Saturday, March 14, 2015

Zoom for Curved Sensors

When Sony curved sensors were presented on summer 2014, there were concerns about their compatibility with zoom lens. Palo Alto, CA-based Optiz proposes a solution. The company believes that curved sensors will be the next big wave in mobile imaging:


Optiz does not say much about the principle of its approach, other than usual marketing words:

"Optiz's proprietary Zoom solution realizes Zoom function in the mobile imaging system without the drawback of conventional solutions."

"Optiz's proprietary pixel architecture technology includes pixel pattern design, photodiode structure and advanced imaging algorithms. With inventive thinking, we have re-imagined these three crucial aspects of the pixel technology and made great advancements in zooming, luminance and color accuracy. All doing so without creating additional complexity to the imaging system while enabling zoom function and reduce module size."

"Our innovative solution offers an optical interface structure that is built into the image sensor chip, thus imaging system need fewer lenses, simplifies the overall optical design and improves image quality."

"Optiz's Zoom solution offers a set of proprietary technology to enable greater sensitivity, lower module profile, superior zooming and dynamic focusing capabilities all without moving parts."

"The innovative pixel architecture provides significant advantages such as reduced cross talk between the pixels at the periphery of the sensor, substantial improvement of Quantum efficiency."

The company has been founded by Vage Oganesian, former Tessera VP R&D, and Wang Wei, CEO of WLCSP. Few pictures from the company site might give some hints on its technology:


Update: I received an email from Optiz saying:

"In regards to the comments on the blog post. We want to clarify that our curved sensor with varying size photodetector does not require physical bending/deformation of the image sensor. We took a conceptually different approach in obtaining the curved structure of image sensor during the fabrication process; So that our image sensor can remain stackable as most of the commercial BSI sensors of today."

16 comments:

  1. I think this has nothing to do with optical zoom. From the pictures I would judge that they have low resolution in the outer area of the sensor and in the center they have higher resolution (half pixel pitch) for high quality digital zoom.

    IMHO I do not see the big benefit compared to an array with the entire pixels in full resolution (i.e. half pixel pitch). (The benefits from bigger pixels only apply to the outer region of the image - who wants a picture where the quality of the center is worse than the out areas? It will also be interesting to generate a image without a visible difference at the border between both areas, especially during movement.)

    ReplyDelete
    Replies
    1. The benefits are there. Lets take monitor that can resolve only 2MP. A 20MP (130MP) sensors at full resolution are useless. If "zoomedin" only a small region will be shown, if "zoomedout" binning must happened. Now the level of zooming and resolution are limited by the pixel size and sensor size and number of pixels. Benefits of "vary pixel" is to maximize the zooming range. Benefits of "curved" is optics simplification towards aberration corrections. Great if there is a product like that!

      Delete
    2. You can do the downscaling of the whole area also in digital processing (for the center area you have to do it anyway), then nothing changes in respect of aberration correction, etc.

      With above solution you have the benefit of lower required bandwidth (and maybe lower power consumption), but you cannot e.g. shift the ROI ("Digital Pan and Tilt") and get the issues I tried to explain above (two areas of the sensors with different pixels, different demosaicing, etc., that should have no visible border in the image).

      Delete
  2. I can't agree with curved sensors being the next big thing after stacked sensors. Stacking has been a holy-grail of focal-plane image processing architectures for decades, with various ways to accomplish stacking explored over the years, from hybrid IRFPAs, Z-plane technology, deposited overlayers such as a-Si, organic films and quantum dot films, and now with TSVs - the latter used in mass production. I think we are just at the beginning of exploiting stacked architectures and they will have long legs. Perhaps curved stacked structures will be made in the future, but I think it would be silly to show the decline of stacked and the rise of curved sensors that this pitch slide shows,

    ReplyDelete
  3. Eric, do you think it is possible to use curved sensor and stacking technique at the same time ?

    ReplyDelete
    Replies
    1. Curved sensors usually rely on thin silicon so that they can be deformed. Stacking goes in the opposite direction. I won't say it is not possible, just a lot harder.

      Delete
    2. Hi Eric, What do you mean by "Stacking goes in the opposite direction"?

      As far as I know, both curved and conventional BSI sensors require a holding substrate. In the stacked case, the holding substrate includes circuits as well and is connected to the BSI silicon through TSVs. So, i don't see any major thickness differences. Or am I missing something here?

      Please correct me if I'm wrong...

      Delete
    3. Well, as far as I know, stacked chips are made from flat components. Curved sensors are thinned silicon on non-active-silicon (or other material) curved substrates.
      Maybe it is possible to thin a 2 or 3-layer stack and then curve it. I think it just a lot harder as I said. In any case, the subject slide shows a decline in stacked devices, and a rise in curved sensors. This is what I object to. They also show a decline in BSI, which seems silly since stacked devices all use BSI. All in all, it is just a silly pitch slide. Probably not done by an image sensor specialist.

      Delete
  4. Would it be possible to take the processing die, hollow it out on the back (inactive) side, and attach the thinned imaging die to that? Will be major trouble with alignment of the TSVs. alough this might be 'fixed' by using multiple very fine pitch contacts above the TSVs and selecting a number of 'landing pads' that are in the correct position via switches. Or use a one-dimensiona, or not-so-fine-pitch stack interconnect.

    ReplyDelete
  5. Hi, There is a way to fabricate a curved sensor without thinning and bending the silicon, see: US8932894 B2. The disclosed approach has also problems: The grey-tone lithography can lead a rough surface.

    ReplyDelete
  6. I received an email from Optiz saying

    "In regards to the comments on the blog post. We want to clarify that our curved sensor with varying size photodetector does not require physical bending/deformation of the image sensor. We took a conceptually different approach in obtaining the curved structure of image sensor during the fabrication process; So that our image sensor can remain stackable as most of the commercial BSI sensors of today."

    "Is there anything published on that (curved sensor): No, all the development work is performed under development agreement with the key customer and supply chain partners."

    ReplyDelete
    Replies
    1. Looking forward to learning about this new technique.

      In the meantime, does anyone know the radius of curvature for Sony's curved sensors? Or the delta-Z required between center and farthest edge? I always thought it needed to be more like a contact lens (dZ ~ L/2?) but I suppose that is for a large FOV.

      Anyway, guess the fabrication options are the make a flat sensor and then curve it, or make a curved surface by etching or some other technique, and then fab the sensor. The latter seems challenging unless the dZ is small. Perhaps Optiz has a 3rd option.

      Delete
  7. Eric, what I heard is that the curvature of the sensor is introduced by means of the mechanical stress present in one of the top layer, probably silicon nitride.

    ReplyDelete
    Replies
    1. Thanks, that falls in the make a flat sensor then curve it department. Any idea about the amount of curvature? BSI v. FSI?

      Delete
  8. Based on the received email (see update at end of post) it seems that they're not bending the sensor at all. Instead they are using uncorrected lenses (such as with a curved sensor) and placing photosites so that the end result corresponds to a corrected image. Think of the light rays passing through a curved sensor and hitting a non-curved image plane behind; that would imply larger photosites at the edges where ray angle and lens vignetting normally cause trouble. Quite clever if it's feasible to manufacture, though I suspect that cell phone sensors are the most likely application.

    ReplyDelete

All comments are moderated to avoid spam and personal attacks.