVAC, electrical faults — so when I started reading about what happens inside a fab, I didn’t head for the lithography. I headed for the plumbing.
It turns out the plumbing is one of the most extreme engineering systems in any industry. This is what I found in a week of research sprints, sources included, mistakes mine.
Water so pure it barely conducts
A fab’s process water isn’t clean the way drinking water is clean. It’s clean the way a vacuum is empty.
The measure is electrical resistivity. Pure water barely conducts electricity, so the purer it gets, the higher its resistivity climbs — up to a hard physical ceiling of about 18.2 megohm-centimeters at 25°C, the point where the only ions left are the ones water makes by itself, tearing its own molecules apart. Semiconductor ultrapure water is specified essentially at that ceiling: 18.2 MΩ·cm at the point of use, with organic carbon down at parts-per-billion levels. Anything dissolved in the water — a salt ion, a speck of silica, a fragment of a dead bacterium — is a potential defect on a transistor measured in nanometers.
Getting there is a train, not a filter. City water is pretreated, then driven through reverse osmosis, where membranes reject roughly 95 to 99 percent of dissolved salts. Then it hits my favorite machine in the whole system: electrodeionization, a continuous polisher that combines ion-exchange resin with a DC electric field. The field keeps regenerating the resin electrically, which means no shutdown cycles and no tanker trucks of regeneration chemicals. Then more polishing — ultrafiltration, UV, degasification — and a distribution loop that never stops moving, because ultrapure water standing still starts dissolving its own pipes and growing biofilm.
The scale is hard to hold in your head. A federal environmental assessment for one three-fab campus in Arizona put its estimated water demand at 14 million gallons a day; IEEE Spectrum’s rule of thumb is that a single large fab can run up to 10 million gallons daily. Leading fabs now reclaim and recycle a majority of their process water, which is its own engineering discipline on the back end.
The dirty half is chemistry
What comes out of a fab is as interesting as what goes in, because it isn’t one waste stream — it’s several, segregated by chemistry, each with its own treatment line.
Etch and scrubber water carries hydrofluoric acid and its cousins. The standard treatment is elegant, old chemistry: neutralize the pH and react the fluoride with calcium — lime or calcium chloride — so it precipitates out as calcium fluoride solids, the same mineral as fluorite. Vendors build entire treatment lines around pH neutralization, fluoride removal, and slurry effluent.
Polishing steps and copper plating produce water carrying dissolved copper, which has to come out before discharge. The toolbox runs from hydroxide precipitation through ion exchange to electrochemical cells that plate the copper back out as recoverable metal — turning a compliance problem into a small mining operation.
The people who run it
Here’s the part that got me. All of this — the RO trains, the polishing loops, the neutralization systems — runs 24/7 under PLC controls and SCADA interfaces, with continuous instruments watching resistivity, organic carbon, and temperature around the clock. Somebody walks rounds past those pumps and membranes every shift. Somebody hears a bearing starting to go before any sensor votes on it. Somebody logs the work order, pulls the sample, decides whether that pH excursion is a sensor drifting or a process moving.
New York State takes this seriously enough that wastewater treatment operators carry graded state certifications under 6 NYCRR Part 650, with plant grades set by a point system and operators qualified level by level.
I recognize these people, because I spent ten years being one in a different building. A BMW cooling system and a process-cooling loop are cousins: pressure, flow, temperature, a pump that cavitates, a sensor that lies, a closed loop that punishes you for guessing. The diagnostic method — symptom, instrument, isolate, verify — doesn’t care whether the fluid is coolant or eighteen-megohm water. The domain knowledge is different, and I won’t pretend otherwise: fluoride stoichiometry and resin chemistry are new to me, and the people who run these plants have earned knowledge I don’t have. But the discipline is the same discipline, and reading their world felt less like tourism than like visiting a neighboring shop.
Why I do this
I run research sprints like this for the same reason I rebuilt engines before I was allowed to diagnose them: you can’t reason about a system you’ve never traced end to end. The method is simple and repeatable — pick a system, chase it from inlet to outfall, write down what the primary sources actually say (standards bodies, government filings, the vendors who build the equipment — not content farms), and flag what you couldn’t verify. Two things from this sprint I couldn’t pin to a primary source at the rigor I wanted: exact fab-by-fab recycling percentages, and the full parameter list on any specific plant’s instrumentation. So those stay qualitative above.
Curiosity with citations is a habit, not a credential. It’s also, I’ve come to think, the most transferable skill I own.
Sources are linked inline. If I’ve gotten something wrong, I’d genuinely like to know — the contact page works.