Sharp's new sensor - just curious

If you like to take a 50mp landscape at ISO 800 F22 hands free, you can wait till 2525.

The rewiewers here are promoting ISO 800 and 6mp that means 10 images (with overlapping) for 50 mp.

We are talking about reducing diffraction, that is needed at present over F11-16.

And you take 200 landscapes per day at ISO 800 to be printed all at 40"x30" ?

Without tripod ?

What kind of photographer are you ?

Buy an Hasselblad back at 39mp ISO 400, use it handheld and let me know.

paulbod3 said:

We are talking about reducing diffraction, that is needed at
present over F11-16.

Click to expand...

Clarity at last! You should have mentioned this earlier.

If diffraction at f/16 is all you are worried about, then you don't need the 0.1 micron jots Eric Fossum described. The resolution of current digicam sensors (such as the Sharp sensor cited by the OP) is more than adequate for this purpose.

At f/11, the diameter of the first dark ring of the Airy diffraction disk is 15 microns. 60% of the energy falls within an 8 micron diameter. Images with 1.9 micron pixel pitch (2.7 micron for green pixels) have sufficient spatial resolution to deconvolve the diffractive blur, and would allow a useful resolution enhancement. The higher the deconvolved resolution, the noisier the image.

HTH
--
Alan Robinson

paulbod3 said:

And you take 200 landscapes per day at ISO 800 to be printed all at
40"x30" ?
Without tripod ?
What kind of photographer are you ?
Buy an Hasselblad back at 39mp ISO 400, use it handheld and let me
know.

Click to expand...

Interesting speculations about my photographic habits. Did you actually read any of the posts in this thread? All I said about my own photograhy was that I mostly shoot at ISO 50 or 100, and that I often shoot landscapes near dawn or dusk.
Most of my other comments related to the practicalities of image enhancement.

Time to move on, as far as I am concerned.
--
Alan Robinson

Agree with you, my only point is that we can work on the image with a precision of 0.1µ whatever post-processing we like to make.

Noise depends on various factors, and do not feel can be worse than with today's sensors.

What I like of the idea of digital film, or in any case of smaller photosites is that :

in any case we are going to have excess processing power soon
also technology will allow smaller photosites

we can find an use for these conditions by clustering as needed the jots
Dr Fossum's paper is a good indication of how this can be done

Digital film can give a lot of flexibility, it is clear that for higher ISO the resolution might be reduced, but still might be redundant or anyhow better than today's sensors.

Using larger photosites because :

noise is lower
diffraction limit
etc
is not needed because the smaller ones can be combined in many different ways.

Perfection will be never achieved, progress is a never ending effort.

It is not that nature is made as science describes it; nature is as it is, and science is the convenient approach to understand and produce better.

I am sure you are an excellent photographer.

http://www.faculty.fairfield.edu/jmac/sj/scientists/grimaldi.htm

http://en.wikipedia.org/wiki/Francesco_Maria_Grimaldi

FM Grimaldi was a far relative to the Genoa-Monaco family.

http://www.scienzagiovane.unibo.it/English/scientists/grimaldi-1.html

http://www.scienzagiovane.unibo.it/English/scientists/grimaldi-2.html

http://www.scienzagiovane.unibo.it/English/scientists/grimaldi-3.html

It is time to find a new approach.

D Q E said:

D Q E wrote:

If you read every jot, wouldn't the clock have to be at a VERY high
frequency?

Click to expand...

I am not sure what "VERY" means to you. Being binary data, the readout would be very DRAM (memory) chip-like. But, this assumes no processing is done on chip. Using column-parallel readout, one has read out of the order of 32k jots in a readout period. If readout is done in 32 msec, then the jot scan rate is 1 MHz which in fact is rather slow for binary data.

If one does the digital development on-chip, that is, turning jots into exposed and unexposed grains, then there is probably at least 10x in data reduction. Getting 100 Mgrains readout out of the chip in 32 msec requires 3 GHz for a single bit port, and if the port is 16b wide, then 200 MHz. This is not blazingly fast either although it will account for most of the power consumption by the chip.

If the conversion of grains to pixels is done on chip, then the data readout requirements are that much lower.

If you need to read out all jots off chip, then some on-chip lossless compression might work well. Scan the chip once to develop a code book, and the second time for read out. On the other hand, in 10 years having 30 GHz data ports on chips will probably be not much of a challenge.

Heh...so maybe we need to think of more jots per sensor. After all, at 0.1 um spacing, the gigapixel array (if square) is only 3.2mm x 3.2mm. Maybe 10 gigapixels would be more fun!

cheers,
Eric

Thanks again for the discussion. I appreciate the calculation too.

