Discussion in 'Article Discussion' started by bit-tech, 15 Mar 2018.
You'd think if a power outage had the chance to cause damage that you'd run some sort of UPS system, i guess the cost of batteries and generators outweighs the cost of damaged wafers.
Fabs use tens to hundreds of megawatts of power, not just for the FEoL/BEoL equipment, but for all the chemical processing (for process chemistry) and the massive amount of environmental processing the cleanrooms need. If it takes a few hours to complete a fab cycle and secure the pods in a clean area (most fabs don't have a dedicated clean holding area, using racks stored on the ceiling for the robots to easily access and avoid wasted floor space) you need to keep all that support equipment running for hours to avoid spoiled wafers. If cleanroom spec is dropped because that mighty HVAC equipment shut down, then you need to consider all wafers in that environment to be contaminated and nonviable (to avoid them spreading contamination inside the fab equipment). Hundred-Megawatt-hour UPSes are effectively power stations in and of themselves.
3.5% of March output destroyed
Waits for 6% price increase in April
Not really, going on this (PDF warning) article written five years ago on saving power in large semiconductor fabs they use around 100MWh per hour, that's well within the capabilities of a UPS.
Normally backup generators in these types of installations will power critical and safety systems only, so will not power the whole plant if required. It could be that there was a backup in place, but it didn't work as planned? I haven't read anything other than the Bit-Tech article, so don't know the details.
Technews.tw article which is the source doesn't provide more details, so you didn't miss much:
How the chuff did you find that?! I spent half a bleedin' hour this morning playing around with find-in-page, Google Site Search, and Google chuffin' Translate trying to track that down.
I'll add a link in the piece - ta!
Found the link on guru3d.
Uh, 100 MWh per hour is 100 MW. Even assuming perfect power-factor, that would be a 100,000,000 VA UPS. 2kVA/u seems about standard, so that would be 50,000U of racked UPS units, or 1,190 racks for UPS units alone.
You don't use standard UPS units in large buildings, you have a bank of batteries supplying power during the time it takes generators to automatically fire up.
Biggest battery, built by Tesla in Australia, is 100MW (129MWh), previous to that there was 20 MW/80MWh one also built by Tesla, so about what UPS are you talking about? Its even out off reach for on-site backup generators (diesel ones top out at ~4-5MW), for that amount of power you would need medium sized gas turbine. If the power outages are rare, then loosing 11% of your production is probably more economically viable than building a small power plant + national grid level balancing battery.
This tbh, the cost of mitigating for power outages probably outweighs the cost of losing a few wafers in a [rare] outage. If outages were frequent enoguh to warrant mitigating against then they'd probably move somewhere else.
It's the same with security, you get to a point where the cost of being broken into is less than the cost of ensuring you don;t get broken into *Looks in TalkTalk's direction*.
You can have more than a single generator and they would only need to supply power to the system needed to prevent the destruction of the wafers, however i get your point that it's probably not cost effective, some fabs do away with the worry of power cuts by building there own power station so Samsung could always do that.
The problem is the systems required to do that are all the systems. Turning off the vending machines in the canteen means nothing when your plant equipment is sucking down Megawatts.
Most of the fab equipment I'm familiar with (here at ASML Netherlands) does have internal UPS systems. These UPS systems, as far as I know, are only to keep the main computer alive, so when power comes back it can report individual wafer status back to the Wafer Information System (WIS). This system can (potentially) be used to determine which wafers can be recovered/reworked.
A good part of the wafers could be recovered. Think of wafers that were undergoing processing in coat/develop tracks or during exposure in litho tools (like our ASML scanners). Wafers undergoing things like implant or etch will be scrap. The question is though, and as I do not work in a fab I can't say for sure, what's the cost of trying to recover wafers?
IMHO a big pro would be that you don't loose your entire run and half of the fab will not be idle waiting for the new wafer-starts to reach tools further up the line.
A con might be potential lesser quality output for a couple of days, it's up to the fab to decide.
Separate names with a comma.