Stratosphere and x4

A chapter in the Stratosphere saga closed recently when SanDisk settled legal proceedings with two former msystems’ employees. Apparently in exchange for $4 million, Amir Ban and Simon Litsyn dropped their claims against the company and Dov Moran. Some details were released on the Hebrew site NRG. [a machine translation is included at the end of this post]

The settlement was reached via arbitration and details will remain private.

The reference to Stratosphere, as another name for x4, caught my attention. I have long suspected that x4 was somehow connected to Stratosphere. Although Dov denied it, it seems Ban et al had a different opinion. Curiously, both may be right.

In any case its past history now.

This week I decided to dig back through my Stratosphere notes. After review, I am more convinced than ever that x4 is a related implementation of Stratosphere.

That said, x4 is a slightly different beast than Stratosphere. Each seems to have its own touchstone patent. Same inventors, but different Assignees. Different emphasis.

Whereas the Stratosphere technological is a breakthrough in physics, x4 appears to be primarily a breakthrough in signal processing algorithms. A related insight, but different emphasis.

This is interesting as it sheds light on aspects x4 never discussed publicly to my knowledge. x4 has been billed as a whole new way of looking at the physics of NAND itself. Now we may have a few hints.

A word of warning. Please don’t take my interpretations too seriously. I have done my best, but beyond 3 dimensions, I struggle. Both Stratosphere and x4 live in a world where very advanced math meets cutting edge memory physics. A 16-dimensional rectangular grid, embedded within a 16 dimensional hypercube is nothing for these guys.

Stratosphere

Stratosphere was conceived in 2000 as a collaboration between msystems and the Tel Aviv University. Key folks were Amir Ban from msystems, and Professor Simon Litsyn from the university. Professor Litsyn’s doctoral students Idan Alrod and Eran Sharon were also involved.

The breakthrough apparently came where information theory meets the physics of flash.

The key Stratosphere patent appears to be U.S. patent #6,469,931, “Method for Increasing Information Content in a Computer Memory.” This patent was applied for in 2001 and issued in 2002. Inventors are listed as Amir Ban, Simon Litsyn and Idan Alrod. Assignee is msystems.

The key idea is that groups of flash memory cells can store information above and beyond the information stored in the flash memory cells themselves. With a few twists, a top-up in memory capacity is possible for “free.” The image below, from the patent, illustrates a schematic description of a two-dimensional optimal encoding scheme.

strato1.jpg

One example given in the patent is for single level cell flash memory. Today, 16 cells of SLC NAND can store 16 bits of information. Using the Stratosphere approach, 20 bits of information can be stored using the same 16 cells of SLC, for a 25% gain in capacity.

Capacity gains increase as the bits/cell increase. In other words 2 bits/cell MLC is better than SLC, and 3 bits/cell MLC better yet, and so forth.

According to Amir Ban, a prototype of Stratosphere was developed and this Stratosphere technology was shown to a very-interested Toshiba in 2004, which was “fascinated” and sought to enter negotiations over the technology.

A note on Amir Ban, who must be quite the character as well as a rather smart guy: in his spare time he writes world-class computer chess programs. Along with fellow chess lover Shay Bushinsky, he is responsible for Deep Junior, the computer chess program that played Gary Kasparov to a draw in 2003 .

Amir Ban is on the left and Shay Bushinsky on the right in the photo below.

banbushinsky.jpg

x4

In June 2006 msystems’ CEO Dov Moran announced x4, claiming the the technology would prove to be the first viable 4 bit per cell flash storage solution, which doubled the storage capacity of current MLC solutions at roughly the same cost.

Dov explained that msystems had been working on x4 for 5 years in secret. The same time frame as Stratosphere. At its one and only msystems’ analyst day, the detailed x4 presentation was made by msystems’ chief scientist, Future Technologies Group, Simon Litsyn.

