Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/dresden-date-2012-trip-report-top-10-reasons-intel-will-not-succeed-as-a-foundry.1393/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Dresden (DATE 2012) Trip Report: Top 10 Reasons Intel will NOT Succeed as a Foundry!

Daniel Nenni

Daniel Nenni

Admin
Staff member
Per my previous blog, GLOBALFOUNDRIES Dresden Fab 1, my trip to DATE 2012 was very enlightening. I did meet quite a few old and new friends and my IQ certainly increased as a result of this trip. My guess is there were close to a thousand people attending.

First up Tuesday morning, the keynotes by Klaus Meder of Bosche Automotive and Mojy Chian of GLOBALFOUNDRIES. Interesting presentations talking about the industry in general and the respective companies in detail.

View attachment 3174

The most interesting comments I have come from the Executive Session I moderated on What Roles will the Foundries and Fabless Houses Play in Advanced Technology Nodes?

Executives:
Douglas Pattullo, Director of Technology Support, TSMC, NL
Yves Mathys, CEO, Abilis Systems, CH
Gerd Teepe, Vice President, GLOBALFOUNDRIES, DE
Robert Cadman, General Manager & Vice President, eSilicon, UK

Moderator:
Daniel Nenni, Founder, SemiWiki.com

Organiser:
Yervant Zorian, Synopsys, US

View attachment 3175

The continuously technology scaling in advanced nodes can dramatically impact business performance of the semiconductor industry. It can also significantly affect the age-old COT flow, fabless design and pure play wafer manufacturing flow. The executives in this session will discuss future trends and upcoming changes in the semiconductor industry and their impact on the roles to be played by of the foundries, the fabless houses and the rest of the value chain.

The session started with me doing a quick introduction where I talked a little about SemiWiki and the analytics behind it. Since a significant amount of traffic comes from search engines and key word searches, it is an interesting view into what is trending in the semiconductor ecosystem. As of today, 225,972 people have visited SemiWiki viewing 1M+ pages since launch in January of 2011. That is a very nice sampling size for an industry of less than 500,000 people.

As I introduced the speakers I tried to condense their biographies to speed things along but these guys had very long and impressive pedigrees! After they did their individual presentations we had a question and answer session for the remaining 45 minutes. I broke the ice with questions on two top trending SemiWiki topics: 3D IC and 28nm Yield.
TSMC has the most complete answer for 3D IC with partners Xilinx, Qualcom, and GUC. 3D IC will clearly be a competitive advantage for FPGA’s and mobile devices in the very near future. eSilicon also chimed in on the importance of 3D IC but I think they are a bit behind GUC as I blogged in Semiconductor Packaging (3D IC) Emerging As Innovation Enabler!. It is definitely worth a read if you have not read it already.

View attachment 3176

TSMC, GFI, and eSilicon all had comments on 28nm yield. TSMC denied 28nm problems as reported by a tabloid semiconductor site last week. I blogged about it in TSMC absolutely did NOT halt 28nm production!. There is a very interesting discussion in the comment section that you should check out. The bottom line from the panel discussion is that yield presents different challenges at every node, and 28nm is no different, so it is subjective. Both TSMC and GFI insisted that the 28nm yield ramp is on track and meeting customer expectations. eSilicon also has 28nm designs at both TSMC and GFI in progress. I will however wait until the TSMC Q1 investors call to see what 28nm wafer revenues are before I do my “I was right” victory dance which, according to my wife, is not a pretty site!

The other interesting question was about Intel and the foundry business. The answers from the panel were pretty soft on this so let me offer you something more concrete with my top ten reasons why Intel will not succeed in the foundry business:

10. Intel does not support a commercial design ecosystem today
9. Intel does not support a commercial IP ecosystem today
8. Intel has said that they will not manufacture ARM based SoCs due to the competing Intel Atom processor
7. Arm has said that a manufacturing licensing agreement with Intel would be “complicated” due to the competing Intel Atom processor
6. Intel is making too much noise about its two very small first foundry customers
5. The two very small first foundry customers are both FPGA versus SoC companies which drive the foundry business
4. Intel has a financial stake/financial influence in its two very small first foundry customers
3. As history has repeated itself, Intel has ADD in regards to NON microprocessor related projects
2. Intel will experience the same “competing with customers is bad” issue that is haunting Samsung

and the number 1 reason why Intel will not succeed in the foundry business:

1. Because I said so that’s why!

Correct me if I'm wrong here..........

D.A.N.

<script src="//platform.linkedin.com/in.js" type="text/javascript"></script>
<script type="IN/Share" data-counter="right"></script>
 
Last edited:
Hi Daniel,

Thanks for sharing the DATE report from Intel perspective. How was the overall event?

