Taiwan Semiconductor Manufacturing Company (TSM) is the world's leading contract semiconductor foundry. The firm has a virtual monopoly on the 28nm process node, as competitors have failed to turn their "powerpoint" process nodes into anything shipping in any real volume, and I believe that as far as foundries go, I don't expect Samsung (SSNLF.PK) to catch up, as it seems that Samsung is having such a hard time ramping 28nm that its own "Galaxy" flagship products are forced to use Qualcomm (QCOM) designed, TSMC built SoCs.
Financially, this company is sound and will continue to print money (so I'd actually seriously consider this on a pullback). However, my issue with TSMC, though, is that I do not believe that the claims that they are making regarding their future technologies can be taken seriously, and represent at best a knee-jerk reaction to the massive process technology lead that Intel (INTC) has and the pressure that its customers are putting on the company to deliver.
The FinFET Bomb
When Intel announced that it had gotten Tri-Gate transistors prepped for high volume manufacturing in 2011, the world was shocked. My discussions with engineers, even within Intel, lead me to believe that this largely caught everybody off-guard. Indeed, as I began researching Intel earnestly back in 2011, I remember going through some of their technical presentations and seeing that FinFETs were an option being considered for 2015-2016 timeframe - if it were even to be "the" next generation option.
Of course, once Intel stunned the world, the foundries began claiming that they had this technology, too, and that it was coming "really soon". TSMC and the other foundries now claim that their own FinFET based processes will be coming just a year after their as-of-yet unreleased 20nm planar, high-K metal gate processes. Who are they trying to fool?
TSMC's 28nm Was Supposed To Come In Q4 2010
Go back in time and check out the initial press releases around the 28nm process. TSMC had claimed that 28nm would begin production in Q4 2010. Of course, 28nm production did not begin in volume until Q1 2012, and we still have not seen the low power, high performance "HPm" 28nm process that utilizes high-K metal gate in shipping products as of 5/29/2013, and I believe it won't be until Qualcomm's Snapdragon 800 and Nvidia's (NVDA) Tegra 4 hit the market in Q3 2013 that we see this process in commercially shipping products.
If you go back to previous conference calls with management, you will see something peculiar...
Q4 2010 Conference Call (January 2011)
First, TSMC's 28-nanometer is industry-first and ready for primetime. Customer products are already taped-out and in prototyping. We have superior performance, superior reliability and density, the density being 2x over 40-nanometer with our gate-last high-k metal gate process.
So, in January 2011, 28nm HKMG was "ready for prime time". Let's fast forward to April 2011...
Q1 2011 Conference Call (April 2011)
Now I, let me talk about our advanced technologies. First, on 28 nanometers, we completed a very important milestone. We completed full technology qualifications for both oxynitrite, which is our 28 LP and high-k metal gate, which is our 28 HP. Now the significance of this milestone is that the way is now clear to volume production on the 28 LP and 28 HP, both the oxynitrite and the high-k metal gate versions.
Wait a minute, I thought 28nm was "ready for prime time" three months prior to this call? The company mentions shipping chips to customers, but in all likelihood, these were not production ready parts. Let's move on to the Q2 call...
Q2 2011 Conference Call (July 2011)
Now, I'd like to report on our technology progress specifically first on 28-nanometer. We reported earlier that we had tape outs for 89 individual products and the tape out of each of those is on schedule. The first silicon of every tape out was fully functional would consistently satisfactory. In fact density reduction is on plan. The ramp of 28-nanometer however is taking longer than expected due to the softening economy and the demand outlook of 2011.
Okay, so now by July 2011, we hear that customers are now just getting back first silicon on 28nm. This does NOT imply to me high volume, "ready for prime time". This, to me, says that the design houses and TSMC were both working together to ready the process for high volume shipments a fair bit later.
Of course, I will save you the trouble and fast forward now to the January 2013 call, in which Dr. Morris Chang (CEO) states unequivocally that 2012 was the first year of high volume production of the 28nm node, and that 2014 is the first year that the 20nm node goes into high volume manufacturing:
And enough discussions have taken place with enough customers with large requirements to lead us to believe that in both its first and second year of production, in both the first and second year production of 20 SoC and that first year will be next year, 2014. The second year will be a year after that, 2015. In both those years of 20 SoC production, the volume of 20 SoC will be larger than 28-nanometer in its first and second year of production, which were last year and this year.
