X Files

xfiles

The 2008 NAND market mix is about 95% MLC (2 bits/cell or x2) and 5% SLC (1 bit/cell or x1). Mighty close to a one product market.

By 2012 this homogeneity will likely be gone. x3 (3 bits/cell) is expected to take over as the industry workhorse accounting for over half the market at 53%. x2 MLC will have slipped to 25%. x4 will be significant at 17%. And x1 SLC will remain virtually unchanged at 4%. (numbers from Forward Insights).

If this is how it plays out, pretty clearly x3 and x4 will be the name of the game — accounting for 70% of the NAND market.

This could be a big problem for the competition. First there are the IP issues. x3 and especially x4 look to be system solutions where SanDisk IP is particularly strong. Secondly there is manufacturing and design expertise where it looks like SanDisk/Toshiba has taken the early lead.

Samsung, along with the other big NAND players, has been talking about x3, but SanDisk/Toshiba is the only one shipping commercial x3 today, the only player with an announced aggressive ramp.

x3

In Q4 2008 SanDisk expects x3 to account for approximately 15% of total output. In 2009 x3 will grow to about 50%. x2 will be fading from there on out.

As 2010 dawns, SanDisk will be riding a new horse. x2 was a fine steed- a reliable and faithful friend for years. x3 won’t be faster, but it will be effectively 20% cheaper and that adds up to real money in this high volume game.

SanDisk is in position to make a seamless transition from x2 to x3, because the performance of SanDisk’s x3 approaches that of today’s conventional x2 MLC. No mean feat.

And easier said than done. It will be interesting to see how the competition copes.

SanDisk’s success with x3 performance can be directly attributed to advanced programming algorithms and its patented ABL, All Bit Line architecture.

Probably the most important thing to remember about ABL is that it is the architecture of the innards of the chip itself. It’s not the interface or the architecture for connecting chips together. That other stuff gets to be added on top of what the chip itself can do.

Using ABL, SanDisk/Toshiba chips appear to have an edge. For x3 this edge appears pronounced. Regardless of rumors to the contrary, not all NAND is created equal. But again time will tell. The competition has yet to step up and show what their x3 can do.

In the first quarter of 2009 SanDisk expects to start customer shipments of 32-gigabit NAND with 3 bits per cell architecture that achieves a smaller die size than IMFT’s 34nm NAND with 2 bits per cell architecture.

If SanDisk can pull this off, the high ground will have been secured.

x2

analyst-day-slide-21

This last February SanDisk and Toshiba presented ABL at the International Solid-State Circuits Conference, the ISSCC. Eli’s comments from SanDisk’s 2008 Analyst Day:

“I want to show you three articles that SanDisk and Toshiba together published just this month, at the beginning of this month at the ISSCC. ISSCC stands for international solid state circuits conference. It is a once a year conference. It is always held in San Francisco. It is attended typically by about 3000 circuit designers from the world over. It is the premier conference for circuit designers. We published for the first time three very important papers on flash memory.

The first one was on 34 MB/s. I’ll explain what the significance of the red is. 34 MB/s program throughput 16 Gb MLC NAND with All Bit line Architecture (ABL). All Bit line Architecture is a design architecture that was invented by SanDisk in the last 6 to 7 years. Has been implemented in the last three generations in our MLC products. And is what gives MLC the kind of speed that we have achieved in our chips, that other people need to use SLC, single level cell. Basically gives you very high speed, very reliable MLC.

We have always said that our MLC is ahead of other people’s MLC. This is one of the main reasons why. This is the first time that we are actually publishing that concept. Of course it is highly patented.

All three papers in fact use ABL. In fact every single one of our designs today uses ABL. I’ve said in the past that SanDisk uses 98% of all the bits that we ship are MLC including Ultra and Extreme. Very high speed products that use MLC because of ABL. Its a breakthrough. We’ve kept it under wraps where now we are bringing it out.

120 square mm 16 Gb MLC NAND on 43nm.  This is going into production as we speak and we are expecting revenues building up in the second quarter and it will become a very significant part of our revenue in the second half of the year. This is a significantly smaller chip than anybody else. We think that the rest of the industry is going to have some challenges getting into that kind of technology. 120mm sq chip compares with the industry’s smallest today is about 169 or 170mm square.

The third paper is the industry’s first commercial 3 bit/cell NAND and that too is a  SanDisk design in cooperation of course with Toshiba. That technology too is going into production as we speak now and uses ABL.  If you don’t use ABL architecture in 3 bits per cell, your chip will be very slow. That means that you can only use it at reduced capability part.”

