Sony patents honeycomb image sensor

Pin It

Sony filed a patent for a honeycomb type image sensor. The sensor's photodiodes are shaped like hexagons (regular sensors have square photodiodes) which should improve the sensor's performance. Few years ago Fuji worked on a honeycomb type "Super CCD" sensor.

The Google translation is sufficient to get a basic idea about this patent:

  • Patent Publication No. 2010-161200
  • published 2010/07/22. Filled 2009/01/08
  • Distance of the pixel is equal with all courses in the case of a honeycomb
  • Less number of transistors used
  • Low cost
  • It can increase the number of the pixels
  • Good sensitivity
  • It assumes seven pixels one unit
This entry was posted in Sony and tagged . Bookmark the permalink. Trackbacks are closed, but you can post a comment.
  • observer

    just a little correction: those are hexagons (6 sides), not octagons (8 sides)

    • Edgar

      On the other hand, the Fuji Super CCD did use octogons.

      And, BTW, this Sony sensor uses a honeycomb lattice. The Super CCD did not.

      • observer

        No wonder it did not – you could hardly make a honeycomb pattern with octagons. Need to add squares in between. Enough geometry, though.

        • Teun

          Fujifilm used the squares and filled it with smaller pixel which was less sensitive and thus allowed for a greater dynamic range to be captured.

  • Jivee

    Could this be for a sensor that is about to be relaesed? Like in the A560/580 ect? I ask beacuse of the referencce to “good sensitivity”, and the Nikon Rumor of the D90 replacement having “noise like D700″.

    Also, what are the smaller areas 103 and 104? Fuji is doing Phase Detect Autofocus by implementing AF sensors on the ssensor itself. If the rumored D95 and these Alphas have such great AF performance….could these maybe be some sort of AF sensor sites?

    • Sky

      Sony AF good AF is due to pellix design, not incorporated AF sensors.

      And I think it’s impossible that this sensor gonna see daylight anywhere in this year. I doubt it will even in next. Patents are mostly filled long long before actual technology makes it’s place in consumer market.

    • Teun

      AF sensor need to be quite a bit bigger than the size of a pixel, so it cannot be a classic phase detection system.

  • Carlos R B

    Sorry for my ignorance, but in practice , what does it mean?

    • Teun

      That more detail can be captured with the same amount of megapixels through some smart form of interpollation.

      • Carlos R B

        Thanks…nice

  • Paolo d’Alessandro

    I am professor at the Math Dept. of Third university of Rome.
    In 2007 and then 2008 I published two papers:
    P. d’Alessandro , “On Global Geometry of Image on Eye’s Back”, LNCS 4729, BVAI 2007, Springer Verlag, Berlin, 2007;
    Revised version: “Retinal Curvature and geometry of image formation” Brain Research, August, 2008.
    where this sensor was proposed and mathematically justified. Whereas I do not pursue patents but rather free science it is a moral obligation I believe to aknowledge proper credit

    • Anonymous Coward

      Could non-academics get access to your papers without an IEEE or similar membership?

      Secondly, I appreciate that there are hard technical problems that you probably dealt with in your publications, the idea of using a honeycomb arrangement is really not such a “why didn’t I think of that?” idea. Ideas are cheap, the devil is in the detail.

  • Anonymous Coward

    I would like to know if this design would still require an anti-aliasing filter. I guess the moire effects would be quite different on this design.

    Secondly, if a normal Bayer filter has the pattern :
    RGBGRGB
    GRGBGRG
    RGBGRGB
    GRGBGRG

    What kind of pattern do they use on this? I guess I have to go read the patent.

    • Goldfires

      I’m not familiar with Sony’s patent, but during the 1980′s (yes!) I thought a honeycumb pattern would be logically to arrange like this:

      R G B R G B R G
      B R G B R G B R
      R G B R G B R G
      B R G B R G B R

      This gives each of the colours 1/3, or 33.3% of the pixels. The Bayer pattern gives 50% to the green, while red and blue get 25% each, in this pattern:

      G R G R
      B G B G
      G R G R
      B G B R

      Assuming the pixel width are the same, real honeycumb pattern (Fuji’s Super CCD is NOT honeycumb!) should give room for 15,47% moore pixels per square unit.

