TSMC's 3nm fab uses more water than a small city

I want to talk about water.

Not ocean water or drinking water or the poetic kind of water. Industrial water. Ultra-pure water. The most refined substance in semiconductor manufacturing, more important than the silicon itself in some ways, and the resource that might be the true bottleneck of the chip industry.

TSMC’s new 3nm fabrication plant in Tainan, Taiwan will consume approximately 156,000 tons of water per day. That’s roughly the daily water consumption of a city of 300,000 people. For one factory. Making one type of product.

And nobody’s talking about it.

Why fabs need water

Semiconductor manufacturing is, at its core, a cleaning process. You deposit a layer of material. You pattern it with light. You etch away what you don’t want. Then you clean the wafer. Rinse it. Clean it again.

Each chip on a 300mm wafer goes through hundreds of processing steps. Many of those steps require ultra-pure water (UPW), water so clean that it contains fewer than one part per billion of contaminants. For reference, tap water contains thousands of parts per billion of dissolved minerals. UPW is water stripped of everything except hydrogen and oxygen.

A single wafer can require 2,000 to 8,000 gallons of UPW during its manufacturing journey, according to SEMI industry data. A modern fab processes tens of thousands of wafers per month.

Multiply those numbers and you get to 156,000 tons per day pretty quickly.

Taiwan’s problem

Here’s where this gets uncomfortable.

Taiwan is experiencing its worst drought in decades. Reuters reported that reservoir levels dropped below 20% in some regions. The Taiwan Water Corporation has been rationing water to agricultural users to prioritize industrial supply.

Read that sentence again. A country is choosing to reduce water for farms so that chip factories can keep running.

That’s not a hypothetical future scenario. That’s happening now. In 2021. The choice between food and semiconductors is already being made, and semiconductors are winning.

TSMC has been investing in water recycling and reclamation. They claim to recycle about 90% of the water they use. That’s impressive, genuinely. But 10% of 156,000 tons per day is still 15,600 tons of fresh water consumed daily by a single facility.

And they’re building more fabs. Not fewer. The demand for chips is growing exponentially. Each new process node requires more water per wafer, not less. The 3nm process is more water-intensive than 5nm, which was more intensive than 7nm.

The hidden cost of every chip

I’ve written before about how the entire world runs on silicon made by a handful of companies in East Asia. The geopolitical concentration of chip manufacturing is a recognized risk. People talk about it. Governments are responding with plans for domestic fabs.

But the water issue is barely discussed.

Every chip in your phone, your laptop, your car, your refrigerator consumed thousands of gallons of ultra-pure water during manufacturing. Every GPU training an AI model. Every processor running a self-driving car. Every chip in every device on Earth passed through a factory that uses more water than a small city.

And that water came from somewhere.

In Taiwan, it came from reservoirs that are running low. In Arizona, where TSMC is building a new fab, it’ll come from a state that’s already fighting over water rights from the Colorado River. In Europe, where Intel is planning a mega-fab, water sourcing will depend on a continent that just experienced record heat waves and drought.

The paradox

The chips we need to solve climate change (sensors, renewable energy controllers, EV power systems, smart grid processors) are manufactured in a process that puts enormous pressure on water systems that are already strained by climate change.

I don’t have a clean resolution for this paradox. The world needs more chips. Making more chips requires more water. Climate change is reducing water availability. The chips are needed partly to address climate change. Round and round.

IEEE Spectrum published an analysis of the semiconductor water footprint last year that should have been front-page news. It wasn’t.

What I think about

When I think about the chip industry, I used to think about transistors. Nanometers. Moore’s Law. The beautiful precision of lithography. I still do.

But now I also think about water. About a river in southern Taiwan feeding a purification plant feeding a fab feeding your phone. About the farmer downstream who got less water this year so that the fab could keep running.

The chip shortage I wrote about last year is a supply problem. But underneath the supply problem is a resource problem. And underneath the resource problem is a question about what we prioritize when resources get scarce.

So far, the answer is chips.

I’m not sure that answer scales.

Related thinking: