Toward the end of last month a new Dual-Pixel AF patent was published.
Image Sensor and Image Capture Apparatus - US 9,165,964 B2 - published Oct. 20, 2015
http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-bool.html&r=40&f=G&l=50&co1=AND&d=PG01&s1=%22Image+Sensor%22.TTL.&OS=TTL/%22Image+Sensor%22&RS=TTL/%22Image+Sensor%22
It is an image sensor design which "makes it possible to arrange focus detection pixels without generating pixels unavailable as image capture pixels, thus attaining both a high-quality image and a focus detection function of the phase-difference scheme."
My guess is that this is going to be in the Canon EOS 80D (possibly scaled up to full frame for 5D Mark-IV or 6D Mark-II)
Compared to the 2000 DPAF patent:
The new patent allows all pixels to be used for both focus detection and image capture, no interpolation is required to represent the the pixels which would have been made unavailable for image capture.
The new patent mentions that this also leads to improved image quality as there is no longer low sensitivity zones which are susceptible to light loss at particular angles of incidence. These would traditionally need the use of interpolation and exposure brightening to fix dark vertical/horizontal/cross-shaped patterns in portions of the image which are out of focus. (Cleaner bokeh. Perhaps this noisy bokeh issue is why DPAF has not been included in the 5Ds / R)
Compared to the 2008 patent:
The 2008 patent relied on an upper layer to obtain image capture signals (with short wavelength components of incident light) and a lower layer to obtain focus detection signals (with long wavelength components of incident light). Incident light was divided by differences in spectral absorbance, which depends on the thickness of the Silicon layer. An example... the resulting B:G:R components of light mix with eachother in the ratio 6:3:1. Similarly, when the 2008 design is inverted (focus detection signals in upper portion and image capture in lower portion) the problem of poor colour separation remains (e.g. G:R in a ratio of 4:6) Furthermore, with the limitations imposed by silicon thickness (needed for color separation by spectral absorbance), the depth of photoelectric conversion units could not freely be adjusted, limiting dynamic range.
The new patent does not rely on differences in spectral absorbance for phase detection as both functions will occur at the light receiving surface of the pixel. As a result I assume that a more precise colour filter array can be employed.
From what I can gather, the new sensor design could offer:
1.) Noticeably improved colour separation
2.) Slightly improved SNR. I think it also paves the way for higher resolution DPAF designs as it nullifies the inefficiencies that older DPAF implementations had regarding the light capturing surface.
3.) Lastly there is also a possibility to vary the depth of the photoelectric conversion units to increase dynamic range
What I'm unsure about...
There appears to be some mention of a global shutter type function, for synchronization of transfer gates. I assume to allow sub-pixel charges to be accumulated. (I might be wrong, perhaps someone with more of a technical background can clarify.)
Image Sensor and Image Capture Apparatus - US 9,165,964 B2 - published Oct. 20, 2015
http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-bool.html&r=40&f=G&l=50&co1=AND&d=PG01&s1=%22Image+Sensor%22.TTL.&OS=TTL/%22Image+Sensor%22&RS=TTL/%22Image+Sensor%22
It is an image sensor design which "makes it possible to arrange focus detection pixels without generating pixels unavailable as image capture pixels, thus attaining both a high-quality image and a focus detection function of the phase-difference scheme."
My guess is that this is going to be in the Canon EOS 80D (possibly scaled up to full frame for 5D Mark-IV or 6D Mark-II)
Compared to the 2000 DPAF patent:
The new patent allows all pixels to be used for both focus detection and image capture, no interpolation is required to represent the the pixels which would have been made unavailable for image capture.
The new patent mentions that this also leads to improved image quality as there is no longer low sensitivity zones which are susceptible to light loss at particular angles of incidence. These would traditionally need the use of interpolation and exposure brightening to fix dark vertical/horizontal/cross-shaped patterns in portions of the image which are out of focus. (Cleaner bokeh. Perhaps this noisy bokeh issue is why DPAF has not been included in the 5Ds / R)
Compared to the 2008 patent:
The 2008 patent relied on an upper layer to obtain image capture signals (with short wavelength components of incident light) and a lower layer to obtain focus detection signals (with long wavelength components of incident light). Incident light was divided by differences in spectral absorbance, which depends on the thickness of the Silicon layer. An example... the resulting B:G:R components of light mix with eachother in the ratio 6:3:1. Similarly, when the 2008 design is inverted (focus detection signals in upper portion and image capture in lower portion) the problem of poor colour separation remains (e.g. G:R in a ratio of 4:6) Furthermore, with the limitations imposed by silicon thickness (needed for color separation by spectral absorbance), the depth of photoelectric conversion units could not freely be adjusted, limiting dynamic range.
The new patent does not rely on differences in spectral absorbance for phase detection as both functions will occur at the light receiving surface of the pixel. As a result I assume that a more precise colour filter array can be employed.
From what I can gather, the new sensor design could offer:
1.) Noticeably improved colour separation
2.) Slightly improved SNR. I think it also paves the way for higher resolution DPAF designs as it nullifies the inefficiencies that older DPAF implementations had regarding the light capturing surface.
3.) Lastly there is also a possibility to vary the depth of the photoelectric conversion units to increase dynamic range
What I'm unsure about...
There appears to be some mention of a global shutter type function, for synchronization of transfer gates. I assume to allow sub-pixel charges to be accumulated. (I might be wrong, perhaps someone with more of a technical background can clarify.)
