ankorwatt said:
jrista said:
ankorwatt said:
read what I write, the real improvements are around 1,1 to 1,4 um sensel size
and there are no APS or 24x36 from Canon or others yet= with that small pixel size
BSI cost about 30% more than FSI
Eric Fossum:
Improvements like BSI typically improve image quality mathematically and from a perception point of view, by increasing QE and reducing effects orginating from pixel stack height, when comparing two pixels of equal size. At 1.4 um pixel pitch the improvement offered by BSI is small. By 1.1 um pixel pitch, BSI offers a substantial advantage, unless some FSI breakthrough is made. BSI costs more to make so there is motivation for the FSI breakthough
It really depends on the photodiode size. A 7D has 4.3 micron pixels, but the actual photodiode is smaller than that. The entire pixel is surrounded by 500nm (.5 micron) transistors and wiring, which would mean the photodiode...the actual light sensitive area embedded in the silicon substrate, is only about 3.3 microns at best (and usually, the photodiode has a small margin around it...so closer to 3 microns). A 24.4mp sensor would have pixels in the range of 3.2 microns, however with a 500nm process, the actual photodiode pitch is closer to 2 microns.
Canon has already demonstrated that larger pixels can be huge for overall SNR (and therefor actual light sensitivity) with the 1D X. Despite the fact that the 1D X is a FF sensor, it benefits greatly from a larger pixel, and thus a larger photodiode size...as the gain is relative to the square of the pixel pitch. Production of a BSI APS-C 24.4mp sensor would mean that it could have 3.1 micron photodiodes that perform at least as well as the 7D's 18mp sensor, as total electron capacity is relative to photodiode area. A 24.4mp BSI 7D II could then be roughly as capable (~21,000 electrons charge FWC @ ISO 100) as an 18mp FSI 7D.
Personally, I find that to be quite a valuable thing. Especially given that the 7D currently performs about as poorly as one could expect by today's standards. A 2 micron photodiode in the 7D II would mean SNR suffers even more, which is going to have an impact on IQ, especially for croppers, so I can't imagine Canon doing that.
you are mixing up things, why do you think Im saying that Canon needs 180 or smaller tech?
The real benefits of BSI you find in very small sensel, and I do not think it is a good idea to talk to much what you believe or think when Eric Fossum have shown when the benefits starts of a BSI construction.
And that is around 1,4 micron and smaller
A tipping point for BSI will be the 1.1 micron pixel node where FSI will likely be unable to achieve the market-required performance – necessitating a transition to BSI for applications that require this smaller pixel."
There are some benefits of BSI and larger pixels and that is with wide angle lenses and corners, as for example SLR+ wide angel lens and incident light angle
I am not mixing anything up. The primary benefit of 180nm is that you have more area per pixel to dedicate to the photodiode. In the case of 1.4 micron pixels, use of a 500nm process is already a non-option...you would have already passed the limit you claim would be reached with 1.1 micron pixels on a 180nm process...the photodiode of a 1.4 micron pixel on a 500nm process would be maybe .3 microns (300nm). You have to translate from a 180nm 1.4 micron pixel to a 500nm 3.2 micron pixel. The wiring and transistors in a 500nm process take up a lot of space. That space could be put to better use...and assuming one does not change from a 500nm process....well, then BSI DOES have value.
Instead of taking up ~1 micron of pixel pitch for wiring and other logic, you take up a quarter of a micron if you moved to a 180nm process on APS-C. That means, for a 4.3 micron pixel pitch, the actual photodiode could be ~3.95 microns, rather than 2.1 microns. That increase in area is where you gain the greatest potential for an improvement in IQ. Now, with 180nm transistors, you can pack more of them in. Canon could stick with a 2.1 micron photodiode, and have a lot more logic circuitry around it with a 180nm process. That would allow them to add more sophisticated noise reduction logic, maybe drop in some on-die ADC, etc....simply because each transistor and all the wiring consumes less space. But fundamentally, photodiode area is the key thing from an SNR standpoint, and a higher SNR leads to less noisy images.
When it comes to Canon's read noise, the primary issue there is high frequency components and binned pixel processing on an off-die component. The longer the signal remains analog, and the closer any pixel processing is to a high frequency component (a DIGIC processor is a CPU...the whole thing is a high frequency component), the greater the chance that read noise will interfere with shadow detail. It doesn't matter what the fabrication process is...Canon could move to 180nm, and keep using their Digic processors with off-die ADC. They will continue to have shadow noise problems, despite the move to a better process. If they move the ADCs on-die, and do something akin to what Exmor does, by moving the PLL, Clock, and other high frequency components to an isolated area away from those ADC units, then Canon could reduce their shadow noise.
That only affects low ISO, however, and a lot of Canon users care more about high ISO. Using a BSI design, even in APS-C, allows photodiode area to remain large. Canon could also still add more advanced per-pixel logic in a BSI design even if they stay on a 500nm process, as they would have the full photodiode area on the front side to utilize for logic (i.e. additional noise reduction circuitry...one of their patents described a power-source free CDS system that decoupled the power input while performing CDS, as keeping the power coupled continued to add dark current noise.)
It is not NECESSARY for Canon to move to a 180nm process, or only use BSI with small form factor sensors having 1.4 micron pixels or smaller, in order to continue innovating and improving IQ. As far as I am concerned, for the kind of high ISO work I do, I would LOVE to see Canon produce a FF BSI sensor. That would allow them to increase photodiode area, particularly in a shared pixel architecture, by another micron. Right now, in the 1D X, photodiode pitch is around 5.8 microns, while the actual pixel pitch is 6.95. I think it would be awesome to see a 1D XI with a BSI design that had 6.95 micron photodiodes. That is a 43% increase in total photodiode area, an increase that would have a measurable improvement in high ISO performance (imagine an actual usable ISO 25600 and maybe 51200 for wildlife and birds.) Again, Canon could move to a 180nm process, and either pack more logic into each pixel and improve readout NR (i.e. CDS), or reduce the logic, increase photodiode area, and move the ADC on-die, which at the very least should increase the maximum readout rate and possibly improve read noise performance. There are a whole lot of options...Eric Fossum isn't the only source of CIS innovation, nor the bible of what is and is not possible with CIS devices. Eric Fossum has done a lot of research in the area, however so has Canon (remember, it wasn't that long ago that Canon had the best sensors in the digital camera arena...they certainly have the knowledge and knowhow...I think their current reliance on 500nm is more of a business and financial matter than a lack of ability.)
I think moving to BSI, even if Canon sticks to 500nm, is a better option. It frees up the entire front side for logic, and the entire back side to light sensitive photodiodes. It is something Canon could do with their current process, potentially freeing up a billion dollars for other purposes (R&D, greater production capacity, whatever.)