Seeking Alpha

Russ Fischer

 
View as an RSS Feed
View Russ Fischer's Comments BY TICKER:
Latest  |  Highest rated
  • Intel: On-Processor Chip Memory - It Can Be Done At High Margins [View article]
    "Russ and Cincinnatus may be the only ones in the world that proclaim TSMC has been lying about its 20nm and 16nm progresses."

    I'm OK with that:)
    Jul 12 02:13 PM | 1 Like Like |Link to Comment
  • Intel: On-Processor Chip Memory - It Can Be Done At High Margins [View article]
    markitos,

    Interesting.
    This article claims that the foundries have not yet bought equipment for finfet production.

    http://ubm.io/1ohMSQZ

    If you can believe that then believing finfet from the foundries before mid to late 2015 is pretty tough.

    I understand that Intel modified standard equipment for their finfet production. Maybe the foundries did too.

    I see foundries spending ~$20 billion to get to finfet. That means they have to have $40 billion in sales in the bag. Where? Intel has the much maligned $50 billion, 75% margin PC CPU business to support the CapEx. What do the other guys do?

    I say they quit the bleeding edge or charge ridiculous prices of their customers, who will, in turn, have to charge ridiculous price for their products.
    Jul 11 07:08 PM | 5 Likes Like |Link to Comment
  • Intel: On-Processor Chip Memory - It Can Be Done At High Margins [View article]
    drchet,

    Yes, I am able to do just that.
    This industry came very close to being totally dependent on Korean suppliers for mobile DRAM.
    Prior to Elpida, Micron didn't/couldn't/wouldn't do mobile DRAM. How would you like to be running Apple staring at the prospect of Samsung (or cousin Hynix) being your only mobile DRAM supplier?

    Today Apple and all those using Qualcomm chips that are DRAM/AP PoP assemblies use one supplier for the AP and another for the DRAM. I can only imagine the finger pointing on quality issues that that arrangement generates. At least in the beginning Apple had only one place to point; Samsung. Not so today.

    From a supply chain streamlining standpoint, I would guess that Apple would pee in their pants to have one supplier of both parts, all assembled, tested and done up in a bow and delivered to their sub-contractor.

    Look around....there is only one company in a position to supply both DRAM and AP on a single chip or package....that would be Intel because it is only becoming feasible at 14nm. Do you see TSMC doing a monolithic AP/DRAM? No, because TSMC only does fab for fabless (design) semiconductor companies.....and they are two nodes behind Intel.

    Intel can do pretty much whatever it wishes without regard to patent issues since they offer, and everyone accepts a giant patent cross license agreement from them.

    BTW, there is no DRAM equivalent JV with Micron as there is with IMFT on NAND.
    Jul 11 06:55 PM | 5 Likes Like |Link to Comment
  • Intel: On-Processor Chip Memory - It Can Be Done At High Margins [View article]
    From your link:
    "The pad ring extends 0.5 mm (500 µm) into the chip, so the core area is 9 x 9 = 81 sq. mm."

    So the pad ring on a 100 mm^2 chip is 20% of the area. Those I/O pads have to exist on the processor as well. All that RAM pad area on the DRAM AND the processor go away. The driver circuitry on both parts also goes away.

    It IS doable today; it IS cost effective today. Whether it WILL be done is anyone's guess.

    The reality that it CAN be done should be of concern to anyone who owns Micron. That is part of the reason that I bailed. Can you imagine what would happen to Micron's stock price the day after Intel even mentioned DRAM-on-mobile-chip?

    The simple fact of the amount of "stand-by" capacity at Intel has most semiconductor executive scratching their heads and looking over their shoulders.

    Intel is the master of yielding large die sizes; a 100^2mm die is not much of a challenge to Intel.

    BTW, the article was talking about 90nm processes.
    Jul 11 06:19 PM | 6 Likes Like |Link to Comment
  • Intel: On-Processor Chip Memory - It Can Be Done At High Margins [View article]
    Jaret,

    you do know that DRAM is in-line fixable through redundant rows and columns, so the addition of a reasonable amount of DRAM shouldn't affect yields as you suggest.

    Think single chip microcontrollers with the amount of memory moved a few decimal places to the right. Something like that will be needed for the IoT anyway.

    I'm just trying to think of possibilities to use the extra Intel capacity.
    When we find out what Intel has up their sleeve it will probably be a head-slapper and all my guessing will be wrong.

