In this patent application (JP2023-166867) Canon looks to improve the performance and accuracy of a Quad Pixel autofocus capable sensor.
If you have not been following sensor developments, Canon right now splits the pixel in half and uses the difference in the signal from the one half and the other half to determine how far away the focus is. Using this information Canon can calculate the distance the lens has to move to focus on the target at that pixel. However, this method is prone to be very inaccurate if the lines of contrast that you are focusing on, are aligned to the split of the pixels, since then the signal difference would be minimal. This means that usually these sensors are better suited to focus on vertical lines, versus horizontal which they can struggle with.
There are two ways of combatting this, one method is to change the direction of the split for half of the pixels. So in other words, half the pixels are vertically aligned and the other half are horizontally aligned. The other method is to simply split each pixel into 4, by splitting each pixel horizontally and vertically. This method method is what Canon is describing in this patent application and is generally called Quad Pixel.
This patent specifically deals with the fact that many sensors are not actually square and there are issues you focus and read the pixels at the extremities and those issues are different in the horizontal and vertical directions since the distance from the center is also different. Canon is specifically identifying the sensor in question as being a full-frame sensor (36x24mm). Dealing with a larger sensor and these small pixels, light incidence, and the varying amount of cross-talk, etc can cause more of a problem. With the larger full frame 3:2 aspect ratio sized sensors, this is more of a problem than smaller and closer to 1:1 aspect ratio sensors such as micro 43s. It should also be noted that Canon specifically mentions an interchangeable lens camera – but this can obviously vary and does not mean this is the only application for such a sensor.
From the patent application, Canon notes the problem.
Most conventional image sensors do not have an aspect ratio of 1:1 . Therefore, when focus detection pixels using the image plane phase difference method in the vertical direction and focus detection pixels in the horizontal direction are arranged on the image sensor, focus detection using the image plane phase difference method is difficult to achieve, especially in the periphery of the image sensor. Performance is biased in either the vertical or horizontal direction. The image sensor described in Patent Document 1 changes the separation state of the photoelectric conversion unit depending on the saturation state of the pixel. Therefore, depending on the aspect ratio of the image sensor, the state of charge crosstalk (a phenomenon in which charge leaks to the adjacent photoelectric conversion unit) between the photoelectric conversion units in the vertical focus detection pixels and the horizontal focus detection pixels respectively occurs. It is not possible to control the above problems and cannot solve the above problems.
Furthermore, the image sensor disclosed in Patent Document 2 does not have a configuration that controls charge crosstalk when the photoelectric conversion section is not saturated and also controls charge crosstalk between the photoelectric conversion sections depending on the aspect ratio of the image sensor. Since there is no mention of the talk rate, the above problem cannot be solved.
And ultimately, what Canon hopes to resolve using this patent application;
The present invention has been made in view of the above-mentioned problems, and it is an object of the present invention to improve the performance of focus detection in focus detection using an image plane phase difference method that utilizes signals output from image sensors having different vertical and horizontal lengths
TL;DR? Make Quad Pixel AF sensors work well for autofocus at the edges of full frame sensors.
While all patent applications are just a glimpse into what Canon is researching and may not ever make it into a final product, applications such as this make me feel that Canon will indeed come out with a quad-pixel AF arrangement, sooner than later.
Japan Patent Application 2023-166867
take 45mp R5 sensor, that now becomes a 180MP sensor for AF operations. That's a hard problem. There's sensor efficiency, heat, AF performance in the outer zones, etc,etc,etc. to deal with that you just don't have to deal with as much when it's a mere 90MP.
Maybe that just means portrait vs landscape.
my no gut feeling is that i'm not sure that would cause quasi-AF less optimum zones or problems with tracking. consider if you lock focus on a vertical, then pan to a horizontal, it could report back a bad AF value and so forth, it could certainly confuse the tracking algorithm.
whereas if they are all horizontal and vertical, it would make the calculations for tracking easier.
On the other hand, multiple processors would spread heat hot spots and given the relatively low quantities used, it may be cheaper overall to use two in parallel rather than multiple types of processors. eg the volume for R1 would be very low but sharing a processor amongst a number of bodies (2 in R1) would mean having one SKU to order and keep in stock.
The recent Apple M3 series shows at least 10 variants and they have sufficient volume to warrant them but the volume of M3 max 16 CPU/40 GPU/16 neural cores with 128GB shared memory is going to be much lower than the more basic versions.
M3 => 8/16/24GB shared memory => 3 SKUs
M3 pro => 12 core (2 memory options), => 2 SKUs
M3 max => 14 core (2 memory options), 16 core (3 memory options) => 5 SKUs
Reducing lithography line width makes processors more power efficient (and more transistors per area). Canon is relatively less power efficient compared to Sony (battery life using CIPA measurements). Mirrorless was always going to draw more power than DLSR but it would be good to get through a day of shooting with the R5 on one battery (YMMV).
A step improvement in this regard would be appreciated by future users.