GlobalFoundries introduces 22nm FD-SOI process technologies

[Snafuh]

GF showed a chart about performance and cost

More slides about use and benefits of 22nm FD-SOI here: http://www.golem.de/news/22fdx-glob...e-chip-fertigungstechnik-vor-1507-115195.html

They also say it is scalabe to 10nm "ensuring that FD-SOI has a bright future"

 

[Idontcare]

GF showed a chart about performance and cost

More slides about use and benefits of 22nm FD-SOI here: http://www.golem.de/news/22fdx-glob...e-chip-fertigungstechnik-vor-1507-115195.html

They also say it is scalabe to 10nm "ensuring that FD-SOI has a bright future"

Ouch, projected to reach that point nearly 4 years from now!? (Q1 2019)

A lot can happen in 4 years. Like 7nm and EUV.

 

[Tuna-Fish]

What customers can we expect on 22 nm FD-SOI, and why?

Isn't everyone heading for 14/16 nm?

The vast majority of silicon chips made are not on a leading edge node. 14/16nm have very high upfront design/mask costs. This 22nm process family is specifically designed to have low upfront costs (at the cost of lower peak performance and higher per-unit costs). This means it's bad for the kind of high-end mass-market products that go on 14/16nm, but it's meant to be a much better choice for all those small design firms that make chip runs that are a lot smaller than what Intel, Apple, Samsung, Qualcomm or AMD do.

It's still an open question if the node will be a success among those companies -- there is probably a lot of investment into TSMC-specific tools in them and GF will face an uphill battle to attract customers.

 

[DrMrLordX]

Holy CRAP it worked.

Scary.

 

[Tuna-Fish]

Ouch, projected to reach that point nearly 4 years from now!? (Q1 2019)

A lot can happen in 4 years. Like 7nm and EUV.

They promise easier design and half the mask cost today with somewhat higher per-unit costs for now and "totally going to down in the future pinky swear".

For the customers that go for this the upfront costs are what matter and 7nm and EUV are not going to bring those down...

 

[NostaSeronx]

- STMicroelectronics foundries is being tooled for 22FDX. // 3x 300mm -> Crolles 2 @ STM, Fab 1 @ GlobalFoundries, Fab 8 @ GlobalFoundries.
- Gate-First FUSI HKMG makes a comeback.

22FDX Metal Stack Examples;
-LP-
M1-2 -> 64nm Pitch
M3-7 -> 90nm Pitch
M8-9 -> 720nm Pitch
-HP- +// UHP Extension
M1-2 -> 64nm Pitch
M3-8 -> 90nm Pitch // UHP -> M7/M8 -> 180nm
M9-10 -> 180nm Pitch // UHP -> 320nm
M11-12 -> 720nm Pitch // UHP -> 1440nm

The BEOL is 14nm(STM)((first two layers)) + 28nm(STM)((everything beyond first two layers)) rather than just 28nm(STM)((whole stack)).

Should probably point this out as well;
FEOL is not STM's 14nm FDSOI, but STM's 14nm FDSOI plus. STM slides say 60% higher speed @ iso power with 14nm FDSOI plus over 28nm FDSOI, rather than 30% with normal ole 14nm FDSOI.

It is also a bit silly that it says FinFET-like performance, but when you check Ingenic Semi's benchmarks it is;
- w/o BB -> 200 MHz slower, 2 Hrs longer battery life.
- w/ FBB -> 200 MHz faster, 8 Hrs longer battery life.
// http://www.ingenic.cn/en/?product/id/2.html <-- This ported down to 14nm FinFET(2.8 GHz-11 hrs) / 22FDX(2.6-13 hrs/3.0 GHz-19 hrs).