TSMC 28nm Yield called into question? Again? Really?

[Daniel Nenni]

Did I read this right? Did Future Horizon’s CTO just question Dr. Morrris Chang’s integrity? And EETimes prints it? How else would you interpret this article published by last Friday (1/20/2012)?

TSMC's 28-nm process in trouble, says analyst
LONDON – Mike Bryant, technology analyst with Future Horizons Ltd. has said that foundry Taiwan Semiconductor Manufacturing Co. Ltd. is in trouble with its 28-nm manufacturing process technologies, which are not yet yielding well. Bryant referenced un-named contacts made with multiple companies waiting for designs to be produced by TSMC on 28-nm processes. Bryant said that there are 10 designs in manufacture from seven companies. "We're now hearing none of them work; no yield anyway,"

Ah, the old un-named contacts ploy, tabloid journalism 101. This is déjà vu 40nm ramping. The 40nm yield ramp was longer than 65nm so it must be TSMC’s fault, right? Of course the only people questioning 28nm yield are so called industry analysts and editors, not the people who actually do 28nm design and manufacturing.

Let me tell you this from my personal experience, there is working TSMC 28nm silicon all over Silicon Valley if you care to look. 28nm FPGAs are already shipping to customers. Other working 28nm silicon includes a microprocessor, GPUs, and several wireless SoCs from industry leading providers. The word on the street in Silicon Valley, from the people who actually taped out in 28nm, tracks with what Morris Chang said in the Q4 conference call which you can read HERE.

I will say a few words on the status of our 28-nanometer ramp. Our 28-nanometer entered volume production last year and contributed 2% of 4Q '11s wafer revenue. Defect density and new progress is ahead of schedule and is better than 40-45-nanometer at the corresponding stage of the ramp-up. We expect 28-nanometer ramp this year to be fast and we expect 28-nanometer will contribute more than 10% of total wafer revenue this year. Our tape-outs on the 28-nanometer, we have so far completed 36 individual tape-outs and have scheduled another 132 individual product tape-out in 2012. While three versions of the 28-nanometer technology, the LP, the HP and the HPL have entered volume production, the fourth version, the HPM, has entered risk production this quarter and is expected to begin volume production in the second half of this year……… Dr. Morris Chang

Anybody else bothered by this? Is there really a place for tabloid journalism amongst semiconductor professionals? I would unsubscribe to EETimes for printing it but I already did that after the 3D IC debacle.

D.A.N. (Blogger, not a Journalist)
<script src="http://platform.linkedin.com/in.js" type="text/javascript"></script>
<script type="IN/Share" data-counter="right"></script>

 

[LinkedIn]

Ernest M I agree that 28nm yield is important. Contrarian viewpoint ~ Maybe increasing capacity has impacted yield, yet whatever the yield, this story indicates throughput is up: TSMC Ships More 28nm Wafers than Expected, Projects Rapid Ramp - X-bit labs

 

[CharlieD]

EETimes got duped again, clearly TSMC 28nm is yielding. If a product is NOT yielding at this stage it is more likely a design or IP problem. Lets see how 28nm revenue ramps this year because 10% of TSMC's revenue in 2012 is a lot of wafers. Expect a retraction or clarification from Future Horizons?

 

[Xtoa]

A CTO say this? What are they yield problems? With no more information the news is useless. What is this guy selling?

 

[Daniel Nenni]

A CTO say this? What are they yield problems? With no more information the news is useless. What is this guy selling?

"Future Horizons provides semiconductor market research, analysis and consulting for use in opportunity analysis, business planning and new market development. Its industry information and semiconductor training seminars and Forums are widely considered to be the best of their kind. Emphasis is placed on the world-wide microelectronics and electronics industry, and the European market environment. Contact and affiliate information."

I do not see how making inflammatory statements helps their business. It will be interesting to see how this plays out.

D.A.N.

 

[kingmouf]

Isn't it the same company that is preparing a report for EU on the feasibility of 450mm wafer manufacturing in Europe? I'm wondering, if they try to show that TSMC is not doing well, then there is a chance for such an effort in Europe.

Even if it is not the case, the bad thing that is constantly happening is that analysts are not held accountant for what they say. And in an economy that is based so much on speculation this really translates to bad business, panic, layoffs and other negative things.

 

[LinkedIn]

Daniel,
I did read the same article and honestly I don't know who should I believe. For the guys at Future Horizon bad news is a good occasion to write something, and on the other hand Dr Chang is a TSMC guy so his statement cannot be 100% objective.
I am wondering what is real yield and I think the only way is to wait. If you have any information that can be shared I would really appreciate. Regards.
Posted by Pascal Lo Ré

 

[Daniel Nenni]

Daniel,
I did read the same article and honestly I don't know who should I believe. For the guys at Future Horizon bad news is a good occasion to write something, and on the other hand Dr Chang is a TSMC guy so his statement cannot be 100% objective.
I am wondering what is real yield and I think the only way is to wait. If you have any information that can be shared I would really appreciate. Regards.
Posted by Pascal Lo Ré

As an officer of a public company on a quarterly conference call, I would hope that Morris Chang is telling the truth. If not, he could risk legal action, share holder law suits, even jail if he intentionally misled Wall Street. I have found Morris to be genuine on these calls and have no reason to doubt his words, ever.

