you can only do that with two exposures, or a much longer exposure that does a charge reset part way through (essentially 2 or 3x times). basic electronics here. a pixel fills with electrons. those electrons are then moved to the capacitor and stored. there is no longer any electrons in the pixel.
I'm just quoting information YOU SHARED, as best I remember it. If that patent described multiple wells, not multiple capacitors, I'm happy to say I may have gotten the wrong microdetail of the exact component used for storage. Guess what, the exact details of the storage weren't really the main point of my comment. The macro detail of how it works, and what the advantages are, are accurate.
I'm cutting out your insult because this isn't that kind of forum.
you need a subsequent shot to fill the other capacitor, you also need it at a different exposure time. any time you "slice" a time, you get ghosting. it's impossible to not get ghosting on a stills image no matter how quickly you take slices.
As I said, picture it switching 1ns for one sensor 999ns for the other. That's a total of one microsecond. If the hardware can switch at that rate (I don't recall as it was many months ago) you'd get 125 slices into each set of buckets even in a 1/18000 sec exposure, and the technique would normally be used for far longer exposures as one of the sensors is 10 stops underexposed. (I don't recall the exact slice time from the article, but the principle of switching between two buckets, with something like 1000:1 ratio in exposure times, was the main point and I portray it accurately here.)
I'm not sure what you mean by ghosting: motion blur not being smooth but rather being discrete? Granted! I'll bet you anything that if you use this mode to shoot something with extremely fast subject motion, it won't be quite smooth. I didn't say it
would be smooth for all possible subjects. Neither did the patent. And yet consider how fast something would have to move. At the swapping rate I gave as an example, the highlights-only half-exposure would show those dots as a connected line if the subject moved up to 125 pixels. At 1/8000, what subjects are moving more than 125 pixels? I grant there are some but not that common... and this mode will show some artifacts in this case. What percent of photos
don'ti have that rate of subject movement? 95%? 99%? 99.99%? I think I've only taken one photo in my life with substantial subject blur at top shutter speed. But the camera is full of features that can only be used in limited circumstances, right? I don't see this as an argument that you couldn't have one more feature that "only" works for 95% or 99% of your shooting.
while this isn't an issue for film at all (ghosting in film exposure is an entirely different animal which is what Canon was looking at with that white paper), it's a definite more hard issue for stills. which is why all these tricks are employed on what? cinema cameras. that discussion document that Canon did many small timeslices, but again, it would still have issues with stills photography. there's simply no way around this.
Fair enough. The rest of us will enjoy it while you sit there in your Sony T-shirt saying it's impossible :-D
Seriously, I haven't heard of a cinema camera using this kind of setup either,
but that is the point here. This is something new.
also if you have two capacitors per pixel and a corresponding light shield over your capacitors your overall quantum efficiency is quite low (literally 1/3).
If the storage buckets for the pixels are half normal capacity, though, then perhaps they individually have half the area, so two of them fit the same area. Note my recent post showing the R5 DR at 11.85 stops. Cutting area in half will trim that by a stop or so. Leave some room for slop and extra wiring and it could be that one set of these buckets only has 10 stops by itself. But then make an equal exposure between the two sets of buckets and you're back up to 11 even without using the built-in HDR-style unequal exposures.
Again, canon has no BSI or stacked tech in production, so there's only so much real estate on a FSI sensor. So in instances where you have to turn this "off" your DR drops down to pre 2008 era sensors.
Again,
this is something new. By definition, they are not doing this new thing yet. It is something that is about to start.
Down to 8 stops? Half the current area, all things being equal, would be 1 stop less, not like 3. And the paper's unequal-split exposure method, to get 20 stops, would need 10 stops (or more) from the buckets individually, and underexpose the second set of buckets by 10 stops.
I mean, if you don't think that paper you shared last year was serious or possible... why did you share it? Now that there's a press release of a sensor with the exact capabilities we'd expect from that paper, why do say it can't possibly be what was in the paper?
How do YOU think they're jumping from an absolute firm best of 11.85 to 20, if you don't think their published method to get another 10 stops is what they're doing?