This might be a bit of a long shot for this venue, but: I'm looking into sensors for a scientific imaging application where the key is excellent sensitivity (quantum efficiency and noise performance) combined with low readout latency.
I don't need a lot of pixels, something like 64x8 or even a 32x1 line sensor would be just fine. However, readout of these pixels must be quick, not in terms of frame rate (throughput), but in terms of latency. Ideally, the readout latency would be well below 100 µs, so that together with an exposure time on the order of 50 µs, I'd get the result within less than 100 µs of the trigger event (on a custom FPGA board, so the exact interface does not really matter). Also, the short exposure times, compared with the fact that the observed process is rather faint in intensity, necessitate a very good sensitivity.
Traditionally, photo-multiplier tubes would be used for detecting the process, but my goal is (spatially resolved) imaging instead of just counting photons. For this, electron-multiplying CCDs (EMCCDs) seem to be an interesting option.
To give you an idea what this is all about, take a look at Andor's iXon 897 Ultra camera. It would be a near perfect match for the application, if it was not for the long readout latency of >500 µs. The main cause the long latency seems to be the fact that the CCD operates in frame-transfer mode, so that even if only a small region of the camera's 512x512 sensor is used, the image still has to be shifted through the whole storage area.
Are there EMCCD sensors that do not use a frame transfer setup? Or sensors in an oblong/line shape? The target application uses wavelengths just below 400 nm, so good quantum efficiency on the short end of the spectrum would be necessary, even though latency is the key factor. Whether it is a finished camera product or a bare sensor is ultimately not that important, as throwing together some readout electronics for this specific application would be doable.
Another product I stumbled upon is the FirstLight OCAM² camera. It seems to use a custom EMCCD sensor design manufactured by e2v with several readout stages to reach a latency of under 60 µs. However, the used sensor only has a very mediocre quantum efficiency in the near-UV, and while the website mentions that custom developments are possible, custom-ordering a sensor might be prohibitively expensive.
I still think that I'm missing a much more obvious solution, though.