Quantum Film Might Replace CMOS Sensors

An anonymous reader writes "Quantum film could replace conventional CMOS image sensors in digital cameras and are four times more sensitive than photographic film. The film, which uses embedded quantum dots instead of silver grains like photographic film, can image scenes at higher pixel resolutions. While the technology has potential for use in mobile phones, conventional digital cameras would also gain much higher resolution sensors by using quantum film material." The original (note: obnoxious interstitial ad) article at EE Times adds slightly more detail.

Comments

Re:Sensitivity is not Resolution

[forkazoo]

Couldn't one lead to the other? Would averaging 4 noisy pixels give you a better light sensitivity than just having the one?

To a certain extent, yes. But, there is a certain minimum overhead for every pixel. The more pixels you cram onto a sensor, the more space on the sensor is dedicated to overhead instead of picking up light. Consequently, there are real limits to how much resolution you would want to have on a sensor.

Re:Sensitivity is not Resolution

[Bigjeff5]

There is a physics problem when your image sensor is too small - photons have size and mass, and there is a point at which you cannot collect enough light to take a good picture.

That's why expensive cameras have larger image sensors - they aren't packing more pixels per square inch, they are actually packing fewer pixels per square inch. A high end 10 mega-pixel will have an image sensor that is 10x bigger than a pocket-sized 10 mega-pixel camera, and it will take phenomenally better pictures.

This is the source of the GP's confusion about what the summary means - is "quantum film" more sensitive to light? Or are they simply able to pack more sensors in a smaller area? If they are actually able to collect accurate color information from fewer photons (i.e. more sensitive to light), then you can shrink the size of high end image sensors and still maintain quality. If it simply allows them to pack more pixels onto a sensor without being able to collect accurate color data with fewer photons, then quantum film is absolutely worthless. It offers no benefit to the quality of images in that case, even if they can crank a camera up to 30 megapixels it will still look like shit.

Re:Sensitivity is not Resolution

[ceoyoyo]

No, it doesn't. The lens system of the camera only has a certain resolving ability. Once you pass that point, you can make the sensor as high resolution as you want and you're just wasting your time because the lens isn't passing information at that level of detail anyway. Basically, you're measuring blur more and more finely.

Take a picture from anything less than a high end SLR or medium format camera and zoom in until you're actually looking at one image pixel to one screen pixel. Now tell me how good the image looks. Pretty crappy, hey? That's because the lens isn't capable of producing a decent image at even the resolution of the current sensor, never mind a better one.

Re:Sensitivity is not Resolution

[dgatwood]

This is about the laws of physics. I'm sure somebody will correct me if I'm not explaining this very well, but...

There's a limit to how precisely a lens can focus light. Now, in theory, as the aperture gets smaller, the diffusion decreases, so you might think that the small lenses would be result in a more precise image than larger ones. However, with those smaller lenses come smaller image sensors, which means that even if the lens can focus light to a smaller point, the pixels are also smaller, thus canceling out much of this improvement.

The bigger problem is that the smaller the lens, the greater the impact of even tiny lens aberrations on the resolving power of the lens. A speck of dust on a 1.5mm lens makes a huge difference, whereas it can be largely ignored on a lens with a 72mm diameter.

Also, as resolution increases, light gathering decreases. That's pretty fundamental to the laws of physics. Think about the bucket analogy. You have four square buckets measuring 1 foot by 1 foot. You place them side by side during a thunderstorm. You get another bucket that is two feet on each side. You place it beside the others. The same amount of rain (approximately) falls onto the four small buckets as the single large bucket, thus the large bucket has four times the amount of water in it that any one of the smaller buckets does.

The same principle applies to pixels. All else being equal, resolution and light gathering are inversely proportional. Small cameras are already hampered pretty badly by light gathering because of their small lenses. Increasing the resolution just makes this worse. I can tell the difference in noise between my old 6MP DSLR and my 10MP DSLR. I can't imagine what 20MP in a camera phone would look like. :-D

I think the real question should not be whether we can make smaller cameras, but rather whether we can make existing small cameras better by improving the light gathering. This technology might do that---whether it will work better than some of the newer CMOS sensor designs that already move the light-gathering material to the front remains to be seen---but at some point, making things smaller just means that they're easier to lose. I think we're at that point, if not past it....

Sensationalist, almost rubbish

[AliasMarlowe]

Image quality is limited by several factors. The sensitivity of the detector is only one, and is the only one that quantum dots can address. In this instance, the sensitivity increases only by a moderate amount, so the improvement in signal level (or reduction in pixel size preserving signal level) is also moderate.

