Interesting things about Nikon

Discussion in 'Digital Photography' started by Chuck, Jul 8, 2004.

  1. Nikon uses Sony CCDs, Canon's use their own CMOSs.

    Sony is the biggest manufacturer of Bayer sensors in the world, while
    Canon is one of the the smallest CMOS manufacturers (though there are
    lot of "real" CMOS chip makers to contend with, like National
    Semiconductor, Intel, AMD, and Motorola). Whatever the reason,
    Canon's CMOS manufacturing prowess is dismal by current standards,
    they still use 486-era chip fabrication processes (0.35 micron) even
    in DSLR bodies selling for near $10,000 MSRP.

    Perhaps the main advantage of CMOS chips is that in theory they keep
    pace with Moores Law, and CCDs do not since they require specialized
    fabrication. Of course when a copier company like Canon makes the
    CMOS, that isn't worth much.

    Canon's old CMOS chips have given CMOS sensors a reputation for high
    noise compared to CCD chips. But that has all changed now that a real
    chip maker, National Semiconductor, has entered the market with their
    own PentiumIV-class (0.18 micron/60M transitors) Foveon Pro 10M
    digital camera sensor. Here is the result...

    http://www.pbase.com/image/31155134/original
     
    Georgette Preddy, Jul 11, 2004
    #21
    1. Advertisements

  2. Crawl back in your hole, o master of bullshit.
     
    Randall Ainsworth, Jul 11, 2004
    #22
    1. Advertisements

  3. (Georgette Preddy) wrote in
    Hi George,

    As usual - you read what you want. It does of course not
    have ISO 200 only. That it says so on those pages does not
    mean that it is true, you know. It has ISO 200-1600.


    /Roland
     
    Roland Karlsson, Jul 11, 2004
    #23
  4. Chuck

    john Guest



    I don't have a problem with the cost of the wireless remote (it's only $17),
    and it's superior if you want to use it to start the timer for a
    self-portrait or group shot).

    My problem is that it is *STILL* backordered. Awful planning!
     
    john, Jul 11, 2004
    #24
  5. If you get a programmable remote, you might
    be able to get it programmed by someone that
    have a Nikon remote.

    Or maybe - your ordinary remote just works?

    Try it!


    /Roland
     
    Roland Karlsson, Jul 11, 2004
    #25
  6. (Georgette Preddy) wrote in
    Probably true
    Totally bogus. There is no need for any super duper computing ability
    when you make sensors.
    That is a misunderstanding. Moores law is based upon the fact that the
    price per square mm of wafer is more or less constant and that we make
    smaller and smaller transistors all the time. Then the price per
    transistor decreases. Smaller transistors are also faster and
    generate less heat.

    Sensor elements cannot be decreased according to Moores law.
    Most Japanese companies have a very broad spectrum of products.
    Something that is not possible in US with your anti trust laws.
    One company (and yes - Finland is a kind of Japan) called Nokia
    (very big mobile phone maker you know) also made rubber boots
    just until recently. And Yamaha makes high class music instruments,
    motor bikes and gardening tools.
    Rightly so. They are rather good at that - low noise that is.
    National Semiconductor has entered no market at all. Read here:
    http://www.foveon.com/faq_oem.html#FAQ_oem_8. It does not say
    the NS has entered any market. It says that the South Portland
    facility (of National Semiconductors) has been very helpful. Just
    as they should be as NS are co-founder with several other
    companies.
    Totally bogus information. The Foveon chip is not a processor.
    Interesting comparison ... can you give a link to the review
    page that made this comparison?


    /Roland
     
    Roland Karlsson, Jul 11, 2004
    #26
  7. All the chip has is computing power. It doesn't do anything else.

    I think you are confused since this is not centralized computing power
    but distributed computing power. Both attributes arise from having
    powerful chips overall. You need enormous computational power to
    process then assemble 10.3M RGB values in a small fraction of a
    second.

