Friday, February 22, 2013

ISSCC 2013: Sony Stacked Sensor Presentation

Tech-On published an article and few slides from the Sony 8MP, 1.12um-pixel stacked sensor presentation at ISSCC 2013. The 90nm-processed pixel layer has only high voltage transistors and also contains row drivers and the comparator part of the column-parallel ADCs, with TSV interconnects on the periphery. The bottom logic area is processed in 65nm with LV and HV transistors, and integrates the counter portion of the column ADCs, row decoders, ISP, timing controls, etc.

Sony does not tell the exact details of its TSV processing, but shows the stacked chips cross-section:

The sensor spec slide mentions 5Ke full well - very impressive for a 1.12um pixel:

Tech-On article shows many more slides from the presentation.


  1. RGBW? So Sony are shipping RGBW cellphone sensors?

    It was just one RGBW sensor that didn't work out last year (and reverted to RGB)?

  2. Can we use these stats to interpolate other sony images sensor stats? This is the first time I remember them giving "standard" sensitivity, saturation signal,noise and conversion gain data. It would be good if we can compare them to the sony marketing stats and use that data to interpolate sony's other image sensors performance. I am not an image scientist so go easy on me.

    As said before, it is interesting to note they mentioned the white pixel sensitivity and RGBW coding after their press release late last year saying they dropped RGBW.

  3. Regarding RGB-W: the paper submission deadline was mid-September 2012. The actual paper submission might have been done earlier than that. May be by that time Sony has not make a decision to abandon RGB-W filter.

    Another possibility is that Sony keeps working on RGB-W improvements and we'll see it re-appearing in future products.

  4. Ah, good point Vladimir on the submission deadline though the announcement of the pull was mid August but still if they were being efficient they could have already submitted the paper so this paper could have been measured on prototypes that never shipped.

    I would presume that if only the CFA changed then the green sensetivity and other parameters are still the same.

  5. So, comparing the green and white sensitivity values... The white pixel only buys you half a stop of light? No wonder they ditched it! I wouldn't want to go to all the debayering mess just for half a stop of light.

    Seeing this, RGBK makes a lot more sense: take away one green pixel and put a dark grey filter instead (an ND that is 2 or 3 stops stronger than your darkest color hue) so you can get a couple of extra stops of colorless DR in the highlights.

  6. In principle you are right about the ND filter. Unfortunately in that case you throw away sensitivity and that would be a pitty. The RGB-W is the only DR extension I know of that increase the DR in the low lights, while the RGB-ND extends the DR in the highlights like many other solutions.

    1. Well, yes, but I'm guessing those other solutions involve reduced exposure times, which makes them non-ideal for video. RGBK wouldn't have that problem (and video is what I'm interested in), and the reduction in sensitivity is not that huge, I'm just doing away without one of the green pixels, the other one stays right there (again, I'm thinking video, where an oversampled sensor -more photosites than what you actually record- is relatively common).


All comments are moderated to avoid spam and personal attacks.