1DX2: Dynamic Range

[GuyF]

At the risk of starting a thread that will possibly degenerate into name calling and acknowledgement that DxO are the masters of measurement ( ), I'll expand here on a recent post I made regarding an article in UK magazine, Amateur Photographer.

Written by Bob Newman, a Prof. of Computer Science at the University of Wolverhampton, he talks about the changes in approach taken by Canon and Nikon to their analogue-to-digital converters (ADCs).

I'll paraphrase the salient points:

Previously, Canon's approach resulted in lots of residual noise when using low ISOs compared to Nikon. The reason being that Canon's ADCs were not on the same chip as the sensor. With a limited number of communications channels between them (usually 16 in the top-end bodies), the camera had to process the data very quickly before moving on to the next frame.

There is a trade-off between speed and noise. The read noise would lurk in the shadows and when boosted, would offend the eye (my words, not Bob's).

Nikon had taken the route of placing the ADC on the same chip as the sensor which allowed them to have more of them; 24 on the D4 and thousands in other cameras which used a column-parallel arrangement allowing the ADC(s) to be built into the sensor array. This allows for slower processing and lower read noise.

At high ISO, the implimentation of the ADC is less critical as the boosted signal is much greater than the electrical noise inherent within the circuitry. Therefore Canon's high ISO performance wasn't held back but low ISO had limits to the dynamic range.

Now with the 1DX2, Canon have gone the route of using these column ADCs to get improved low ISO DR whereas Nikon have gone the opposite way to give the user lower DR at low ISO but things improve once you get to high ISO.

So there we have it. The 1DX2 kicks ass at low ISOs but the D5 will take the DR lead at high ISO. Recent examples on the web show Canon's high ISOs are quite smudgie compared to the D5.

Now over to DxO for their take

 

[J.R.]

Had the opportunity to check out the D5 recently and have started shooting with my 1DX2 as well. I can say that both are excellent cameras.

If one can't get good enough shots using either of these two cameras and has gripes that the other one is better, he needs a check-up from the neck-up.

 

[9VIII]

Yesterday I spent a few minutes looking at the D500 tests on DPR.
It's interesting that the D500 seems to have accomplished something even the D5 couldn't, it looks good at high ISO (better than any other crop sensor, barring Fuji's Lack of moire), but also has excellent Dynamic Range. So at a glance it seems like the D500 doesn't make any sacrifices on either end (it's also funny how Sony themselves seem to manage to do so much worse than Nikon when they make all the sensors).
And at the same time, all these gains also still seem a bit underwhelming compared to Full Frame. The 6D is obviously the superior camera at high ISO, and even the 5D2 still looks better under many circumstances, you just can't go over ISO 3200. Within the ISO range that you do have, older bodies don't actually look like much of a compromise.
And that goes double for any of the more recent low end Nikon bodies. As far as landscape and portrait use is concerned the D500 isn't much different from a D3300.

 

[IglooEater]

Yesterday I spent a few minutes looking at the D500 tests on DPR.
It's interesting that the D500 seems to have accomplished something even the D5 couldn't, it looks good at high ISO (better than any other crop sensor, barring Fuji's Lack of moire), but also has excellent Dynamic Range. So at a glance it seems like the D500 doesn't make any sacrifices on either end (it's also funny how Sony themselves seem to manage to do so much worse than Nikon when they make all the sensors).
And at the same time, all these gains also still seem a bit underwhelming compared to Full Frame. The 6D is obviously the superior camera at high ISO, and even the 5D2 still looks better under many circumstances, you just can't go over ISO 3200. Within the ISO range that you do have, older bodies don't actually look like much of a compromise.
And that goes double for any of the more recent low end Nikon bodies. As far as landscape and portrait use is concerned the D500 isn't much different from a D3300.

Good points . One has to admit that D500 sensor is something else..

 

[kaihp]

At the risk of starting a thread that will possibly degenerate into name calling and acknowledgement that DxO are the masters of measurement ( ), I'll expand here on a recent post I made regarding an article in UK magazine, Amateur Photographer.

Written by Bob Newman, a Prof. of Computer Science at the University of Wolverhampton, he talks about the changes in approach taken by Canon and Nikon to their analogue-to-digital converters (ADCs).

(snip)

So, paraphrasing the paraphrase, Bob is saying that:
1) Nikon's approach of putting the ADCs on the sensor so you can have more ADC channels (so you have more time sampling each pixel) is a benefit for lower read-noise at low ISO levels
2) Canon has (or had) a better ADC architecture, which results in a better high-ISO image.