    Another main advantage of CMOS designs that I didn't mention is
    embedded image processing ability, the sky is the limit since the
    sensor itself is a traditional processor.
    You obviously don't know that Moore's law was created by the man who
    concieved Foveon about the chips he now builds at Foveon, see...

    http://www.chipscalereview.com/archives/0104/article.php?type=dept&article=assembly_lines

    You need to review how a CMOS sensor works. The fabrication is
    traditional. Read Mead (inventor of the CMOS fabrication process,
    Moores Law, the Bayer light sensing element, digital video, and the
    Foveon sensor), his insights will straighten you out in a hurry.
    Still, a copier company can't compete with National Semiconductor when
    the goal is building high tech CMOS chips. The NSC plant that makes
    their Foveon Pro 10M is well beyond the means of a non-specialized
    company like Canon.
    Incorrent. CMOS sensors have long held a reputation for extreme noise
    compared to CCDs, because early designs did have extreme noise
    compared to CCDs. The 1Ds sensor is a prime example, noise is very
    high compared to a modern CMOS design like a Foveon. The reason the
    1Ds wasn't laughed at like it is today was because it had such large
    sensor pitch, so relative noise was still reasonable compared to
    all-smaller pitch CCD chips.

    Today, Foveon chips have larger sensor pixels than the 1Ds, and they
    use modern 0.18 micron fab with tons more on-chip processing power.
    The $8,000 1Ds sensor is a dinosaur. The analogy of a 486 compared to
    a PentiumIV actually isn't an analogy at all, it is the direct
    relationship between the two chips. Noise charts and relative price
    is all you really need to understand in order to "get it". See...

    http://www.pbase.com/image/31155134/original
    NSC owns 50% of Foveon. The other 50% is distributed among private
    investors.
    Of course it is! A CMOS is a CMOS.

    NSC's Foveon Pro 10M is significantly more powerful than a PentiumIV,
    with about 25% more transitors. Yes, it is a highly specialized
    computer processor, but so are old, slow, expensive and outdated Canon
    CMOSs.
    dpreview.com

    Phil Askey is too smart to EVER explicitly show the ancient 486-class
    $8000 Canon flasgship right next to the sleek new PentiumIV(+) class
    Sigma DSLR. Canon would financially cut him off at the kneecaps
    before sundown!
     
    Georgette Preddy, Jul 12, 2004
    #27
  8. (Georgette Preddy) wrote in
    Most of the area is used for detecting light. A large part
    of the rest is used for shifting charges. The parts of the
    chip that does the computation you are describing is located
    along the edges of the chip. If that functionality was found
    at each cell, then the computation should take microseconds
    and not a fraction of a second.
    Yes - that is a potential.
    Moore's law is invented by Moore. Bayer sensor elements is invented
    by Bayer. Mr Meade has not invented CMOS. And even if Meade has invented
    everything in the world (including the wheel and fire) was my description
    of Moore's law correct. If you read the article you see that.
    There is prbably nothing wrong with the NSC plant. And there
    is probably nothing wring with the plant Canon uses. Canon does
    not use the "copier" facility to make the CMOS chips; they
    use some profesional plant in Asia somwhere I assume.
    Utter nonsens. The Canon CMOS chips are very noise free. The reason
    for CMOS being noisy and how Canon has solved that has been described
    (by me and others) to you several tims. But you don't listen.
    I ask you again. What review is this taken from. Links please?
    A CMOS? What is "a CMOS"?
    A memory chip has lots of transistors. No one calls a memory chip
    a processor chip, except mayby you?
    That is not a link to the comparison. Phil has written tons of stuff.
    A real link to the actual comparison that shows the noise examples is
    needed.
    Then show us how this smart guy has shown this implicitely.

    And why you are at it, show us your proof that Phil really is
    bought by Canon.


    /Roland
     
    Roland Karlsson, Jul 12, 2004
    #28
  9. Chuck

    adm Guest

    A couple of other points for Preddy to chew over as well:

    1) Canon and Nikon make some of the world's most advanced semiconductor
    processing equipment. The reason Canon makes copiers is the same reason they
    make cameras - OPTICS.

