Confirmation Bias and Image Sensors

If you want to discover some photographer's "religion," just ask a question about which image sensor is best. You'll almost immediately be rabbit-holed into a warren of comments about dynamic range, color model, high ISO capability, read noise, well size, and a host of other things for which the person making the response probably has no idea how the details they worship actually work.

D3 sensor

For years now we've had the debate about whether Nikon designs its own sensors. My response has pretty much always been that "they do," and that they often do so in some form of coopetition with Sony. Some people who should know better have disputed my contention. 

Now comes David Etchell's sponsored article about Nikon's sensor design group which gives us new details, and wham, the whole confirmation bias informed debate has started all anew.

I'll start with an important note: not a single person who is making all those comments or describing what is happening in image sensors—including myself—knows for an undisputed fact what is or isn't going on within the deep-levels of sensor design in Japan. The exact relationship between Nikon and Sony, Fujifilm and Sony, Olympus and Sony, and any other pairing of vendors that comes up for image sensors involves (apparently) highly sensitive information and IP ownership that is only known to the two parties.

I put apparently in quotes in the previous paragraph for a reason: in Japan, culture comes into play. For Nikon to say there was any Sony technology in the Nikon sensors would be to lose a bit of face, and that would tantamount to publicly acknowledging shame, something you try to avoid in Japan. Likewise, for Sony Semiconductor to say that there was any Nikon technology that worked its way into the Exmor chips would be the same. 

Almost certainly, both those things have happened. I'm also sure that neither party will ever acknowledge that. 

Something we do know: Nikon does not have a plant in which they manufacture image sensors. They have used Renasys, Sony, and Toshiba for that in the past for DSLRs, and perhaps others. Sony Semiconductor also is known to manufacture sensors designed by others. That's even more true after they bought the Renasys and Toshiba fabs. 

In terms of physical manufacturing of an image sensor you need a fab to do that, and the number of companies who have fabs that currently produce the final image sensors for dedicated cameras is pretty small. Mostly Canon, Sony, and TowerJazz (CMOSIS appears to be fab-less; I'm not sure about whether TSMC is producing dedicated camera sensors).

And this is where a lot of the confirmation bias starts to enter the picture. In particular, Sony Imaging (camera) users seem to think that because Sony Semiconductor makes sensors for others, that only Sony Imaging gets exclusive access to all technologies and the "best sensors." Put another way, many Sony enthusiasts are claiming that only Sony cameras get the best sensors produced by Sony Semiconductor.

First, it's clear that while Sony Semiconductor has done "exclusive deals" for specific sensors—the 36mp sensor first used in the Nikon D800 comes to mind—those don't all accrue to Sony Imaging. But I'm much more amused by the folk who point to a specific "test" result as an indication of "best." This is classic confirmation bias: find something that supports your position and claim that this is "proof." 

For example, recently some Sony users have been using DxOMark's Sports score to indicate that "Sony cameras are better than Nikon cameras using the 'same' sensor." Hmm. What happens if you sort by DxOMark's Landscape score? Oops. Suddenly the Nikon sensors generally seem to be better.

When you isolate the entire characteristics of a sensor down to one arbitrary test, you get arbitrary results. Hey, if that supports your confirmation bias, you win! 

Nonsense. 

I've been analyzing sensor performance—actually: maximum achievable image quality—for almost 25 years now (yes, that's a correct number). There's much more nuance to that than you'll ever see in most discussions of sensors on the Internet, and I know of no single score that is an absolute determinant of whether one sensor is better than another. Like lenses, sensor designs are ultimately a compromise between a lot of different competing factors.

The D5 sensor, for instance, is way down at #32 on DxOMark's "Overall Score." (For you Sony fans, the A7Sm2 is #44, which might tell something about DxOMark's bias. ;~) Yet for images taken in the ISO 3200-12800 range, I'm not sure I'd pick a camera other than the D5. Moreover, most of us shooting a D5 regard it as an incredible sports camera, yet in the "Sports" category for DxOMark it's #30. What? 

