The Big Semiconductor Water Problem
If you want to watch the video, it is below
Today, I write this right now in the midst of a massive deluge in Taipei. I heard a statistic last week that there has only been 6 days of sun in the city.
I got a lot of emails from people on this video, challenging my off-hand comment about water being taken away from the farms to funnel to the semiconductor fabs. I don’t really have much to say to them about this. After all, it’s a localized matter. But the good thing is that recycling technologies are a lot better today than they were many years ago. They have to, because you can’t simply draw water from a place in an unlimited manner. Even if that place isn’t a desert.
As I am writing this, Taiwan is suffering through one of its worst droughts in many years. The northern-western part - Taipei, New Taipei City - is generally fine. There are enough reservoirs.
But other areas like Tainan, Kaohsiung, Taichung, and Hsinchu are having some issues. Their water infrastructure is not as resilient or their water demand is greater. There are reports in the media of TSMC budgeting for millions of dollars to truck water down to its fabs in Tainan.
Ah, and TSMC along with Intel and a bunch of other chip companies are building fabs in Arizona. A land inundated with clean, fresh drinking water.
The company - and the industry as a whole - has long had to deal with water usage issues at their facilities. And the problem is only going to get worse as we progress forward.
Water in Chip Fabrication
When people think about water shortages, they think about faster showers, less flushing, and the like. And saving water in the home is important. But water usage in the agricultural and industrial spaces far exceeds home residential usage. And semiconductor manufacturing in particular uses a lot of it.
Much of this water is for washing wafers. 30-40% of the steps in modern wafer fabrication involves some form of wafer cleaning.
For instance, there is a step called "wet cleaning", where sequential wash steps are applied to remove organics, metals, and particles from the wafer. Between each of those washing steps, water is used to rinse off the chemicals from the previous steps.
And semiconductor companies cannot use regular tap water out of your sink to wash these wafers. Instead they must use ultrapure water. Yes, that is a real industry term.
Ultrapure water is water. It is used in various other applications like nuclear energy or chemical manufacturing. But the cleanliness standards in the semiconductor industry are stricter than almost all the others.
In the industry they are trying to remove what are called “killer particles”, particles large enough to affect a chip's nano-architecture. These can be microorganism waste, organic particles, dissolved metals, gases and more. As much of these as possible must be scrubbed out of the water.
It is a lot of work to generate and verify the purity of ultrapure water. There are over a dozen steps and a number of those steps are repeated over many loops.
It takes about 6,000 liters or 1,600 gallons of city water to create 3,800 liters or 1,000 gallons of ultrapure water. And a single 200-mm or 8-inch wafer can use over 5,600 liters or 2,000 gallons of ultrapure water.
The Chips Matter Too
The types of circuits made also affect water usage. Wafers are made in layers. And as I mentioned earlier, each time a layer is made, it needs to be washed and cleaned before the next layer can be applied.
Logic circuits like microprocessors have more layers and are more complex than other types of semiconductors such as memory ICs, display panels, or solar panels. Thus, their manufacture requires more water and energy.
In 2020, TSMC noted in their corporate social responsibility report that they were not able to reach their water consumption goals, set a few years back. They had wanted to reduce the liters of water consumed per 8-inch wafer equivalent layer by 27%. They ended up doing just 5%. The reasons for this are due to increased demands for process cleanliness.
In other words, the more advanced the node process is, the thirstier it tends to get. It means more layers to wash. More ultrapure water needed. And more stringent standards for that ultrapure water. With each new process node, the threshold for something to be a killer particle gets a little smaller, making it that much harder to scrub out.
So the cutting edge chips of today take more water to fab than they did 10 years ago, despite all the advances made in water reuse and conservation. And it is likely the trend will continue.
We can’t just only think about all the water coming in. We also got to consider what’s coming out.
These processes generate a lot of wastewater and much of it needs to be treated. In 2013, TSMC’s fabs in Tainan generated 19 million tons of wastewater. Tainan the entire city that same year generated 29 million tons.
Wastewater left behind from this wet washing contains a high amount of pollutants like ammonia, phosphate and fluoride. There are studies of semiconductor workers getting cancer from exposure to these chemicals.
And not all wastewater is the same. Some of it needs more expensive and extensive treatments than others. For instance, ultraviolet light which is used for disinfection, is relatively cheap. But others like reverse electrodialysis, where salt ions are transported through a membrane using the power of friendship, cost substantially more.
So to properly deal with all these different types of wastewater, large foundries separate the streams and give each one their own special treatment, addressing over 25 chemicals in total.
Some of the final treated water output is reusable, meaning that you can put it back into the system and generate more ultrapure water with it.
