Fab Updates

February 6, 2011

This post started off as a Q4 wrap, but has morphed into a fab update.

For anyone interested in the Q4 2010 transcript, it can be found here or at the bottom of the Pages box to the right.

There is a lot of interesting info in the conference call. Sanjay’s style is very different than Eli’s. Not as many sound bites, if you will, but very forthright and informative.

Most of the important points from the conference call have already been covered in analyst reports and general media coverage. The fab info, on the other hand, has just been skimmed over, so I’ve decided to concentrate there.

In the big picture, it’s all good news. The Fab 4 expansion is on track. This expansion is leading edge, significant and generally under-appreciated.

Plans for Fab 5’s initial ramp were revealed. No surprises, but a couple of puzzles.

Fab 4

Once Fab 4 is fully provisioned mid 2011, SanDisk’s total production output from Fab 3 and Fab 4 combined will be 2 million 12 inch wafers per year, or 166,667 wafers per month (wpm).

From mid 2010 to mid 2011 about 62,000 wpm will have been added in SanDisk/Toshiba JV fab capacity, SanDisk’s share is 31,000- Bank of America’s (BoA’s) numbers. See graphic below.

This is a lot of wafers. And given demand, they’re all going to be needed.

BoA’s numbers look close enough and cross-check with SanDisk comments. Jun-11 is pegged at 333 K wpm. 333 K wpm x 12 = 3996 K wafers per year – rounding up- 4 million. SanDisk’s share is half that- or 2± million wafers per year.

In parallel with the wafer expansion, SanDisk/Toshiba has been converting existing production from 32 nanometer to 24 nanometer.

By the end of 2011 all SanDisk’s production is expected to be 24 nanometer, “except for any 32 nanometer requirements that we still may have related to certain customers.”

The big deal about 24 nanometer is that it will be the primary cost reduction driver for 2011, deployed on both two and three bits per cell architectures. Going from 32 nanometer to 24 nanometer will give SanDisk a chip cost reduction of 30% to 40%. A most interested interchange from the the Q4 conference call:

“Atif Malik, Morgan Stanley: OK and one last one. If I look at your history of cost reduction in the last three years, you have been very strong and you’ve always upsided your cost reduction targets in the 3-bit per cell or technology migration. But just according to physics, if you scale 32 nanometer wafer to a 24 nanometer, you should be able to get a 45% kind of cost reduction just doing the 24 square over 32 square. And then you have this 3 bit, you have a knob of 3 bit per cell just getting embedded in OEM applications. So I mean why should we believe that the cost reduction should be less than 40% this year?

Sanjay Mehrotra: Let me just provide you some help here. The result, the technology is not only about scaling the flash memory cell but it is also about scaling the periphery transistors, the circuits inside the chip. And those two tend to scale at somewhat of a different rate. When we look at 32 nanometer to a 24 nanometer, wafer comparisons at mature yields, the 24 nanometer can give you a benefit in the range of 30% to 40%.

Again, at the wafer-level, with mature yields, on both the 32 nanometer wafer, as well as 24 nanometer wafer. However, for our yearly cost reduction, as Judy pointed out, there are many factors. One of them is certainly 24 nanometer ramp up. We will be ramping up 24 nanometer during the course of the year. And other factors, of course that play a role in our overall cost reduction are the non-captive supply, as well as Fab 5 related start up costs and the impact of exchange rates.

Atif Malik, Morgan Stanley: Should we model a linear kind of a ramp for 24 nanometer for this year?

Sanjay Mehrotra: Yes, the 24 nanometer ramp will be fairly linear during the year.

Atif Malik, Morgan Stanley: And Fab 5 we’ll start with 24 or 18?

Sanjay Mehrotra: We are not disclosing the plans for Y5 [Yokkaichi 5- Fab 5] at this point. As I said, that 1x technology will start late this year. Y5 will start in the August timeframe with production in the Q3 timeframe.”

A couple of things jump out, aside from the 30% to 40% wafer-level benefits. First, the 24 nanometer ramp will be relatively linear. Second, SanDisk is being coy about the technology or node, that Fab 5 will be starting up with.

Fab 5

Fab 5 will be a mega-fab. It will be a beast- about the size to Fab 4, which has an official capacity of 210,000 wafer starts per month and is one of the largest 300mm facilities ever built.