As I posted, the information I have heard directly from the companies I work with in Silicon Valley track with what Morris said: TSMC 28nm is ramping as expected and will be in full production by year end. 28nm FPGAs are already shipping to customers, there is no question about that. Other 28nm products are on the way so Future Horizons already has egg on their face.

We will know for sure in the coming months but my bet is on Morris being right and Future Horizons paying the price.

D.A.N.

 

[Russ Fischer]

I have known Morris to be honest honest and forthright, sometimes to a near fault.

Obviously Morris is deeply concerned about TSMC's ability to keep up with Intel and this is good example of hes honesty:- CNA ENGLISH NEWS

Also, if 28nm is 2% of revenue it is probably .2% of wafers out. These are many designs in a new process in prototype volumes. Price would be something like 10 times the average wafer price for TSMC. That's about 740 wafer starts per month. If they are shipping 28nm parts to lots of customer, nobody is getting much more than sample quantities of anything. If the are 29 nm pars bouncing around the valley they are probably the same parts being shown over and over again.

 

[LinkedIn]

Thank you for your opinion. I apologize for being suspicious, but I've learned not to believe company officials (I live in Japan, so the Olympus case is quite fresh in my mind).
As you said we will know for sure this year. I'll wait for new information.
Thank you again.
Posted by Pascal Lo Ré

 

[Jasopin]

All the product companies cut back their semiconductor manufacture order in Q4, 2011, to be on the safe side. I guess Future Horizons Ltd was confused with wafer order and yield data!

By the end of Q4, 2011, TSMC has a good three-year on 28nm development, ramping, and yield improvement. If I recall the status of 45nm at the end of three-year time, the yield was reasonable high. If Dr Morris Chang said "Our 28-nm entered volume production last year and . . . . . . Defect density and new progress is ahead of schedule and is better than 40-45-nm at the corresponding stage of the ramp-up. . .", there is no doubt about it. Again, I guess Future Horizons was confused with reduced wafer order and yield data in Q4 2011!

 

[Daniel Nenni]

TSMC 28nm will be one of the most successful process nodes.......

Also, if 28nm is 2% of revenue it is probably .2% of wafers out. These are many designs in a new process in prototype volumes. Price would be something like 10 times the average wafer price for TSMC. That's about 740 wafer starts per month. If they are shipping 28nm parts to lots of customer, nobody is getting much more than sample quantities of anything. If the are 29 nm pars bouncing around the valley they are probably the same parts being shown over and over again.

TSMC has publicly stated that 28nm production "started" in Q4 2011 with 28nm @ 2% of revenue. TSMC Q1 28nm revenue percentage has been forecasted at 5%, this comes from 36 products in production. There are 100+ more 28nm designs going into production this year so expect TSMC 28nm revenue to exceed 10% in 2012.

Future Horizons: "However, there are recent comments of major yield problems with their 28-nm process actually being even worse than at GF [globalfoundries=," Bryant told the audience.][/globalfoundries][globalfoundries=

This is complete nonsense. Lets see where the Q1 numbers end up, but my prediction is that TSMC 28nm will be one of the most successful (technically and financially), one of the most widely adopted process nodes we will experience for some time, believe it!

Who else is releasing 28nm numbers? Certainly not Samsung. TSMC has set the "transparency" standard here. Samsung is still lurking in the shadows fueling rumors of 28nm success. Samsung is not now, nor will it ever be a pure-play foundry. They do not today have the necessary transparency to succeed in this business. Comparing TSMC to Samsung is apples to potatoes.

Future Horizons: Pressure to keep up with Intel had caused some miss-steps by TSMC, while Samsung Electronics Co. Ltd....

Also complete nonsense. TSMC fabs THOUSANDS of different products, Intel fabs microprocessors. Comparing TSMC to Intel is apples to a golf ball.

I have no problem with TSMC holding themselves to an impossibly high standard of manufacturing, but a pure-play foundry is not an IDM, nor is an IDM a pure-play foundry.

I have no problem with the industry holding TSMC to an impossibley high standard of manufacturing, but lets not forget what TSMC has done for the semiconductor industry, they have literally made it what it is today, an open market where anybody can design a modern semiconductor device.

I understand so called industry analysts and journalists are under pressure to increase readership and sell reports, but at what cost to the industry? Are inflammatory remarks like this from "professionals" really for the greater good of the semiconductor industry?