But some of that water won’t be reusable, so it is most often put into the cooling tower where it is evaporated away to help maintain temperatures. This helps control the environment within the massive fab, the single biggest use of the facility's energy and water after direct manufacturing.
Such a system is complicated and expensive. So for many smaller companies, it is more cost effective to simply aggregate the water, broadly treat it, and return it to the city once it meets minimum health and safety requirements. They would then get new raw water from the city to process into ultrapure.
Large fabs like TSMC, Samsung and Intel have to implement these water reuse and recycling technologies because of their scale. They have to make so many chips that without these water recycling techniques they would run out pretty quick.
This explains why TSMC's recycle rate in 2017 was 87%, but per a Singapore study also done in 2017, the average semiconductor fab recycles less than half of that: 42%.
Finding Water in Tainan - TSMC
In 2019, TSMC fabs in Tainan alone consumed 50,000 metric tons or 50 million liters of water a day.
Sourcing all that raw water to be used is a challenge. The area draws from the Nanhua and Zengwen reservoirs, which as of right now are dangerously low in capacity. And Taiwan's south is also its agricultural heart, which needs water to feed its population.
To keep from draining the area's water supply, the company's fabs had to optimize their water draw and usage.
In addition to the water recycling technologies we talked about earlier, the company built a large rain collection system to capture rainwater and store it in 700 ton tanks. After running some checks, the rain can be processed, used for landscaping purposes or sent to the sewers.
Meanwhile, the company has signed tenders to build water reclamation plants near its fabs. These would recycle industrial wastewater using reverse osmosis, hollow fiber ultra-filtration, sand filtration, and more. These procedures are kind of like Singapore's NEWater.
The first ones to come online are in Tainan Science Park. Additional reclaimed water capacity is expected to arrive in the coming years from future plants in the Yongkang and Anping areas of Tainan.
Water in Arizona
Considering all we have learned thus far, Arizona might seem a funny place to put a fab. Because as you might notice, much of the state is a desert.
Currently, 40% of total water demand is met by nonrenewable groundwater. The Colorado River and other surface water sources, 57% and reclaimed water provides the rest. So basically where the Colorado falls short, the shortfalls gets pulled out of the ground. This is not quite sustainable.
For the most part, the state government has managed to control water usage by transferring water away from the farms to the city. Agriculture devours the vast majority of water in the American southwest - almost 90% of total demand in 1980. Turning farmland in cities and suburbs over the years has helped curb overall water usage.
The state has attempted to come up with a coordinated plan for corralling water demand. But results have come up dry and people are still pumping groundwater. It is a concern.
Why Arizona Then?
So how does it make economic sense for a rational government to want to put a semiconductor fab there? There are two prevailing reasons.
The first is that high volume semiconductor manufacturing is such a profitable and high-value add industry that the state is willing to take on the cost. Even if that means shelling out to build water infrastructure or taking someone else's water allotment away.
The state gets paid a lot more for having foundries on its land than with other usages. A 2012 paper compared the economic value of four land usages in the city of Chandler, Arizona - where the fabs will be. The four usages were high value semiconductor manufacturing, industrial manufacturing, single story office buildings, and retail.
They found that the economic benefits from semiconductor manufacturing outweigh the costs of providing the water. Governments gets paid a lot more on a per-acre basis with semiconductor manufacturing than with other land usages. They are very much incentivized to bring these facilities here.
Much of this revenue enters the state coffers through property taxation. The industry brings less high paying jobs on a per-acre basis than with the other land usages. It is likely due to the fact that most of the work is automated.
The second reason has to do with these water recycling and reuse technologies. The industry is rich enough to be able to invest in water saving, reusing, and recycling techniques. That means a high percentage of water - far higher than with the other land usages - is theoretically returned to the wastewater system.
Thus for the foundries themselves, the other advantages that Arizona provides - a ready supply of educated talent, tax incentives, seismic stability among others - win out over the simple lack of water.
High value semiconductor manufacturing is so valuable and profitable that governments are more likely to find water for it by taking away someone else’s share - usually a farmer’s. That’s been the case in both Taiwan and Arizona, where farmers are finding their water allocations cut and diverted to tech companies.
Agriculture is the number one usage of water virtually everywhere. There is an argument to be made that most agriculture as it is practiced is not an economically efficient use of precious water. Especially if it comes out of the ground.
But there are political consequences associated with doing this. Farmers in the USA are some of the most valuable constituents around. Losing the farmers is a fast way to losing your reelection. So governments have to strike that balance delicately.
The semiconductor industry’s thirst will not abate. Moore's Law will continue pushing ahead. Water recycling, reduction techniques, and reuse technologies have to keep up. If not then at some point, there is not going to be enough water to go around for everyone.