One thing to keep in mind, this sensor does not require some nearly-impossible technological breakthru. The only part that needs work is the single photoelectron high gain detection. There are many possible ways to do this, but jot electronic device is still to-be-demonstrated. (I think it won't be that hard.) The rest is straight forward.

The other thing I will just remind our gentle readers about is that I am assuming that pixel sizes will continue to shrink to sub-diffraction limit (SDL) levels. I just wanted to think about what we could do differently if the Moore's Law juggernaut continues. All I have done is propose a digital simulator of film. We know film works great on many levels, so it should be no surprise that if we simulate it using silicon hardware, the digital film sensor will have the combined benefits of film exposure qualities AND electronic digital readout.

-Eric

alanr0 said:

alanr0 said:

Jot solution
----------------
Jots per pixel: 32000
Readout noise: none - this is the crucial, possible false, assumption

Click to expand...

Click to expand...

Readout noise should be zero, or close enough to zero to ignore.

alanr0 said:

alanr0 said:

Photon noise: sqrt(electrons)

Click to expand...

I would expect there to be some leakage effects, particularly as
the potential barriers will be so narrow. Perhaps we could model
this as an equivalent leakage current that lights up a small
proportion of jots at random? Quantum tunneling? Difficult to
estimate without some idea of the device structure.

Click to expand...

Crosstalk is always an issue, and colorized cross talk is really ugly. See the color comment below.

alanr0 said:

Another thought.
How thick does the CFA need to be? At 0.1 micron or below, we
can't afford much space between the filter, the microlens array (if
we have one) and the sensor.

Click to expand...

I think Foveon has effectively proven that any wavelength dependent absorber can work. So pick something where a small thickness variation can make the difference between a "blue" pixel and a "red" pixel. Even silicon could work, but a narrower bandgap material might work better...like Si-Ge? (note this is a passive layer and not electrically active).

alanr0 said:

How about a coarse (0.5 micron?) sub-pixel array. Each sub-pixel
handles a single colour, collected from a number of smaller jots.
We still have fast read-out from the same total number of binary
jots, but we accumulate photon counts from each of the sub-pixels.
This works either for single or multiple read-outs.
We can choose to develop a conventional image by merging variable
numbers of sub-pixels depending on the SNR. Alternatively, we can
develop Eric's digital grains with numbers and/or size depending on
the exposure.

Fun to speculate in any case.

Click to expand...

Yes, good thoughts. And as I have said elsewhere, this is not a wild-eyed idea. It is close to being practical, I believe. However, I am not aware of anyone working on it at this time.

-Eric

paulbod3 said:

The definition in cinema can be improved to a certain extent also
with an higher image rate, in case the single image is too "poor",
and this would not imply a total change of machinery.

On the other side, starting at 2k and then forcing the users to buy
4k, 16k and so on might be a bit expensive for some markets.

1920x1200 is an accepted standard for video, streaming, and
complies fully with the old Zeiss dof calculations, allowing an
easy workflow, but, from what I have seen is a bit lacking in a
large theater, unless with extra processing as you suggest.

Click to expand...

I' not suggesting extra processing, just to maximise the performance of a 1920x1080 raster size - HDTV as we know it uses subsampled colour, drastic compression and component colour models - it's nowhere near the full potential of 1920x1080p.

--
mumbo jumbo

1920x1080 x3?

Interesting concept. I've actually been playing around with it in my head for a few months and yes it looks promising. I think though that you will also generate a variable sensitivity for each jot due to variations in the substrate. Tuning each jot to have a uniform threshold for the single-> small# of photon flip may prove an insurmountable engineering trouble. I still think the idea has great merit though. subsampling is quite interesting especially if you can find some sort of interesting way to mount the sensor to readout per pixel for a full frame read.

If its possible to make the sensor and readout vertical you could then fabricate full wafers of only the sensor then test for defects and carve the sensor up and mount it in some sort of flipchip fashion to a gigantic dumb memory array that does your integation for you.

I'm also not sure off the top of my head what the signal for 1-10photons is compared to the energy of a thermal excitation. This may present problems as we try to accurately measure photon incidents at varying temperatures.

As for colors, there has been recent work at my university with dye tagged nanoparticles that can be size controlled invariant to the color. these nanoparticles seem to self assembly leading to thoughts that one could allow for a random self assembly with some ensemble of colors and then calibrate some sort of ROM to map the spacial sampling back into the color space using colored light to excite some fraction of the photosites(jots). Thanks for the paper it was quite an interesting read and paralleled some of my thoughts on the matter.

As a note though... there seems to be a typo when you suggest .25^2micron to .1^2 micron as I read it as 1^2micron which is larger than the .7 you cited from prior art.