Here’s part of that reason.
That uses some water, but I’m fairly sure that the numerous washing steps in IC production dominate the water use.
Really it should all be recycled. Or perhaps purified to the point where it can be used for other purposes. Use it for concrete and it might not even need to be purified.
Neither does your car’s ECU. It resides under the dash for the most part. The boards in my 1992 BMW were in a cast aluminum case and had connectors that looked like they belonged in a fighter plane. The box had it’s own fans connected to outside air by ductwork.
Did you mean hardly seems surprising? I’m pretty sure lots of consumer chips run at higher volumes.
Do you know the relative chip sizes? My contact with TSMC was for huge processors with a pitifully small number of parts per wafer.
Perhaps not the engine control module, but there are plenty of other chips that are in the engine compartment. Two examples from my car: The ABS module and the generator regulator.
Foundries produce wafers. While the foundry will perform some per-wafer testing to ensure that the wafer was processed within specified limits, they do not test the individual die. Wafers are sent to one of more different facilities, frequently in Malaysia or Singapore, for probe testing, dicing, assembly, and final package testing.
The foundry gets paid per wafer, they do not particularly care how big the die is or what the end market is. The cost of a wafer will depend on the particular process flow used (affects number of processing steps and the variety of processing equipment used), the number of wafers of a particular design that are processed at one time (setup time for swapping out reticles) and the total number of wafers your company purchases per year (how big a customer you are).
Not true in our case. Instrumentation was placed on the wafers, and certain results meant that the wafer as a whole was rejected, and we weren’t charged. But we did significant wafer probe (not at speed) on individual dies before the dies were packaged. Then the wafers were sent to another place in Taiwan (for us) for packaging and package testing, along with a wafer map showing which dies to package. Our tests were so long that they were split, and there was also testing at different temperature corners, plus post-burnin test.
Wafer probe was done inside the foundry, which makes sense since I assume they used the feedback. Our parts were often process drivers, though.
Our packages were very expensive, and our yields were relatively low because of die size and aggressive technology, so it would make no economic sense to send untested wafers to be packaged.
I didn’t mean that die are packaged untested, just that probe test of the actual die is usually done at a different location. Probe testing for wafer acceptance is done using test structures places in the scribe lanes between dice. This testing is done by the foundry before the wafers are shipped.
Our probe was done within the walls of TSMC, but I’m aware that anything can get shipped all over the place. When I was in Bell Labs wafers were manufactured in Orlando and shipped to Singapore for packaging.
For context: an olympic-sized pool contains 2.5 M liters of water according to my googling, which is just about 2 acre-feet. So 120 acre-feet/day. Or 44,000 acre-feet/year. In the US, an acre-foot is roughly the amount of water a family of four consumes in a year, so that much water would support a population of 175,000 people (domestic water use only - not the agriculture or power generation to support them.) Taiwan’s population is 23.5 million.