Image sensors are a semiconductor, and as such they're a bit prone to Moore's Law. Not completely, as the photo diode portion of the sensor hasn't tended to gain from miniaturization of the electronics, but certainly the rest of the heavy lifting that an image sensor does benefits from process size reduction.
Process size reduction has helped drive the ADC (Analog-to-Digital Circuit) from off-sensor to on, has increased bandwidth rates, has given us dual and multi-gain, has made changes to read noise and well saturation capabilities, and much more.
Other semiconductor techniques, particularly those centered on precision and purity as well as yield improvements, have made our sensors very low in PRNU (Photo-Response Non Uniformity, or pattern) noise, and cheaper to build.
While there's a lot of speculation and excitement about completely different sensor approaches—for example, Fossum's Jot experiments, now turned into a startup at Gigajot, that try to capture where a photon came from and when it arrived—the big steps forward happen fairly far apart and don't always give us exactly what we're looking for when they first arrive.
Indeed, the industry is really still working on getting everything they can out of the switch from CCD to CMOS structures, coupled with moving to smaller process fabs and a few other innovations, such as BSI (back side illumination) and stacked semiconductors.
In the near term, therefore, we're still in a bit of the "more of the same" type of situation.
Overly generalized, we're approaching the limits of single sample Bayer-type capture. At the highlight end there's a limit to how many electrons you can store, especially as we size cells smaller, while at the shadow end we're already mostly limited by the randomness of photons. That's not to say we won't continue to see movement forward on dynamic range, but the increases now don't tend to be dramatic, but smaller movements upward, often made by small tweaks to the on-chip circuitry.
Sony's recent publication of the road map for the medium format sensors—100mp for the smaller size currently used by Fujifilm, Hasselblad, and Pentax, and 150mp for the larger size, with BSI slowly moving into the line—is actually fairly indicative of where we are right now.
Yep, more pixels. More movement towards BSI and advanced chip-making techniques that started with smartphone image sensors.
Those two things go a bit hand in hand.
In front side illuminated sensors (FSI) some of the top area of an individual photosite is not collecting light due to the way the data and power lines and on-site electronics are created and take up space. Light ends up having to go down a vertical corridor to be captured. In back side illuminated sensors (BSI) you basically make something similar to your FSI sensor, flip it over, and then grind off the extra silicon so that the full photosite area is collecting light, and basically at the surface or the sensor.
Making smaller photosites (more pixels) generates less light hitting the photo diode in FSI sensors, but making them BSI lets more light hit the photo diode. Or you could keep the same light-gathering area of FSI and put more pixels on the chip using BSI. In other words, if you want to increase pixel count or improve the basic dynamic range capabilities, you need something like BSI. Balanced carefully, you can gain some pixels and some DR (dynamic range) on a BSI sensor compared to an FSI one.
Thus, when you look at Sony's chart for medium format, you see BSI (they shorten it to BI) slowly coming into the sensor.
I suspect we'll see this same move replicated in full frame, as well. Indeed, the BSI 42mp sensor used in the Sony A7RII is indicative that this approach has already started. Couple that with other small changes in the sensor—copper wiring, multi-gain, etc.—and the camera makers are essentially giving us more pixels while keeping or improving the basic capture attributes of the sensor.
So don't be surprised when we see new 46-70mp sensors appear in full frame cameras in the near future. Maybe we'll hit 100, but I think 70mp is a safe guess of top end for the moment, as it is well within the predicted bounds of the technologies working their way up the sensor size ladder (e.g. BSI started with phone sensors).
Thing is, the sensors aren't getting less expensive to make. BSI and Stacked designs add silicon and steps to the sensor creation process, increasing costs, at least until careful fab optimization eventually leads to higher yields and faster creation.
The three primary purveyors of full frame cameras—Canon, Nikon, and Sony—all are moving forward with the more pixels approach, though with some caution. The question is at what point does the pixel count really net you useful gains for your primary customers? Diffraction, after all, eventually takes a bit of a toll, basically putting an anti-aliasing function back into cameras that have had that taken out. In theory, this is something that should keep medium format separate from full frame, just as it did in the film era. A 50mp medium format sensor is going to have a diffraction advantage over a 50mp full frame sensor—though lens choice could impact that slightly, as medium format doesn't tend to have as many very fast lenses.
Still, it seems clear that customers will still snap up more pixels, just as they snatch up more lens reach at both ends, too. More is more, after all.
So for full frame, the march continues: more pixels and more DR if they can scratch that out. The costs of doing that for a platform that isn't high volume means that full frame prices for state-of-the-art cameras aren't coming down any time soon. We may get simpler sensor designs—e.g. the Sony 24mp that's in a number of products—in cameras that come down in price. But if you're pursuing ultimate quality in full frame, I would tend to say to expect small price increases.
But look at APS-C (DX).
You'll note that everyone seems to be stopping in the 20-24mp pixel count realm for the moment (Samsung briefly gave us 28mp). First, that gives you some product differentiation for the top end full frame cameras. By not continuing to rapidly ratchet up the pixel count in APS-C, this is giving some breathing room for cameras like the 5Dr/s, D810, and A7RII and their successors.
More importantly, not pressing the APS-C (DX) sensors quite as hard up the technology tree as full frame means that you can keep pricing down. I believe that the camera companies have finally figured out that they should have flat fours and V6's in their lineup (with a few giving us V8s and V12s). The benefits of making less complicated, less powerful imaging engines give the camera companies excellent lower cost models (APS-C/DX), while there are always those who will buy the "give me all you've got" approach (full frame/FX). Plus, then there's the relatively small group that will opt for the expensive exotics (medium format).
But there's nothing stopping the same pixel count push from returning to APS-C. You can certainly drive to 36mp or even 54mp in the crop sensor size. It's just that the gains start to be less visible. That can be compounded by the fact that you're trying to keep the crop sensor cameras reasonable in price, and thus have also created "reasonable" lenses, not super high performance ones.
Of course there are sensors smaller than APS-C/DX. The problem with that, as I wrote over a decade ago, is the knife-to-a-gunfight one. This is additionally complicated by the smartphone makers now embracing multiple-sensor cameras, which gives them larger sensor performance in a smaller sensor, just multiples of them.
The law of volume works for the smartphone makers. We're now up to billions of smartphones shipped. In the US, we're somewhere north of 55m smartphones sold a year. Compare that to 3.1m compact cameras and 2.8m ILC cameras. Almost an order of magnitude volume difference (and it's slightly more than that because I'm using North America figures for cameras, which includes Canada, but only US numbers for smartphones).
Simply put, the push for better sensors happens at the bottom now because that's where the money is. Already this has completely distorted the compact camera market. Even the 1/2.3" sensors don't hold up well against smartphones these days if all they're put behind is a plain vanilla camera and lens. Thus, the remaining "market" in compact cameras tends to be in waterproof cameras and cameras with huge lens ranges. And the former will go away within a generation or two of smartphones is my guess.
We're now in a world where 1" is the minimum size for compacts. Nikon just told the world that they're not sure that even the 1" compact world is profitable and protectable by cancelling their much-anticipated DL series. I think they were a little premature with that proclamation, but certainly a few years down the line it will be true.
All of which is making 4/3 the smallest viable sensor size. That should scare the daylights out of Olympus and Panasonic. They've already seen their compact camera business gobbled up by smartphones, now they're once again on the front lines of the war. Sony, too, has plenty of troops on that front, though they've been trying to defend themselves by upping the technology ante aggressively. Sony brought BSI to 1" sensors early, and now has added the even more expensive stacked sensor to the mix.
So here's where I think we are:
- Compacts — Long focal length lenses will soon be the only safe zone, because it's the only "more" the camera makers will be able to claim. The complication here is that making really long zooms is easier and cheaper to do with smaller sensors, so you essentially give up on trying to beat the image quality of smartphones and instead only give a huge optical range to beat them. Note to camera makers: you need something better than SnapBridge to get those images over to the smartphones or else even the long-lens market dies.
Outside of long lenses, the only rational choice moving forward is APS-C/DX or larger sensors with highly competent lenses. The Fujifilm X100 and Ricoh GR approach, basically. Nikon almost joined the crowd with the Coolpix A, but managed to overprice that product and under equip it. Sony joined the crowd at the highest possible end with the RX1 (and Leica with the Q). What we haven't gotten is a basic 24-70mm (or even 35-70mm) zoom in such a product. In other words, a more flexible large-sensor compact. - Mirrorless — If you haven't noticed, mirrorless managed to try everything from 1/2.3" (Pentax Q) sensors to medium format ones (Fujifilm GFX, Hasselblad X1D). At the moment, only two Japanese makers—Fujifilm and Sony—have a multi-size sensor strategy going in mirrorless. Long term, I think that's the only viable approach: you need a smaller sensor size for "affordable" cameras and a bigger sensor size where you stretch all the technologies to the max for the "high-end" cameras. Anything else and you're a one-horse wonder, and we all know how that story ends.
- DSLRs — Realistically we've only got two viable entries here, Canon and Nikon, and both have the same problem: they've chosen APS-C/DX for their "affordable" line and full frame/FX for their "high-end" line. Note where I write about APS-C/DX stalling at 24mp. That has to be intentional. Because there's only a one-stop differential between the two sensor sizes, the performance differential isn't obvious unless you start separating the two arbitrarily. That's being done in pixel count at the moment, but it really should be done in all forms, I think. APS-C/DX is already good enough to hold serve in its present form—especially if you get the lens set right, buzz, buzz—while the larger size needs to stretch its advantage out a bit more.
