FX sensors cost 20x DX sensors? More info?

Discussion in 'Nikon' started by ana_stasi|1, May 10, 2012.

  1. I read the following on Wikipedia:
    Production costs for a full-frame sensor can exceed twenty times the costs for an APS-C sensor.[citation needed]​
    Does anyone know where I can find more detailed information on the production of camera sensors and why the above is the case?
     
  2. Search photo.net. This was discussed (more than) a couple of years ago.
     
  3. Such sensors frequently have flaws (a bad pixel or two or a hundred) as they are manufactured. It's unavoidable. They're made on large wafers, multiple chips/sensors per wafer. The odds of a given small sensor having such a flaw are much lower than the odds of a physically larger section of the parent wafer having a similar flaw.

    So, you get a higher failure rate with the larger sensors during manufacturing. Combine that with the fact that smaller sensors sell many, many times more frequently than larger ones, you have the economy of scale that goes into producing millions more of the small ones. This contributes to the lower costs on the smaller sensors (or, a smaller customer base for the large sensors contributes to the higher costs, depending on how you want to look at it).
     
  4. Yield is the name of the game. The higher the yield, or % of good parts, means lower unit prices given the same production quantities. The same applies to ALL semiconductor devices.
    Try google. :)
     
  5. Another cost parameter is the wafer usage. Wafers have fixed dimensions (circular). The smaller the chip is, the more efficient real estate usage is possible. When producing 36x24mm chips, there's a lot of unused space on the wafers.
     
  6. Ana, here's a video description of how CCD sensors are made by Teledyne Dalsa.
    http://www.youtube.com/watch?v=bqJksXwrx7U
    A factory built to produce sensors can cost $500M or more, and along with its R&D costs, must be recovered through sales.
    http://news.cnet.com/8301-10784_3-9747468-7.html
    Full-frame sensors currently lag APS-C sensors in sales by a wide margin which in part accounts for the price-ratio difference. Another contributing factor is yield - more complex sensors of larger wafer size might have a lower yield during manufacture due to defect-rejection. The manufacturing process also takes longer than smaller sensors.
    There are other contributing factors not within the normal course of business such as Sony's sensor factory flooding:
    http://image-sensors-world.blogspot.ca/2011/10/sony-thailand-ccd-plant-flooded.html
    This is not an exhaustive list but is generally accurate; others may input corrections or additions.
     
  7. So are medium format sensors, such as the one for the Pentax 645, another cost leap ahead? More expensive yes, but not such a huge leap I think?
    There must be a huge 'failure' rate for such large chips and a real low number of sales relative to FX. DX or CX.
     
  8. Let me try to put the failure rate in perspective.
    Semiconductor yields are based on average defects per unit of area (square inch or whatever).
    Suppose your process has an average defect rate of one per square inch.
    If your sensors are 1/10 square inch in size, you can expect nine good sensors out of every ten (90%).
    If your sensors are 1/2 square inch, you can expect one good sensor out of every two (50%).
    If the sensor is a full square inch in size, you can expect zero good sensors (0%).
    Another point to consider is the quality level required for different sensor sizes.
    Consumer-grade (small) sensors can tolerate a few bad pixels because consumers don't care.
    Professional-grade (large) sensors have much tighter quality requirements, generally rejecting any sensor
    with even one single defect. This contributes to lower usable yield.
    It all boils down to economics. It costs the same to process one wafer regardless of what's on it.
    - Leigh
     
  9. If you have an average defect rate of one per square inch, there will still be some square inches which are defect free (and some with multiple defects). If the number of defects per area follows a Poisson distribution with expectation = 1 per square inch, then the probability of a given square inch to be defect free is about 37%. So 37% yield in this case. (Disclaimer: it's bedtime and my stats are rusty.)
    Also, since dead/hot pixels can be mapped in software I doubt the sensor has to be completely defect free when it is manufactured.
     
  10. Semiconductor manufacturering operations are based on statistical methods.
    When you're making parts by the hundreds of thousands, the time required to test each becomes
    prohibitively expensive if the yield drops below a certain threshold, perhaps 30% or 20% or whatever.
    There will always be some good parts even from a bad process, but finding them is not cost effective.
    As I mentioned before, the definition of "defect" varies with the size of the sensor. A minor problem that
    would pass inspection for a consumer-grade sensor might be considered a hard failure at pro-grade. So
    the defect rate increases as the sensor size increases.
    You can mask a bad pixel in software, but you cannot replace the information that was lost due to the error.
    - Leigh
     
  11. slight math issue - if you have 1 defect per sq inch - then you can product 100 x 1/10 inch sensors, with a 99% yield
     
  12. No.
    I said the sensors were 1/10 square inch, not 1/10 inch square (= 1/100 square inch).
    - Leigh
     
  13. ShunCheung

    ShunCheung Moderator

    Clearly this is a rapidly changing field so that any price figure can get out of date quickly, but Thom Hogan has been suggesting for a few years that roughly speaking, the cost for a DX sensor is about $50 while for FX is about $500, a 10x difference. Regardless of whether it is actually 10x or 20x or somewhere around there, there is little dobut that FX costs a lot more to manufacture. And it is typical that final retail price is something like 3x to 5x of manufacturing cost (with plenty of exceptions).
    Therefore, unless a camera company is willing to take a loss on a camera model (and some people believe that Sony did with the Alpha A900 and A850, but we have essentially no way to verify), "affordable" new FX digital cameras is not going to be reality for a (long?) while. If you want an FX DSLR and don't want to pay much, the best alternative is to go after older models that are out of favor in the used market.
     
  14. Suppose your process has an average defect rate of one per square inch.
    If your sensors are 1/10 square inch in size, you can expect nine good sensors out of every ten (90%).
    If your sensors are 1/2 square inch, you can expect one good sensor out of every two (50%).
    If the sensor is a full square inch in size, you can expect zero good sensors (0%).​
    Er, no. If you have an average defect rate of one per square inch, you'd expect 50% of full-inch sensors to be good. Look up poisson distribution.

    It's true that, if 90% of 1/10 square inch sensors are good, only 81% (90% of 90%) of 1/5 square inch (2 x 1/10 square inch) sensors are good. A 1/2 square inch sensor is equivalent to five 1/10 square inch sensors, and for all of them to be good, that's .9 x .9 x .9 x .9 x .9 or ~59% chance of being good. A one inch sensor has about a 35% chance of being good (90%10) by the same logic.

    An FX sensor has 2.25 x the area of a DX sensor. One might expect the failure rate to be higher by a power of 2.25. I suspect there are more factors involved that make them more expensive (for a start, you only get 1/2.25 the amount out of a wafer, so they're that much more expensive even if they all work...) There was a time when Canon had to make a special stepper in order to produce a full-frame sensor; I suspect those days are past.

    Even so, 20x sounds like a lot, and I suspect economy of scale is the vastly dominant factor. That would explain the medium format step-up not being equivalently huge. Although, if there's a D600, I'll be interested to see what effect it might have. It'll probably mostly make FX lenses cheaper...
     

Share This Page