Here is a slide from the SanDisk/Toshiba ISSCC presentation which does a nice job of illustrating the 240% speed gain of ABL MLC (34 MB/s- blue line: “ABL MLC Full Sequence”) compared to conventional MLC (10 MB/S- dotted yellow line: “MLC Conventional”):

through-put-slide

There is a lot more to say about ABL, but that is going to have to wait for a later post.

Suffice it to say that thanks to ABL, x2 MLC program- throughput of a 16Gb NAND chip can be boosted 240%, while at the same time providing an energy efficient system and a lower die size. Icing on the cake is that ABL apparently scales nicely to 43nm and below.

x4

x4 is a riddle, wrapped in a mystery, inside an enigma.

The wraps are scheduled to come off soon though- in the second quarter of 2009 when SanDisk expects to start customer shipments of the industry’s first 64 gigabit die employing 4 bits per cell (x4) technology, manufactured on 43nm technology.

This WSJ take says a whole lot about the potential of x4:

“The ruling, whose details haven’t been previously reported, discloses Samsung testimony that SanDisk’s patents to an invention known as x4 could cost the South Korean company billions of dollars. A panel of arbitrators in May found that a SanDisk unit was within its rights in canceling a contract that gave Samsung rights to the technology, a ruling recently affirmed by a federal court in New York…

Most investors have been unaware of a separate struggle affecting the x4 technology, which is not yet in production. It was invented by msystems Ltd., an Israeli company that SanDisk purchased in 2006.

Many companies are advancing flash memory technology. But x4 is regarded as a breakthrough that could double the capacity, lower the cost and potentially extend the useful life of the widely used variety of flash memory known as NAND.

“There isn’t really anything to compete with it,” said Jim Handy, an analyst at Objective Analysis, a market research firm in Los Gatos, Calif.”

If SanDisk’s x4 performs as advertised, and if Jim Handy is right in that there isn’t really anything to compete with it– and if Forward Insights is right and x4 grows to 17% of the market by 2012, no wonder Samsung wants to buy SanDisk.

17% of say 30 billion dollars [pick your own number- 2008 looks like $14 billion±] is a tad over $5 billion/year. Nice little niche- to own.

Here is a link to my last post dedicated to x4 speculation. Proceed at your own risk.

x1, SLC

Toshiba recently announced a new 43nm SLC NAND lineup which will come to market in the first quarter of 2009.

Maybe Toshiba is going to use all this SLC production itself. Somehow, I doubt it. Toshiba and SanDisk are joined at the hip fab-wise. Besides SLC has its uses.

SLC may be more expensive, but it does some things very well, thanks to its superior reliability and fast read and write times. ABL SLC even more so.

Toshiba said that its new “SLC devices will offer support for mobile phones, office automation equipment and servers, all of which require high levels of read and write speeds and reliability.”

The same should hold true for SanDisk.

SSD enterprise server applications seem particularly interesting. In the Q1 SanDisk conference call, Eli singled out the enterprise SSD segment as an opportunity for SanDisk:

“ One new market that we believe will provide opportunities to differentiate ourselves is the emerging Solid State Drive (SSD) market where we have deep expertise and strong IP.  This market appears to be developing into two significant segments: a large commodity segment in the notebook space where price per megabyte will be key, and the enterprise space where premiums could be significant.  Recently, there has been growing interest in embracing SSDs in the enterprise space to augment HDD storage arrays with high I/O bandwidth and low power Flash memory.  Intel is strongly promoting SSD adoption in computing platforms, and Sun and others are reporting dramatic performance and power gains achieved through incorporating SSDs in new enterprise server storage architectures.  These systems are sold on cost per I/O and cost of ownership rather than cost per megabyte.  For this market segment, the higher price of Flash SSDs is not an impediment, and we expect accelerated deployment of Flash SSDs in enterprise storage in 2009 and beyond.”

No doubt the new Toshiba/(SanDisk) SLC will be ABL, shown on the through-put chart above reaching about 60 MB/s (green line).  Again this 60MB/s program-throughput is the chips themselves.

It will be interesting to see what the overall SSD system performance will be when such program-throughput is optimized with a high speed interface and corresponding system design.

Today in these tough times, production cost structures are more important than ever. The NAND business is going to have to hunker down and live within its means. An end of an era and the beginning of another.

Over the next four to five years, as shrinks become more difficult and costly, SanDisk’s prospects along with the rest of the industry will hinge on mastery of x3 and x4. x3 and x4 IP will become critical.

x1 will have its story as will x2 and SSDs. Performance and reliability at each geometry will be more important than ever as shrinks slow.

How this story plays out remains to be seen, but the pieces in this game are already in play.  The truth is out there.

