Monday, September 19, 2016

Omnivision Announces Deep Well Motion Artifact-Free HDR Pixel

PRNewswire: OmniVision announces the 1080p30 OV2775 built on 2.8um OmniBSI-2 Deep Well pixel technology, which delivers a 16-bit linear output from a single exposure with best-in-class low-light sensitivity. The OV2775 is aimed to automotive imaging applications, including next-generation ADAS, rear video mirrors, camera monitor systems (CMS), and dash cameras.

"The challenging imaging conditions in automotive applications require high dynamic range (HDR) images and fast motion capture, but unwanted effects that often accompany HDR images are not just cosmetic shortcomings — they are functional concerns," said Bahman Hadji, senior automotive marketing manager at OmniVision. "The OV2775's Deep Well™ pixel technology provides a linear 16-bit output to achieve 94dB dynamic range, compared to traditional sensors with only 12-bit linear output. This 94dB output comes without HDR combination artifacts and has no sudden drops in signal-to-noise ratio across the scene. When operating in dual-exposure mode, the sensor can further expand dynamic range to more than 120dB using a second 'very short' exposure to minimize motion artifacts."

The OV2775 comes in an AEC-Q100 Grade 2-qualified, compact 6.7 x 5.9 mm chip scale package. It contains an advanced set of features to enable ISO26262 ASIL B-rated camera systems. The sensor can be used with OmniVision's ISP companion chips (OV491 and OV495) for display-based automotive applications.

The OV2775 is currently available for sampling and is expected to enter volume production in Q2 2017.

As far as I can judge from the raw format description in the flyer, the Deep Well is a sort of combination of two 12b images, possibly with dual gain pixel or something similar:

13 comments:

  1. They can shift the pixel data 6-time left-hand, then they can get 16-bit deep well :)-

    ReplyDelete
  2. If my calculation is correct, 94dB would result in ~50ke FWC which is an awesome value for 2.8um pixel pitch (and single exposure). However, "single exposure HDR-16 bit combined RAW" sounds like something needs to be combined (e.g. two different gains, as you assumed) although OV claims linear response with no SNR drop or HDR artefacts. Would be interesting to learn more about OV Deep Well technology.

    ReplyDelete
    Replies
    1. 50K supposing 1e read noise ...

      Delete
    2. Correct... but even 25K (supposing 2e read noise which is almost realistic) is quite a good number.

      Delete
    3. If you suppose 2e- read noise you need 100Ke- FWC.

      Delete
  3. Maybe related to this Omnivision patent which describes single exposure with double, dual conversion gain, readout:
    https://www.google.com/patents/US9402039

    ReplyDelete
    Replies
    1. The problem is how to store 50Ke or more electrons inside a small PPD.

      Delete
  4. Its a split pixel design.

    ReplyDelete
    Replies
    1. I'm not sure if OV uses split pixel technology in this sensor since "single exposure" wouldn't profit of split pixel. The "dual-exposure mode" would need split pixel if the exposure of the sub-pixels must start/end simultaneously (to reduce motion blur). Otherwise staggered exposure would be sufficient (but with increased portion of motion blur) and OV's statement "second 'very short' exposure" seems to go in this direction.

      Delete
    2. Although exposure of two sub-pixels with different CG can also be considered as "single exposure" which would expain why OV talks about "combined".

      Delete
  5. Its dual conversion gain.

    ReplyDelete
    Replies
    1. Dual CG would imply dual exposure but OV is talking about "single exposure". I see only the "dual-exposure mode" which could make use of dual CG (but not necessarily).

      Delete
  6. Can also be a dual diode in every pixel like the one of ST, presented this week at AutoSens. Notice that the ADC is only 12-bit, so somewhere it needs to do something DUAL.

    ReplyDelete

All comments are moderated to avoid spam.