Did Bob say anything about why points #1 and #2 couldn't be combined?

Surely a CMOS sensor process is non-optimal for implementing an ADC in, but there's no reason not to put two independent dies on the same sensor substrate and put (lots) of connections between them. In fact, this is the norm in the flash memory (stacks of 16-32 identical chips to boost capacity) and smartphone processor business (5+ chips in a single package to deliver CPU, flash memory, DRAM, GSM modem etc).

 

[GuyF]

Did Bob say anything about why points #1 and #2 couldn't be combined?

No, and no measurements either. I don't know enough about chip fabrication techniques to speculate why we can't have the best of both worlds. They probably can but want to keep that selling point for the next version!

 

[frankchn]

1) Nikon's approach of putting the ADCs on the sensor so you can have more ADC channels (so you have more time sampling each pixel) is a benefit for lower read-noise at low ISO levels

I suspect the main contributor to read noise in this case may be the length of the electrical pathway between the sensor and the ADC. On-chip ADCs have lower read noise because the signal has much shorter distance to travel to the ADC before being digitized.

Did Bob say anything about why points #1 and #2 couldn't be combined?

I think cost would be a big reason. An FF sensor die is already quite expensive to make because of the size constraint (it must be 36x24mm and no smaller) and integrating 4 pretty big ADC chips (like in the 1DX, see this image from the 1DX teardown -- the Analog Devices chips are 4-channel ADCs) in package via an interposer of sort is going to be really expensive.

For instance, nVidia is pricing its Tesla P100 compute card, with a 600mm^2+ primary die and stacks of HBM2 memory on a ~50mmx50mm passive silicon interposer at >$10,000 each. To be fair, Canon and Nikon are not using leading edge 14nm fabrication techniques for any sensor silicon, but even assuming an 80% cost reduction, it would still be cost prohibitive for Canikon to do this, given that NVidia is probably making a lot more GP100 dies than Canon or Nikon will make 1DX2 or D5 sensors.

 

[Eldar]

Does anyone know how Sony handles this issue with the new 100MP monster chip, used on the new Hasselblad and Phase One? They claim to have 15 stops of DR.

 

[john1970]

Had the opportunity to check out the D5 recently and have started shooting with my 1DX2 as well. I can say that both are excellent cameras.

If one can't get good enough shots using either of these two cameras and has gripes that the other one is better, he needs a check-up from the neck-up.

A very good point to make. If one cannot get the photography they desire using modern pro-grade cameras then one needs to work on their technique.

 

[arthurbikemad]

Went out with my partners 1200D yesterday and took some of the best images I have taken in a long time just shows you, or rather, shows me... :

Can't wait for the Mk2 to start filling the stores, just need a little more feedback from daily users before I punt my money into my hobby shooting, but after 20 or more years I am up for a pro body once in a lifetime.

 

[GuyF]

Went out with my partners 1200D yesterday and took some of the best images I have taken in a long time just shows you, or rather, shows me... :

Yup, I've always said you can take a great image with any camera (within certain limitations!).

...after 20 or more years I am up for a pro body once in a lifetime.

I feel the same. Put my name down for a 1DX2 this morning. Will trade in my 5D3 and do the rest on 24mths interest free credit so can't complain. You only live once so I might as well do it while I can still carry the heavy stuff!

 

[jrista]

At the risk of starting a thread that will possibly degenerate into name calling and acknowledgement that DxO are the masters of measurement ( ), I'll expand here on a recent post I made regarding an article in UK magazine, Amateur Photographer.

Written by Bob Newman, a Prof. of Computer Science at the University of Wolverhampton, he talks about the changes in approach taken by Canon and Nikon to their analogue-to-digital converters (ADCs).

(snip)

So, paraphrasing the paraphrase, Bob is saying that:
1) Nikon's approach of putting the ADCs on the sensor so you can have more ADC channels (so you have more time sampling each pixel) is a benefit for lower read-noise at low ISO levels
2) Canon has (or had) a better ADC architecture, which results in a better high-ISO image.

Did Bob say anything about why points #1 and #2 couldn't be combined?

Surely a CMOS sensor process is non-optimal for implementing an ADC in, but there's no reason not to put two independent dies on the same sensor substrate and put (lots) of connections between them. In fact, this is the norm in the flash memory (stacks of 16-32 identical chips to boost capacity) and smartphone processor business (5+ chips in a single package to deliver CPU, flash memory, DRAM, GSM modem etc).

