Yesterday morning (Sep. 21st, 2016) I attended a few sessions of AutoSens 2016 in Brussels. It is a new conference organized by the people who started and grew Image Sensors Auto at the time they were still working for Smithers. AutoSens was very well attended and was very well organized in a great setting, namely in the Auto-Museum in Brussels. Excellent choice !
In the morning sessions, there were 3 papers related to image sensors. Pierre Cambou (YOLE Development) talked about new developments in mobile and their spin-off to automotive applications, Daniel van Nieuwenhove (Softkinetic) gave a nice overview of 3D imaging technologies and how the Softkinetic’s solutions fit into this landscape, and Tarek Lule’s (ST) presentation was about a HDR flicker-free CMOS image sensor. The latter one had the most technical information, although Tarek did not give any details about the pixel architecture. But what I understood from his talk is the following : the pixels are making use of multiple photodiodes :
- a large photodiode is capturing information in an continuous mode during the exposure time, with a high sensitivity and is basically “looking” after details in the darkest parts of the image,
- a small photodiode that is capturing information in a chopped mode : during the exposure time the photodiode is active during short periods of time and is inactive during the remaining time of the chopping period, it sums the signals obtained during these short periods of time. In this way, information is collected AT THE SAME TIME as the large photodiode, but because of the size and of the chopping, the diode “looks” after details in the mid-range of the image, motion artefacts can be avoided,
- a small photodiode that is capturing information in a chopped mode : during the exposure time the photodiode is active during VERY short periods of time and is inactive during most of the time of the chopping period, and it sums the signals obtained during these very short periods of time. In this way, information is collected AT THE SAME TIME as the large photodiode, but because of the size and the chopping, the diode “looks” after details in the high-range of the image, motion artefacts can be avoided. The chopping frequencies for the two smaller diodes is the same, the only difference is the duty cycle of active and non-active.
Apparently the pixel needs three photodiodes, but because of the chopping, the work that needs to be done by the two smaller photodiodes can be done by a single one in combination with an appropriate time-multiplexing between the short and very short active times within the chopping period. So, the pixel is based on two photodiodes in combination with a few storage nodes. More information was not revealed …
Conclusion : clever idea to make a flicker-free imager with a high-dynamic range (quoted 145 dB). Not many performance numbers were given, but the overall working of the device was shown by means of a video. Looking forward to learn more about this device !
Albert, 22-09-2016.
Hello Mr. Theuwissen,
thanks for reporting from AutoSens.
Regarding ST image sensor I guess it’s kind of a split pixel with three photodiodes which share a single floating diffusion.
Assuming that each PD has a transfer gate, the exposure of the PDs can occur simultaneously but the readout will happen sequentially.
Still I’m not sure whether this technology is really removing or “just” reducing the flicker.
Pulsing/chopping the small photodiodes is certainly a clever approach which on one hand controls the sensitivity of the sub-pixels, on the other hand avoids big intensity variations (frame-to-frame) of LED light sources.
But the combination of frame rate, exposure time and flicker frequency still may produce small (or not so small) intensity variations of LED light sources – especially in daylight condition this might be challenging.
Although these intensity variations may be visible by human eye, the image processing algorithms may not suffer to much.
Just an example: even if the intensity of a red traffic light (LED) is oscillating in certain range, the processing algorithms can detect it as red light.
However, this approach seems quite interesting and I’m also keen to read more about it.
Kind regards
Goran
Goran, first of all thanks for your comments. It is always very much appreciated to get some feedback from my readers.
Well I thought I clearly understood that the pixel is using 2 PD’s, one for the short exposure and one for the two short multiplexed exposures. If that is the case I think the pixel needs to have at least 2 FD nodes, because the multiplexed sampling (by means of 1 PD) needs two storage nodes to store and accumulate the collected charge. But once these two signals are readout, the FD nodes are becoming available for reading the “full-length” exposure. So I think 2 PDs + 2 FDs can do the job. Whether this is also what is used in the ST pixel remains a question mark. Hopefully we will soon get the answer at one or another technical conference. Regards, Albert.