Has Canon developed a new 21mp sensor for the Canon EOS R system? [CR1]

There are fundamental core reasons why stacked is better for global shutter - it's not just a "nice to have", the fact is that unless you either go BSI, or you go stacked because you only have so much silicone space. On a normal sensor for efficiency, you need all that space taken up for photodiode wells. Anything that gets in the way of that will cost you stills dynamic range performance. That's just well, physics. Sony's done several different novel solutions (and even Canon has come up with a few) but all of them are based upon stacked designs. Usually three-layer designs so you can incorporate memory. Canon's done more R&D on global shutter stacked designs than global shutter non stacked designs. To get around using stacked you have to use novel solutions such as time slicing. but it's imperfect for both video and stills. you can get around it in video by making the slices smaller and smaller, but that's still not a good solution for stills.

To get too stacked sensors, Canon first has to develop a BSI sensor. They haven't even gotten that far yet. So no, when looking a "past development trends" there's really nothing out that that suggests Canon is about to shove a global shutter sensor into what is a stills camera first and foremost.

I've seen articles that mention the biggest deterrent to this tech (in 2017) was the inherent loss of DR, but nothing about Son'ys stacked sensor design. I imagine the stack sensor may be inherently better, but I haven't read anything that suggested it or similar tech was needed. I don't understand the tech well enough to know whether the stacked sensor poses any hindrances based on the architecture their global sensor, but I don't think that's a stretch either.

From the article, GLOBAL SHUTTER IS COMING TO CANON DSLRS ...EVENTUALLY, Written by George Shaw

From Canon, "...However, its dynamic range is narrow. To solve this problem, Canon made two major improvements. " The improvements Shaw refers to, "...Canon has improved this issue by doubling the saturated electrical charge coming off the photodiodes and then reading the data twice during processing. The result is that the sensor can read up to 16,200 electrons at 60 frames per second, while the original sensor data is 8100 at 120fps ".

Canon has also added a so-called “light guide structure,” which reflects more light onto the photodiodes in order to read the maximum amount of image data off the sensor. In order to accomplish this feat and not fall behind like their rolling shutter cousins, the new global sensor uses Canon's A-D (analog-digital) converter called "SSDG-ADC" to keep up. Both these improvements provide greater dynamic range, with less noise and under .45w of power consumption, easily better than a CMOS sensor reading the same image using rolling shutter.

Again, this is not a theoretical sensor, they have current distributors in place for the industrial CMOS sales now. This is tech that is well down the road with a little maturation. Unfortunately, I can't find the white paper Canon published on it a few years back, but I don't think that is necessary to prove the point. Canon has been busy.
 
Upvote 0
GS sensor will come but it'll be expensive. so no surprises but GS sensor in R line up is questionable? people are complaining R5 is expensive so how is Canon going to price this. it most likely will end up in speciality product and high end cine line where they can ask for high price. I can see them putting this in R when competition like Sony start doing it and use GS sensor in their A9 series. Sony started doing this by using stacked sensor in A9 series and ask for a premium price. I can't see Canon using this in their 1DX equivalent R. they'll have to maintain a certain price point for this series. Canon might create a new series for R.
 
Upvote 0

koenkooi

CR Pro
Feb 25, 2015
3,569
4,109
The Netherlands
[..] The microprocessor company ARM for instance subcontracts all of its manufacture out. [..]

I have to step in here, Arm Ltd. sells designs and architecture licenses, they don't sell actual silicon. So Arm doesn't have any manufacturing to subcontract out. However, a lot of Arm licensees do use fab-less design, so they have e.g. TSMC produce the design they have licensed from Arm.
 
  • Like
Reactions: 2 users
Upvote 0
Sep 20, 2020
3,065
2,395
120fps? How long does it take to fill up a 2TB CFexpress card?
I do not hear photographers complaining as much about the cost of media as I do from videographers.
My theory is that film was expensive while videotape was pretty affordable.
Film is also one and done where CFExpress cards can be used over and over.
 
