Sunday, January 22nd 2012

28 nm struggles: TSMC & GlobalFoundries

Making silicon chips is not easy, requiring hugely expensive fabs, with massive clean-room environments and at every process shrink, the complexity and difficulty of making the things goes up significantly. It looks like TSMC and GlobalFoundries are both having serious yield problems with their 28 nm process nodes, according to Mike Bryant, technology analyst at Future Horizons and this is causing a rash of non-working wafers – to the point of having nothing working with some chip designs submitted for production. It seems that the root cause of these problems are to do with the pressures of bringing products to market, rather than an inherent problem with the technology; it just takes time that they haven't got to iron out the kinks and they're getting stuck: "Foundries have come under pressure to release cell libraries too early – which end up with designs that don't work," Bryant said. In an effort to try and be seen to treat every customer equally, TSMC is attempting to launch ten 28 nm designs from seven companies, but it's not working out too well: "At 45-nm, only NVIDIA was affected. At 28-nm any problems for TSMC will be problems for many customers" said Bryant.
GlobalFoundries are also struggling, although perhaps not quite as badly, according to Bryant: "However, there are recent comments of major yield problems with their 28-nm process actually being even worse than at GF [Globalfoundries]". Their 32 nm & 28 nm nodes are struggling, because they are using problematic gate first processing and this is worsened by the fact that they are using processes from two companies, AMD and IBM. Trying to debug two processes at once is causing serious headaches, compounded by a lack of cooperation between bases in Dresden and the US which appears to be caused by bad management.

Note that Bryant's assertions are at odds with what TSMC's CEO and chairman Morris Chang said when he spoke to analysts the previous day about the company's fourth quarter financial result: "Our 28-nm entered volume production last year and contributed 2 percent of 4Q11's wafer revenue. Defect density and new progress is ahead of schedule and is better than 40-45-nm at the corresponding stage of the ramp-up. We expect 28-nm ramp this year to be fast and we expect 28-nm will contribute more than 10 percent of total wafer revenue this year." It will soon become apparent who is right.

Regardkess, the pressure isn't going to let up for these companies, since Intel have been successfully manufacturing at the 32 nm level for a year, making their Sandy Bridge processors. On top of that, in April, Intel will introduce their significantly smaller 22 nm-based process technology in the form of their Ivy Bridge CPUs which have been demonstrated to work very well indeed. These are based on Intel's proprietary Tri-Gate 3D transistor technology too, which gives further performance increases, hence upping competitive pressures significantly.Sources: TG Daily, EE Times
Add your own comment

33 Comments on 28 nm struggles: TSMC & GlobalFoundries

"At 45-nm, only NVIDIA was affected. At 28-nm any problems for TSMC will be problems for many customers" said Bryant.

So I take this as a hint that NVDA will delay once again their line of next gen gpu's. Or something they(or anyone) can use to say that it's not entirely their fault i presume.
Posted on Reply
And besides, nVIDIA and 45nm?
Is there a 45nm GPU by nV? Must've missed that...
Posted on Reply
largon said:

I guess by "development" you only meant the lenght of the development cycle, not technological advancement, shrinkage of fab processes, etc.
The SOI department are already testing 8nm-5nm FD-SOI FinFet UTBOX

Time to Volume is longer because the cost to mass produce the tools is expensive

largon said:

Yet, AMD's chips are slower and consume considerably more power.
Slower is an architecture thing(They choose to go to FMA(Then Round) in a FADD(Then Round)+FMUL(Then Round) world) but by design it is above Intel designs where 32nm/22nm are still stock @3.5GHz Stock for the i7 2700K/i7 3770K

Trinity 3.8GHz @ 100W
Komodo/Vishera 3.9-4.2GHz @ 125W

Sandy Bridge Integer Pipeline 14
Bulldozer Integer Pipeline 15

The clock rates wouldn't be possible if the fabrication process was doo doo, We didn't get 65nm again(You know Phenom I @ 2.6GHz and less, Imagine FX @ Opteron 4200 Clocks while @ 140Ws)

xenocide said:
Intel's is more expensive because it is a better product with more reliable yields.
I don't see Intel designs getting any overclocking records with x86 CPUs

So Reliable my bum
Posted on Reply
seronx said:
I don't see Intel designs getting any overclocking records with x86 CPUs
They held the record before--with a Netburst Celeron. Clock Speed doesn't mean shit if the Performance per Clock is bad, and Intel outperforms AMD clock-for-clock. If I recall, Intel also locks down their CPU's to ensure they have longer lifespans and run more reliably where as AMD just mass produces them and says HAVE AT with the BE CPU's. While BD holding the OC record is impressive in it's own rite, it is not indicative of actual performance.

I also meant Reliability in terms of their yield and manufacturing capacity. If Intel says they will have a product ready by a certain time frame, they always are. Alternatively GF producing low yield (whether or not it was their fault remains to be seen) delayed BD quite substantially.
Posted on Reply
I don't believe this. I think they are saying this just to justify higher prices; they did the same on 40nm node.
Posted on Reply
No surprise here... Been saying this all along about such production issues (those are somewhat behind us now Mike Bryant is many weeks late in his statments), and that TSMC big bump in 28Nm chip pricings, things are what they are.

AMD development and design permits them to get into TSMC first, but with that they get to field and encounter more of TSMC set-backs as they surface. Nvidia comes along several weeks later, although there’s their own teething pains’; they have fewer as some process problems are already sort through.

We’re all astounded at AMD 7970 price, but Nvidia has still yet made it into the batter’s box... How will it play out? Nvidia needs to come with a GK104 first just to be reavant on price. Sure while the GK110 will have performance, the price is probably shockingly prohibitive in this current economic landscape. Waiting a few months will numb us and provide opportunity to get better yields.

Nvidia is making a smart move, wait for the graduation/summer buy season to drop their bomb.
Posted on Reply
st.bone said:
You seem to know the details. I've had this info before, do you by any chance know where i can find Intel's read map, i used to find them on their website but nowadays i cant seem to see them, so if you can find the road map please post it or post a link preferably PDF
I didn't find this on a road map from Intel specifically.
That's a site for Intel's architectures for DT/MB till skylake.

I do not remember which site had the info for intel mobile phones, but what I do remember is that it had 3 boxes with the architecture names, with their respective years.

seronx said:

I don't see Intel designs getting any overclocking records with x86 CPUs

So Reliable my bum
Reliable doesn't necessarily mean that it has to beat overclocking records.
Posted on Reply