Can Intel beat TSMC?

By Zvi Or-Bach, president and CEO of MonolithIC 3D

Intel CEO Brian Krzanich, in the company Investor Meeting, presented company expansion focused on a foundry plan on Nov. 21, 2013. “You will see us focusing on a broader set of customers,” said Krzanich. “If somebody can use our silicon, and make computing better, than we want it to run better on Intel. It’s inclusive, it’s all-inclusive,” Krzanich added, as covered by Barron’s blog  Intel: Competitors Have Given Up ‘Scaling’ Advantage in Moore’s Law

Intel clearly believes that it can beat the pure play foundries by an ongoing reduction of transistor cost while improving performance and power with dimensional scaling – essentially maintaining the trend of Moore’s law just as in the past. Intel will “not take our foot off the pedal” of process technology, Krzanich explained, and he expects the company to be making parts as small as 10 nanometers in transistor size by 2015, versus today’s 22 nanometer parts. He was followed by Bill Holt, Intel’s EVP and head of semiconductor manufacturing, showing the following slide describing Intel expectations to drive down the cost per transistor.

Fig 1

Maintaining dimensional scaling is in-line with Holt’s previous slide presented at the Jefferies May 2013 Analyst Meeting:

Fig 2

Here we observe the first discrepancy where Intel says they are “continuing to scale while others are pausing to do FinFETS,” while the other foundries say that their transistor cost will not be reduced for nodes below 28nm. This was made very clear by GlobalFoundries in it recent seminars and is nicely illustrated in this ASML Semicon West 2013 slide:

Fig 3

This has also been generally accepted by analysts. Below is a slide from IBS’s Handel Jones presentation at the CEA-LETI day in June of this year:

Fig 4

Some may argue that Intel will have a hard time competing as a foundry due to potential customer concern of Intel as a competitor. This is a valid point, but it did not stop Apple to buy cell phone devices from Samsung.

Some may argue that Intel will have hard time competing due to the lack of broad EDA and IP support. This is also a valid point but Intel does not need to win all fabless designs. If Intel wins just the few super high volume designs, it may well win the war.

Some may argue that “Intel announced their high volume mobile SoFIA chips are mask fabricated at external foundry and do not use Intel internal manufacturing for at least next 2 years (2014-15). ALL of Intel’s production for standalone modem chips today is outside Intel. Conclusion being Intel still does not have the right silicon technology for mobile computing which is why X86 less than 0.1% of Smartphone market,” as one commenter at Intel Nears Foundry Inflection Point blog. This might be why Holt presented Intel’s plan to develop foundry-type processes.

“Those products were optimized primarily for performance, and so Intel had avoided the problem that can crop up when transistors are packed more densely, namely that performance of the wires connecting transistors, the interconnects, can degrade. We didn’t scale the wires as much as we could have, because the products we were building didn’t demand that.” Now, he said, “the company’s technology would be focused more on those interconnects as Intel takes the scaling lead. The result would be the ability to more nimbly move between transistors optimized for performance, on the one hand, as in server and desktop chips, and transistors optimized for low-power mobile devices.” as illustrated below:

Fig 5

It would seem that if Intel could scale transistor cost as they have done in the last 40 years then they could win these super high volume consumer-oriented designs where cost is extremely important. And TSMC is clearly taking this seriously. As was made public after they lost Altera to Intel, TSMC aligned itself to face head-on Intel’s challenge by expediting the development of FinFet technology.

As TSMC’s P/E is 14.42 while Intel’s P/E is only 12.87 the market should have responded very well to these presentations but apparently it did not — and in reverse to NASDAQ trend, Intel stock fell more than 5% the day after:

(Click to view full screen.)

(Click to view full screen.)

Nor did Altera’s stock perform well since announcing the move to Intel as a foundry, especially when compared to Xilinx who choose to stay with TSMC, as the stock price chart below illustrates:

(Click to view full screen.)

(Click to view full screen.)

The Stock market might be wrong, as it been wrong many times before, but then there are other concerns:

Why did Intel feel the need to put so much money in the ASML EUV program if they can do just as well without EUV? Does Intel reduced cost per transistor account for its escalating cost of R&D, which in 2013 averaged more than 20% of revenue vs. less than 14% in 2005? Does Intel reduced cost per transistor account for its escalating cost of capital, which, per their balance sheet on Depreciation/Depletion, averaged in 2013 more than 26% of revenue vs. less than 10% in 2011?

It is not clear what the Intel proprietary technology is that allows it to do so much better than the foundries to produce a per transistor cost reduction. It does seem that their fab equipment and especially lithography is the same. And it also unclear why the Intel per transistor costs are not impacted by the much higher cost of lithography with the double and quadruple litho steps needed in manufacturing these advanced process nodes and the extra development and process steps required.

There is one more important issue that seems to be ignored. For SoC applications, the embedded SRAM is a key factor because it dominates the die area, as we recently presented in our blog Are we using Moore’s name in vain? If Intel’s embedded SRAM is scaling each node as before, then it would represent an important advantage over the foundries. Yet Intel recently announced integration of DRAM into Haswell and promised future Xeon and Xeon Phi models that integrate memory atop processors in 3D packages instead. Will these be aggressive enough to keep the on-system memory costs scaling?

Fig 9

In short, if Intel could keep the traditional 30% cost reduction per node from 28nm to 10nm, and the foundry’s cost per transistor is staying flat, then Intel would be able to provide their foundry customers SoC products at a third of ther other foundries cost, and accordingly Intel should be able to do very well in its foundry business.

POST A COMMENT

Easily post a comment below using your Linkedin, Twitter, Google or Facebook account. Comments won't automatically be posted to your social media accounts unless you select to share.

4 thoughts on “Can Intel beat TSMC?

  1. Ian Dedic

    Anyone who thinks that Intel can provide foundry SoC at a third of the cost of TSMC, I’ve got some magic beans to sell you…

    In the “area scaling” graphs, the reason the other foundries have a flat bit between 20nm and 16nm is because these both use the same double-patterned 64nm pitch metal stack.

    The reason Intel doesn’t have this blip is that their 22nm process is single-patterned, their 14nm one uses the same 64nm pitch double-patterned stack as everyone else.

    So the graphs are compete Intel PR bull**it — they’re bigger area at 32/22nm and the same at 14nm/10nm.

    On cost, everyone in the world buys the same fab equipment with the same performance from the same munufacturers, the only way to get lower cost is higher yield or lower running costs (labour, power, water, building) — and if anything TSMC has an advantage here due to location and scale.

    Intel may have claim to have an advantage due to intensive yield improvement because they only make a very small number of chips (and these can use very restrictive design rules), but this will disappear if they join the foundry club.

    The same laws of physics (and equipment cost) apply to everyone, including the wafer cost increases with double patterning at 14nm and triple at 10nm — together with the higher mask count for FinFETS, this is why everyone who is being honest says the cost per gate no longer drops with process beyond 28nm, or even increases slightly. You get more density and lower power, but no cost saving per chip with the same function — and the chip development costs are going up rapidly.

    For the moment this *is* the end of Moore’s law as he defined it (the cost per transistor goes down — *not* that the density goes up). New process nodes now mean that you can still integrate more, but that if you do the cost goes up — in other words the market will have to get used to paying more for extra functions, not the same.

    Reply
  2. Bill Kohnen

    Agree with what Ian points out and would add it simply is not in the Intel DNA to accept the role of a subservient supplier which is required to keep customers happy – even when they as the supplier are correct. TSMC and even Samsung know how to manage this role.

    That being said as an American I have always thought that the technical capabilities and competitive importance Intel gives the US are greatly undervalued so it does not make me feel good to see Intel forced into the less noble world of being a subcontractor.

    Reply
  3. RG

    poor Intel, when all you have is a hammer, every problem looks like a nail.
    yes, they have only 0.1% of the mobile market, despite the world’s best silicon process technology, but that doesnt automatically make a business plan.

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

NEW PRODUCTS

VaporSorb filter line protects advanced yield production from Entegris
10/21/2014Entegris, Inc. today announced a new product for its VaporSorb line of airborne molecular contamination (AMC) filters. ...
Next-generation nanoimprint lithography technology
10/21/2014EV Group (EVG) today introduced its SmartNIL large-area nanoimprint lithography (NIL) process....
SEMI-GAS broadens gas mixing capabilities for highly corrosive gases
08/28/2014SEMI-GAS Systems, a provider of ultra-high purity gas delivery equipment, recently broadened the capabilities of its custom Xturion ...