An ADC is not a very complex circuit. Quite simple, actually. The primary benefit of moving the ADC onto the sensor die, is you can then have one ADC unit per column of pixels. The primary source of noise in a digital camera is ADC noise, and that is usually due to high operating frequency (and it gets worse the longer the traces from pixel to ADC unit are).

Canon's architecture uses a handful of high frequency ADC units off the sensor die. Usually 4 or 8. So, you have both problems...high frequency and long traces. Neither are good for read noise, however you eventually hit physical limits as you increase gain, so high ISO read noise with a Canon is little different than high ISO read noise with any other brand of camera.

The Sony Exmor architecture, used by Nikon, uses column-parallel ADC units on the sensor die. They operate at a significantly lower frequency (because they can, as each one only has to process a few thousand pixels rather than millions of pixels). They are on-die, so traces are short. Sony's architecture also allows for each ADC unit to optimize for the given column, eliminating vertical banding. The ADC units also incorporate digital CDS, which helps suppress another source of noise. Finally, the clock unit which times the ADC is remotely located on the sensor die, away from all other components, which avoids any high frequency jitter from the clock adding any additional noise in the ADC units.

Due to the whole complex of innovations in Sony's Exmor sensor, read noise generally tops out around 4e- at base ISO for most full frame sensors, compared to the 30e- or more that Canon full frame sensors have at base ISO. That is a very significant difference, since noise terms add in quadrature, and read noise is already an RMS, you have to square the read noise term before combining it with any others. So a Sony sensor adds a 4^2 noise term, or 16e-...while a Canon sensor adds a 30^2 noise term, or a whopping 900e-.

If you have a faint signal of 20e-, your SNR with an Exmor would be somewhere along the lines of 20/SQRT(20+4^2), or 3.33:1. Your SNR with a Canon would be along the lines of 20/SQRT(20 + 30^2), or 0.66:1. The signal with a Canon sensor is mired in noise...while it actually has decent SNR with the Exmor. You still have an SNR over 1:1 with a mere 5e- signal with the Exmor sensor...a Canon sensor would completely bury the signal in read noise with a pitiful SNR of 0.17:1. You would need a signal of 33e- to even reach an SNR of 1:1, over six times stronger than the minimum signal allowed by Exmor's innovative on-die architecture.

Some people don't really care about shadow details, others do. If you do care about shadow details, Canon's architecture, using higher frequency off-die ADC units, is plain and simply noisier. A lot noisier. Moving the ADC units onto the sensor die would certainly help Canon's shadow IQ. Thing is, they know that. They actually designed a column-parallel ADC architecture for their prototype 120mp APS-H sensor that was demoed years ago. They demonstrated similar technology in a couple of other prototype sensors they have demoed since then. Perhaps the 1D X II is finally employing the same technology, and is the reason it has better DR. Maybe the 5D IV will also use similar technology, and finally close the DR gap with it's competitors. I truly hope so, as someone who cares about shallow signals myself, it would be extremely nice to see Canon finally entering the modern age with better noise management in their cameras. Their established architecture is archaic and needs to be retired.

 

[GuyF]

Jrista,

Thanks for the input and clarifying the process involved in getting the signal off the chip.

I wonder if they'll add the same ADC architecture to the 5D4 or just keep their best stuff for the flagship.

Question now is, what will DxO say? ;D

 

[nvsravank]

An ADC is not a very complex circuit. Quite simple, actually. The primary benefit of moving the ADC onto the sensor die, is you can then have one ADC unit per column of pixels. The primary source of noise in a digital camera is ADC noise, and that is usually due to high operating frequency (and it gets worse the longer the traces from pixel to ADC unit are).

SNIP...

These are the kind of posts that make me come back to canon rumors (Once the fervor over new equipment dies down).
Thanks Jrista

 

[Mr Bean]

An ADC is not a very complex circuit. Quite simple, actually. The primary benefit of moving the ADC onto the sensor die, is you can then have one ADC unit per column of pixels. The primary source of noise in a digital camera is ADC noise, and that is usually due to high operating frequency (and it gets worse the longer the traces from pixel to ADC unit are).

SNIP...

These are the kind of posts that make me come back to canon rumors (Once the fervor over new equipment dies down).
Thanks Jrista

+1

@Jrista, perhaps a dumb question (its morning here in Australia and I'm only on my second coffee.....), but, does the current architecture being used by Canon allow for other benefits such as faster FPS, etc? Or, which I suspect, is it simply a path they took some years back in manufacturing, and now its a slow process to change.

 

[Don Haines]

At the risk of starting a thread that will possibly degenerate into name calling and acknowledgement that DxO are the masters of measurement ( ), I'll expand here on a recent post I made regarding an article in UK magazine, Amateur Photographer.

Written by Bob Newman, a Prof. of Computer Science at the University of Wolverhampton, he talks about the changes in approach taken by Canon and Nikon to their analogue-to-digital converters (ADCs).

(snip)

So, paraphrasing the paraphrase, Bob is saying that:
1) Nikon's approach of putting the ADCs on the sensor so you can have more ADC channels (so you have more time sampling each pixel) is a benefit for lower read-noise at low ISO levels
2) Canon has (or had) a better ADC architecture, which results in a better high-ISO image.

Did Bob say anything about why points #1 and #2 couldn't be combined?

Surely a CMOS sensor process is non-optimal for implementing an ADC in, but there's no reason not to put two independent dies on the same sensor substrate and put (lots) of connections between them. In fact, this is the norm in the flash memory (stacks of 16-32 identical chips to boost capacity) and smartphone processor business (5+ chips in a single package to deliver CPU, flash memory, DRAM, GSM modem etc).

An ADC is not a very complex circuit. Quite simple, actually. The primary benefit of moving the ADC onto the sensor die, is you can then have one ADC unit per column of pixels. The primary source of noise in a digital camera is ADC noise, and that is usually due to high operating frequency (and it gets worse the longer the traces from pixel to ADC unit are).

Canon's architecture uses a handful of high frequency ADC units off the sensor die. Usually 4 or 8. So, you have both problems...high frequency and long traces. Neither are good for read noise, however you eventually hit physical limits as you increase gain, so high ISO read noise with a Canon is little different than high ISO read noise with any other brand of camera.

The Sony Exmor architecture, used by Nikon, uses column-parallel ADC units on the sensor die. They operate at a significantly lower frequency (because they can, as each one only has to process a few thousand pixels rather than millions of pixels). They are on-die, so traces are short. Sony's architecture also allows for each ADC unit to optimize for the given column, eliminating vertical banding. The ADC units also incorporate digital CDS, which helps suppress another source of noise. Finally, the clock unit which times the ADC is remotely located on the sensor die, away from all other components, which avoids any high frequency jitter from the clock adding any additional noise in the ADC units.

Due to the whole complex of innovations in Sony's Exmor sensor, read noise generally tops out around 4e- at base ISO for most full frame sensors, compared to the 30e- or more that Canon full frame sensors have at base ISO. That is a very significant difference, since noise terms add in quadrature, and read noise is already an RMS, you have to square the read noise term before combining it with any others. So a Sony sensor adds a 4^2 noise term, or 16e-...while a Canon sensor adds a 30^2 noise term, or a whopping 900e-.

If you have a faint signal of 20e-, your SNR with an Exmor would be somewhere along the lines of 20/SQRT(20+4^2), or 3.33:1. Your SNR with a Canon would be along the lines of 20/SQRT(20 + 30^2), or 0.66:1. The signal with a Canon sensor is mired in noise...while it actually has decent SNR with the Exmor. You still have an SNR over 1:1 with a mere 5e- signal with the Exmor sensor...a Canon sensor would completely bury the signal in read noise with a pitiful SNR of 0.17:1. You would need a signal of 33e- to even reach an SNR of 1:1, over six times stronger than the minimum signal allowed by Exmor's innovative on-die architecture.

Some people don't really care about shadow details, others do. If you do care about shadow details, Canon's architecture, using higher frequency off-die ADC units, is plain and simply noisier. A lot noisier. Moving the ADC units onto the sensor die would certainly help Canon's shadow IQ. Thing is, they know that. They actually designed a column-parallel ADC architecture for their prototype 120mp APS-H sensor that was demoed years ago. They demonstrated similar technology in a couple of other prototype sensors they have demoed since then. Perhaps the 1D X II is finally employing the same technology, and is the reason it has better DR. Maybe the 5D IV will also use similar technology, and finally close the DR gap with it's competitors. I truly hope so, as someone who cares about shallow signals myself, it would be extremely nice to see Canon finally entering the modern age with better noise management in their cameras. Their established architecture is archaic and needs to be retired.

well said with one caveat
An A/D converter CAN be a very simple circuit.... but some of the stand alone A/D converters are quite complex circuits, and although I don't know for sure, my bet is that the external A/Ds that Canon uses fall into the complex circuitry realm.....