Upvote 0
Sep 20, 2020
3,065
2,395
Camera MP are a lot like CPU Ghz used to be in the late 90's. Intel and AMD used to chase each other to the next 0.1 Ghz speed increase and brag loudly each increment. If your computer was more than 0.5 Ghz behind you felt like you had a clunker. But they hit 4Ghz around 2005 and since then haven't gone anywhere. Current models are all in the 3 to 4 Ghz range 15 years later.
20.x MP is the 3.5Ghz of camera tech.
Now that has been replaced with nm.
People keep comparing Intel 10 nm, AMD 7nm, and ARM 5nm even though their architectures are so much different.
 
Upvote 0

addola

Sold my soul for a flippy screen
Nov 16, 2015
155
148
Global shutter on a stills camera sensor? Seems fake.

Global shutter will avoid banding when using flash, and allows much higher shutter speeds with flash.

In 2017, Canon published a research paper on that 2/3" sensor. It was published in an IEEE Journal, and titled "A 1.8e-rms Temporal Noise Over 110-dB-Dynamic Range 3.4 μm Pixel Pitch Global-Shutter CMOS Image Sensor With Dual-Gain Amplifiers SS-ADC, Light Guide Structure, and Multiple-Accumulation Shutter".

The rumors is very believable, but whether it is practical or ready for everyday use in a consumer product is another story.
 
  • Like
Reactions: 2 users
Upvote 0

Traveler

EOS R6
Oct 6, 2019
158
201

Did you notice that the older post about the 2/3” sensor mentioned two numbers of the DR?
HDR of 111dB (18 stops – quite high number, too), and
NDR of 79dB which are more realistic 13 stops.
Would it mean that the “real” DR of this new sensor would be also two stops higher, something around 15 stops?
 
Upvote 0

addola

Sold my soul for a flippy screen
Nov 16, 2015
155
148
Did you notice that the older post about the 2/3” sensor mentioned two numbers of the DR?
HDR of 111dB (18 stops – quite high number, too), and
NDR of 79dB which are more realistic 13 stops.
Would it mean that the “real” DR of this new sensor would be also two stops higher, something around 15 stops?
I am a grad student, so I have access to the IEEE document for the 2/3" sensor. I can't share it for copyright concerns, but It seems that there are two modes: Multiple-Accumulation procedure and HDR procedure. The paper says: " As a result, the dynamic range increases from about 79 dB in 60 fps with the multiple-accumulation procedure to 111 dB in 60 fps for the HDR procedure. ".

I don't know what they mean by "multiple accumulation readout procedure", but it does sounds something similar to "stacked sensor"?

Who knows, but it will be huge if Canon put something like that into their product line.
 
Upvote 0

Traveler

EOS R6
Oct 6, 2019
158
201
I am a grad student,...
...The paper says: " As a result, the dynamic range increases from about 79 dB in 60 fps with the multiple-accumulation procedure to 111 dB in 60 fps for the HDR procedure. ".
That’s what I read in the screenshots, too. But does it mean that the HDR mode is something that doesn’t have any disadvantages or can it be something as unusable as multiple shots?
 
Upvote 0
Sep 20, 2020
3,065
2,395
I am a grad student, so I have access to the IEEE document for the 2/3" sensor. I can't share it for copyright concerns, but It seems that there are two modes: Multiple-Accumulation procedure and HDR procedure. The paper says: " As a result, the dynamic range increases from about 79 dB in 60 fps with the multiple-accumulation procedure to 111 dB in 60 fps for the HDR procedure. ".

I don't know what they mean by "multiple accumulation readout procedure", but it does sounds something similar to "stacked sensor"?

Who knows, but it will be huge if Canon put something like that into their product line.
That does not sound like a stacked sensor.
HDR reads the sensor multiple times for each image frame.
My guess is that adding the signals from each readout together boosts the decibel amount.
I believe that is how it works with recorded sound so maybe recorded light works the same way.
 
  • Like
Reactions: 1 user
Upvote 0
Apr 25, 2011
2,509
1,884
Global shutter will avoid banding when using flash, and allows much higher shutter speeds with flash.

In 2017, Canon published a research paper on that 2/3" sensor. It was published in an IEEE Journal, and titled "A 1.8e-rms Temporal Noise Over 110-dB-Dynamic Range 3.4 μm Pixel Pitch Global-Shutter CMOS Image Sensor With Dual-Gain Amplifiers SS-ADC, Light Guide Structure, and Multiple-Accumulation Shutter".