Right now the building shell is under construction.

Once the shell is complete, Fab 5 will be fitted-out in two phases: Phase 1 and Phase 2, each of which is expected to be of approximately equal size.

On the Q3 2010 conference call, SanDisk said that no decision had been made on the timing of Fab 5’s startup of Phase 1.

Decisions have now been made.

SanDisk feels comfortable moving forward with Fab 5’s ramp in the third quarter of 2011.  No surprise there.

Up to now in SEC filings and industry commentary, SanDisk has consistently been saying that “that Fab 5 will increase its 2011 wafer output by less than 10%.”

As of the Q4 conference call, this has dropped to less than 5% for 2011. Sanjay’s exact words: “The incremental contribution from Fab 5 to our total wafer output is estimated to be less than 5% for 2011.”

This change from less than 10% to less than 5% is no cause for concern. First, those paying attention understand there many variables. Second, most aren’t watching this closely.

That said, one has to wonder, if this slight slip isn’t part of a larger effort to accelerate the transition to 1x, the next node down below 24 nanometer- which would be most welcome news.

1x

The graphic below- SanDisk’s NAND Roadmap- is from the SanDisk Investor Day 2010.

Transitions roll right along 43nm ⇒ 32nm ⇒ 24nm, then there is a year+ gap, from late 2011 to late 2012, before 1x and then nothing thereafter. The gap from the end of 24 nanometer -late 2011- to the beginning of 1x nanometer -late 2012, has been puzzling.

In the Q4 conference call, quoted above, Sanjay says that 1x production will now begin late 2011- not late 2012.

Looks like we’re going to be getting a new slide on 24 February, this year’s Investor Day.

When pressed, Sanjay wouldn’t comment on what node 1x will start or whether Fab 5 will start on 1x.

Eli has described 1x as 20nm or a little bit under and 1y as closer to 15nm.

We should know soon enough, but my guess is that for SanDisk, 1x will be 20nm to start. This might set up one more node down before extreme ultraviolet lithography, or EUV, around 15nm, and 1y.

Up until now the SanDisk/Toshiba party line has been that Fab 5 will start with “leading-edge 20-nanometer generation, with subsequent generations to follow.” Translation: 24 nanometer.

Maybe this is still true, but one has to wonder why Sanjay just didn’t say so. Also if 1x will be starting only a few months later, why wouldn’t Fab 5 just start off with 1x- whatever that node may be?

However this turns out, it’s all good news.

Thanks to the Fab 4 expansion and the Fab 5 start-up, SanDisk will soon have access to a lot more NAND, at leading edge geometries and at best costs.

While not discussed in this post, EUV seems to be making steady but slow progress. ASML Holding NV has reportedly shipped the world’s first ”pre-production” extreme EUV lithography tool to Samsung.

Toshiba will be getting its EUV lithography tool from ASML by mid 2011, just in time for Fab 5 R&D.

From the look of it, EUV won’t be ready for high volume production until 2014 at the earliest. There are still issues with the power source, resists and defect-free mask technology.

In the meantime, double patterning, plus various computational lithography techniques might just do the trick.

SanDisk appears to have a plan and the required facilities.

How this translates to the bottom line in 2011 has yet to be determined. The pieces seem to be in place, but in the Q4 2010 conference call, Judy wasn’t reassuring about the bottom line. Guidance was decidedly tepid.

Q1 appears to be looking OK, but guidance for 2011 left a lot to be desired.

It’s not like this hasn’t happened before. A year ago today for example.

The Q4 2009 conference call took place on 28 January 2010. SanDisk closed at $28.78. SanDisk guided most conservatively. The next day SanDisk dropped over 11% to close at $25.42.

The stock gradually recovered passing its 28 January 2010 price on 17 February.

SanDisk 2010 investor day took place on 26 February 2010. SanDisk closed at $29.15. Guidance was raised. The next day the stock gained over 11% and closed at $32.63.

SanDisk’s 2010 Q1 cc took place on 21 April 2010. SanDisk closed at $37.59. The next day the stock closed at $42.22, a gain of over 12%.

It will be interesting to see if the cycle repeats in 2011. So far we have the lukewarm yearly guidance and the ensuing drop.

Next stop will be SanDisk’s Investor Day on 24 February.