    I'm just having a little brain fun here and making some money at it.

    I see that you have a technical education, but nothing about behind the scenes semiconductor work history. That would help.

    BTW, I found your article insulting an demeaning to me.

    I suspect you are one of those guys who insults without even knowing you're doing it.
    Jul 11 05:51 PM | 19 Likes Like |Link to Comment
  • Intel: On-Processor Chip Memory - It Can Be Done At High Margins [View article]
    pb,
    I guess the chip would be 50-60 square mm bigger....Maybe $20 more money. If I'm even close, DRAM on-chip should cost about $5 per GB and sell for $15 to???/GB.
    I'm guessing that it only becomes possible at 14nm. If the other guys can't get there, DRAM on-chip becomes proprietary.

    DRAM on chip shouldn't affect yield much. They use extra rows and columns so the memory can be "fixed" during test.
    Jul 11 05:37 PM | 1 Like Like |Link to Comment
  • Micron And DRAM Cost [View article]
    The IMFT joint venture doesn't include DRAM or the HMC for that matter. Intel would buy DRAM from Micron at market prices.
    Even with IMFT there is no prohibition against having individual capacity. Micron has wholly owned NAND capacity as well as JV capacity. Presumably Intel could do the same thing if it made business sense.
    Jul 11 12:00 PM | 3 Likes Like |Link to Comment
  • What Are Intel's Next Disruptive Technologies? [View article]
    Perhaps a page out of the toy industry marketing?

    That would be: Limited availability of new goodies for Christmas forcing purchase of the "older" stuff (Haswell), makes for a good 4th qtr. Then a flood of the new stuff (Broadwell) for 1st qtr to make the 1st qtr good as well.

    The toy guys have been doing this, to good advantage, since Christ was a corporal.
    Jul 11 10:20 AM | 3 Likes Like |Link to Comment
  • What Is Intel Telling Us? [View article]
    Barinder,

    Yes I do.
    Jul 11 09:58 AM | Likes Like |Link to Comment
  • What Are Intel's Next Disruptive Technologies? [View article]
    Arnold,
    Another well done article.
    Jul 11 12:04 AM | 4 Likes Like |Link to Comment
  • Micron And DRAM Cost [View article]
    Jaret,

    Nice try on the yen thing.
    http://on.wsj.com/TVdrSp

    The Micron play was great; it made me, you and a bunch of other people a lot of money.

    The events that made it great were:
    1. The Elpida acquisition that didn't look like a good idea in the beginning.
    2. The unprecedented rise in DRAM prices.
    3. The Hynix fire.
    4. The lying about the recovery time from the Hynix fire.

    I bought about 20% of the way through the DRAM price increase and watched the other pieces of the perfect storm assemble themselves.

    It's done now. We now know that Micron (with Elpida), at full output, at today's prices, can do $16 billion per year at about 18% net profit, or approaching $3 per share. We have two quarters of the same revenue with a forecast for about the same revenue for next qtr.

    Now the spending starts. Both of these companies have not been able to spend to keep facilities up to date. CapEx will likely double, so depreciation will double.

    Micron PP&E has remained about the same from 2011 to the latest qtr, which confirms that the Elpida assets had been depreciated down to close to zero. From here that Elpida cost advantage will decline.

    Employee costs will increase as people play catch up, since pay raises have been non-existent for the past 10 years. Hiring to fix the NAND problem has already begun.

    All in all, operating expenses will begin to weigh on costs and earnings going forward. The move to smaller and unified nodes in DRAM will help efficiency and offset the headwinds to earnings. The only way that margins expand significantly from here is with an actual shortage of memory which will hurt the entire industry. Big players will not let that happen.

    In a recent Goldman report, they claim 30% operating margins on DRAM will attract peripheral suppliers, even mentioning Intel as a potential supplier of jelly bean DRAM!

    The Intel part of your argument is hard for me to buy. Intel has moved static ram to on-chip cache and steamrolling the SRAM suppliers in the process and no one seems to have noticed. Hell, Micron used to be a big SRAM supplier, not any more.