Somehow I feel Intel may able to take the foundry business to new scale. My understanding is

1. The new fab process has upfront investment of US$ 5-6 Billion. This figure may be very high for 20nm or below process node. It will be difficult for fab houses to justify the investment.

2. The Ultrabook ( 22 nm IVY Bridge) and probably "Atom" or equivalent for Smartphone may have significant market in 2014-16. To keep the ASP for these processors down, Intel may like to avoid "TSMC" or any other fab house.

3. Using fab like TSMC, global foundries etc, it will be difficult to bring down ASP for smartphone processor below $ 15. At this price point there will be huge market for feature phones and low cost smart phone.Same issue with the Ultrabook processors. currently $150 has a bottleneck for bringing ultrabook price down. I expect ASP $50-60 will bring lot of cheers to motherboard vendors

4. If you scan the balance sheet of Intel the cash reserve has crossed $10.6 Bn in 2011 up from $7.6Bn in 2010. so you have a reason to spend :)

5. Similar kind of steps have been taken by IBM and global foundries.

The above is my observation. I may be wrong in my analysis.

Best regards,
Sanjeev
 
Hi Sanjeev,


Thank you for the reply. I agree with your points 2 and 3 which is why I believe Intel will fab mobile designs using Atom chips only. FPGAs help ramp new processes so I understand the first two foundry customers as well.

Other than Atom SoCs and these FPGAs, I do not see Intel opening up their process technology secrets to the world of fabless design.

Just my opinion of course,



D.A.N.
 
Hi Daniel,

I believe you have blurred the two main reasons that Intel may not succeed in the foundry business.

One of these is the same as a reason that they seem to be going into foundry in the first instance: one interpretation might be that the motivation is largely as a spoiler against processor competition.
The other is financial: Intel is accustomed to margins that are way beyond what is being achieved via foundry. Unless they can offer something very special that will make foundry returns look comparable to their core businesses, it seems unlikely that management will be able to justify continued "diversion" of resources to support foundry. (You describe the consequence as ADD - but I think it's more rationally based than that).
The only way I can see Intel foundry surviving is if Intel splits out the foundry businesses so that the foundry sees only foundry-level profits from business provided by other business units - and then finds a way to continue to support its extraordinary level of semiconductor R&D without over-insulating own-developed processes from other users. Beats me how his can be done without losing a lot of what currently makes Intel such a powerful force.

GS
 
Hi George,

You are correct. TSMC has the highest foundry margins at 60%. Other Fundries are 40% on down to 20%. Samsung can dump wafers since they do turnkey product designs with other Samsung products at higher margins. In fact, Samsung has a track record of product dumping even today. So why would Intel want to go head-to-head with Samsung in the IDM-Foundry business? Or even TSMC in the pure-play foundry business?

I say ADD because Intel has made many non microprocessor related acquisitions that have failed, many more than have succeeded. Intel also failed in the FPGA business on at least one occasion that I remember. Margin may also have had something to do with that as well.

Thank you for the input......... It will be interesting to see how this plays out but my view is that Intel is an IDM-foundry to promote Atom SoC designs and battle ARM. Anything beyond that is marketing and PR theatrics.

D.A.N.
 
Last edited:
Do we need 10 reasons ? ... And does it mean the top foundry will close the technology gap ? Regards ...
Posted by Eric Bouche
 
That is the best part of the Intel foundry announcement, it has motivated the semiconductor ecosystem to push forward with 3D transistors and 20nm will come much faster had Intel not announced. So we all owe Intel a salute of thanks. Other than that it is a bunch of marketing and PR hype.
 
Just a question concerning Intel. To this point, Intel has announced only two foundry customers for its high-end process. They both happen to be FPGAs which are great for process ramping and as you said Intel has stake in both companies. I am wondering, would it be possible for Intel to use FPGA companies for its latest and greatest node and provide foundry services for its previous nodes? Considering that most of the time, Intel is approximately 18 months to 2 years ahead of competition, I am wondering if it would be attractive for customers not competing directly with Intel to use its previous nodes (e.g. for now 45nm and 32nm)? Of course the key question, is which customers are not directly competing with Intel and would choose Intel as a foundry... but hey! Samsung and Apple are suing each other to death and Apple is Samsung's largest customer!
 
I think this goes to the question of margins. Foundry business for older nodes is much lower margin than new nodes, as low as 20 percent. This would not be good business for Intel. FPGAs and Atom mobile SoCs will require the node advantage Intel offers. TSMC will start 20nm tape-outs this year so the Intel node advantage is shrinking. Intel, however, is an IDM foundry so normal pure play foundry logic does not apply!
 
"Intel has a financial stake/financial influence in its two very small first foundry customers"

Influence maybe, depending on the fab deal. But I have not found any public (searchable) evidence of an equity stake by Intel Capital or other division of Intel.
 