What David's article on Nikon's sensor team—I'm not sure David actually was given access to the full team, as I believe there is a group working purely on future IP that doesn't seem to be included in his article—says about Nikon is what I've known to be true for a long time. For every camera Nikon has produced, there is a dedicated team working at extremely low levels to make sure that all the sub-components—lens, filtration, sensor, ADC, post processing—are optimized to the design goals of the camera itself. 

I'll acknowledge that a fair amount of what Nikon disclosed to David was work at what I call the "toppings" level. That's the Bayer filtration, microlenses, and the UVIR/AA filtration layer that lives above the silicon. It's less clear what is happening within the imaging chip itself. Given Nikon's demonstrated use of a CAD program—though the images on those machines in David’s articles are all of older FSI sensors, not current work—they could be working with low-level circuitry component libraries (photo diode, ADC, etc.) whose source of origin is Nikon, Sony, someone else, or a combination of these. A lot of behind-the-scenes licensing goes on in the semiconductor world, and most of that happens completely out of the visibility of outsiders. 

I've never held the position that Nikon has created every element that goes into its image sensors from scratch. Clearly that's not true. The dual gain thing first showed up in an Aptina sensor for the Nikon 1 and based on patents, was Aptina's intellectual property. The phase detect capability in those Nikon 1 sensors is a little more difficult to decipher, but it could have come from outside Nikon, too. So what? 

For some reason people don't get into this debate about Apple's iOS processors. Those chips—currently at A11 Bionic—have licensed CPU cores from ARM, and used to have licensed GPU cores from Imagination Technologies. The chips themselves are manufactured these days by Taiwan Semiconductor Manufacturing Company (TSMC). Do people claim that these are just ARM chips, or TSMC chips? Nope.

And for that same reason you shouldn't really say that Nikon image sensors are just Sony chips. Just as Apple is twisting the ARM design library (and now its own GPU design) to specific product goals, I believe that Nikon has long been twisting image sensors to their product needs. Sometimes that's with relatively simple changes, mostly in the toppings, but often enough in the deeper etchings in the silicon itself. 

Frankly, while I report what I know about the various Nikon sensor designs, I don't much care about who did what. What matters to me, and should matter to you, is whether the camera built around those lower level components actually performs in ways that make it unique and desirable. 

Moreover, I think it unwise to discount all the contributions that Nikon has made to the overall sensor market, even when it wasn't necessarily all Nikon IP in the underlying product. People seem to forget that the much maligned D2H was actually one of very first usable CMOS-type image sensors (Canon had one, too) and that it was probably the D2x that brought Nikon's best CMOS work over to Sony Semiconductor. Likewise, phase detect on sensor and dual gain were first in ILC cameras from Nikon. 

Curiously, it wasn't long after the D2 series that Nikon's sensor design group seemed to go into "secret mode" and stopped publishing papers at imaging conferences. Prior to that, they had been fairly visible in the sensor seminar world and in technical publications. 

I suspect some of this had to do with the "shame" of the D2h not being a "success." Frankly, I think it was the 4mp bit that was the real disappointment in that camera, not the CMOS design. Several key individuals were moved to an office with a view after the D2h, and I think that all the noise (pardon the pun) that Nikon had made about the characteristics of the D2h sensor was seen as a mistake. Since then, Nikon has been much more discrete about saying anything about image sensors, even when it was clear they designed them themselves (e.g. D3, D3000, etc.).

Nikon's opening the kimono a bit for David is a sign that perhaps Nikon is starting to believe they need to make more emphasis on the sensor work they do to make the best possible imaging system with each new product that comes out. 

Nikon used to have a marketing slogan: At the Heart of the Image. I always felt that one was one of their better slogans. The problem, of course, is that they didn't do enough to "prove" the slogan. As David clearly describes, Nikon's sensor design team is right at the heart of the images the Nikon cameras produce. It's time that Nikon's marketing department start stitching together the bigger picture and nursing that heart back to health.

The sponsored article is a start. 

text and images © 2018 Thom Hogan
portions Copyright 1999-2017 Thom Hogan-- All Rights Reserved
Follow us on Twitter: @bythom, hashtags #bythom, #dslrbodies