The Promise and Challenge of Matrix 3D

As “King Flash” gets long in the tooth, potential successors to the non-volatile flash memory throne are already jostling for position. The best known contenders are Phase Change Memory, MRAM, Nanotubes, and Milipede, but there are many others, including Sandisk’s Matrix 3D.

For the next five years high-density NAND flash’s crown seems secure, but after that all bets are off. Sandisk’s NAND Technology Roadmap only extends to 2012. Probably, not just by coincidence. 20 to 30 nanometers and 4 bits/cell might be the end of the NAND road not just for Sandisk, but for the competition as well, if the competition can even make it that far: roadmap.jpg .

What edge does Sandisk have that its competitors don’t?

For starters, Matrix 3D is shipping commercially today; the basic technology is understood; and it can be manufactured with traditional CMOS processes using standard equipment and materials. It also appears to be scalable below the limits of NAND with higher bit scaling equivalencies. So, where’s the downside?

The not-so minor problem is that today’s Matrix 3D is only one time programable (OTP). It stores data just fine, but only once. That data will be retained for 100 years, but the chip is not re-programable.

The challenge for Sandisk is to turn this OTP technology into read-write (R/W). Sandisk thinks it can be done, but, not surprisingly, isn’t talking. This is a high stakes game. Better to keep the competition guessing.

Judging from Sandisk’s tone in this year’s analyst meeting, progress is being made. Last year when asked about Matrix 3D, Judy Bruner, Sandisk’s CFO, replied curtly that no roadmap existed and ended the discussion.

This year in discussing Matrix 3D, CEO Eli Harari, not known for careless phrasing, nonchalantly noted that he is optimistic that it could take just 5 to 7 years to achieve. One slide from analyst day even indicated that 3D R/W switching is expected in next three years: nvmtech.jpg .

While Sandisk won’t talk about the basic 3D Matrix technology, Matrix did plenty of talking before it was acquired by Sandisk in 2005. Also Matrix had the attention of the big guys from the start. No doubt it didn’t hurt that its founders had already made names for themselves on the tech scene.

Matrix was founded in 1998 by Thomas Lee of AMD/ Rambus; Mark Johnson of AMD/ Rambus/ MIPS/ Transmeta and Mike Farmwald, founder of Chromatic Research (and a co-founder of Rambus).

Matrix received a total of $80M in strategic funding, with big chunks coming from Sony, Eastman Kodak, Nintendo, Seagate, and Microsoft, among others.

Matrix’s 3D chip technology is something entirely different from NAND. It is not a variation of floating gate or charge trapping, but something referred to as antifuse. Instead of storing electrical charges, the chip has gazillions of microscopic fuses. When info is read to the chip, fuses are either blown or left alone, storing the info permanently (up to 100 years.)


Have no idea how SNDK plans to turn this technology into R/W.

What makes the Matrix technology particularly exciting is that it is truly three dimensional. It has multiple layers of memory arrays within the chip itself. In other words, active circuitry is not confined to the silicon surface. It extends vertically as well. More memory can be accommodated in the the same footprint. Same principle as in real estate: when land prices get really high, build upwards to maximize usable space.

Easier said than done. While many have tried & made claims to have succeeded, only Matrix has succeeded in manufacturing and shipping commercial product. The key, apparently, to Matrix’s success, is that Matrix’s engineers adapted the normal manufacturing processes used for LCD screens to memory chips. Layers of polycrystalline silicon are laid down on top of a base chip and then re-crystallized, forming active 3D matrices.

The early generations of Matrix memory only had two layers of memory cells. Current chips have four. Suspect SNDK is working on even more, as it has said that it sees 3D R/W as x4 to x8 equivalent.

In 2006 SNDK had $76M in 3D chip sales, mostly for gaming. This was up from $25M in 2005. These third generation 3D chips are a combination of 130 nm (memory layers) and 150 nm (base logic layer) process geometries.

Fourth generation 3D chips at 80 nm are scheduled to go into production in the second half of 2007. Sandisk is also planning to package these chips as consumable one-time use memory cards. The target market is the convenience user and the archival user.

Pilot programs have already begun, but Eli has said that, because of recent NAND price declines, the 80nm 3D will not have a cost edge on NAND.

These current uses, while not all that exciting in themselves, should pave the way for the next generation of 3D chips, which is where things could get very interesting.

45 nm 3D OTP chips are expected to arrive about 2008/2009. These could be the lowest cost-per-bit memory chips on the market and ideal for content distribution for mobile applications and DVD replacement.

Interestingly, SNDK does not make its own 3D chips. Today they are most likely being fabricated by Taiwan Semiconductor Manufacturing Company, which had been working with Matrix before the acquisition. On analyst day SNDK noted that it is planning to secure captive 3D manufacturing once volume exceeds the current foundry source.

Eli said a decision will be made in 2008 as to how to proceed with in-house 3D chip production. My guess is that SNDK is waiting until it has a much better idea of progress being made with the technology. Probably the biggest unknown is potentially the biggest deal: R/W.

If R/W looks viable and/or OTP for content distribution looks promising, SNDK will likely move ahead in a big way and, probably, with a partner, or partners, to help with the upfront costs, engineering ramp, etc. The likely candidate would seem to be Toshiba, but this probably isn’t a given.

Toshiba recently made an announcement about its own 3D technology, which appears to be an innovation in NAND stacking technology. The technology is entirely different from Matrix 3D and still in its initial stages.

Over the long term, Sandisk is targeting a 70/30 split between captive and externally supplied 3D chips, similar to today’s NAND supply. For this to occur, others besides immediate fab partners will have to adopt this technology.

