Thursday, March 22, 2007

Micron Predicts Death of Fabless, Again

ElectronicsWeekly: “What makes a great difference is manufacturing. Companies using fabless are really struggling because the technology is all in the processing,” said Sandor Barna, imaging senior director at Micron.

This is at least 7-year old claim. The reality check shows the difference is in execution, rather than fab vs fabless. So far Micron execution was excellent, while, for example, Magnachip was not. So Micron proliferates, while Magnachip struggles, both being IDMs.

12 comments:

  1. I don't agree with the exact statement, but a price war is coming and ODMs have a lot more margin buffer than fabless companies. I predict a year of declining sales for Omnivision and other "mainstream" fabless image sensor providers.

    ReplyDelete
  2. I agree with your prediction, but for different reasons. Looking into the history, the rise of IDMs coincided with industry transfer to 4T pixels. The reason is that 4T pixel requires quite an intimate device and process knowledge. I mean active knowledge, that is ability to simulate the pixel and understand why simulation and measurements differ and what to do with it.
    Historically fabless lacked this type of knowledge. What's more, the foundries do not like to share this type of information with their customers. Big customers, like Omnivision and PixArt got higher level of cooperation, but still they lacked the in-house knowledge. As a result, IDMs have an upper hand in 4T imagers.
    But the situation might change. There is just a handful of engineers keeping the Micron pixel expertise. If Omnivision manages to grab some of them, they close a technology gap in a year or two.
    As for the price competition, it's not sufficient to be IDM to get better costs. There is a good reason why TI, Freescale, NXP and many others opted for fab-lite strategy. Far East foundries are damn effective and price competitive these days.
    The real problem of Omnivision is the lack of key technical talent. If they find it, they are back in the game again.

    ReplyDelete
  3. You are not correct. First of all, it was Photobit (fabless) engineers that taught Micron how to make 4T cells. (and taught some of the foundries a bit before the acquisition). That is the reason for 4Ts and IDMs occurring at about the same time. Second, in a commodity market price war, the company with the lowest costs wins. Most of these companies you list have avoided the commodity market. Micron, however, has been in the commodity business a long time and was the only US survivor of a price war with Japan in the memory business many years back. Lastly, Omnivision already has a fine process team and between them and TSMC, they know perfectly well how to make 4T cells performance-competitive with Micron.

    Lastly, I meant to say declining market share, not necessarily sales. The market is still growing for a while yet.

    ReplyDelete
  4. Interesting discussion. You say Photobit did know how to make 4T pixel? Well, so why Photobit-Motorola-Kodak alliance was largely unsuccessful?

    No doubt, Micron execution excellence is a living legend now. Its commodity business knowledge can not be underestimated. However, I doubt Micron can compete on price with 2nd tier Chinese foundries. If these foundries learn how to make pixels, fabless jump ahead again. The "only" problem is how they get this knowledge.

    Getting back to Omnivision's "fine process team", how come its 1.75um pixel is not announced yet? My guess is its team needs some more "refinement".

    ReplyDelete
  5. Kodak wanted to be a Chinese wall between Moto and PB. PB never was able to get design rules, process information or talk directly with Moto. In fact, the wall was so effective that by the time the first chip came out of Kodak, PB was already ahead 2 generations. Kodak was so slow that Moto started their own CMOS APS effort. Interestingly enough, without Kodak or PB, Moto, an ODM, was not able to make a decent image sensor.

    Anyway, the first 4T APS came out of the Kodak-JPL alliance, and was a joint invention between JPL and Kodak. The sensor was reported for the first time in 1995 and the IP reported in NASA Tech Briefs shortly thereafter.

    As far as the Chinese wafer fabs go, if they are so price competitive, why dont they go into the memory business immediately? Maybe because equipment costs the same in China as it does in the US and Korea, and labor costs dont matter so much for this kind of business.

    You make a lot of assumptions as a blogger. But, you don't seem to know too much image sensor history or business. I like your blog anyway. Thanks for doing it.

    ReplyDelete
  6. Everybody knows his portion of history, me and you included. The full picture can be build with all these bits combined, but who knows all the bits.

    Talking about Photobit-Kodak-Motorola work, I've heard two dozen versions of why it failed. The same story looks very different from different angles. For example, one of the versions tells that Photobit was a very minor player in the alliance and its contribuition was insignificant. Compare this with your claim that Photobit has all the knowledge on 4T technology and taught others. On the other hand, some even do not mention Photobit as doing any work there.

    Why Chinese fabs do not compete in DRAMs? Look, 90nm is ancient process in DRAM world. At the same time most 2nd tier factories do not have even this. How can they compete in DRAMs?
    As for the process cost, it's a common statement that labor is just 1% of the process cost, thus fabs in China are supposed to have no price advantage. However, have you seen a quotation from any tier 2 Chinese fab? If you have, you know what I mean. Give them pixel knowledge in hand and you see Micron struggling to sell at cost.

    ReplyDelete
  7. Interesting comments on Photobit's role. I wonder why they were included in such an alliance if they played a minor or forgettable a role. Surely Kodak and Moto went on to become powerhouses in the CMOS image sensor business. (not). And Photobit merely laid the foundation for Micron becoming the world's largest image sensor house within a few short years of the acquisition. Undoubtedly a coincidence...

    ReplyDelete
  8. OK, look by yourself. Here is the original PR on Kodak and Motorola cooperation. Do you see any mention of Photobit?

    The product announcements include a nice wording on Motorola and Kodak patents, but nothing about Photobit. Look on this, for example:

    "This wafer process uses patented pinned photo diode and unique pixel design technology from Motorola and Kodak."

    How come that Photobit involvement is not even mentioned in these official PRs?

    Not that I'm denying Photobit's role, but it seems that both Motorola and Kodak keep it well hidden. Why they did so?

    ReplyDelete
  9. Why don't you google "Kodak Motorola Photobit" and you will find several articles on this.

    The Moto-Kodak alliance came into place becaue George Fisher, previously CEO of Moto, became CEO of Kodak.

    It is easy to understand why huge companies dont necessarily mention the little guys in their press releases.

    ReplyDelete
  10. Yes, many mention Photobit as a part of the alliance, but I'm unable to find any official PR on that. Incidentaly, 1997 article on Photobit-Kodak-Motorola also says:

    "most OEMs will agree, in that the foundried image sensors are said to be not up to snuff in resolution and overall image quality."

    Sounds familiar? History swang back and forth since then. I would take these claims with a grain of salt today.

    ReplyDelete
  11. Is there any problem between Micron and OmniVision?
    Hope to know a interesting story.
    Thanks.

    ReplyDelete
  12. To my knowledge Micron sued Omnivision for a few patent infringements, Omnivision counter-sued. Then it all ended in some form of cross-licensing agreement after about 1.5 year long litigation.

    Anybody can add details? Honestly, I doubt anybody reads this old thread, other than two of us.

    ReplyDelete

All comments are moderated to avoid spam.