This is the same Professor Simon Litsyn from Tel Aviv University, that developed Stratsphere with Amir Ban and the same Simon Litsyn that is currently splitting the $4M settlement from SanDisk. Personally I hope that SanDisk finds a way to entice Mr. Litsyn into working with them (if he isn’t already) as he seems to have proven he has the right stuff.

litsysn.jpg
Simon Litsyn

Whereas the Stratosphere technology is a breakthrough in physics, x4 appears to be primarily a breakthrough in signal processing algorithms. A related insight, but different emphasis.

The key x4 patent appears to be U.S. patent #7,023,735, “Method for Increasing the Reliability of a Flash Memory.” This patent was applied for in 2004 and issued in 2006. Inventors are listed as Amir Ban, Simon Litsyn and Idan Alrod. Assignee is Ramot at Tel-Aviv University.

The main idea is that by programming flash memory using reference voltages, fractional reference voltages, and redundancy bits, the reliability of info stored in MLC flash memory can be improved. Again, just like for Stratosphere, each memory cell is not considered only by itself. Reliability can be improved by taking advantage of the frame of reference offered by groups of memory cells.

The patent illustration below illustrates the voltage bands of an eight level flash cell including fractional reference voltages.

x4patent.jpg

x4 has been described as a system solution, including a special x4 chip, a special very complex x4 controller, and special x4 software. All are essential. This patent #7,023,735 seems to address the issues of signal processing algorithms and hence the x4 controller and software.

I really don’t have a sense of the x4 chip itself, other than no special equipment is purportedly required for fabrication and that without the proper x4 controller the chips themselves are considered worthless. The x4 chip itself has never been described along the lines of the Stratosphere technology as far as I am aware, but the x4 controller fits pretty much exactly.

While the 735 patent is theoretical, the x4 controller is very real. Working prototypes exist. It may take over a million gates, but it works purportedly as advertised. As Simon Litsyn said in 2006, when Dr. Boaz Eitan questioned x4:

“I don’t know how I can respond to such comments, but I’m prepared to say that we laughed out loud at his comment. We had the product sitting on the desk, and functioning like clockwork.”

SanDisk x4 Roadmap

At the recent 2008 analyst day, SanDisk described its NAND roadmap, illustrated below.

sndkroadmap.jpg

x4 is still on schedule for a 2009 introduction at 43nm as a 64G chip. Same schedule as a year ago.

43nm is the technology node, and 2009 is the year, where SanDisk plans to get serious about moving beyond 2 bits per cell. It will be interesting to see how the competition responds.

In 2008 SanDisk will begin the production of 3 bits/cell at 56nm, but the majority of the production in 2008 will remain 2 bits per cell technology. This is because 43 nm 2 bits/cell is lower cost than 56nm 3 bits/cell.

2009 is when 43nm 3 bits/cell technology will become a significant driver of SanDisk production output. About 50% of total bit production in 2009 is projected to be 3 bits/cell 43nm technology. 2 bits/cell production will be less than 3 bits/cell production. x4 production in 2009 will be less than 2 bits/ cell production.

Basically 2009 will the year of introduction of x4. No comments were made about 2010, but x4 is projected at 3xnm.

My suspicion is that SanDisk’s engineering controller resources are currently focussed on SSDs @ 2 bits/cell and x3 @ 3 bits/cell. Next up will be tuning x4, as per schedule.

Notably absent from 2008 Analyst day discussion was any commentary related to x4 cooperative development with either Toshiba or Hynix. msystems had been working with both. My guess is that the dynamic has changed with SanDisk.

**** machine translation of 2008 NRG article below ****
[This translation was passed along by an acquaintance]

Compromise between SanDisk to the keys of the disk on key Ban and Litsyn, that developed drives of memory of the flash that are small when worked whether M-systems, will receive 4 million dollar in the frame of agreement of thaw that is secret. In return they will erase the claim against SanDisk
7/1/08

At the start of this week was signed agreement of compromise between the company SanDisk, that acquired M-systems, and Amir Ban and professor ‘ Shimon Litsyn, that developed drives of memory of flash are small ) disk on key (when worked with M-systems. By right of the use in this technology is sold if M-systems SanDisk in exchange for 1.5 billion dollar. According to agreement of thaw, Ban and Litsyn will receive about 4 million dollar. In return, is erased the claim that served Ban against parent company M-systems and founder of the company Dov Moran in the assertion of question violated their obligation towards him Onicso to themselves you are profitable the invention. The agreement itself will not be served for approval of the court, inasmuch as that will be prepared secret. The agreement was obtained in the frame of procedure

Ban was project manager ” stratosphere ” of M-systems, that his different name pursuant for 4x and that business in the development of the disk on key. According to him, if M-systems obligated to establish a company to the project, but after became clear her that the project of winner to the great success and that the company Toshiba is impressed from him very – changed her mind and requested to the asset to herself every fruits of the project. Accordingly sued Ban is a third from the rights that produced from the invention if M-systems and has her shares. Also concluded age, that joined the project in the stage more late, served a claim against if M-systems, SanDisk, Amir Ban and Dov Moran.

http://www.nrg.co.il/online/10/ART1/681/925.html

**** Marker.com from 2004 below ****

Former worker claims M-Systems owes him rights in Stratosphere technology
10.10.2004 | 14:47
Eynav Ben Yehuda

A former employee of M-Systems (Nasdaq:FLSH) claims to essentially be one of the company’s founders, having held high-ranking positions there in the 12 years almost since its establishment. In practice M-Systems was launched as a startup in 1989 and Amir Bann joined in 1992, but he claims to have played a key role in making it a half-billion dollar company.

Bann quit in October 2003 but claims he is owed 33% of the rights in a joint company that M-Systems agreed to establish. He is now suing at the Tel Aviv Labor Tribunal for those rights.

Not so, M-Systems responded in a letter to Bann’s lawyer, before the suit was filed: there was no agreement.

From 2000, writes Bann’s lawyer Itzhak Goldberg, Bann and others had been engaged in developing a technology called Stratosphere. M-Systems had shown no interest in it. Nor did it provide financing for development, which was collected from external investors.

After resigning, Bann suggested M-Systems and the Stratosphere inventors, including himself, establish a joint company that would commercialize the patented technology. M-Systems was persuaded and asked Bann to prepare a business plan, writes the lawyer.

The resultant business plan incorporated a plan for the distribution of shares and rights in the putative company. (Bann did not append his business plan to his lawsuit, because it contains confidential information, his lawyer wrote.)

In late December M-Systems decided to accept the plan, Bann claims, and he and others made plans for the future accordingly. For one thing M-Systems asked him to delay leaving the company and he agreed to accept a half-time position.

Under the agreement he claims to have with M-Systems, Bann said he should have received 33% of the shares in the joint company.

Believing the joint company would be established, Bann claims, he acted to recruit manpower, to update and execute the business plan and interest venture financiers, and mainly continued to develop the technology, resulting in a prototype.

To create the prototype, Bann developed a new technology after December 2003, to increase the memory capacity in the flash compartment at relatively low cost. Bann notes that he developed the technology for the new company, not for M-Systems itself, based on the agreement he says was reached in December 2003.

Meanwhile, great interest was sparked in the new technology, Bann continues, adding that M-Systems has business ties with Toshiba. And the technology was presented to Toshiba, which was fascinated. It sought to enter negotiations with M-Systems over the technology, Bann claims.

At that point, M-Systems acted in bad faith and cancelled the agreement, claims Bann, and sought to deprive him of his rights.

Contacted by attorney Goldberg in mid-2004, M-Systems responded that the claims represented a distortion of reality and said it rejected Bann’s claims. Not only did M-Systems not cancel or breach an agreement, it stated in its letter, which Bann appends to his lawsuit: there had been no agreement. Nor had Bann been a founder, having joined the company well after its establishment. Also, M-Systems claims to have fully financed the development of Stratosphere by Bann and others.

M-Systems agreed to consider establishing a joint company, it says, but never actually agreed to do so, and the business plan Bann presented did not constitute any such agreement. While M-Systems considers spinoffs from time to time, it rarely carries out the plans, it added.

Obviously a subject as weighty as establishing a joint company to develop a new technology would be thoroughly covered by agreements, M-Systems pointed out, which was not the case here. And Bann’s claims to ownership over the technology contravene the terms of his employment agreement, which he did not formally terminate in October 2003 as claimed.

Advertisements

13 Responses to Stratosphere and x4

  1. Amir Ban says:

    Interesting and intelligent analysis, but the 7,023,735 patent is not the “X4 patent”. Indeed, a search for an “X4 patent” must come up empty-handed, since X4 is a straightforward application of Stratosphere. Stratosphere envisions taking some baseline flash and increasing bits/cell, using methods described in 6,469,931 and later patents. X4 means using the same methods when the baseline is 2 bit/cell MLC and the target is 4 bit/cell.

    The term X4 was coined by my Stratosphere group in a meeting with Toshiba in March 2004 to describe the specific cooperation proposal we were putting forward to them. This was the same meeting whose smashing success brought Dov Moran to “repossess” Stratosphere, ultimately leading to legal proceedings.

    Anyway, the 7,023,735 patent, originally filed June 2003, was too early to support msystems’ “X4 is not Stratosphere” claim. That claim was first made only in 2006, on much flimsier grounds. From the legal point of view, the claim was a blunder which would not have withstood the scrutiny of a court proceeding, and may have contributed to Sandisk’s reluctance to reach a courtroom.

    Amir

  2. savolainen says:

    Greetings Amir,

    Many thanks for the comment.

    Back in 2006, I suspected that x4 was Stratosphere, but the description of x4 didn’t include programming groups of flash cells, which to my limited knowledge was a most important Stratosphere angle.

    Also from the 6,469,931 patent, my original reading was that the maximum gain (Stratosphere) was 58%. On re-reading this I see I may have under-estimated the power of Strato:

    “Higher-Order Hamming Codes

    The method detailed above using s, which has been demonstrated on codes for 16 cells, with an information gain of 25%, is directly applicable to higher order s for 32, 64, 128 etc. cells. For higher orders, the proportional information gain increases regularly and can be shown to reach a limit of around 58%.

    To demonstrate this, consider that there are Hamming Codes for blocks whose length is a power of 2: H.sub.ex (2.sup.m, 2.sup.m -m-1, 4) and n=2.sup.m. The check matrix for this code is 2.sup.m.times.(m+1), so that the syndrome has (m+1) bits. A co-set of the code is defined as the set of words with a given syndrome. The set of code-words is the co-set whose syndrome is comprised of all zeroes, and any other co-set is a translation of the code-word set by some fixed bit vector (by translation we mean a bitwise addition modulo 2, or a bitwise XOR operation).

    From the construction of the sphere arrangement given above by example for `n=16`, it is apparent that not only the , but also any of its co-sets, could serve as a basis for the sphere packing construction. Co-set based sphere packing construction arrangements may yield a different number of spheres (because of the difference in the ways 0’s can be modified to 2’s in the co-set words). However, since: The total number of points in the n-dimensional voltage space is 3.sup.n. The number of distinct co-sets is equal to the number of possible syndromes, which is 2.sup.m+1 Each point is a center in the sphere construction for one, and only one, co-set. For at least one co-set, the sphere construction has more sphere centers than the average for the sphere constructions for all co-sets.

    Therefore, there is a sphere construction with at least 3.sup.n /2.sup.m+1 spheres. The number of information bits this construction is able to represent is the integer part of log.sub.2 (3.sup.n /2.sup.m+1)=n log.sub.2 3-m-1. Remembering that the original information content of the n cells was `n` information bits, the proportional gain is log.sub.2 3-(m-1)/n, which (as n=2.sup.m) increases monotonically with a limit of log.sub.2 3=1.58496 . . . as n.fwdarw..infin.. Therefore, 58% information can be gained through the construction. A gain of 50% is attainable for n.gtoreq.128. ”

    From your comment it seems that using baseline 2 bit/cell MLC, 4 bit/cell can be reached with Strato. Is this 100% gain from additional refinement (other patents) or did I simply misinterpret the 58% information gain comment and 100% gain is possible with 2 bits/cell?