When Sandisk acquired Matrix, it not only acquired Matrix’s 100+ engineers, it also acquired Matrix’s 100+ patents on 3D chip technology. This IP may be prove to be even more significant than Sandisk’s MLC patents for paving Matrix 3D’s path to the crown and securing SNDK’s supremacy in the field. However, without R/W capability Matrix 3D will be merely a pretender to the NAND flash throne.


9 Responses to The Promise and Challenge of Matrix 3D

  1. sambatyon says:

    Hi Savo,
    Glad I found you, I have been reading your postings on the FLSH board religously
    now that you mentioned MRAM I wonder what is your opinion on this Co. NVEC
    they claim to have patents for this technology.


  2. flashwave2000 says:

    What do you think happened to the gruvi card….i thought they had big hopes for this, but i haven’t seen this go anywhere?

  3. Poofypuppy says:

    Hi Savo,

    Thanks for the post, great info as always.


  4. Carrborough says:

    Fantastic to find out about this blog. I always studied the FLSH board and learned alot; I never had anything to teach anyone so I just lurked (since ’99). Thanks, Savo, for getting this going.

    Still lurking after all these years,

  5. savolainen says:

    Hi Sambatyon, gal2k, poofy, and Carrborough,

    Sambayon, As far as MRAM goes, it seems more of a candidate to replace NOR, DRAM or SRAM. Currently it doesn’t have the cost structure to displace NAND.

    MRAM is a story worth watching though. It is fast, doesn’t degrade over time as flash does, and has low power consumption.

    We will know that some folks are really serious about MRAM when a backer is willing to put up the billions of dollars needed for a fab. Then it is going to take a couple of years for construction and ramp and MRAM will require a system level solution.

    Don’t know anything about NVEC.

    gal2k, believe the Gruvi card was Matrix tech. Although not a winner itself, it very might well be the forerunner for future success. Don’t know the specifics, but assume that the Gruvi cost didn’t cut it and neither did the concept of delivering audio via a relatively high-priced card.

    Once the 45 nm 3D tech is out in 2008/2009, suspect we will see movies etc on similar cards at quite reasonable prices. This could be really really big. Of course, depending on the specifics. SNDK appears to be working on trying to build relationships with movie/TV studios to build the required alliances. Kate Purmal seems to be the key point person. SNDK’s emerging digital content strategy is probably worth a post.

    poofy & Carrborough, thanks for the feedback. Find it interesting to know who’s out there reading this blog. shows me hit counts by day. Yesterday there were 468 views. Highest ever FWIW, probably because SNDK had a good day and someone posted a link on Yahoo.

    Counts don’t really matter to me, but find the phenomenon of how info moves around the world on the internet very curious. It would be interesting to get a breakdown by country, but I don’t. Suspect most of those views came from the USA, then Israel, Europe and other. But it could be something else entirely.

    Today trying to track down whether the iPhone had Samsung’s MDOC-clone OneNAND inside (which I suspect it does), I used Google to translate some relevant Chinese comments on the iPhone.

    Maybe some folks in China are reading this and translating in reverse.


  6. sambatyon says:

    Hi Savo ,
    Thank you for your informative reply.
    I read somwhere that Freescale is working on MRAM I would appreciate if you could
    enlighten us on this.


  7. brunoi cipolla says:

    you write “However, without R/W capability Matrix 3D will be merely a pretender.”

    Savo, Given a low enough price OTP can replace all flash, like i did eliminating R/W CDs when the price for OTP CDs went low enough. (And all paper is OTP and recyclable)


  8. savolainen says:

    Hi BC, when I wrote: “However, without R/W capability Matrix 3D will be merely a pretender.” was really just talking about the successor to NAND itself. I should probably change that sentence to read “… pretender to the NAND flash throne.” or something like that.

    There is lots of potential in just OTP especially should another viable successor to NAND not emerge for a while.

    Before writing this post on Matrix 3D, I went back through my transcripts looking for Matrix 3D references. Interestingly when talking about Matrix 3D replacing NAND, Eli quite often adds that it also has the potential to disrupt/displace the hard disk drive business as well. HDD would be an example of an application where OTP wouldn’t work. Eli seems to have high hopes.

    It will be very interesting to hear in 2008 how Sandisk is going to proceed with in-house production of 3D chip production. On slide 156, Forecasted Capital Investments from analyst day, “potential investment in 3D capacity” is specifically excluded. These numbers could be pretty big.

    If SNDK decides to go forward in a big way, SNDK might find it self in the delicate position of needing to reveal technological progress to rationalize the large investment, while not wanting to release any info for competitive reasons.
    Recently ran across some interesting speculation here (towards the bottom of the Page): Sandisk Tinkers with Matrix 3D Memory

    Although this was written in early 2006, still seems relevant. The author doesn’t see how 3D R/W could work based on the OTP design by Matrix. Apparently some (ex) Matrix employees don’t think it possible either.

    Then the author goes on to speculate that Sandisk might be working with MRAM which apparently has some similarities. Then he goes on to say that MRAM wouldn’t allow the same densities at 3D, noting that if they were to figure that out they should win a Nobel Prize.

    “High-ranking Matrix employees noted to me that it would be impossible to revert a Matrix OTP fuse element fabricated at TSMC once programmed or “fused”…

    Maybe SanDisk engineers are considering MRAM? At least there would be a similarity based on the cells sitting between metal lines. If Sandisk makes an MRAM at the same bit cell density that the Matrix 3D OTP memory allowed, let me be the first to nominate them for the Nobel Prize.”

    Matrix 3D is a very interesting SNDK subplot. If they get the breakthrough wouldn’t be at all surprised to hear that other technology or technologies are involved. Maybe even something from FLSH. Always suspected that more was going on than just x4.


  9. franklin says:


    I am trying to understand Unity and how it relates to Matrix patents..

Leave a Reply

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

You are commenting using your 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: