11 ms·

Samsung and Omnivision Claim Smallest Camera Pixels

36 points by sohkamyung 5y ago

giuliomagnifico 5y ago
> CMOS camera chips with pixel pitches of just 0.56 micrometers Wow. For comparison the latest Leica M11 with a full frame sensor with 60.3MPX has a pixel pitch of 3.76 μm. I’m curious about the ISO sensitivity but, since this sensor will use a lot of “computational photography”, like any modern smartphone, the ISO/light sensitivity should be less important than the power of the chip matched to this CMOS.
ck2 5y ago
Does this translate to astronomy for better imaging of very remote faint objects? (btw analog film apparently was 1.5um)
- acchow 5y ago
  You want larger pixels to collect more light for astronomy.
- mnw21cam 5y ago
  Unless you have a telescope with an extremely wide aperture, then no this sensor doesn't help. Pixels that small mean it is diffraction-limited if the telescope has an aperture of F/1.0. That's plausible in the most funky of camera lenses (but expensive), but completely unheard of in telescopes. The widest aperture telescopes are the RASA-type which have an aperture of around F/2.0, and they are expensive enough. They are regarded as fantastic light-sucking monsters, great for dim targets but not for small targets. Most of the time, the image sensor on a telescope actually slightly over-samples the sharpness of the image that the telescope is able to project. So having smaller pixels won't help. But to actually answer your question, these small pixels will be worse at low-light applications than larger pixels.
IndrekR 5y ago
We are talking about pixels smaller than the wavelength to be captured here. 560nm is bordering yellow and green. This does not make too much sense.
- amelius 5y ago
  Why not if you can use software to reconstruct the original image?
- mnw21cam 5y ago
  Yes, it means that i'll be diffraction-limited, even if the lens is F/1.0. It'll be diffraction-limited for red light around F/0.9. It's possible to make lenses that open, but hard. That's why camera phone lenses are so funky.
boondaburrah 5y ago
AFAIK you want the biggest pixels you can for the resolution you want in order to cut down on noise, no? The only exception to this (which proves the rule IMO) is that Nokia 40 megapixel phone that came out a while ago. It deliberately recorded 5mp photos, using the 40mp of data to intelligently average the pixel information, rejecting noise it could detect in software. It was unfortunately married to a less-than-stellar smartphone so I don't think it sold well despite the pictures being beautiful (IIRC at the time it's low-light performance couldn't be beat).
- goosedragons 5y ago
  A lot of modern flagship Android phones use pixel binning. The Galaxy S21 Ultra has a 108MP camera to take 12MP photos for example. Sensor size still matters and part of why those Nokias were so good at the time is they just had significantly bigger sensors than most phones.
- phh 5y ago
  > AFAIK you want the biggest pixels you can for the resolution you want in order to cut down on noise, no? You want the biggest sensor, but nowadays, we're targeting smallest pixels at the same time. The basic idea is that every pixel will have its own quantum and thermal noise, the noise on every pixel is independent. This means that to have the best measurement of the "real" light brightness at one place, you make the median of all the surrounding pixels (that's called binning). This obviously reduces resolution, hence we're making smaller pixels. (Nowadays we make better algorithms than median, but median is still pretty good) Lately, one of the major reasons we've been losing light on the sensors, is the RG/RB matrices, we loose light at every transition from one color to another. So, you combine those two pieces of information, about pixel binning and about color filtering, and that makes you modern sensors: 108MP (in Samsung Galaxy S22 Ultra case), where the RG/RB matrix only has 12M"spots", pixels being 9-binned (so you "create" one pixel, by measuring 9 at the same time). So Galaxy S22 Ultra's 108MP is kinda "just" 12MP with much better light sensitivity. In real life, it's a bit better than just 12MP, because even if you are in a R-only region, you can still find the changes of global brightness, and have sharpness further than 12MP. Hence Samsung touts "combining 9-binned image with non-binned image"
nynx 5y ago
I hope we eventually get phased array optics on smartphones (or just in general). Imagine the whole back of the phone is a sheet of nanosized optical components able to determine the angle of incoming photons. Off the bat, that could capture 3D environments, but with a large enough sensor, you could digitally zoom in basically indefinitely. Phased-array optics exist for larger wavelengths, but it’s likely they’ll eventually exist for visible light as nanomanufacturing improves.
etaioinshrdlu 5y ago
The cross-section structure kind of reminds me of animal rod and cone cells. They are on the same order of magnitude size, as well. I estimate a rod cell is 4x wider than one of these pixels.
lm28469 5y ago
Phone camera aren't resolution limited for a while now. Fitting more and more smaller pixels on the same sensor doesn't improve anything at all (lower DR, lower low light perf, bigger files, &c.), phone lenses can't even resolve them They'd better work on "normal" sized pixels with extra performance rather than smaller pixels that are barely as good as the previous gen. Alas, big numbers sell
- acchow 5y ago
  Typical big numbers game from Samsung. As you said, it definitely sells.
- userbinator 5y ago
  Omnivision are not exactly known for image quality either. They're known more for being cheap and (mostly) decent for the price, but tend be noisier than e.g. Sony's sensors.
- DeathArrow 5y ago
  >Fitting more and more smaller pixels on the same sensor doesn't improve anything at all (lower DR, lower low light perf They improve sharpness if optics can resolve them. Also, having more data will might ease autofocus algorithms and computational photography techniques. Low light and DR depend solely on the quantity of light falling on the sensor. That depends on scene illumination, sensor size and lens aperture,not on pixel size. However, having a huge resolution is a waste of storage and processing power if the optics can resolve much of those pixels. And it could slow things down.
  - infinityio 5y ago
    > Low light and DR depend solely on the quantity of light falling on the sensor. To my understanding - If the pixels are smaller, the gaps between pixels are likely to make up a larger proportion of the sensor, so you do lose some light from this process, although there are definitely upsides to extra pixels if there is more detail to be resolved
    - 5y ago

