Thursday, June 21, 2012

Samsung on CIS Trends

EETimes published Samsung white paper "CMOS Image Sensor: Current Status and Future Perspectives" by Seok-Hee Hwang. The paper talks about image sensor trends for mobile, DSC, DSLRs and automotive applications.

Mobile sensors trend is a pixel shrink while maintaining the performance by advanced processing, such as BSI, new color filters, etc.:

Pixel and process evolution in Samsung
Sensitivity trend versus pixel size
(a) light loss ratio at each component
(b) spectral response of Samsung 6th and 7th generation CF
Camera module temperature vs power density for CIS SOC type

Not that much said about DSC trends. DSLR sensors use 130nm and 180nm process and, possibly, offer global shutter in future. Automotive sensor adoption is expected to be driven by many safety laws:

Laws and regulations affecting the in-car camera market, by TSR 2011

In the end paper talks about different automotive interfaces and importance of HDR. All in all, it's quite interesting and easy read targeted to wide audience.

18 comments:

  1. Well summarized and easy to understanding. Good job Mr. Hwang.

    ReplyDelete
  2. Sensitivity as a function of pixel size graph looks like typical marketing BS created to entice customers to buy ever higher megapixel sensors. For a given pixel design, if you increase pixel size, you should achieve higher sensitivity (more photons will be captured). This graph hides a time variable (design and process improvement) to make a marketing claim.

    ReplyDelete
  3. BSI is not enough for 0.9um generation.
    A breakthrough must be necessary.

    ReplyDelete
    Replies
    1. You have to invent photons in your pixel :)

      Delete
    2. Yeah, I am trying now. (^^)
      But, I feel so-called CMOS image sensor process is not a sollution.
      Speciality could be required.

      Somebody say standadization(CMOS) and Customization(CCD) are repeated alternately.

      CMOS (BSI) life time is too short, anyway.
      Some concept will be succeeded, though.

      Delete
  4. Hmm, wonder what happens when the pixels shrink below the wavelength of visible light in 2020...

    ReplyDelete
    Replies
    1. QIS for capture what, noise??

      Delete
    2. @ "Hmm, wonder what happens when the pixels shrink below the wavelength of visible light in 2020..."

      Solid immersion lens may help.

      Delete
    3. @Anon: Surely you don't think the photons cannot penetrate the silicon just because the pixel size is smaller than the wavelength of the light! The photons entering BSI surface don't know what the pixel size is. So, this is really more of an effective resolution issue.

      Delete
    4. well, in this case, most of QIS pixels will capture nothing just noise and consume electrical power for nothing. This is not very green approach :)

      But seriously if you can make a single-photon detection element with no static power consumption, that will really help. Do you have any idea, Eric?

      -yang ni

      Delete
    5. I think the standard CMOS APS pixel will get to deep-subelectron read noise before long (deep means <0.1 e- rms -- I just made that expression up but I like it)

      Delete
    6. Some feasibility studies on sub-wavelength regime have been done.
      However, no one has a clear picture. Literally as well.
      In particulary, business.
      Me neither. (^^)

      Delete
    7. Hi all,
      As you may all know, there are many ways to describe the sensitivity of an imaging system to light. I'm having difficulties to compare sensors when the given information is quite different. For example, in the datasheet of the sensors, it is possible to find sensitivity information as responsivity (V/lux.sec) which makes perfect sense, or in lux as minimum detectable light and in some cases just in voltage as typical, minimum or maximum. This is a big confusion for me because it is not stated how this value is calculated or what minimum, typical or maximum refer to. Also there is the saturation signal also given in mV which less than the maximum sensitivity value, I have no idea how does it work. I look at the datasheets of the Aptina MT9V032 (CMOS) and Sony ICX274AL (CCD) image sensors. They both have the same expression of the sensitivity in (mV). Does anyone know how to come out with the responsivity value from these sensitivity values in mV?
      Thank
      Cheers

      Delete
    8. Short answer: You don't. Sensor datasheets of different manufacturers are notoriously hard to compare. Sensitivity measurements based on Lux contain the spectral characteristics of light source and sensor, both of which are different in those datasheets. Going to a comparable value from there includes so much guesswork to render the result unreliable.

      Comparing a CCD directly with a CMOS sensors is in itself questionable, most CMOS sensors come with build-in ADCs and do not have an analogue output, so any measurements given in mV should be taken with a few grains of salt. The quality of the CCD image is highly dependent on the driving and ADC circuitry. Almost every digital CCD camera overclocks the sensor nowadays, so do not expect a camera to hold all the promises of the sensor. Here are some values from actual cameras, build with these sensors:
      ICX274: QE 51%; Dark Noise 8e-; Saturation capacity 9000e-
      MT9V032: QE 42%; Dark Noise 23e-; Saturation Capacity 12600e-

      JH

      Delete
  5. What is the view on scaling sensors for digital light detection, e.g., a four-color scheme. How far down do you think sensor size can be pushed if the only requirements are a) on/off discrimination with say 0.1% or better error and b)no requirements for color fidelity, just RGB discrimination.

    ReplyDelete
  6. Aren't some of the figures of the presentation copied from the paper of theuwissen "CMOS Image Sensors: State-Of-The-Art and Future Perspectives" without any citation ????

    ReplyDelete
  7. Albert TheuwissenJuly 9, 2013 at 9:20 PM

    Yes, you are right. Shamelessly, simply copied and no reference or citation ...

    ReplyDelete

All comments are moderated to avoid spam and personal attacks.