Increasing the signal level will improve the S/N ratio for readout noise, assuming the readout is comparable to that available in today's cameras. Readout noise has been aggressively tacked by camera manufacturers, and is already very low. The principal source of noise in conventional images is shot noise (photon noise), and this is unrelated to the detector sensitivity. Shot noise depends ONLY on the number of photons arriving at each pixel, and is the reason that darker areas of digital images tend to be noisier, or require information-destroying denoising operations in postprocessing. Other forms of noise, such as dark current and dark noise, are relevant only in special applications, such as astrophotography.

Shot noise is intrinsic in the statistics of photon fluxes. The number of photons arriving at a pixel from a radiance which is "uniform" in time and space is Poissonian: the standard deviation is the square root of the mean. The signal to noise ratio is the mean divided by its square root, which is the square root of the number of photons which arrived in that sampling interval (exposure). If 10,000 photons are expected to arrive at a pixel in a given exposure time, then the shot noise will be about 1% when comparing multiple "identical" exposures of that pixel. Changing the detector sensitivity raises or lowers the readout signal level, but does not change the signal to noise ratio in the signal from shot noise.

Reducing the shot noise requires more photons arriving at each pixel. Getting more photons per pixel requires either (i) bigger pixels on the detector, (ii) better illumination of the subject, or (iii) better optics. This is why professional cameras have larger pixels than prosumer cameras, which tend to have larger pixels than pocket cameras, phone cameras, etc. Better lenses also help (but large apertures also affect depth of field). For given lighting conditions and optics, bigger pixels result in lower image noise, unless the readout circuitry really sucks.

So, quantum dots will result in a higher signal level than conventional CCD/CMOS/CID detectors under similar imaging circumstances. The improvement is probably limited to improving the ratio of signal to readout noise, which is already pretty good. Quantum dots will not magically increase the number of photons arriving at the detector, and if used to reduce pixel size, will result in worse signal to noise ratio for the shot noise (biggest noise problem in most photography). Result: not a dramatic improvement, although detectors giving horribly noisy images (needing heavy destructive denoising) may get even smaller.

Just send the bums some money, so they'll shut up. The potential of quantum dots in imaging sensors has been known for years.

Re:Sensationalist, almost rubbish

[hvdh]

Reducing the shot noise requires more photons arriving at each pixel. Getting more photons per pixel requires either (i) bigger pixels on the detector, (ii) better illumination of the subject, or (iii) better optics.

(iv) Increase photon capture efficiency.

The article says that in conventional CMOS sensors, three quarters of the incident photons are either absorbed by a metal layer or hit a spot between photo diodes, not contributing to photo diode charge and read-out signal. The new coating can convert those photons into charge, increasing the signal by a factor of four without changing pixel size, optics or illumination. Noise will be lower.

If it works as advertised, this is a good thing.

Re:Sensationalist, almost rubbish

[AliasMarlowe]

The article says that in conventional CMOS sensors, three quarters of the incident photons are either absorbed by a metal layer or hit a spot between photo diodes, not contributing to photo diode charge and read-out signal.

You are referring to areal efficiency or "fill factor" of detectors. CMOS had low areal efficiency some years back, but no longer. Both CMOS and CCD detectors are almost always equipped with integrated microlenses nowadays, which direct almost all of the incident light on the whole detector onto the active photosites. Some light is still lost at boundaries between the lenses, and due to the efficiency of the lenses. The ineffective regions between photosites receive hardly any light at all. Here's a quote from the Wikipedia article on CCDs:

Microlenses can bring the fill factor back up to 90 percent or more depending on pixel size and the overall system's optical design.

Integration of microlenses onto the chip is a major reason why CMOS detectors have caught up with CCD detectors in image quality. Compared to CCDs, a smaller fraction of a CMOS detector consists of photosites. Both benefit from provision of microlenses, but CMOS benefits rather more, and reaches almost the same areal efficiency as a comparable CCD. With less than 10% of incident photons lost, there is only a limited scope for improvement, by quantum dots or other methods. Those claims in TFA were reminiscent of fresh bullshit.

Good CCDs can exceed 85% in quantum efficiency at some wavelengths, such as in the icx285 which is typically used in industrial devices. However, efficiencies are lower at other wavelengths, and CCD and CMOS detectors used in consumer devices often peak at below 60% quantum efficiency. So there is room for improvement here, but not nearly as spectacular as the claims of TFA.

Keep in mind, as I mentioned in the earlier post, that increases in detector sensitivity (through areal efficiency or quantum efficiency) will elevate the signal level, but will not affect the ratio of shot noise in the signal. For that, you need more incident photons though bigger pixels and/or better subject illumination and/or bigger lens apertures and/or longer exposure times. TFA smells a bit like marketing hype. Quantum dots may lead to improvements in detector fabrication & price, but not so much in image quality...