To me, the big monoculture risk is Sony's image-sensor fab. As I understand it, Sony, Fuji, Nikon and many others all rely upon the output of one company's manufacturing process.
There's a good reason for it -- they appear to make the very best sensors (edit: in terms of dynamic range and noise) in the commercial-camera market. Canon makes their own sensors, which are good in absolute but not relative terms, which appears to hold them back in performance. I don't know if it is patent-related or process-related.
While I use Canon imaging systems, I don't see the choice to use other brands as better or worse. Each manufacturer's line has strengths and weaknesses; you pick the best tool for the job. Essentially every camera and optical system on the market, in absolute terms, is a wonderful instrument for imaging.
This reality is intensely freeing -- it means you can focus more on the image and less on the instrumentation. "f/8 and be there" is still wonderful advice.
Sony's image sensor fab business is a bit different.
First of all, there's two companies: Semiconductor and Imaging. Sony opens some of their technologies to other customers by transferring Imaging unit's patents to Semi and allows them for re-licensing. So, not everyone is using the same sensor even if they are coming out from the same fab.
Second, there's customer tailored sensor business. Sony can provide a customer a baseline to start and, customer can customize this sensor according to their needs and got it produced for themselves.
Lastly, there's strict division between customers in semiconductor side. So Sony Imaging can't access to specs and design of other customers' sensors.
How it works in real life?
- Sony A7 series use Sony's sensors with all their secrets at their disposal but, not with Nikon's and Fuji's secret sauces.
- Nikon orders customized, "Designed by Nikon" sensors to Sony Semi. Sony Imaging cannot access them (unless they buy and disassemble a Nikon Z).
- Same for Fuji. They have less rolling shutter in XT-4 for example, and it's again built by Sony, for Fuji. No, Sony Imaging can't play with them.
- Canon R5's sensor is "Designed by Canon". I'm not sure they are producing it. Sony or Tower Jazz maybe producing it. I don't know.
At the end of the day, the fab is almost one but, designs are many and diversified and, nobody can see each other's design. Like TSMC and Global Foundries.
It is even more tenuous than that. If just one of a company's fabs goes down much of the imaging industry will have problems. A few years ago when the earthquake affected Sony's Kumamoto sensor fab it caused some serious issues for some camera vendors. Nikon for example had announced their DL 1" sensor compact line to compete with Sony's 1" sensor cameras and reportedly due to the earthquake they delayed and eventually completely scrapped the project.
Are sensor fabs functionally similar enough to CPU fabs that if China tried to take back Taiwan, or just made it hard for Taiwan to ship chips to the West, the Japanese sensor fabs could retool for chips?
Absolutely not. The lithographies, processes, and packages are completely different. Some sensors are still made on 300nm processes, useless for modern CPUs. 45 nm is considered cutting edge for imaging sensors, but would be a huge setback for computing. Intel, GlobalFoundries and even IBM fabs would be much better.
The increase in CPU power doesn't become apparent until you start to use vectorization heavily or use your CPU for floating point-heavy scientific computing.
Some of the code I've written can speed up 3x just by using newer SIMD instruction sets and bigger caches. It's unbelievable until you see it.
I also manage HPC clusters, so we see strange effects when developers excessively hand-tune their code for a specific architecture.
Well of course hpc can benefit from 13 years of cpu advances. (And gpus). Noting that sse4 was even available on it. My point was we still wouldn't be relegated to stone age if we were stuck on 45nm processes, and we could also benefit from better architectures.
In bigger semiconductor business we have three main fabs.
Intel, TSMC, Global Foundries. Intel is closed to outside customers, so again we have a duopoly there.
It's not very very different in the bigger picture if you ask me.
Also, Glofo stopped developing cutting edge nodes (7nm or smaller) after AMD had ditched them a couple of years ago. It's now all but a two-horse race between TSMC and Samsung.
There are lots of fabs, but not all specialize in techniques used for imagers (like backside illumination). I worked in LCOS and we had one vendor qualified that could produce a shinier metal layer -- something most fabs wouldn't care about, but directly affected our brightness & power consumption.
Agreed. There's also the Fujifilm X-Trans sensors which try to recapture some of the magic of film. However I really wish more people knew about Sigma and their Foveon sensors[1].