36 points by sohkamyung 5y ago

> CMOS camera chips with pixel pitches of just 0.56 micrometers Wow. For comparison the latest Leica M11 with a full frame sensor with 60.3MPX has a pixel pitch of 3.76 μm. I’m curious about the ISO sensitivity but, since this sensor will use a lot of “computational photography”, like any modern smartphone, the ISO/light sensitivity should be less important than the power of the chip matched to this CMOS.

ck2 5y ago

Does this translate to astronomy for better imaging of very remote faint objects? (btw analog film apparently was 1.5um)

acchow 5y ago
You want larger pixels to collect more light for astronomy.
mnw21cam 5y ago
Unless you have a telescope with an extremely wide aperture, then no this sensor doesn't help. Pixels that small mean it is diffraction-limited if the telescope has an aperture of F/1.0. That's plausible in the most funky of camera lenses (but expensive), but completely unheard of in telescopes. The widest aperture telescopes are the RASA-type which have an aperture of around F/2.0, and they are expensive enough. They are regarded as fantastic light-sucking monsters, great for dim targets but not for small targets. Most of the time, the image sensor on a telescope actually slightly over-samples the sharpness of the image that the telescope is able to project. So having smaller pixels won't help. But to actually answer your question, these small pixels will be worse at low-light applications than larger pixels.

IndrekR 5y ago

We are talking about pixels smaller than the wavelength to be captured here. 560nm is bordering yellow and green. This does not make too much sense.

amelius 5y ago
Why not if you can use software to reconstruct the original image?
mnw21cam 5y ago
Yes, it means that i'll be diffraction-limited, even if the lens is F/1.0. It'll be diffraction-limited for red light around F/0.9. It's possible to make lenses that open, but hard. That's why camera phone lenses are so funky.

boondaburrah 5y ago

AFAIK you want the biggest pixels you can for the resolution you want in order to cut down on noise, no? The only exception to this (which proves the rule IMO) is that Nokia 40 megapixel phone that came out a while ago. It deliberately recorded 5mp photos, using the 40mp of data to intelligently average the pixel information, rejecting noise it could detect in software. It was unfortunately married to a less-than-stellar smartphone so I don't think it sold well despite the pictures being beautiful (IIRC at the time it's low-light performance couldn't be beat).

goosedragons 5y ago
A lot of modern flagship Android phones use pixel binning. The Galaxy S21 Ultra has a 108MP camera to take 12MP photos for example. Sensor size still matters and part of why those Nokias were so good at the time is they just had significantly bigger sensors than most phones.
phh 5y ago
> AFAIK you want the biggest pixels you can for the resolution you want in order to cut down on noise, no? You want the biggest sensor, but nowadays, we're targeting smallest pixels at the same time. The basic idea is that every pixel will have its own quantum and thermal noise, the noise on every pixel is independent. This means that to have the best measurement of the "real" light brightness at one place, you make the median of all the surrounding pixels (that's called binning). This obviously reduces resolution, hence we're making smaller pixels. (Nowadays we make better algorithms than median, but median is still pretty good) Lately, one of the major reasons we've been losing light on the sensors, is the RG/RB matrices, we loose light at every transition from one color to another. So, you combine those two pieces of information, about pixel binning and about color filtering, and that makes you modern sensors: 108MP (in Samsung Galaxy S22 Ultra case), where the RG/RB matrix only has 12M"spots", pixels being 9-binned (so you "create" one pixel, by measuring 9 at the same time). So Galaxy S22 Ultra's 108MP is kinda "just" 12MP with much better light sensitivity. In real life, it's a bit better than just 12MP, because even if you are in a R-only region, you can still find the changes of global brightness, and have sharpness further than 12MP. Hence Samsung touts "combining 9-binned image with non-binned image"

nynx 5y ago

I hope we eventually get phased array optics on smartphones (or just in general). Imagine the whole back of the phone is a sheet of nanosized optical components able to determine the angle of incoming photons. Off the bat, that could capture 3D environments, but with a large enough sensor, you could digitally zoom in basically indefinitely. Phased-array optics exist for larger wavelengths, but it’s likely they’ll eventually exist for visible light as nanomanufacturing improves.

etaioinshrdlu 5y ago

The cross-section structure kind of reminds me of animal rod and cone cells. They are on the same order of magnitude size, as well. I estimate a rod cell is 4x wider than one of these pixels.

lm28469 5y ago

Phone camera aren't resolution limited for a while now. Fitting more and more smaller pixels on the same sensor doesn't improve anything at all (lower DR, lower low light perf, bigger files, &c.), phone lenses can't even resolve them They'd better work on "normal" sized pixels with extra performance rather than smaller pixels that are barely as good as the previous gen. Alas, big numbers sell

acchow 5y ago
Typical big numbers game from Samsung. As you said, it definitely sells.
userbinator 5y ago
Omnivision are not exactly known for image quality either. They're known more for being cheap and (mostly) decent for the price, but tend be noisier than e.g. Sony's sensors.
DeathArrow 5y ago
>Fitting more and more smaller pixels on the same sensor doesn't improve anything at all (lower DR, lower low light perf They improve sharpness if optics can resolve them. Also, having more data will might ease autofocus algorithms and computational photography techniques. Low light and DR depend solely on the quantity of light falling on the sensor. That depends on scene illumination, sensor size and lens aperture,not on pixel size. However, having a huge resolution is a waste of storage and processing power if the optics can resolve much of those pixels. And it could slow things down.
- infinityio 5y ago
  > Low light and DR depend solely on the quantity of light falling on the sensor. To my understanding - If the pixels are smaller, the gaps between pixels are likely to make up a larger proportion of the sensor, so you do lose some light from this process, although there are definitely upsides to extra pixels if there is more detail to be resolved
  - 5y ago