People seem to think that image sensors have reached a plateau, and from that they now also believe that we won't get any more advances. That's not true. But to understand that we have to look at the history of image sensors. In a nutshell, the technology behind the photon to electron conversion has moved through a number of phases over the past 40 or so years:
- CCD
- CMOS
- BSI CMOS
- Stacked CMOS
- Global Shutter CMOS
Underlying that have been other technologies, such as the interface from the sensor to the camera's processor, which these days is likely to be SLVS-EC 16 lane. (That acronym stands for scalable low-voltage signaling with embedded clock.) Prior to SLVS-EC we used to have LVDS (low-voltage differential signaling) or other less-capable technologies handling the transfer. Those of you with really long memories will recall that during the CCD to CMOS transition I wrote that one of the reasons why CMOS was going to be the eventual winner was because of all the added circuitry that everyone wanted to add to the image sensor itself. The transmission circuitry is one of those, and has progressed from simple to complex in short order.
I bring up the transmission protocols because there are quite a few different technologies incorporated inside an image sensor. At one point, we had external ADC (analog-to-digital), but today we have internal column-based ADCs that have been going through several of their own technology marches forward (such as the introduction of dual gain).
Intel used to refer to their chip cycles as Tick-Tock. In Intelspeak, a tick was a big architecture change, a tock was a process size change. There's a lot of tick-tocking going with image sensors, though we have to redefine the Intelspeak a bit so that a tick is a clear technology change, while a tock would be an incremental improvement to a given technology.
The above bullet list tends to be ticks. But incorporating 16 lane versus 8 lane (for SLVS-EC) would be more of a tock, and generally speaking, we don't get the tock marketed to us in the camera world, so you might not even be aware of changes that are happening at that level.
Much of the tocking that's been going on recently has been around iterative refinements. Data gets moved a little faster, gain circuitry is finessed, wafer purification is a little better, and so on. Additionally, the fabs work to tock on yields, as that increases throughput of production and drives profitability of production. But none of the tocks are likely to be ones that you'll hear about, since they all tend to produce any visual gains that are so small you aren't going to notice them (yes, even you pixel peepers).
So one reason why everyone believes that we're on an imaging sensor plateau is that there hasn't been a Big Tick lately. That doesn't mean that there aren't any new technologies in the wings.
The Big Tick that everyone seems to expect to happen next is direct photon counting: every individual photon detected when and where it arrives. In the version called Quanta Image Sensors (QIS), there isn't an electron well that has a maximum capacity and has to be measured, drained, and refilled. Dynamic range of a QIS sensor is nearly one photon to however many you can detect while the shutter is left open.
QIS had gotten to 40mp at 30 fps with a 96dB dynamic range back in 2022 from a company formed by Professor Eric Fossum (Dartmouth) and some of his students (Gigajot). If you recognize the Fossum name, that's because it has been on over 150 patents in the image sensor space, including seminal ones for CMOS and BSI CMOS.
Curiously, Gigajot went completely dark shortly after they made that 2022 release, including removing their Web site from public view. Yet the key employees still all show as being at Gigajot on LinkedIn. That would tend to tell me that they were acquired (or invested in by a Big Company) and their new investors don't want them publicizing their work until it's fully marketable (a 40mp camera was available briefly to purchase at research pricing).
Canon has been working on something similar, using a technology called SPAD (single photon avalanche diode). They even have a specialized, US$25,000, interchangeable lens camera called the MS-500 that utilizes this technology (3.2mp, 1"). It has enough photon discrimination to photograph under starlight (0.001 lux).
So it's not that there aren't any new image sensor technologies working their way through the channels. It's that they're still early in their development and not yet proven cost effective enough that they'll produce US$1000-5000 consumer cameras. Moreover, most of your images don’t need to discriminate a few random photons, so there’s a question of how applicable some of the new technologies might actually be to consumer imaging. But the thing is, don't bet against tech. First, tech has to figure out the problem to be solved. Then it has to fiddle with potential solutions. Then it has to work to a proof of concept. Next it has to be functionally produced in low volume (which is sort of where QIS and SPAD have gotten to). A few more steps down the way someone will get to "cranking up the presses" and start popping out new technology sensors in affordable quantity. And at that point, we get new Tick technology in our cameras.
That said, we're still in the BSI to global shutter range for technology producible at consumer level prices that do consumer type of imaging. Stacked CMOS and Global Sensor CMOS are in the middle of their tock phases, where incremental improvements and pricing drops are what you see from each new generation. We haven't yet hit the best level of what those two technologies can do, though the primary improvements from a photographic standpoint tend to all center around speed (fps or more pixels or both).
The bottom line is that as far as I can tell, image sensor iteration is still happening with regularity and at the heart of our cameras are sensors that are tocking away generation after generation. We're marching to the Stack/Global drummer right now, but I'm pretty sure we'll have a new drummer to listen to soon.