Advertisements

5 Responses to X Files

  1. bullseye says:

    Savo,

    Can you elaborate on the reliability of X3 and X4 compared with MLC? As more volt levels (8 levels for X3) are squeeze into a cell, the distribution has to be even tighter, otherwise, a failure would occur and ECC algorithm has to come to rescue. A metric of reliability used by this industry is the number 0f cycles a device is capable to sustain before it fails. For example, SLC is capable of 100K, MLC 5K, etc. Do you have this data for X3? If X3/X4 improves the die efficiency at the cost of reliability, its price has to be lower than MLC in order to compete. If X3 has to be sold cheaper than MLC, then this would undercut the cost benefit it brings.

  2. Poofypuppy says:

    Great post, thanks Savo. Lots of interesting info to chew on.

  3. bullsfusion says:

    Savo,

    Can you elaborate on the reliability of X3 and X4 compared with MLC? As more volt levels (8 for X3) are squeezed into a cell, the distribution has to be even tighter, otherwise, a failure would occur and ECC algorithm has to come to rescue until it runs out of spare cells. A metric of reliability used by this industry is the number 0f cycles a device is capable to sustain before it fails. For example, SLC is capable of 100K, MLC 5K, etc. How do X3 fare in this type of test? If X3/X4 improves the die efficiency at the cost of reliability, its price has to be lower than MLC in order to compete. If X3 has to be sold cheaper than MLC, then this would undercut the cost benefit it could bring.

  4. Hapa says:

    Greetings Savo, and thanks for another good one.

    If I was Eli, I would address the enterprise SSD market yesterday. It’s all about $$$, where corporations are the remaining players with cash and the urgent need for power savings in data centers. I’m including some excerpt from a link below:

    “…. There’s good reason to do so, according to the U.S. Environmental Protection Agency, which last year estimated that data centers, along with servers, will double their energy consumption to 100 billion kilowatt-hours by 2012 (see Data Centers Could Hit ‘Resource Crisis’).

    That could cost data-center owners $7.4 billion a year in energy costs, and require building the equivalent of 10 new power plants to meet those increased power needs, the EPA reported.

    And while the economy may be slowing, there was about $19 billion in datacenter construction projects undertaken by 21 large customers in the U.S. in June, according to Dean Nelson, who runs global lab and data center design services for Sun (see Data Center Power Consumption: By the Numbers).”

    http://seekingalpha.com/article/105363-sun-micro-flash-memory-goes-corporate

    /hapa

  5. Poofypuppy says:

    I agree with Hapa, it is important for SanDisk to get heavy traction soon in the emerging enterprise SSD market. While storage for cell phones and music players is obviously a big consumer market, enterprise storage is where the big boys (e.g. large financial institutions, Fortune 500 companies, the Pentagon, etc.) play. Every time they purchase a new or replacement SAN (storage area network) from EMC or HP or Hitachi, it is frequently 40TB or 80TB or more that is being sold along with multi-year support/maintenance for the component disk drives. These organizations are more than willing to pay top dollar if they can get guaranteed reliability, higher performance, and reduced lifecycle costs. If SanDisk could gain a strong foothold here, it would be a clear signal that its SSDs are simply the best, able to store mission-critical data for businesses and governments (instead of just for consumer gadgets).

    Unfortunately, SanDisk is seldom mentioned (at least, not by name) in major articles on the topic of enterprise SSDs. EMC, the world leader in SANs, has teamed up with STEC, and is open to using SSDs from Intel-Micron and Samsung…
    http://phx.corporate-ir.net/phoenix.zhtml?c=120870&p=irol-newsArticle&ID=1096121
    http://computerworld.com/action/article.do?command=viewArticleBasic&taxonomyName=hardware&articleId=9088158

    Sun has teamed with Samsung to optimize SLC NAND for SSDs (where SanDisk IP may not hold much sway)…
    http://searchstorage.techtarget.com/generic/0,295582,sid5_gci1332114,00.html

    The only enterprise win by SanDisk that seems noteworthy is the announcement last year that they would provide SSDs for IBM blade servers…
    http://news.soft32.com/sandisk-to-provide-ssds-for-ibm-blades_4717.html
    Haven’t heard much else since that time, however.

    Perhaps SanDisk could team with Seagate to attack the enterprise (data center) space (where Seagate is a leader), but then it seems SanDisk would need to share/split the profits with Seagate. If SanDisk does not become a major player in enterprise (data center) SSDs, then the only other major growth avenue would be the area of single-drive PCs and laptops. But enterprise SSDs is where the buyers (businesses and organizations) are willing to pay a premium price, so it would be disappointing if SanDisk can’t capitalize on this opportunity.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: