Leica M10 Monochrom: Bifurcated Sensor Defect, Might be Unfixable Hardware Problem (UPDATED with comments), also Affects Leica M10-R, Leica SL2
Update Jan 3: I’ve added more examples of the sensor bifurcation, and a bit more clarity to my discussion.
Back in 2009 I reported on a bifurcated sensor problem with the Leica M9. Leica ignored me when I reported it and provided examples, and I expect the same this go-round. Part of a pattern over the years.
Imagine my surprise in 2020 to come across the same phenomenon in the Leica M10 Monochrom. IMO, this renders the camera totally unacceptable, even if it is seen only occassionally, being mostly a problem in smooth tones. But isn’t that one big appeal of a monochrome camera—gradients and smooth tonality? A 1-point hiccup cannot be unseen once seen, at least under the right range of luminance. Of course, I cannot say or not say whether it happens only in some cameras and not others.
Four examples with analysis, and I’ve detected the problem in other images also:
The full-res image and the crops make the problem obvious, if you cannot see it here.
UPDATE Jan 5: I’ve confirmed the issue even with bright-white clouds. In very bright areas, it is not visible unless one applies an extreme Curve. But it shows that the issue is not confined to sky and that seeing the issue is a matter of human perception.
It seems probable that the Leica M10-R suffers from the same defect, since its sensor is presumably the same technology, but with a color filter array applied. However, it’s possible that the Bayer matrix demosaicing process will mask it better, and that only ultra-smooth gradients (like a blue sunset sky) would show it.
Update Jan 11: see Sensor Bifurcation, Stitched Sensors: More Insight From Readers.
Roy P writes:
Wow, that’s amazing. Never seen anything like it before. The entire right half of the frame is subtly different. I looked at the midway point vertically also, but that looks clean. If this were printed, like in a 16x24 size, it would be visible from a distance on a wall.
I’m not sure what’s happening here. Regular ICs, even very biggest SoC like the Apple processors, are 40% or less than the area of a full frame 35mm image sensor. So a photomask can accommodate multiple dies on a single reticle with 4x feature sizes.
I’m not sure what process they use for making image sensors. But it is odd that Leica not only has this problem today, but they also had it in 2009! Clearly, this has never come up as an issue for Sony sensors, which are not only used in 35mm cameras but go all the way up to full MF. If the Leica M10M sensor was imaged twice, a P1 IQ4 sensor would have be imaged four times, and if there was a systemic problem, surely someone would have spotted it in P1 files.
The only thing certain is, Leica is using the same sensor manufacturer they used for the CCD sensors for the M9. I don’t know if that’s CMOSYS or Jazz. But something is odd, for sure.
DIGLLOYD: the only camera brand I’ve seen this issue with is Leica.
Brent N writes:
A friend recently contacted me about your review of the Leica M10 Monochrom due to the fact that I've been using one for over a year. I was unaware previously of any white dot issues and of the bifurcation problem you referred to.
I subsequently posed the question to a Leica forum member who is widely acknowledged to be one of the most knowledgeable members about digital imaging technology.
Here is his reply to me, which I thought you might appreciate
“These sensors are not "stitched." They are forged as one chip on one piece of silicon. But like all modern sensors they have dual readout channels designed-in (like using two aisles instead of one to get people out of a single theater faster.) Half the sensor's output is routed out through contacts on one side, and half out the other side. Some sensors for high-speed pro sports cameras like Nikon D2 have FOUR channels, for even more speed clearing the electrons out of the sensor for the next exposure.The channels are balanced somewhere in the process between the raw silicon and the final camera leaving the factory - so that they match. I have no idea exactly how and where that is done (may be at the sensor factory, may be a firmware tweak per camera at Leica).But sometimes that is not done perfectly, or gets out of whack later. Not that different from a hot or stuck pixel in terms of seriousness.Sometimes the imbalance is very minor, and only shows up with extreme image manipulation (trying to "rescue" an underexposed high-ISO picture) and sometimes it is obvious even at lower ISOs.At any rate, it requires some correction by Leica, anything from rebalancing the existing electronics to replacing something (the sensor, the motherboard, one of the A/D converters, etc.). Or possibly just a firmware fix.Standard warranty item. It is certainly possible that a bad batch of sensor chips was produced and/or misbalanced, but same answer: standard warranty replacement.”
