Sunday, April 15, 2012

Toshiba Pixel Performance

New Toshiba catalog dated by Feb. 2012 shows an interesting pixel performance table:

10 comments:

  1. If we look at the 1.4um pixels FSI & BSI, FWCs are the same, the sensitivities are the same. But Why YSNR10 are different????

    ReplyDelete
  2. * Anonymous said:
    > But Why YSNR10 are different?

    Because of different spectral curves, perhaps. For example, they may have the same QE at 540nm (used for the sensitivity measurement) and the same QE for 5100K light. The SNR10 is taken under 3200K light, however, which requires significant gain in the blue channel in order to white balance for neutral color. Therefore, noise in the blue channel and by extension the QE at 475nm, can have a significant affect on SNR10.

    ReplyDelete
    Replies
    1. in BSI, the blue QE is smaller than QE in red. So the result should be reverse to what is stated. And I cannot see why the readout noise is smaller in BSI configuration.

      Delete
  3. They may have lowered conversion gain to use the same column circuit.
    Noise may have been lowered through backside process.

    ReplyDelete
  4. Another interesting note is that 1.4um full well and DR numbers suggest that the read noise is in range of 5-6e. This is higher than 1-2e reported by other companies. However, it does not seem to prevent Toshiba from reaching SNR10 of 60 lux.

    ReplyDelete
  5. DR seems to be defined by sensitivity instead of full well.

    Assuming conversion gain of 100uV/e, read noise may be around 1.7e.

    ReplyDelete
  6. Looks like Toshiba 1.4um pixel has better YSNR10 performance compared to Samsung by quite a bit. Just comparing numbers. Wonder if they both use the same calculation methodology..

    ReplyDelete
    Replies
    1. You are right, Toshiba 1.4um pixels perform very well.

      Delete
    2. Yes also, their 1.1um pixel YSNR10 looks to be a little too stretched..I am so far not aware of such performance from any vendors even including Sony (atleast from data publicly available at conferences <100lx is almost ruled out unless RGBC is used)...

      Delete
  7. Does not mean too much in absolute terms except for trend analysis. An SNR figure alone may not be telling unless the CCM and color accuracy are disclosed.

    ReplyDelete

All comments are moderated to avoid spam.