That said, with all the advantages that on chip A/D's add to the mix, I am equally sure that a relatively simple row or column A/D on-chip will outperform a high speed serial A/D of chip, particularly when you also take into account that you now need a far more complex analog switch matrix to route the signals off-chip than the simple linear switch to route the on-chip row or column analog signals.....

This also makes me wonder how long it will be until we get to the end-game, which is counting electrons per pixel and doing away with A/D altogether.....

We live in exciting times....

 

[jrista]

An ADC is not a very complex circuit. Quite simple, actually. The primary benefit of moving the ADC onto the sensor die, is you can then have one ADC unit per column of pixels. The primary source of noise in a digital camera is ADC noise, and that is usually due to high operating frequency (and it gets worse the longer the traces from pixel to ADC unit are).

SNIP...

These are the kind of posts that make me come back to canon rumors (Once the fervor over new equipment dies down).
Thanks Jrista

+1

@Jrista, perhaps a dumb question (its morning here in Australia and I'm only on my second coffee.....), but, does the current architecture being used by Canon allow for other benefits such as faster FPS, etc? Or, which I suspect, is it simply a path they took some years back in manufacturing, and now its a slow process to change.

Technically speaking, you should be able to get even faster frame rate by using a hyper-parallel ADC architecture. For example, Canon's prototypical 120mp APS-H achieved an amazing 9.5fps! If you halved that megapixel count to 60mp, the camera should then be able to achieve around 19fps. Halve it again to 30mp, and you should be able to get around 38fps. Halve it again, to 15mp, and you should be able to get 76fps. (In reality, things probably wouldn't be quite that good due to other real-world factors...but to keep things simple. )

Now, I don't know how useful 76fps is...but, just to put the megapixel count in line with existing high speed sports cameras. You can see the potential benefits of using a highly parallel ADC architecture. At pixel counts similar to current cameras, getting high frame rate would actually be the least of Canon's problems. They would actually have more problems moving the mirror fast enough, or focusing fast enough. Of course, then there are other benefits of using hyper-parallel ADC...you could run them at an even lower frequency...which should introduce even less noise.

Some prototypical sensors in development over the last couple of years even use per-pixel ADC now. Or as it is used in practice, you usually have one shared ADC per 2x2 grouping of pixels. You get the same general benefits as going to per-column ADC, only on a much more massive scale. ADC units that only need to convert data from four pixels can run at a significantly lower frequency and you can still read out the sensor just as fast (assuming you have sufficiently fast and parallel image processing circuitry paired with the sensor.)

Anyway. I think Canon has just stuck with their existing architecture for so long because it worked, and they did not feel enough pressure to change it (even though they were clearly working on more advanced technology, have been for many years, well before the introduction of the 5D III even.) It isn't surprising Canon is starting to work in improving their technology...however, I wouldn't expect to see radical ground-breaking innovations from them any time soon. They demo a radical sensor every few years...then get back to business and the new technology usually never sees the light of day. It's just Canon's way.

 

[jrista]

well said with one caveat
An A/D converter CAN be a very simple circuit.... but some of the stand alone A/D converters are quite complex circuits, and although I don't know for sure, my bet is that the external A/Ds that Canon uses fall into the complex circuitry realm.....

That said, with all the advantages that on chip A/D's add to the mix, I am equally sure that a relatively simple row or column A/D on-chip will outperform a high speed serial A/D of chip, particularly when you also take into account that you now need a far more complex analog switch matrix to route the signals off-chip than the simple linear switch to route the on-chip row or column analog signals.....

This also makes me wonder how long it will be until we get to the end-game, which is counting electrons per pixel and doing away with A/D altogether.....

We live in exciting times....

Fair point! I'm sure the off-die ADC units Canon has been using thus far are a lot more complex than an on-die ADC unit might need to be. Assuming they are self-contained units, they probably have a clock generator and all the other stuff that goes along with keeping clocked signals properly timed and all that.

I am hoping the end game ain't that far off. I don't know if you have heard of sCMOS? The CCD world, something I've become a lot more familiar with since moving into astrophotography, is slowly dying. It's dying a lot faster now that Sony is going to stop manufacturing them. A new technology, sCMOS or Scientific CMOS, is already taking CCD's place in the scientific world, where the demands for maximum quality are extremely high.