D.A.N.


<script src="http://platform.linkedin.com/in.js" type="text/javascript"></script>
<script type="IN/Share" data-counter="right"></script>][/globalfoundries]

 

[DigiHound]

"I have no problem with the industry holding TSMC to an impossibley high standard of manufacturing,.."

On a high horse, much?

Sometimes anonymous sources are kept anonymous because that's the only way they'll talk to anyone. There's no reason to think TSMC's 28nm ramp is as disastrous as 40nm was -- at least not yet -- but neither TSMC or GF has exactly been aggressive in shipping 28nm parts in high volume yet.

 

[Daniel Nenni]

"I have no problem with the industry holding TSMC to an impossibley high standard of manufacturing,.."

On a high horse, much?

Sometimes anonymous sources are kept anonymous because that's the only way they'll talk to anyone. There's no reason to think TSMC's 28nm ramp is as disastrous as 40nm was -- at least not yet -- but neither TSMC or GF has exactly been aggressive in shipping 28nm parts in high volume yet.

Okay, point taken, but I would rather be on a high horse than a donkey, and that article is a donkey.

 

[JustDave]

TSMC is Yielding @ 28nm HP

28nm HP yield is better than expected which is what matters. TSMC has met our expectations. End of story.

 

[Daniel Nenni]

TSMC returns fire over 28-nm process issues?!?!?!?!

Returns fire? Really? Another DONKEY story by EETimes:

TSMC returns fire over 28-nm process issues

"Maria Marced, president of TSMC Europe, repeated what has been said before by herself and other TSMC executives; that defect density reduction is on track for the 28-nm node and ahead of where TSMC was with 40/45-nm process technology at an equivalent stage in its roll out."

View attachment 2739
Anything for a click eh Peter? Time for a vacation? How about a trip to Taiwan to see some 28nm silicon........

Not one mention of the analyst's name and company he works for, he had better get used to that.





D.A.N.
<script src="http://platform.linkedin.com/in.js" type="text/javascript"></script>
<script type="IN/Share" data-counter="right"></script>

 

[Daniel Nenni]

I told you so! More to come......

TSMC's 28nm is great, says Xilinx - 2/15/2012 - Electronics Weekly

 

[DigiHound]

If we're going to have a factual conversation on the merits of TSMC's yields (or GF's, or Samsung's, etc) we apparently need to affirm that everyone understands a basic rule of fabrication: Not all lines and designs scale equally.

It's entirely possible for *both* of these stories to be true. Take a look at AMD, Nvidia, and Xilinx. We know NV's yields have been far lower than the company would've liked; this is the first time in a decade or more that the company is leading with a midrange next-gen part. AMD had to push back the 7000 family launches. So on the one hand, TSMC yields are having "problems." The fact that AMD got the HD 7000 series out with a minor delay whereas NV had to delay Kepler by a half-year is proof that design plays a major role in rampability.

Now Xilinx says "Hey, yields are fabulous." But Xilinx builds a much simpler, smaller, and entirely different type of part than AMD and Nvidia. It's using HPL; the article notes that Altera is using HP and LP. Xilinx may have worked closely with TSMC from Day 1 to ensure that its designs were well-suited to TSMC's processes in a way that other companies didn't. It's completely fair to note that the vast majority of TSMC's revenue is derived from established processes, with over 50% of it coming from 65nm and up, meaning that these early troubles, even if significant, are not a sign of major instability or financial issues at the fab.

Finally, 40nm comparisons are disingenuous. 40nm was so legendarily bad, it's easy to look back and say: "We're doing 3x better!" That doesn't actually tell us anything. It's like Intel comparing CPU power consumption improvements to 2005 Prescott-based Xeons, which were notoriously worse than anything else.

I'm not advancing any particular viewpoint here; I don't know the extent or nature of any troubles TSMC has had at 28nm. The point is, these situations are more complex than any single article makes them sound.

 

[Russ Fischer]

Expecting Xilinx to tell the truth about TSMC 28nm process is like asking Col. Sanders to babysit your pet chicken.

A good freind of mine who is a very prolific user of Xilinx parts tells me a different story. He can't get even samples of the 28nm parts and is told that they are like "hen's teeth".

 

[DigiHound]

Expecting Xilinx to tell the truth about TSMC 28nm process is like asking Col. Sanders to babysit your pet chicken.

A good freind of mine who is a very prolific user of Xilinx parts tells me a different story. He can't get even samples of the 28nm parts and is told that they are like "hen's teeth".

I just noticed the donkey pic. That's horrifying.

The bigger question is when Xilinx is planning to ship these things. I don't see that it actually matters much -- GF isn't exactly setting the world on fire at 28nm. Intel is a year away from 22nm SoCs. I'm not saying that delays are good, but as long as they don't descend into 40nm layers of hell, things will be alright.