      Honeycumb also seems more “harmonic”. Each pixel share the same distance to all six neighbours, which all have a different colour than the one in the center.

      Among other things, diagonal lines should be easier to even out by interpolation.

      • Goldfires

        Darn!, the honeycumb RGB pattern did’n turn out as desired. I’ll give it another try:

        |R G B R G B R G
        | B R G B R G B R
        |R G B R G B R G
        | B R G B R G B R

        • Goldfires

          or another:

          |R G B R G B
          | B R G B R G
          |R G B R G B R
          | B R G B R G

  • Nathan

    Ooh… No antialiasing filter would be required, because aliasing happens because of grid arrangement…
    Hexagonal arrangement would appear to benefit greatly from rear illumination to keep the traces out of the picture, and it sure does look like it would lend itself well to the insertion of smaller photosites (trisect the center element to make 3 RGB photosites of less sensitivity for greater dynamic range during clipping, and interpolate the center photosite as an average of its surrounding elements if there isn’t much light). This could lead to a serious increase in dynamic range and an overall reduction of noise. I hope megapixel counts don’t get too high, I wouldn’t mind a 10 MP (10 million element) hexagonal arrangement on APS-C or 14MP on APS-H. It looks like it would produce natural looking images with a more film-grain like noise profile once max ISO had been reached.

    • Edgar

      Aliasing can happens with any periodic arrangement. On a honeycomb lattice it may be somewhat harder to describe, it it is still here.

      Basically, any spatial frequency outside the range accessible to the sensor can lead to aliasing. In the case of a square lattice, the accessible range is a square in Fourier space. In the case of a honeycomb lattice it is an hexagon. The hexagon is rounder than the square, which means that the spatial resolution should be less dependent on the orientation of the lines in a resolution target. Sounds like a good thing.

  • Richard

    Just eyeballing the appearance of the hexagonal site, it appears that it would be possible to fit more photodiodes of a given diameter on a sensor than would be possible with square photodiodes…which leads to the obvious question: are the corners of the photodiodes basically “wasted space” which make no particular contribution to the signal strength?

    The alternative is that a fixed number of photodiodes with a larger diameter would fit onto a sensor of a given size…presumably a larger photo site would have improved performance. The question is whether there is a “Goldilocks solution” (this one’s just right) somewhere in between which both improves performance and increases resolution?

  • http://www.echenique.com Carlos Echenique

    For the record the Bayer pattern is:
    RG
    GB

    in squares. This system would use an entirely different pattern and would have far more accurate color interpolation as it has seven color values per pixel as opposed to Bayer’s four.

  • http://www.mat.uniroma3.it/users/dalex/dalex.html Paolo d’Alessandro

    Richard, Anonymous,
    The patent presentation betrays it was inspired by my papers. I added a link on my site on Sony’s plagiarism. In the inception of theory of kissing spheres Gauss calculated that honeycomb has 90,69% density (it is the maximum possible) instead of 78,54. Then there is the crucial increase of maximum resolution directions. Libraries should provide any scientific paper. Incidentally, to prevent the next plagiarism, there are many other issues in my papers including demolishing the totally wrong idea that spatial vision is only connected to stereo vision, with here too suggestions to the industry

    • Edgar

      Hi!

      The density figures you quote assume that you are packing circular pixels. This assumption may be justified with old-style microlens arrays. However, with the new gapless microlens technology, the density is practically 100% regardless of lattice arrangement.

      The fact that they are using a honeycomb lattice does not mean they are plagiarizing you. Anyone trying to get a more isotropic (i.e. direction-independent) resolution would find honeycomb to be the obvious solution. It is reasonable to assume that Sony was simply unaware of your work, unless you gave further non trivial details on your paper, like explaining how this arrangement could reduce the number of transistors required per pixel, or the meaning of items 102 to 104 in their drawing.

      Maybe you could provide some kind of preprint of your relevant articles on your web site? The library is not so close…

      • Anonymous

        I agree, honeycomb is ubiquitous. In fact, when I first learned about Fuji’s SuperCCD I thought they used it (and was amazed to see that it was not the case).

  • Back to top