    As a follow-on, if x4 can be reached using a baseline 2 bit/cell MLC, can greater storage capacity be reached using 3 bit/cell? Put another way, in your mind, does NAND/ Strato max out at 4 bits/ cell?

    Best,
    Savo

  3. Amir Ban says:

    Savo,

    58% (log2(3)-1) is the limit gain only when increasing 2 levels (SLC) to 3 levels, which was the example we used in the patent to illustrate the invention. When increasing 4 levels (2-bit MLC) to 16, the gain in the limit would be 100% (log4(16)-1).

    The 2-levels-to-3 case was convenient for illustration as it lended itself to solution using an elementary code (Hamming). Bigger cases like X4 need more complex codes (on which there is a vast literature and body of knowledge, and which is the field of expertise of Prof. Litsyn). Quoting from the patent:

    “Even better results of increased information gain provided by the method of the present invention can be achieved by using other codes as basis for the sphere construction, and by subdividing the voltage window into more than 3 discrete levels. The method above presumed single-bit-per-cell Flash as a base technology. Similar techniques are applicable when using Flash technology with true multi-level cell capability, in which case a preferred embodiment of the method of this
    invention subdivides the voltage window into more levels than the inherent (nominal) multi-level cell capability allows, and improves overall information content by encoding several cells at a time. Known results in the theory of sphere packing suggest that the potential information gain obtainable through the application of this invention grows without limit, albeit slowly (in a logarithmic rate) with
    increasing size of the cell group. ”

    So the capability for X4 was already described in the initial patent, though the details of how to do this in a economically and with a reasonably-sized block remained to be worked out.

    Regarding exceeding X4 on top of 3 bit/cell, the answer is that it is certainly possible. Roughly, Stratosphere gain is additive in bits: if you can increase 2 bits to 4, similar methods would increase 3 bits to 5. Caveats apply though: First, X3 is to my knowledge already using aggressive coding, so is already using up part of its Stratosphere gain potential. Second, 5 bits means reading 32 distinct voltage levels, even 64 levels if employing US7,023,735. That’s no mean feat, and almost down to counting a handful of electrons.

    Possibly NROM+Stratosphere can reach 6 bits/cell. During 2001-3 I tried several times to convince Boaz Eitan to cooperate on Stratosphere. I believed then, and still believe that he made a mistake in turning this down.

    regards,
    Amir

  4. savolainen says:

    Amir,

    Thanks again for the explanation. Strato as x4 is making more sense now. It seems like a potentially very valuable technology.

    I suspect that most folks think of x4 simply as a bulked up version of 2 bit/cell MLC and hence will have very poor performance and reliability. I don’t think these folks have accounted for the fact that x4 includes another technology- Strato – as well.

    I am wondering about Strato’s performance and reliability.

    Dov said, that he expected x4 to have performance very close to current MLC and in some respects better than MLC. He also said that as x4 matures he expects it to be used in just about all applications: memory cards, USB flash drives, embedded flash data storage, MDOC, and SSDs.

    On the other hand, Eli doesn’t sound quite as optimistic. He has simply said that x4 will first be targeting audio/video products, and that eventually he expects it to become a “very very important part of the cost reduction.”

    Do you expect x4 performance (eventually) to be good enough for memory cards and SSDs?

    Regards,
    Savo

  5. Amir Ban says:

    Savo,

    Now you are going into specifics, which I won’t discuss, but I will say that in Stratosphere there is a built-in trade-off between capacity gain and reliability. For X4, the nominal 100% capacity gain originally meant taking a reduction in flash reliability. This is probably what Eli meant by “audio/video products”, i.e. applications that can deal with less-than-industrial endurance, data retention, temp ranges. That should be good enough for memory cards and USB flash drives. Can it be made good enough for SSD’s? I don’t know.