Further, on the call, Dr. Chang points out that the 20nm node will represent ~2% of revenues in Q2 2014:
Morris Chang - Chairman and Chief Executive Officer
What quarter will be the first 2% quarter?
Lora Ho - Chief Financial Officer and Senior Vice President of Finance
Well, based on our current estimation, it will be roughly second quarter 2014.
This means that 20nm only begins to ramp in Q2 2014, which further means that products will not actually hit the shelves with TSMC's 20nm until perhaps Q3 or Q4 2014 in earnest. This is much more realistic, and given the timeframes of the release of the first 28nm parts ito the market, this seems to make sense. The problem is that things get really screwy when these guys start talking about the 16nm node, which is expected to come a mere year after the 20nm node.
The claim that Morris Chang makes is the following:
On 16-nanometer FinFET, we have said several times that this is a change in cadence in our new technology introduction. It used to be 2 years per node and in the case of 16-nanometers FinFET, it follows just 1 year, by 1 year, the 20 SoC. So it is a quickening of cadence and that is because of market request, market requirements, customers' requests.
Okay, so according to this, 16nm FinFET - which is a radical change in the transistor structure and gives a massive improvement in leakage current and other performance characteristics - will be available just one year after the 20nm process! Of course, Dr. Chang made it clear in the January call Q&A that volumes in 2015 of the 16nm process would not be large at all in the call:
Andrew Lu - Barclays
So, do you suggest that what we know - production year 2015?
I think it will be very, very small.
So, very small in 2015, and quite frankly, I would be surprised if any 16nm parts were actually in the hands of end users until 2016. Like we saw with the 28nm, this could simply mean that customers are taping out designs (not test chips, but production designs - big difference) and perhaps getting first silicon by the end of 2015, with volume in earnest during 2016.
We now have two contradictory statements. First that the "16nm" process will come just 1 year after 20nm, and then we also know that volumes for 16nm will be "very small" in 2015.
Further, there are additional questions that analysts tried to articulate, but that management skirted:
- If 20nm planar will be obsolete in just a year, why would anybody bother with it, especially as cost/transistor is going up substantially?
- Why does TSMC expect 20nm volumes in 2017 to be rather large when, once again, it will have rendered that process pointless in "2015"?
- How does TSMC expect to overcome the radical challenges/hurdles of manufacturing FinFETs by 2015, when in mid-2013 it is just now getting 28nm high K metal gate right?
These are the questions that one has to ask. You don't need a PhD in advanced semiconductor physics to understand that this is clearly non-sequitur.
And finally, on 5/29, TSMC and Xilinx put out this, shall we say, "incredibly optimistic", press release claiming that they have worked together on a new initiative called FinFast, where they break the laws of physics and have "16nm" test chips (not design tape outs, but simply test chips) out in 2013 and then by 2014 get 16nm parts out to market!
Wait a minute. Don't you have to get 20nm into production first before skipping on over to 16nm? I think that Xilinx and TSMC were caught off-guard by Altera's (ALTR) 14nm win over at the Intel fabs, and as a result are both simply putting out this press release to save face. Altera evaluated TSMC's 16nm process and for some mystical reason decided that it wanted to pony up the premium for Intel's 14nm process. I wonder why?
TSMC is a great company that is a leader in the foundry space, but they are trying too hard to appease investors/customers with some of these claims regarding FinFETs. On the January call (before Intel took Altera and likely Cisco (CSCO) from TSMC), the claim was "minimal volumes of 16nm in 2015". Now, TSMC is trying to pull a FinFAST one on investors and customers by claiming that 16nm will be in production during 2014, totally bypassing the yet-to-ramp 20nm node.
I'm not buying these claims, and neither should you.
Additional disclosure: I may go long TSM on a pullback, but I still don't buy these FinFET claims.