The rumors is very believable, but whether it is practical or ready for everyday use in a consumer product is another story.
You won't get such a DR with such a pixel pitch on a stills sensor (without multiple exposure artifacts), global shutter or not. Physically impossible with the current technology.

On a video sensor, interleaving the ISOs of the adjacent video frames, it's possible. But no "shutter speeds with flash", of course.
 
  • Like
Reactions: 1 user
Upvote 0
Sep 20, 2020
3,065
2,395
You won't get such a DR with such a pixel pitch on a stills sensor (without multiple exposure artifacts), global shutter or not. Physically impossible with the current technology.

On a video sensor, interleaving the ISOs of the adjacent video frames, it's possible. But no "shutter speeds with flash", of course.
A global shutter would not eliminate HDR artifacts but it should greatly reduce them.
Combining entire frames in less challenging than combining individual lines.
HDR has more artifacts but it also has more detail.
Life is full of tradeoffs.
 
Upvote 0
Its based on a current (for sale) 2/3's, 5.5MP, 3.4um pixel, industrial high speed use (120fps), global shutter sensor. The pixel and sensor size suggest they could use this tech on MUCH higher consumer MP count applications.

I feel like this is more of a when than an if. Will it debut in the R1, or will it be pushed further down the road to the next high end body? The fact that Sony is going to be close behind, as these things tend to all get "found out" at the same time as accompanying tech becomes available to clear past hurdles tells me we'll see it sooner than later.

---

I actually REMEMBER using 1 inch Plumbicon/Saticon Tube cameras that weighed 50 pounds (20 kg+) and more! We did 800 lines up to 1200 lines of vertical resolution. Quality-wise, if it wasn't for the blooming/streaks (i.e. blobs and streaks of white showing up near bright highlights), the quality of the tube cameras was actually BETTER than modern HDTV 1080p CMOS or CCD camera resolutions!

There are companies that STILL sell Saticon tubes (for nightvision use) that go as a high as 12000 lines of vertical resolution (i.e. a 3 inch Saticon-style tube!) used for BOTH Optical Daytime AND night-oriented photo-multiplier purposes so the image quality IS OUTSTANDING since the formulations of the phosphors is now so advanced that blooming and streaking is now so low that it is basically non-existent and can be NOW FULLY REMOVED using modern software/hardware DSP algorithms.

Not many CMOS imagers have 12,000 vertical lines of resolution so in certain cases tube-based image sensors are used because they work VERY WELL in specific environmental conditions such as Space or NO Light Areas! Horizontally, these newest tubes on the X-axis, return about 16000 pixels if sampled properly -- The Y axis is easier to sample than the X axis on a tube-based image sensor because of basic analogue circuitry issues due to analogue signal jitter, slew, rise-time problems! AND in some cases, the blooming and streaking of tube-based cameras IS a desired effect for capturing certain natural phenomenon better than any CMOS or CCD imaging system.

I have a few Plumbicon tubes from a 3/4 Inch camera system which I fire up once in a while! Still working GREAT since they were made in 1977!

V
 
  • Like
Reactions: 1 user
Upvote 0
I think people put too much stock in the stacked sensor design. Canon proved there are other ways to skin that cat, even if they are not as efficient. As far as guessing Canon might bring a tech they've been selling for years, I don't think it was ever a big stretch to see this coming. I personally threw it on my wish list for the R1 months ago when the expected drop was 2022-23. Same as the uber mega pixel 5Ds style sensor and camera pending. Some of this stuff is very obvious when you look at the past development trends.

---

There IS another image sensor technology out there beyond CMOS, CCD or even Tubes! It's on the patent and trade secret books of a certain series of firms here in Canada, USA, Europe and Japan and you COULD kinda call it a Double-Stacked sensor since it has BOTH onboard DSP/CPU circuitry AND the individual Red Green and Blue sensitive substrates NO LONGER EXIST like on CMOS imaging systems, but are rather MERGED into a single photosite multi-photon stream frequency-sensing mechanism. I would say it's WAAAAAY BEYOND Sigma's Foveon technology and has the advantage of NOT requiring the very-hard-to-manufacture multiple photosensitive substrates stacked one-on-top-of-each-other like Sigma Foveon needs!