    Amir

  6. savolainen says:

    Greetings Amir,

    Thanks for sharing what you can. I understand that there are limits to what you can discuss. Stratosphere/x4 is a most interesting story which I expect will eventually dribble out. My guess is that once x4 starts shipping, more attention will be paid to the unique aspects of the technology and story.

    Thanks for mentioning NROM. Back a ways, I had also been thinking NROM + Strato had great potential. Too bad that angle wasn’t pursued. Boaz probably did make a mistake.

    Personally I find it curious that Spansion seems to be making tracks with MirrorBit, while Infineon seems to have given up. Maybe AMD and SNDK will work something out down the road?

    I am guessing somewhat here, but suspect that you were key in the development of TrueFFS, a most critical and important insight. Is this true?

    If this isn’t too specific, would be most interested in your thoughts regarding the decision, way back when, to deliver TrueFFS up as a standard for others to share.

    Regards,
    Savo

  7. Amir Ban says:

    Savo,

    I think the difference between Infineon and Spansion is that the first was interested in data flash, while Spansion seems to be concentrating on NOR-like code flash. Code flash obviously cannot use Stratosphere. Even if Spansion wants to try for a winning NROM/Stratosphere combination for data (as well they might), this is up to Sandisk as well, and I see no compelling reason for Sandisk to cooperate.

    Indeed I did TrueFFS, way back in 1992 when msystems was a garage company, and this is what brought us out of the garage. Later in 1996 I did the NAND-suitable version of TrueFFS, NFTL, which went into the DiskOnChip 2000 and later products.

    TrueFFS became the FTL standard, but that involved giving license only to the PCMCIA community, which never became a major flash market. There was no giving up of rights to the rest of the world (as late as 2003 msystems successfully sued JMTek, a Diskonkey competitor, using the 1993 TrueFFS patent). It was risky strategy for sure, but not competing for the standard also had big risks. In hindsight it worked well. It gave msystems stature and brought Intel to our side. And later, when Intel stagnated, we already had all the it took to become an important NAND player.

    regards
    Amir

  8. savolainen says:

    Hi Amir,

    Thanks for the info and for correcting me. For some reason I thought TrueFFS as a standard, extended beyond the PCMCIA community.

    I don’t know if you have seen this, but it appears that Wikipedia is giving Microsoft credit for your TrueFFS work. As I recall MSFT was working with FLSH at the time (?) and did include FFS in MS-DOS, but the FTL didn’t originate with MSFT as this article claims/implies. As you describe, TrueFFS came from FLSH.

    http://en.wikipedia.org/wiki/Flash_Translation_Layer#Flash_file_systems

    Eli has said that the acquisition of FLSH has given SNDK a head start with SSDs. I would think that in your time at FLSH you must have worked on SSDs or at least been aware of the challenges.

    In your opinion, without being too specific, how did FLSH give SNDK a head start? I am guessing that FLSH developed the framework of a viable system architecture/strategy for SSDs that SNDK has been able to build on. I would also think the years of real world SSD experience would be particularly valuable.

    Thanks again,
    Savo

  9. Amir Ban says:

    Savo,

    The Wikipedia entry doesn’t actually say that Microsoft invented flash file systems, but I can see how it can mislead into that impression, especially since the next thing it mentions is FTL while glossing over TrueFFS.

    In fact TrueFFS preceded FFS2 and went on to become the industry’s standard solution, while FFS2, released a few months later and offering a rather different solution was a technological cul-de-sac and was abandoned. There’s a fine document that Intel wrote on FTL at http://www.embeddedfreebsd.org/Documents/Intel-FTL.pdf which gives the right attribution (near the end of the document).

    I would guess that the “SSD head start” is mainly msystems experience at providing industrial flash disks of various kinds, and especially the FFD.

    Amir

  10. tecinv says:

    Not exactly on topic, but related – I have several unanswered questions:

    1) How come Sandisk is lagging so much in MLC SSDs versus competitors ?
    Competitors already produce MLC SSDs, while Sandisk will have them only in 2009.
    And this is when Sandisk always says that Sandisk “understands MLC better”?
    Are there really limitations (As Sandisk claims) to competitors products that make them not suitable for SSD demands. If so what the limititations are?