    2) Although NSC are well regarded in the Semiconductor world, simply owning
    their own processing facilities does not give them any real advantage. Third
    party merchant Fab companies like TSMC and UMC (www.tsmc.com) have lots of
    capacity available across all of the main processes used today, and enable
    companies that haven't spend a few billion dollars on wafer fab plants to
    compete on an almost level playing field. Even Intel uses these guys to
    build a lot of their products.

    Again - for Preddy: A CMOS device is simply an electronic device built using
    "Complementary Metal Oxide Semiconductor" technology. This means that is
    uses both N-channel and P-Channel based transistors. That is all. A CMOS
    device could be a simple logic gate or any other type of circuit for that
    matter.
    I don't know why Preddy keeps banging on about this anyway. Most digicams
    have processors that are separate from the sensor. This makes sense as they
    are doing different jobs and should be optimised that way. In addition,
    having the processor and sensor as two separate units allows for far greater
    flexibility.

    Sensors need big pixels to keep noise down, but processors need small
    transistors to increase speed.
     
    adm, Jul 12, 2004
    #29
  10. My guess is that Mr Preddy is referring to the (very
    interesting IMHO) potential of CMOS to make more
    computational stuff near to the sensor elements.
    The Foveon sensor makes some differentiation between
    the sensors there. Foveon also talks about making
    compund readings from several sensor elements when
    decreasing the resolution, making it possible to
    increase readout speed e.g. for video camera functionality.
    Don't know if they have implemented it. You could also make
    some logarithmic computation, increasing dynamic range.

    All the above is analog computation though. The actual
    analog to digital conversion is made at the edges of the
    chip (or maybe at another chip). And analog chips do not
    follow Moore's law. It is solely digital chips, like
    processors and memory, that do follow this law.

    BTW - Moore's law is higly overrated. It is a nice observation
    and its principle is useful for economics. It is not as precise
    as some wants us to think - and it has no technical relevance.
    It is economics that decides how fast the development of
    higher and higher density shall be. And economics do believe
    in Moore's law - so it is more or less driving itself.
    Yupp


    /Roland
     
    Roland Karlsson, Jul 12, 2004
    #30
  11. Canon already does this (in video mode) on several of their higher end
    camcorders with the larger sensors, i.e. Optura 30, 40, Xi). Combined with
    the Primary Color Filter, they're achieving color that rivals the mid-range
    3CCD camcorders, without the loss of luminance inherent in a prism based
    design.
     
    Steven M. Scharf, Jul 12, 2004
    #31
  12. And in fact I've seen nothing in several years of this discussion,
    either in this newsgroup or on the Foveon web site, that suggests that
    the Foveon sensor chip does any processing whatsoever.

    Dave
     
    Dave Martindale, Jul 12, 2004
    #32
  13. Also, there are smaller fabs that specialize in digital camera sensors, such
    as Tower semiconductor in Israel (which used to be partially owned by NSC,
    though they sold their share). To be fair, one of the products from Tower is
    nothing to write home about (the sensor used by Kodak in the DCS-14/n and
    DCS-14/c), but their smaller megapixel sensors (i.e. 5 megapixel CMOS) are
    widely used.

    The process used at the South Portland Maine plant where Foveon sensors are
    manufactured is not the optimal process for digital camera sensor.

    A digital camera sensor does not require a 0.18 micron process. This process
    was good for Pentium III era processors (P4 is 0.13u and 90nm). Since a
    digital camera sensor requires larger pixels, with lower noise, it's better
    to use a larger geometry process, and to increase the area of the sensor.

    Putting the processing engine onto the same die as the sensor is
    interesting, but is not the greatest idea. Digital camera processors need
    high clock speeds and there are big disadvantages to putting them on the
    same die. You also often want to build different cameras with the same body
    and processor, but different resolution sensors.
     
    Steven M. Scharf, Jul 12, 2004
    #33
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.