Thursday, April 08, 2010

Samsung 1.75um Sensors Feature EDoF, 70Lux SNR10

Business Wire: Samsung announced two new image sensors – 1.3MP S5K6A1 and 2.1MP S5K5B3 - for use in notebook computers and other devices requiring video capture. Using 1.75um pixels, the new imagers are said to have significantly improved low light sensitivity of 70Lux of YSNR10.

The 2.1MP S5K5B3 has an embedded EDoF to allow bar codes and business card reading without needing mechanical AF support. In February Tessera reported that Samsung is its EDoF licensee. S5K5B3 sensor supports camera modules with full 1080p HD at 30fps capability in a z-height less than 4.5mm. S5K6A1 sensor has been designed for camera modules with 720p HD at 30fps capability in a z-height less than 3.5mm.

Samples for both of these image sensors are currently available with mass production scheduled for Q2 2010.

According to TSR, the notebook and PC camera market is expected to reach 150M units in 2010 and grow at a compound annual growth rate of 12% to reach 200M units by 2013. The camera attachment rate of over 70% now, according to Dojun Rhee, VP of System LSI marketing, Samsung Electronics.


  1. Anyone can update me what the frame rate is used for SNR10 test ??

    Thanks in advance !

  2. If Samsung used Nokia standard test, it should be 15fps.

  3. How would you convert the light sensitivity spec to mv/Lux-Sec?

  4. I afraid there is no way to convert it. Potentially, mV/Lux*s could be estimated if we know noise in the darkness (and make a couple of rough guesses). However, at SNR10 illumination the noise is normally dominated by photon shot noise.

  5. If you go to the Samsung press release, the front package photographs appear to show a front mounted carrier glass, while the back package photographs appear to show TSV's poking out the back (along the perimeter). Does anyone have any thoughts on Samsung WLP? It would certainly explain the low Z-height.

  6. if the SNR10 is photon shot noise limited, the pixel should get 100 electrons. Suppose that the QE=100% and FF=100%, you should have 0.1Lux faceplate at 15fps.

    If you use a F2.8 lens, the scene illumination level should be 4*2.8^2 more, that is 3lux.

    The illumination level for SNR10 is measured at 71lux. So for me, it's not really limited by the photon shot noise.

  7. You might want to check your assumptions and calculations again. For obvious reasons I cannot lead you thru the calculation, but when you get it all figured out, assuming that the sensor is in fact, photoelectron shot noise limited, you will learn a lot.

    Either that, or the rest of us have all missed out on making 3 lux YSNR10 sensors with 1.75 um pixels.

    Do you work for Invisage?

  8. Dear Eric,
    please give us your calculation !
    By the way, I don't work for Invisage :)

  9. Since I consult regularly for Samsung Semiconductor R&D Center (image sensor R&D - not the image sensor products group which is in another division) I cannot say much more. But, the product team is pretty strong these days and they know what they are doing.

  10. I should add that the image sensor R&D we do is sort of next-next(-next?) generation stuff. The product guys of course have their own R&D for stuff closer in on the roadmap.

  11. Eric,

    We don't want you to tell the Samsung's secret ! We would like that you give us your calculation of how many electrons a 1.75um photodiode can generate under 70Lux at 1/15s, that's all !

  12. What parameters do you want to use?
    Illuminant? (A? 3200K BB? D65? etc.)
    Scene reflectance? (18%?)
    lens transmission
    f# (2.8?)
    sensor reflectance?
    effective fill factor?
    CFA transmission spectra?
    QE (photodiode depth)
    collection efficiency?
    integration time? (1/15s ?)

    Maybe there are some other parameters you need...I dont recall right now as I am at an airport. But, you get the point, and I think anon forgot most of these in the 3 lux YSNR10 sensor estimate.

    I think you should ask Albert Theuwissen to write a little blog entry on this calculation if you don't already know how to do this.

    If I get some time some time maybe I will post a spreaddsheet calculator on the website. But, until the Fall, that is unlikely!

    Good luck on your own model. Let us know what you get for a practical 1.75 um pixel.

  13. @ if the SNR10 is photon shot noise limited, the pixel should get 100 electrons.

    Not correct. SNR10 is commonly calculated after color correction matrix (CCM). Depending on the color crosstalk and color response shape, CCM can degrade SNR by a factor of 1.5-2.5. So, photon noise limited pixel should get 150-250e to provide SNR of 10 after CCM.

    With F2.8 lens 1.75um pixel having 100% QE and IR-cut filter very roughly can generate about 170-180e/Lux-s for 20% gray level patch at the target (depends on illumination type). Color filter leaves 1/3 of this number - 55-60e/Lux-s.
    Then, 1/15s exposure time leaves only 4e/Lux.

    So, 70Lux gives 280e of signal for this ideal pixel. And we need 150-250e to get SNR of 10 after CCM, assuming noiseless sensor.

    Note that Tessera mentions that Samsung licenses its UFL solution. UFL allows to improve the lens F# to 2.0:

    This adds some more inaccuracy to our simple calculations.

  14. @ CCM can degrade SNR by a factor of 1.5-2.5

    I meant:

    CCM can degrade SNR10 figure by a factor of 1.5-2.5

  15. Oh yes I forget that this is a color sensor :(
    so my calculation was for B/W and with 100% gray patch ...

    20% * 1/3 (B/W) * 1/1.5-2.5 * 70 lux = 3 lux

    So we are not so far ... !


All comments are moderated to avoid spam and personal attacks.