That new technology is being perfected now by multiple boutique companies and SHOULD be on the market within five years. Because of the TYPE of electrical micro-circuitry it is, it's BASE DSP sample size is almost always set to 64-bits wide per colour channel and usually set to output a 32 bit Nyquist re-sampled bitwise value per RGB channel which can then can be truncated/rounded down to 16-bits, 14-bits 12-bits, 10-bits or 8-bits per colour channel with minimal processing time.

Soooooo, Sony, Philips, Canon, Teledyne-Dalsa, ON, NEC, etc are NOT the only games in town for high resolution image sensing technology -- There are some BOUTIQUE image sensor design and manufacturing houses that build VERY HIGH END imaging solutions that go WAAAAY beyond consumer technology!

Many in that circle ARE indicating that consumer-level applications WILL be entertained because the manufacturing costs can be NOW lowered enough to make it viable for this select group to apply these NEW imaging technologies to the broader consumer market segments. It also helps that some of the NEWER image sensing technologies have MUCH higher resolutions AND MUCH higher dynamic range than contemporary CMOS camera sensors so the BASE advantage makes it an obvious choice to start impinging upon the BIG BOYS of CMOS imaging (i.e. Sony, Canon, etc!).

V
 
  • Like
Reactions: 1 user
Upvote 0
Agreed, of course. I agree Arm's not a B2C company like Canon (or at least, Canon's EF and RF lines of cameras). Still, B2B, B2C, that's not really the core of my point. My point is that ARM designs stuff they can't manufacture. Apple designs stuff they can't manufacture. My personal work history involves Sony, Fujitsu, Toshiba, Hitachi, all MAKERS of PALs, nonetheless designing quite a few PALs they didn't have the capability to make and thus outsourcing their manufacture to Motorola via my old firm Omron.

In fact you'd probably have a hard time finding a firm the size of Canon that DOES manufacture literally everything they design.

And yet while Canon's described a certain sensor in detail in papers, and we now have leaks of the coming manufacture of something fitting the papers' description... nonetheless the website owner is telling us that because he personally doesn't know that Canon has facilities to manufacture this part, therefore it is absolutely utterly impossible that the part is the part described in the paper. And insulting me for even suggesting it.

---

Since we actually DO HAVE CMOS and GaAs substrate manufacturing capability AND that the WORLD'S LARGEST ONE-PIECE CMOS SENSOR at 131,072 by 131,072 pixels (64-bit RGBA colour) on the world's LARGEST single-slab of etched 400 mm silicon was done RIGHT HERE in Vancouver, Canada, I think I can probably make a valid comment that Omron, Canon, Philips or ON Semiconductor DO HAVE the capability to manufacture whatever TYPES of image sensors they want AT THE SAME LEVEL SONY CAN if they put their minds and dollars to the job!

Anyways, Canon has done a 440 megapixel chip the size of a paper sheet for Space/Satellite purposes and I SAW THAT ONE PERSONALLY! Canon also ALREADY HAS 120 megapixel APS-H size sensors and has more than a few 50/60 megapixel and 80 megapixel sensors so they can do almost everything at scale because they are such a large size firm (Market Cap of 18 Billion USD with revenue of 37 Billion USD!), while we have barely hit above a billion dollars only in recent years and make maybe ten of those super-big 400mm sensors per year!

Canon and others can MATCH Sony's CMOS imaging prowess but they NEED to be hungry for market share. Canon is definitely getting BETTER but I REALLY THINK they need to start concentrating on LARGE SENSOR smartphones with 2/3rds inch, 1 inch and APS-C DCI 4K and DCI 8K resolution image sensors for stills AND video. THAT will save their bacon from Sony's onslaught of upcoming high resolution CMOS image sensors which WILL include new large 50 megapixel+ sensors for upcoming super-smartphones !!!

We shall see soon enough!

V
 
  • Like
Reactions: 1 user
Upvote 0

magarity

CR Pro
Feb 14, 2017
283
193
righto.. so the 45Mm R5 would have to make it with 5 frames per second limit or some such according to your logic? it's a 20 fps capable camera.
again, you are incorrect, 20 MP is not a tech limitation but rather a designation at this stage.
I thought it obvious I was making up numbers - the point being that increasing frame rate requires increasing processing from the chips supporting the sensor. Is that incorrect?
 
Upvote 0