Nikon Z6 III Partially Stacked Sensor

Will be published 16th July. Fair old lead time. In the meantime I'm happy to expand here, as long as I don't use the same words my opinions are my own.

You can see from the photo that 'pretty fast' is nowhere near as fast as the ball is going in Jared Polin's shot that you referenced. And facts is facts. The Z6 III's electronic shutter traverses more than four times as fast as the A7r II's. The ball in your shot was simply travelling much slower (as you'd expect, given the amount of impulse likely to be imparted on the ball in each case).

Yes, I'm not sure how Canon does that. If anyone has a clue, I'm all ears.

I think you've confused the R6 II and Z6 II. Different cameras from different manufacturers.

Most modern sensors have multiple pixels, usually four, sharing a common read transistor. This allows them to be aggregated on reading if required. I suspect that the Canon sensors will be arranged so that the dual pixels can be read separately for rangefinding and together for imaging.
If pressed to guess what Canon is doing, I would guess it's electronically similar to the 'partially stacked' solution, but using a technically different method. We can run some numbers. A 24MP sensor using separate ADC chips needs 6000 connections within a 36mm distance. If this was done with normal die connection pads this provides for about 60 microns per pad. I've seen bonding pads down to 35 microns, so this is feasible. So the Canon sensor could be integrated at the package level with separate ADC chips connected by bond wire to the sensor chip. In a way it's the same solution. I have no doubt that Canon is technically capable of die-to-wafer 3d ICs, because like Nikon they produce the required alignment systems - but I suspect that Nikon would have torn down every Canon camera there is, and if any was using a 'partially stacked' sensor they wouldn't have claimed the Z6 III as the first.

As Don suggested, it's marketing hype. What matters is the results that are achieved, not the means used to achieve them. Stacked and partially stacked sensors have no value in themselves. The are a means to achieve a goal, not a goal in themselves.
If I'm right about the multiple dies connected with bonding wires, the main advantage of 3D IC tech would be economy of manufacture. That's not something that the marketing people are likely to make a big fuss about.

Most of the top-end sensors read out top and bottom - though some Canon sensors have the 'column' lines running along the rows. I found this shot of the R5 sensor on a Canon site

What's interesting is the very large number of bonding pads regularly placed top and bottom. Contrast with a similar pixel count FSI integrated ADC sensor (IMX094)

Notice two things. One that it has far fewer contacts top and bottom (and that they are irregularly positioned) and that they are connected directly to package lead-out pads. Those connections are absent in the R5, suggesting that the connect to something else inside the package. Notice also that the top and bottom borders on the IMX094 are larger than the side borders, in order to accommodate the ADC circuitry, whereas all the borders on the Canon chip are the same size (though much larger). I'm not saying it's a slam-dunk, but it does support the idea that canon is using a multi-chip package with the ADC chips integrated at the package level.

As I remember it was a diagonal distortion, which could go both ways.