The fact is that there is a completely different technology on the market which produces very different – and arguably far superior – image quality and colour reproduction compared to 99% of digital cameras on the market. The Sigma system can be clunky and the sensor tech has its drawbacks (eg almost unusable above 200 iso), however given the right conditions the images they produce can be stunning.
Unfortunately, real Foveon is a thing of the past, Sigma switched to a Bayer sensor in their latest "fp" camera and their last Foveon-based Quattro had 3 layers but a 4:2:2 pixel mosaic, making it no longer a "true Foveon". I owned SD1 Merrill and it was the sharpest camera I've ever used, but there is nothing comparable on the market these days.
All true. I do hold out hope for a “true” foveon return. And you can still grab Merrill models second hand, like I did about a year ago.
I mainly use an Olympus em5 mkII so I’m saddened by the OP news, although I will wait to see how it plays out before switching to Ricoh and/or Pentax. Guess I have a thing for the underdogs.
I own and still use a DP2X. When things go right, the result is truly breathtaking. 80 to 90% of the photos I take with this are junk, but the remaining 10-20%, boy-oh-boy, truly memorable results!
I was under the impression that the Foveon sensor had been overtaken by events. Kind of like the Lisp Machines: When it was first introduced its dedicated "no compromises" approach was great, but then the more pedestrian technologies were able to outscale them and they lost their edge.
The Nikon 1 line, with exception of the J5 (which used Sony's sensor), used sensors made by Aptina.
Aptina's sensors were pretty spectacular from an AF perspective in that they had phase-shift AF on-sensor in 2011. The AF was as close as you could get to DSLRs on a mirrorless camera for quite some time. It's just too bad that Nikon came up with a weird set of 1st generation Nikon 1 cameras that made no sense to anybody.
IIRC, Aptina ended up cross-licensing their patents to Sony.
I don't think its quite accurate to say Canon sensors hold anyone back. Canon photographers continue to win awards, continue to create world-class images, etc, etc. Canon gear is very well represented in the professional photography community. They wouldn't be using the gear if they couldn't get their work done. There are several gear "measurebators" online arguing about base ISO DR and +6EV exposure recovery, who produce nothing with their cameras. We can safely ignore them :)
Sony sinks tons of money in their tech because other companies use them as a factory, so that makes total business sense (and along with samsung and a few others, they do have cutting edge imaging tech). Canon is different in that its much more vertically integrated so it rarely makes sense to drop millions of dollars on sensor fab upgrades unless they're going to make the money back. They have dozens of filed patents on all the technology that the competition is using, so I don't think they're lacking on the IP front.
>I don't think its quite accurate to say Canon sensors hold anyone back. Canon photographers continue to win awards, continue to create world-class images, etc, etc. Canon gear is very well represented in the professional photography community.
I don't think photography is where the concern lies these days. All medium/flagship cameras from major brands today are more than good enough for photography, commercial, sports, landscape, advertising, editorial, or whatever...
Agree. Even with video we're spoiled with tech to the point where it takes hundreds of YouTube channels to explain what the heck the tech does and how to best use it. Reminds of a quote - "We are not limited by our tools, just by our imagination". On a related note, I recently re-watched the original Jurassic Park, and the movie quality still blows me away.
I mean, your same argument could be made for TVs/monitors as well. LG, Samsung and AU Optronics make the vast majority of LCD panels. You could buy an LG or a Vizio with the same LG-fabbed panel. In fact, even Samsung TV has been known to use LG panels:
AU Optronics, InnoLux, Infovision, BOE optoelectronics, IVO, LG - these do laptop panels (also see panelook.com ). There is some competition in TN/IPS. But OLEDs or specialized panel situation may be different.
> Canon makes their own sensors, which are good in absolute but not relative terms, which appears to hold them back in performance
It's a little early to state this. Once the R5 and R6 come out we'll have a better idea of their capabilities as those are slated to have their new tech (8K video etc etc).
Which TBF hints at the reason Sony does have all of this nice image sensor technology.
While they've been a newer player in the Camera industry, they've been doing professional digital video for far longer (SW EP2 was shot on Panavision-modified Sony cameras.) That's on top of their long lineage in the pre-digital-recording TV Market.
On an unrelated note, I do believe that Canon has also pulled a patent on quad-pixel AF. Patents being awful aside, quad-pixel AF would be awesome: every pixel becomes a cross-type focus sensor.
(For the uninitiated, focus sensors are usually either horizontal or vertical, and cannot detect features that are perpendicular to them. Cross-type sensors are a combination of the two to increase the detection capability.)
Not disagreeing, but just to counter your point a bit, the Sony A9 has zero cross type sensors, but it has the best on-sensor PDAF system currently on the market. Conceptually, I totally agree with you that cross type sensors are BETTER than non cross type sensors, and dual cross-type are even BETTER, but its rarely as simple as that in the real world.
"Best on-sensor PDAF system currently on the market" is a non-scientific statement, but I'll certainly agree with you that it's great.
However, in less optimal situations (e.g. low light, low contrast, and contrast perpendicular to the detection direction), it's my personal experience that current non-cross-type PDAF solutions fail, requiring 90 degree rotation to focus, which wouldn't be needed for cross-type. That, or use of an alternate focus subject if one is available.
It's more of an issue on smaller, less light sensitive sensors. The super high-end sensors are more likely to find something to focus on even when things are suboptimal.
But, alas, I'm a hobby photographer, and the 4500 USD an A9II costs is a bit more than I'm willing to spend.
Cross-type has nothing to do with low light or low contrast. Sorry, I don't understand what point you want to make. There are dozens of things that affect focusing, or to be more specific - focusing speed, focusing accuracy, focusing repeatibility, etc. Many of these factors are tied to the lens. For e.g. Outer focusing points on a poorly made "consumer" lenses, generally, are not very accurate when you're close to MFD, and the focusing system tends to hunt a lot. STM focus motors are not designed for focusing speed, but they are designed for repeatibility. Also, most low light focusing and low-contrast focusing is hampered because of the high read-noise from the line sensors. I have the A9II and it has the best focusing system I've ever used. I thought the 1DX II was accurate, but this blows it away. Haven't tried the 1DXIII though.
> Cross-type has nothing to do with low light or low contrast.
Of course it does. PDAF needs to observe a feature (a phase difference) in the image, which requires light and contrast.
In strong light, there are usually plenty of features available, making PDAF seem borderline magic, especially for something like DPAF where almost every single pixel can be used as a focus point.
In lower light and contrast, fewer features are available. Subjects are rarely perfectly symmetric, so there are different features are available across the different sensor axis (e.g. a striped shirt).
As subjects are not symmetric, the chances of you running out of features to focus on due to reduced contrast along two axis is much lower than running out along just one. Cross-type allow both axis to be used simultaneously, while non-cross can only use one.
This means that, with the same sensitivity, cross-type is much more likely to be able to detect a feature in an area than non-cross, as there are more features available to it. As features turn to noise and AF becomes difficult, this leads to increased focus ability at lower light.
For non-cross, a 90 degree rotation of the camera is needed to reveal those features.
> For e.g. Outer focusing points on a poorly made "consumer" lenses, generally, are not very accurate when you're close to MFD, and the focusing system tends to hunt a lot.
I see what you're trying to say ("inner focus point is not just cross-type, it's also the sharpest point on the lens"), but that's not really relevant here. I'm making the observation on a camera without cross-type at all, dead-center or thereabout.
One can also use the zebra MF assist feature to highlight this, at least for DPAF cameras.
> I have the A9II and it has the best focusing system I've ever used. I thought the 1DX II was accurate, but this blows it away. Haven't tried the 1DXIII though.
This is the beauty of on-sensor focus - you measure exactly what you see, instead of measuring something else entirely and hoping that the calibration data is good enough...
There's a good reason for it -- they appear to make the very best sensors (edit: in terms of dynamic range and noise) in the commercial-camera market. Canon makes their own sensors, which are good in absolute but not relative terms, which appears to hold them back in performance. I don't know if it is patent-related or process-related.
While I use Canon imaging systems, I don't see the choice to use other brands as better or worse. Each manufacturer's line has strengths and weaknesses; you pick the best tool for the job. Essentially every camera and optical system on the market, in absolute terms, is a wonderful instrument for imaging.
This reality is intensely freeing -- it means you can focus more on the image and less on the instrumentation. "f/8 and be there" is still wonderful advice.