You were apparently correct about the white dot issue, although it's occurrence seems rare. I've been an active member of the Leica Forum since 2003 and there is only scant mention of this problem. Perhaps this was solved with firmware.
You are not correct, in my opinion, about the stitched sensor. I've shot more than 12,000 frames with the M10 Monochrom and I examine my images scrupulously. I have never seen any line of bifurcation. It does not exist with my camera and I have to think if it existed with other copies it would have become the subject of a great deal of heated conversation on the Leica Forum.
DIGLLOYD: I’d be delighted to hear of a firmware fix, but I am dubious that it is possible. Leica?
There are several logical fallacies in the last paragraph. Which is not to say that makes my theory correct, only that rare is often not so rare*. Reader Brent N is not a subscriber and therefore has not seen my examples or my discussion of where it is likely to occur, so he is at a disadvantage in knowing how to detect the issue. So I’ve passed along suggestions to him so he can examine appropriate images to look for the bifurcation and/or look for it by making images.
As I understand digital sensors (not all that well!), it seems unlikely that a camera would be read out as bifurcated left and right halves, vs top and bottom halves, or just interleaving rows, just from a wiring point of view. AFAIK, entire rows are read. If there are 2 or 4 or N channels, I would expect row interleaving, not left right bifurcation. Or at least top/bottom. Maybe my understanding is incorrect as to how sensors work?
Therefore, sensor stitching remains the most plausible explanation—it has been used on 35m and medium format for some years, and that is a demonstrable fact observable by someone who regularly takes apart cameras and has proven it by examination—see below. OTOH, perhaps stitching is no longer used—I don’t know.
Whatever the cause, it is likely that I got the one bad camera out of a many? Rather, my experience is that quality control at Leica varies, so a “rare” exception might not be so rare.
* “Rare” in my long experience means “not very observant users” generally suffering from confirmation bias and cognitive dissonance to which we are all susceptible, but having been forced to confront those things, I scrupulously question my own findings and have had years of practice to keep my findings provable, avoiding those psychological traps. The usual chain of events is denial followed by days or weeks or months of delay, then a mea culpa. It’s the one pattern that stands out after 12 years of reporting on issues that users and manufacturers would prefer not be mentioned. The best example of this was the Sony shutter shake fiasco—a shitstorm that took a long time to brew up—I was the first one to raise the not so “rare” issue, which was denied by Sony and everyone else, but was 100% reproducible.
Dan Llewelyn of MaxMax.com writes:
Well, first guess is that the sensor is a 'stitched' type.
I noticed this first on the Nikon D700. After removing the AA, one could sometimes see a line right down the center. Some D700's didn't exhibit the problem and others did. So took a D700 sensor with the problem out of the camera and shined a 100 watt halogen dental curing light at a sharp angle to the sensor and could see the line.
Talking to some of my industry friends, they told me that for the D700, the foundry didn't have a big enough machine to make the full frame sensor in one pass, so what they would do is make 1/2 the sensor, flip the sensor around and then make the second half. As you can imagine, you have manufacturing variances and sometimes you can see the line and sometimes the line impacts the picture. You will likely find that it is related to the lens and aperture as well.
I have a few medium format cameras where you can clearly see stitched lines. I have an 80MP Leaf were you can easily see 8(!) stitched sensors. Curiously, they are doing something in the software/firmware because I don't see it when I take a picture. Or maybe it is just the combination of the lenses and apertures I have used.
Seeing the line with the sensor in the camera can be hard. With the ICF and coverglass in the way, it is hard to get a high intensity light source at the right angle to see it. Sometimes you can see it looking at the sensor and moving it around. Polarized light probably helps.
If the foundry they are using has a max size of APS, the only way to make a larger sensor is to make it in two or more passes.
The AA filter helps hide the stitch, and some sensors stitch better than others. First time I saw the problem was after a D700 conversion. A customer sent back his camera for me to look at a by chance I noticed the two sides reflected light slightly differently. Then I researched and found that the D700 was being made in an APS foundry. Some of the D700's didn't have the problem or the customers didn't notice. I would guess the problem would be more apparent on wide angle small aperture shots. Something you might want to test. I wouldn't be surprised if you find a pattern.
Here is a picture of a D700 sensor with the sensor line issue that I took 10 years ago. If you look carefully, you can the two sides of the sensor. With most pictures, you didn't see any problems. But sometimes, you could.
DIGLLOYD: hence my working theory on the Leica M10 Monochrom bifurcation.
But are sensors here in 2020 still made in two passes?
Roy P writes with another working theory:
Regarding Brent N’s “channel” theory: then how does it answer the bifurcation you saw in the M9, which did not have dual readout channels?
The architecture of a CCD makes only a sequential read possible, like a shift register. So every row probably had its own readout channel, and each pixel data had to be shifted out in a serial communication style. (Which is what made Live View impossible with CCD sensors, btw.) So a bifurcated image would be extremely unlikely with a CCD sensor for hardware reasons.
At any rate, half the data path going to one side vs. the other half going in the other side is irrelevant. That does not lead to two different kinds of behavior. All the sense amplifiers, buffers and output registers across the sensor chip will all operate within the same process window across the chip. The differences between them, meaning the across-chip line-width variations (ACLV) are purely statistical and random. If there is any systematic variation (e.g., certain lines always print thinner or certain contacts consistently print smaller, etc., those would occur across the chip, and the fab lithographers would see and fix them. So this guy maybe right with respect to the sensor having two readout channels, but that cannot be the reason for the bifurcation.
That got me thinking there maybe something in the way the shutter works in the M cameras. The typical Copal shutter has the front curtain dropping down from above to open the window to the sensor, then the rear curtain doing the same after the exposure time. For high shutter speeds, the rear curtain starts dropping almost immediately, so you have a open slit that essentially sweeps the sensor.
But that movement is typically vertical, so I was wondering if Leica might have implemented a horizontal curtain movement in the M cameras. So I did a search and to my surprise, came across this video that does show a horizontal sweep of the curtains:
My guess is, there’s something systemic in Leica’s implementation of the shutter that could be causing the problem. For instance, an infinitesimal pause in the rear curtain at the midway point, perhaps. There could be some kind of a dampener around the mid-point of the frame to regulate the velocity of the curtains, and if there’s a difference in the way they act on the front and the rear curtains, that could manifest as an exposure difference in one half of the frame. If such a problem existed in the M9 (or an M3, for that matter), it could be a legacy problem continuing in the latest M cameras today.
DIGLLOYD: I like that this shutter hypothesis is at least plausible. Except that it isn’t: the Leica M10M has a standard focal plane shutter that travels vertically, not horizontally. This can be seen by taking of the lens and operating the shutter, as well as in its structure.
Setting the direction of travel aside, even if the shutter were somehow at issue , then it offers up another idea to explain away how many users would not see the issue: it could be related only to a range of shutter speeds. The examples I posted were at 1/4, 1/30, 1/45, 1/60, 1/190, 1/180. If the effect were confined to some range, it would make its occurence less common and thus it would be observed far less often. But that’s a pretty substantial range.
- The Leica M9 shutter is not the same as in the Leica M10M , the M10M and more recent cameras having damping. Yet both cameras show the bifurcation issue.
- Pixel-level precision as to the bifurcation (exact center) across a range of shutter speeds does not seem credible, the shutter being a mechanical device (plus it’s traveling vertically!).
- A shutter speed of 1/4 second seems to rule out the shutter hypothesis, because the shutter will fully open during exposure to the tune of about 1000 times longer than the shutter travel (or something far longer).
Though I’m no longer a paid subscriber to your website, I still check your valuable articles from time to time.
After reading your article on sensor bifurcation, I have checked a lot of images taken with my Leica M10-R. Yes, the issue does exist. In fact, if you push dehaze and clarity to maximum in Lightroom, these’s a good chance than you’ll see the vertical line, especially in a landscape shot with sky across the image horizontally.
Though in real life I will never push this far in LR, and the bifurcation won’t happen every time, it is still very disappointing to have a flawed sensor which Leica might use for at least 4-5 years to come.
I have also checked some images from Leica SL2, and have noticed a similar bifurcation issue, with texture, clarity and dehaze maxed out in LR. Since I don’t have a Panasonic S1R, I am not able verify if this issue exists on S1R as well. You may download the DNG file from the dropbox link below. The image was taken with SL2 and Apo-Summicron-SL 35.
DIGLLOYD: it makes sense that the M10-R would also have the issue since it uses the same sensor but with the color filter array, which would tend to mask things more due to demosaicing.
Any non-Leica cameras with this issue? I’ve never seen it except with Leica...