Google + SkyWater Moving To 90nm For Their Open-Source Silicon Design Initiative

[phoronix]

Phoronix: Google + SkyWater Moving To 90nm For Their Open-Source Silicon Design Initiative

Google and SkyWater have teamed up the past few years with an open-source design kit for allowing projects to see their open-source silicon designs manufactured. This effort started off on a 130nm process node but announced today is the rolling out of 90nm manufacturing...

https://www.phoronix.com/news/Google-SkyWater-90nm

[microcode]

Super cool. If I didn't have more than full time work right now, I'd be working on composable parametric cells for this. How hard is it to buy process time with SkyWater?

[willmore]

It's times like this when I wish I had taken the VLSI design class, but I just couldn't fit it in with all the other stuff I was taking. ;( One of my coworkers in college took it and they got to fab chips. I remember the day he brought it in to work to show us. Lots of geek jealousy that day.

[coder]

This is nominally the same technology used to fab mid-generation Pentium 4 CPUs, nearly 20 years ago.

[qarium]

This is nominally the same technology used to fab mid-generation Pentium 4 CPUs, nearly 20 years ago.

but i think today you get more performance out of this node than 20 years ago ..

thats because 20 years of research and trial and error did in fact result in faster micro-architectures.

for stuff like libre-soc this is a good start.

[coder]

but i think today you get more performance out of this node than 20 years ago ..

Well, we don't know if it's exactly comparable to Intel's process, since the number & types of layers likely differ.

thats because 20 years of research and trial and error did in fact result in faster micro-architectures.

Intel made the Pentium M on the same process node. That was definitely better than the Pentium 4. However, a lot of microarchitecture innovations are only possible with higher transistor-counts. You can't simply transplant things that are working well at 7 nm back to a 90 nm chip.

Anyway, my point was just to put this in perspective. It's a completely noncompetitive node for anything general-purpose. It only really makes sense for special-purpose ASICs and maybe analog chips.

for stuff like libre-soc this is a good start.

Maybe as a learning experience, but not for making useful chips. For that, they ought to get on a similar 28 nm node to what R.Pi is using.

[qarium]

Well, we don't know if it's exactly comparable to Intel's process, since the number & types of layers likely differ.
Intel made the Pentium M on the same process node. That was definitely better than the Pentium 4. However, a lot of microarchitecture innovations are only possible with higher transistor-counts. You can't simply transplant things that are working well at 7 nm back to a 90 nm chip.
Anyway, my point was just to put this in perspective. It's a completely noncompetitive node for anything general-purpose. It only really makes sense for special-purpose ASICs and maybe analog chips.
Maybe as a learning experience, but not for making useful chips. For that, they ought to get on a similar 28 nm node to what R.Pi is using.

well your pentium4 vs pentium M example makes hope that today they get more performance out of 90nm

"simply transplant [...] 7 nm back to a 90 nm chip."

well i think we should really try this. of course you can not do many cores but if you do a single core cpu instead of an 8core cpu then you maybe can do it.

"It's a completely noncompetitive node for anything general-purpose. It only really makes sense for special-purpose ASICs and maybe analog chips."

well the good news is we allready have fully open 14nm power9 cpus...
this means we do not need a general purpose 90nm chip...

and libre-soc is not a general purpose project if you see these texture processing units it is ASIC...

"Maybe as a learning experience"

but if you see this libre-soc project they have ZERO experience so any learning experience should be better.

right ?