    2) Why doesn’t Sandisk go directly to 3D R/W (They have patents on the technology and how to produce it, as I understand). It will solve the performance/reliability and cost problems of X4. BTW: Were there any questions or update about 3D R/W in the analyst meeting ?

  11. savolainen says:

    Thanks again Amir for the info,

    Right now I am working on Intel’s NAND strategy. In an earlier comment you mentioned Intel. I would be most interested in any insight you might have as to why Intel’s efforts with msystems stagnated. Was Intel simply distracted? Too slow and bureaucratic? Short on vision? Have always felt that this might have been an opportunity missed- primarily by Intel.

    Regards,
    Savo

  12. savolainen says:

    Greetings tecinv,

    You bring up two significant SanDisk story-lines.

    To my mind, its still too early to grade SNDK’s MLC SSD efforts.

    SNDK’s first MLC SSDs will be 43nm. Personally I believe this was a strategic decision. My sense is that Eli et al looked around at the lay of the strategic landscape and decided not to waste time where it wouldn’t pay off. Believe they decided it was better to be “late” than to go for the pr. If SNDK delivers, all will be forgiven. If not, well that’s another story.

    BTW I don’t think we have seen the best the competition is poised to deliver.

    I have recently been working on Intel’s NAND efforts/strategy. Intel looks poised to deliver some impressive MLC SSDs- before SNDK shows its stuff. Sometime in July, Intel is touting delivery of MLC SSDs with sustained R/W of 250/50 MB/s.

    By comparison Toshiba’s MLC SSDs are reportedly 100MB/s read and 40MB/s write.

    Intel’s MLC SSDs will purportedly be 50nm NAND. Toshiba’s have to be 56nm MLC, currently the San-shiba workhorse. And as far as specs go, believe this is Intel and Toshiba each talking, not independent third parties.

    As far as 3D R/W goes. Believe this will be a post 2010 story if it lives up to expectations. There was a lot of info in the analyst day. My hope is to put together a dedicated 3D Analyst day post– eventually.

    The most revealing 3D lines from analyst day were Eli’s:

    “ This (Matrix) was a very very very good acquisition. Of course a technology acquisition, but I think very prescient, because I think that in the next several years it will start paying off…

    The other part is developing the switching element that will allow us cycling. That is a very very difficult challenge and although we have some interesting progress, actually quite a lot of progress on a number of fronts, I would not say that we have a manufacturing technology at this point.”

    Note the implication that the SanDisk’s 3D diode R/W technology seems to work as advertised. The issue was deemed manufacturing. I am about to post a transcription of Eli’s Analyst day remarks which include these remarks and more on 3D.

    Happy Easter to all,
    Savo

  13. Amir Ban says:

    Savo,

    Your question conflates two issues: the breakup of the Intel-msystems alliance and Intel’s stagnation and loss of flash leadership. This were two different events that happened at different times, though they may well be connected.

    The problem with the Intel-msystems alliance was that Intel’s vision was msystems nightmare: It saw msystems as basically a software house, whose software technology could be given away to the flash world to make Intel’s flash work, with msystems receiving per-project compensation. We could never get them interested in doing joint products, and the fact that their plan had nothing in common with ours did not seem to register on them.

    And so we parted, and embarked on the fruitful Toshiba alliance.

    This was in 1998. At that time Intel was the world’s no. 1 flash maker. NAND was growing, but NOR was king. In that year they were sampling StrataFlash, the world’s first MLC product. They had a roadmap for 3-bit MLC in 2001 and 4-bit MLC in 2003.

    It didn’t happen, and while NAND was booming and NOR was stagnating, all they could do was make solemn statements that they are not planning a move to NAND. By 2004 they lost no. 1 position to Samsung, and a year later they were finally entering the NAND market, as no. 5 or so, at a point where NAND is fast becoming a commodity. And Intel is a company that by philosophy, strategy and design stays away from commodities, and are not interested in any position except #1. A colossal failure that has not received the attention it deserves.

    Amir

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: