Next week, CMOSIS will introduce the CSI2100 2MP CMOS sensor. Its features include a 1440 x 1440 pixel array with 12 µm B&W GS pixels and a 500 fps frame rate in 10-bit mode. The sensor boasts a a full well capacity of 2 Me-/pixel and is optimized to detect small signal variations in bright images captured at high-speed. This makes it suitable for medical, scientific and industrial applications. The CSI2100 evaluation kit was co-developed with Pleora Technologies.

FWC of 2 Me-/pixel? really?

ReplyDeleteWhy not? In 12um pitch, you can put 200-300fF of cap equivalent to 2-3Me- FWC. They might not be integrating on SFD node.

ReplyDeleteMay I know the power consumption? That will say a lot!

ReplyDeleteIt's a simple 3T pixel.

ReplyDeleteIs this from one of IISW 2015 papers? Which is 700fps, 2Me- and1400*1400.

ReplyDeleteLooks like good news for shot-noise-limited applications where sufficient light can be made available - of which there are quite a few.

ReplyDelete10-bit ADC cannot assure shot-noise-limited applications.

Deletewhy not? SNR_ADC (=2^10/sqrt(12)) > SNR_pixel (=sqrt(2M))

Deletehi, not sure you did the arithmetic.

Deletesnr of ADC is 1024/sqrt(12)=295.6 while sqrt(2M)=1414.

Pretty sure that the 10B ADC cannot provide shot noise limited detection.

signal is 1024 LSBs, noise is 1LSB/sqrt(12), signal to noise is 1024 /(1/sqrt(12))=3550! CLEAR????

DeleteI'm yet to meet a single ADC that has a SNR greater than the number of LSBs it can represent. Most have a quantization noise that is < 0.5LSB and most provide much worse performance, especially on-chip devices.

DeleteI think you should bring this ADC wonder to market ASAP! I'll be the 1st in line to buy it!

correction

Delete... noise that is > 0,5LSB

Well, not all are that bad. For instance, look at Intersil ADCs. Performance of some of them is quite close to an ideal one:

DeleteISLA118P50: 8b physically, 7.99 ENOB - close to the ideal, right? And this is in mass production.

http://www.intersil.com/en/products/data-converters/high-speed-a-d-converters/a-d-converters/ISLA118P50.html

I agree, some ADC's are nearly ideal. And the 7.99 ENOB for an 8bit ADC is the best I've seen. thanks!

DeleteMy beef is the suggestion that a10b ADC could have an ENOB that is bearly 12bits, which is what was suggested by the original poster... and hence could provide a shot noise limited detection capability for a 2Me- full well.

All good.

You still confuse LSB with q. noise and you still confuse Telekom ADCs with IS ADC...Can some experts explain this for all? Cmosis, Albert, Erik?

DeleteAlbert has a blog post which I think covers the confusion. The ADC does not cover the full range of the sensor only the linear portion. http://harvestimaging.com/blog/?p=1238

DeleteI beg your pardon... but since I'm in the business of designing and building these so called "IS" units, I'm almost certain that I do not confuse anything right now.

Deletebottom line is that if you have an ADC with 1024 LSBs, the best it can do is display a full range of 1024:1. Even if the ADC had zero noise, it makes not difference since it can't display anything between 0 and 1, 1 and 2, 2 and 3, etc. Hence... it can't convey information about 1440 levels of brightness.

And since most ADCs don't come anywhere near ideal numbers (12B ADCs have typical ENOB's of 10, 14B ADCs have a bit over 12 and 16B ADCs a bit over 14), and since most CIS ADCs have similar numbers, I don't see how a 10B ADC can ever provide enough dynamic range to handle 2Me.

Measure a number of times and average the result?

DeletePixel noise is the standard deviation measured over number of frames so the noise can be any number between zero and one.

Delete2 Me/pixel sounds hyperbolic! How does one ever reset this photodiode? For reasonable voltages available on the chip it is impossible to fully deplete this PD. And too high voltage triggers other leakage mechanisms!

ReplyDeleteMaybe because it is not a Pinned Diode ?

ReplyDeleteMay I know readout noise of CSI2100?

ReplyDelete