24/05/22EMCCD sensors were a revelation: increase your sensitivity by reducing your read noise. Well, almost, more realistically we were increasing the signal to make your read noise look like it was smaller.
And we loved them, they found an immediate home with low signal work such as single molecule and spectroscopy and then spread amongst microscope system providers for things such as spinning disc, super resolution and beyond. And then we killed them. Or did we?
EMCCD technology has its history with two key suppliers: e2V and Texas Instruments. E2V, now Teledyne e2V, started this rolling with early sensors towards the end of the 1990s but made real strides with the most accepted variant, having an array of 512 x 512 with 16-micron pixels.
This initial, and probably most dominant EMCCD sensor had a real impact and half of this was really pixel size. 16-micron pixels on a microscope collected 6 times more light than the most popular CCD of the time, the ICX285, featured in the popular CoolSnap and Orca series. Beyond pixel size, these devices were back illuminated converting 30% more photons taking that 6 times greater sensitivity to 7.
So effectively EMCCD was 7 times more sensitive before we even turned it on and got the impact of the EMCCD gain. Now of course you can argue that you could bin the CCD, or you could use optics to create larger pixels sizes – it's just most people didn't!
Beyond this, getting read noise below 1 electron was key. It was key, but it wasn't free. The multiplication process increased the uncertainty of the signal measurement meaning the shot noise, dark current, and anything else we had before multiplication was increased by a factor of 1.4. So, what did that mean? Well, it meant EMCCD was more sensitive but only at low light, well that's kind of when you need it right?
Against a classical CCD, it was no contest. Big pixels, more QE, EM Gain. And we were all happy, especially those of us in camera sales: $40,000, please ...
The only things we could have done more with were speed, sensor area, and (not that we knew it was possible) a smaller pixel size.
Then export controls and compliance came, and that was not fun. It turns out that tracking single molecules and tracking rockets are similar, and camera companies and their customers had to control camera sales and exports.
Then sCMOS came, starting by promising the world -and then over the next 10 years nearly delivering it. Smaller pixels getting people the 6.5 microns they loved for 60x objectives and all with lower read noise of about 1.5 electrons. Now this was not quite EMCCD, but against the 6 electrons of the comparative CCD tech of the time it was amazing.
The initial sCMOS were still front illuminated. But in 2016 back illuminated sCMOS arrived, and to make it appear even more sensitive to original front-illuminated versions it had 11-micron pixels. With the QE boost and pixel size increase, customers felt like they had a 3.5 x advantage.
Finally, in 2021 the sub-electron read noise was broken with some cameras getting as low as 0.25 electrons - it was all over for EMCCD.
Or was it ...
Well, a bit of the problem is still pixel size. Again you can do what you want optically but on the same system, a 4.6-micron pixel collects 12 x less light than a 16-micron one.
Now you could bin, but remember binning with normal CMOS increases noise by a function of the binning factor. So most people are happy with their 6.5-micron pixels thinking they can bin their way to sensitivity, but they are doubling their read noise to 3 electrons.
Even if noise can be reduced, the pixel size, and full well for that matter, are still a compromise for real signal collection.
The other thing is the gain and the contrast – having more greys and chopping your signal up smaller does give better contrast. You can have the same noise but when you show only 2 greys for every electron with a CMOS you don't get much to play with when you have just 5 electrons of signal.
Finally, what about the shuttering? Sometimes I think we forget how powerful a tool this was in EMCCD: global shutters do really help and are very light and speed efficient, especially in complicated multi-component systems.
The only sCMOS camera I have seen come even close to the 512 x 512 EMCCD sensor is the Aries 16. This starts with 16-micron pixels and delivers 0.8 electrons of read noise with no need to bin. For signals of above 5 photons (per 16-micron pixel), I think it's the best I've ever seen and about half the price.
So is EMCCD dead? No, and it will not really die until we get something that good again. The problem is, well, all the problems: excess noise, gain aging, export controls...
If EMCCD technology were a plane, it would be a Concord. Everyone who flew it loved it, but they probably didn’t need it and now with bigger seats and flatbeds – just sleep those extra 3 hours across the Atlantic.
EMCCD, unlike Concord, is still alive because some people - a small, ever decreasing number - still need it. Or maybe they just think they do?
Using an EMCCD, the most expensive and complicated widely-used imaging technology doesn't make you special, or an imaging expert - you're just doing something different. And if you haven't tried to change, then you probably should.