    Intel needs a simple process to use as a pipe cleaner for advanced logic nodes...DRAM is great for that purpose. DRAM on chip could take many forms, the most likely being a TSV stacked assembly with ultra-wide, high speed interfaces. I'm expecting and projecting that 4GB of DRAM could be a 160 square mm chip done on fully depreciated lines with very high yields. If that were to happen, I would bet my left 'nad that Micron will not be involved in the DRAM chip. At $35 per 4GB block of memory, Intel would generate $12,000 wafers with a $3000 wafer cost.

    Anyway, Micron is a much better known quantity now. Intel, on the other hand is still a big question mark. Still building new capacity long after it became obvious that PCs had stopped growing.

    For anyone who is concerned about my early exit from Micron. I bailed at 110 times my "investment"....which would have been ~330 times my investment had I stayed in. My gain on Intel is at 4X, so the total is 440x over less than two years. I expect the Intel play to ultimately be a 50 bagger. 110x50 is 5500.

    I'm doing just fine and am really happy for those who scored on Micron. Don't get married to it and if you stay in buy protection around earnings until you are ready to cash in.

    Simpler still, join me in Intel and sleep soundly. Today, down market, down SOX, down Micron, up Intel. Really down ARMH??? Is Intel getting some mobile design traction?
    Jul 10 12:43 PM | 10 Likes Like |Link to Comment
  • Intel: Can The PC Market Keep It Chugging? [View article]
    I think Intel doubles from here by Jan 2016.

    There's about $.80 per share in mobile losses that will go away in the next year and a half, one way or the other. That loss should be replaced with a $.20 per share profit.

    Depreciation is $7 billion per year that should be cut in half when CapEx declines. That's another $.70 per share.

    The R&D associated with expensed CapEx should decline by $.40-60 per share.

    Growth through filling the fabs should add another buck a share.

    We could be looking at Intel earnings in the $5 area in the fairly near future. At even 12X earnings that gives a $60 stock price.

    With a 2-4 year depreciation schedule on equipment, the 14nm fabs will be half depreciated by the time they ship the first parts. That makes it pretty tough on the fabless/foundry guys who have yet to commit spending on finfet processes.

    Just sayin'
    Jul 8 01:07 PM | 12 Likes Like |Link to Comment
  • Intel And On-Chip Memory [View article]
    ote,

    No question that the payoff is bigger in HPC.

    There is more going on here than that the propagation of voltage....wider interfaces, for example.

    To the extent that the "memory wall" exists even in mobile processors, necessitating higher clock rates, nearer memory and wider interfaces would allow reduced clock rates on the processors with attendant lower power.
    Jul 5 06:46 PM | 3 Likes Like |Link to Comment
  • Intel And On-Chip Memory [View article]
    Just heard from Micron IR. The HMC is a collaboration between Micron and Intel, but covered under a different deal and Intel will pay market prices for HMC.
    Given this, I would say that Intel is involved in TSVs up to the hips. TSV memory on processors is the likely path for the short term.
    Jul 4 11:08 AM | 5 Likes Like |Link to Comment
  • Intel, TSMC, Samsung, Apple: Who's On First? [View article]
    Rag,

    I would rather drink a glass of snot than say this:
    I agree with nearly everything you said above. What is going on is a battle for the high technical ground in semiconductors and if the foundries can't find a home for about another $20 billion in sales they will fail on the leading edge simply due the cost.

    Even then the foundry business at 40% GM is doomed against an Intel with a built in $50 billion of 75% GM business.

    I frankly just don't see how the foundries make it ON LEADING EDGE. On anything below leading, the dance continues for the foundries.

    http://bit.ly/1ohMSAH

    Qualcomm is already in the process of "punishing" TSMC for favoring Apple over QCOM AND deviating from the long term business model of not supplying directly to OEM customers. From the above link it appears that TSMC has traded QCOM business for Apple business. The QCOM collaboration with Samsung/GF doesn't sound great either.

    http://ubm.io/1ohMSQZ

    According to the above link the equipment management says the foundries have not yet spent for 14-16 nm finfet. How can you expect volume production of 14-16 nm finfet in 2015 if the money for equipment has not been committed yet?


    If the foundries had14-16nm finfet plancts runing now we wouldn't see good yield production for at least a year.


    The foundries are fighting for the their very lives against a deeply entrenched and incredibly well funded enemy. My bet is with Intel as the ultimate winner. You and Danny-boy have another set of convictions.
    That's OK.
    Jul 4 11:00 AM | 4 Likes Like |Link to Comment
COMMENTS STATS
2,304 Comments
5,371 Likes