Comments on: A Quick Idea – Image Sensor Based on Time-to-saturate

By: Gerbils

Gerbils — Sat, 26 Mar 2011 02:45:44 +0000

See the following, they seem to be along the lines of your idea.

1) “A TIME-BASED CMOS IMAGE SENSOR,” Qiang Luo John G. Harris
http://www.cnel.ufl.edu/hybrid/_private/publications/tcmosimager_01329135.pdf

2) “A time-to-first spike CMOS image sensor with coarse temporal sampling,” Qiang Luo, John G. Harris and Zhiliang J. Chen

3) US Patent 6069377

By: Matthew Saville

Matthew Saville — Sun, 30 Jan 2011 21:34:08 +0000

The place where technology like this might be the most feasible would be in 4×5 large format scanning backs. They’re already kinda throwing shutter speed out the window. ALSO, the entire sensor is just a few rows of pixels, so the physical circuitry and programming involved would both be a LOT simpler. You should contact the Betterlight guys with this idea!

By: Matthew Saville

Matthew Saville — Sun, 30 Jan 2011 21:17:43 +0000

I think the idea is brilliant for landscape photographers, although I often find myself shooting at 1-30 second shutter speeds when the light gets really low. If I’m not mistaken, shooting at 10 seconds and wishing to record an extra 2 stops of detail in shadows would require what, 40 seconds? And that’s not even filling a photosite to 100% full, that’s merely adding ~2 stops of DR. Plenty of times when I’m shooting at dawn, dusk, or especially at night, I could be making a 30 second exposure and STILL have completely black shadows in certain areas of the image. I would hate to think how many seconds / minutes / hours those photosites would take to fill all the way up! Of course, I’m sure your solution of a cap-time would work. It would essentially just be adding a few stops of DR to the image, *NOT* creating a 100% white image with time values for brightness calculations. Dunno if that’s possible though.

And of course, either way it’s DEFINITELY gonna hafta be tripod-only for the type of work I do at f/8 with a polarizer etc, where my average shutter speed is a 1/10 or 10″ even in half-decent light…

I’m sure the main problem will be in designing the circuitry for a sensor that has individually sensitive pixels. Having a different “shutter speed” for each pixel individually could be a monumental programming task…

The closest thing, and an indicator that you may be on the right track, is a sensor like Fuji’s legendary DSLR sensor, with completely different types of pixels on the same sensor. Using the same “shutter speed” but different photosite size / type, you definitely get added DR as Fuji has proven. I think their S3 or S5 still tops the charts for it’s JPG default output DR.

My bet, however, is that this concept is as probable as a digital B&W or digital panoramic sensor. ;-) What we’re more likely to see is, improved DR using different ISO’s, through simple advances in current “Active D-Lighting” technology. If for example I shoot a landscape at ISO 100 or ISO 50, but the sensor can boost shadow sensitivity to 200 or 400, well, that’d be decent. Or if I shoot a D3s at ISO 1600 haha, I could gather better shadows by blending ISO 3200 and 6400 into the shadow pixels… Not perfect, but a small advantage with no jeopardizing of shutter speed for low-light sports / action photographers.

=Matt=

By: Robert

Robert — Sat, 19 Jun 2010 09:06:27 +0000

I think the largest issues are the dynamic scene and the time recording for each pix.
I am also a layman interested in improvement of dyna-range and noise. Wondering if you know how long it takes to complete 4 shots in Fujifilm’s Pro Low-Light Mode? I mean the time starting at the beginning of the first shot till the ending of the 4th shot? Thanks.

By: Sloan Lindsey

Sloan Lindsey — Mon, 26 Apr 2010 20:15:29 +0000

The largest issue I see with this method is in recording the time to saturation.
Using your number of 7.6ns gives us a polling rate of about 131mhz. We need to be able to check the state of the pixel at that rate without adding any adittional noise. I’m not sure that is feasible as we need to cycle the entire chip at that rate which will likely increase the temperature. Its an interesting idea though.

I found your post looking for information of another idea that I had. What if like in your scenario you only measure full saturation but instead of sampling over time you sample over space. Imagine an array of pixels with decreasing saturation limits (for simplicity assume halving). If in any time interval we reach saturation we record a bit. Now we have a 1 bit posterized image of the places where light hit with greater than a certain intensity. Now the section with half as much sensitivity gives us a one bit image of where the light hit with twice as much intensity. We continue building this array with as many steps as the total number of stops we wish to capture letting us choose the amount of dynamic range we capture. Now we have a heavily posterized image (film was binary in this manner as well) we can sub sample with another binary string offset by 1/2 and get twice as many gray scale graduations.

The merit of this idea is that it allows for very high resolution DSLR sensors are not feature size limited (as shown by 1/2.3″ sensors with comparable resolution) the limiting factor for this system is then decoupled from the read noise and dependent only upon thermal noise. Furthermore if we build our sensor in such a way that we can program the initial charge (like memory and likely rather slow) we can then set the sensitivity and the dynamic range (we can choose our significant figure -> range ratio) on the fly per scene.
Truly though I see the strength of my idea simply to be as a novel idea in order to increase resolution. flash memory has much higher densities than sensors so the resolution is possible. This also opens ideas for different modes of sensing instead of using the photovoltaic effects we could use photoresistive materials to act as our photon counters.