Current sCMOS sensors are really holding some powerful cards:

1) They have extremely low read noise, usually less than 2e-, sometimes even in the sub electron range.
2) Many have extremely low dark current, which diminishes to near vanishing when the sensors are cooled.
3) They are increasingly made as BSI these days, rather than FSI, which improves the light sensitive fill factor to nearly100%.
4) They have extremely high quantum efficiency for some of the newer generation, topping 95% Q.E. and allowing pretty good photon counting statistics (ratio of around 1.3ph/e-).
5) They can be read out at very high frame rates, 20-40fps and in some cases (when they use lower resolution and much larger pixels) beyond...and they do this without increasing read noise, so it is still down around 1e-!

Currently, most actual photon counting cameras are EMCCD, or electron multiplying CCD. They rely on extreme gain (often on the order of 40:1) and excessive cooling (-100°C) to massively amplify the photo signal, without any dark current and with read noise in the sub 0.5e- range (some are reported to have 0.1e- read noise, although in actual practice I haven't heard of less than 0.16e-). An EMCCD still tends to cost a couple hundred grand, though. Seems a small tradeoff for that loss of 0.3ph/e- to move to sCMOS...

A few years ago, it cost $200,000 for an sCMOS camera, easy. Today, you can pick one up for as "little" as $17,000. I don't think it will be terribly long before some of the technology those sensors employ finds it's way into consumer grade cameras. Maybe a few years? I certainly hope so. I am at least hoping that sCMOS tech gets cheap enough that I could buy a nice astro camera with an sCMOS sensor for around $5000 (which is what the average astro CCD costs these days as well.)

 

[Don Haines]

Fair point! I'm sure the off-die ADC units Canon has been using thus far are a lot more complex than an on-die ADC unit might need to be. Assuming they are self-contained units, they probably have a clock generator and all the other stuff that goes along with keeping clocked signals properly timed and all that.

Yeah.... We were using a quad of 16 bit, 4Gsps A/Ds on a project at work..... they each came on a 68 pin IC and needed to be heat-sinked and water cooled.... they cost $4000 a chip! Definitely not simple

I am hoping the end game ain't that far off. I don't know if you have heard of sCMOS? The CCD world, something I've become a lot more familiar with since moving into astrophotography, is slowly dying. It's dying a lot faster now that Sony is going to stop manufacturing them. A new technology, sCMOS or Scientific CMOS, is already taking CCD's place in the scientific world, where the demands for maximum quality are extremely high.

Yes, I have hear of them..... very expensive.... On the other hand I remember buying a 10MB hard drive at work for ONLY $9,995 and we thought it was incredible.... under a dollar per K of storage!!!!!! I just bought a couple of 6TB drives to add to my backup... $175 for 6TB.... or a dollar per 34GB of storage.... 7 1/2 orders of magnitude difference! The same will happen to sCMOS and at some point they are going to have dropped in cost to the point where they can be put into the 1DX's of the camera world and once that happens, the spread will be inevitable....

Exciting times and always something good coming down the road!

 

[Mr Bean]

An ADC is not a very complex circuit. Quite simple, actually. The primary benefit of moving the ADC onto the sensor die, is you can then have one ADC unit per column of pixels. The primary source of noise in a digital camera is ADC noise, and that is usually due to high operating frequency (and it gets worse the longer the traces from pixel to ADC unit are).

SNIP...

These are the kind of posts that make me come back to canon rumors (Once the fervor over new equipment dies down).
Thanks Jrista

+1

@Jrista, perhaps a dumb question (its morning here in Australia and I'm only on my second coffee.....), but, does the current architecture being used by Canon allow for other benefits such as faster FPS, etc? Or, which I suspect, is it simply a path they took some years back in manufacturing, and now its a slow process to change.

Technically speaking, you should be able to get even faster frame rate by using a hyper-parallel ADC architecture. For example, Canon's prototypical 120mp APS-H achieved an amazing 9.5fps! If you halved that megapixel count to 60mp, the camera should then be able to achieve around 19fps. Halve it again to 30mp, and you should be able to get around 38fps. Halve it again, to 15mp, and you should be able to get 76fps. (In reality, things probably wouldn't be quite that good due to other real-world factors...but to keep things simple. )
.
.
.

Thanks for the clarification Jrista.