Jeffrey Rittichier Loved the "Top 10" reasons and agree that they are true. You can add confidentiality conflicts of interest to the list along with a paucity of tools that Intel will offer to outsiders trying to design for a particular node.
 
The majority of foundry business comes from mature technology using depreciated equipment - as you pointed out advanced technology will "only" contribute 10% within the first year.
That's fairly typical.

Intel expressed interest to work with some very selected customers that can bring their own IP to the SoC table.
There's plenty of other stuff - not only the processor..wifi, power managment, whatever
 
I haven't heard excessive noise about the FPGA guys. I did hear Ottelini mention the FGPA guys and, "A couple of strategic customers that we can't discuss right now."
Is manufacturing a customer designed chip foundry business or is it, "customer specific manufacturing"

Anyway, the common wisdom is that Intel is doing the FPGA fab to shake down the 22nm process. I think something else is going on. Intel has all the static ram "shakedown vehicles" it needs.
What if Intel were to supply an Atom based SoC that had as many of the standard functions that any SoC would have, plus it gave the user a million gates of FPGA that they could configure anyway they wanted in order to generate some differentiation in their device.

So far, the customer has had to take whatever it is that Samsung, QCOM, NVDA, etc. has to offer. what if he has a clever idea that doesn't exist on standard product SoCs? A million gates would be about 10 million transistors.... we have heard that 10 million transistors would be about the size of this period "." in 22 nm technology....that's pretty amazing particularly for a guy who sold Xlinx 10,000 gate FPGAs with a chip size of my thumbnail.

An embedded FPGA would relieve a lot of "mind reading" on the part of Intel about what their customers really want...let them make what they want.
I think that is exactly what we will see.
 
very good post. it is possible that some FPGAs on the SOC can do a better job than the general applications processor. I think INTC foundry deal is not a real foundry deal. It is more let's try a few things deal.

>What if Intel were to supply an Atom based SoC that had as many of the standard functions that any SoC would have, plus it gave the user a million gates of FPGA that they could configure anyway they wanted in order to generate some differentiation in their device.

excellent thinking. plus deal with other potential threats before they even emerge as threats. It is a good strategic thinking by INTC to do this "foundry" deal with the FPGA customers.
 
LinkedIn said:
Anyway, the common wisdom is that Intel is doing the FPGA fab to shake down the 22nm process. I think something else is going on. Intel has all the static ram "shakedown vehicles" it needs.
What if Intel were to supply an Atom based SoC that had as many of the standard functions that any SoC would have, plus it gave the user a million gates of FPGA that they could configure anyway they wanted in order to generate some differentiation in their device.

So far, the customer has had to take whatever it is that Samsung, QCOM, NVDA, etc. has to offer. what if he has a clever idea that doesn't exist on standard product SoCs? A million gates would be about 10 million transistors.... we have heard that 10 million transistors would be about the size of this period "." in 22 nm technology....that's pretty amazing particularly for a guy who sold Xlinx 10,000 gate FPGAs with a chip size of my thumbnail.

An embedded FPGA would relieve a lot of "mind reading" on the part of Intel about what their customers really want...let them make what they want.
I think that is exactly what we will see.
Click to expand...

Well, Xilinx is already doing this (ZYNQ has two Cortex-A9 cores) and Altera is also doing this. These are the most established players in the domain, their tools are by far more common and there is a huge IP pool to choose from. Although you make a quite valid point, I'm not sure how well it would be positioned in the market. The only way I see it can work is if Intel at the same package can offer a significantly higher performance Atom at significantly lower power envelope and competing prices.
 
I haven't heard excessive noise about the FPGA guys. I did hear Ottelini mention the FGPA guys and, "A couple of strategic customers that we can't discuss right now."
Is manufacturing a customer designed chip foundry business or is it, "customer specific manufacturing"

Anyway, the common wisdom is that Intel is doing the FPGA fab to shake down the 22nm process. I think something else is going on. Intel has all the static ram "shakedown vehicles" it needs.
What if Intel were to supply an Atom based SoC that had as many of the standard functions that any SoC would have, plus it gave the user a million gates of FPGA that they could configure anyway they wanted in order to generate some differentiation in their device.
So far, the customer has had to take whatever it is that Samsung, QCOM, NVDA, etc. has to offer. what if he has a clever idea that doesn't exist on standard product SoCs? A million gates would be about 10 million transistors.... we have heard that 10 million transistors would be about the size of this period "." in 22 nm technology....that's pretty amazing particularly for a guy who sold Xlinx 10,000 gate FPGAs with a chip size of my thumbnail.
An embedded FPGA would relieve a lot of "mind reading" on the part of Intel about what their customers really want...let them make what they want.
I think that is exactly what we will see.
 
You must log in or register to reply here.
Back
Top