Sort of incredible that Germany is getting decade old technology for this much investment.
Germany needs CPUs for the industry. Cars, machines, planes, wind turbines and so on :)
But you cannot build Personal-Computers with them? That failure is caused by Siemens (Siemens-Nixdorf) and AEG. In GDR/DDR they were close to build a 386 compatible CPU. The remaining knowledge is still somewhat available, Global Foundries in Dresden.
But the CPUs for PC? Yes. Yes. Okay. Intel will build a Fab there, too ;)
There are plenty of application were structure size, performance and/or power draw doesn't really matter.
E.g. Automotive / Industrial applications
And it can vary too, mass produced / generic chips can be more cost efficient if they can make more of them on a single wafer.
I thought a big part of the reason why that's the case in general is because the old fabs that house those old processes have already been depreciated - customers basically only pay for maintenance and the of raw materials.
But that doesn't really apply to brand new fabs.
The Chips Act was marketed as an attempt to bring back "cutting-edge chips" to Europe, and to promote research and innovation. In the end it turns out to just be a huge government subsidy for a bog-standard fab which is at least a decade behind what TSMC is building in Taiwan.
Great for automotive companies who want to strengthen their supply lines, but pretty pointless when it comes to doing what it was actually supposed to do. Europe already has plenty of fabs like these which did not require billion-euro handouts.
That's great news for Germany then.
2. It's not as if BEVs are chip-free, and most of the chips used aren't ultra-high performance things that require the very latest generation fab equipment.
You are not going to use some 3nm EUV boondoggle process just to pump out voltage regulators or sub-100MHz microcontrollers, but you need tons of those for pretty much every application.
Sure there might be four or five CPUs on a modern process in a modern car, but there are COUNTLESS simpler/slower/cheaper chips required that are made on an older process and this is not gonna change anytime soon.
Less gigahertz and gigaflop, more kilovolt and kiloampere.
It's not a dumb approach.
I think that would be more credible if Global Foundries didn't already have "Fab 1" up and running that has a mature 22nm process.
This deal with TSMC at least seems like it provides a way forward with people that know how to execute. One that the Global Foundries option can't credibly provide.
But TSMC isn't willing to make those investments in Europe either. It's their goose that lays golden eggs, and keeping it all in Taiwan makes a lot more sense from a geopolitical perspective. The only reason they are now building this fab in Germany is because it is ancient technology to them and they are basically getting a free fab out of it.
There is pretty much no way forward for this. You'd have a point if a European company managed to sign a technology transfer for last year's node, but that's simply not what is happening here.
You'll have to excuse me if I'm being daft, but I assume this is not the CHIPS Act we have here in the US?
I'm aware Europe's trying (or has done) something similar, but if it's called the Chips Act this is the first time I've heard of it that way.
You can't tell the people you just want to subsidise the automotive industry.
Not everything needs the 5nm cutting edge, and to be fair calling
> 300mm (12-inch) wafers on TSMC’s 28/22 nanometer planar CMOS and 16/12 nanometer FinFET process technology,
old is a bit unfair
22nm dates back to 2012, and TSMC has been doing 14nm FinFET mass production since 2014. So yeah, a decade behind cutting-edge can indeed be called "old" when the entire point of the Chips Act was to catch up with Asia and become a center for innovation.
Plus, the only vendor of EUV lithography machines is likely 100% committed for the foreseeable future. DUV machines are much easier to get - you can go to ASML, Nikon, Canon and I assume a few others - while you find some space on the EUV order books.
Besides, there are already plenty of fabs in Europe at these nodes at companies like GlobalFoundries. Getting another one isn't going to be very useful in trying to catch up, especially when it is one from just before a huge paradigm shift. It's not going to teach them anything about EUV, so they would be trying to catch up from 7-8 nodes behind without even using the technology needed to reach the cutting-edge.
It just cannot take the voltages involved and fabbing switching regulators or CAN bus transceivers on a 3nm process would be a waste. Especially when a cheaper and higher voltage capable 180nm or 300nm process is available.
Right tool (and process) for the right job.
No, TSMC 22 has no eFLASH, expensive SRAM, lacking of analog capability, lacks high temperature tolerance, and is tied to multiple patterning design flow.
TSMC 40 on the other hand is the last portable node, with MRAM, eFLASH, high temp, some mixed signal, RF, no MP, and everything else your soul desires.
Maybe they'll add the missing parts as chiplets?
It isn't for anyone besides HN/Twitter/&c. tech fanatics tbh
Having the bleeding edge tech sounds great, but where are the customers that are going to pay for bleeding edge output and the countless billions of investment it takes to get set up to produce it?
Decade old technology sounds very sensible.
It's all good and nice if private companies want to do this, they should know best after all. Spending another 5 billion on taxes on this is flabbergasting, especially in an area of the country that is incredibly foreigner-unfriendly (to say it mildly).
Many chips don't need fancy new tech.