pixel size limit?

[OpticsEngineer]

is there a practical limit to the size of a pixel in a camera sensor?

 

[Jack Hogan]

is there a practical limit to the size of a pixel in a camera sensor?

Three practical limitations that come to mind are

1. node size vs expected wavelength
2. size of PSF vs size of pixel
3. number of expected photons per pixel at expected exposures

I think currently that boils down to about 0.5um pixels, though they tend to be used in binned configurations, so they are effectively 1 or 2 um pixels with the smaller actual size being used for additional benefits like OSPDAF and HDR.

In the limit there are jots.

Jack

 

[Eric Fossum]

is there a practical limit to the size of a pixel in a camera sensor?

Three practical limitations that come to mind are

1. node size vs expected wavelength
2. size of PSF vs size of pixel
3. number of expected photons per pixel at expected exposures

I think currently that boils down to about 0.5um pixels, though they tend to be used in binned configurations, so they are effectively 1 or 2 um pixels with the smaller actual size being used for additional benefits like OSPDAF and HDR.

In the limit there are jots.

Jack

I like to think of a continuum (in 2-D) silicon surface where photons are realized and converted to photoelectrons (and photoholes). Pixels are the sampling sites of those photocarriers (which could be 2-D or 3-D). In this sense, there is no real limit down to near crystal lattice dimensions.

You asked about practical limits so I think it depends on what you want to achieve with your samples. In addition to what Jack mentioned there is the practicality of readout electronics pitch, and the off-chip-transmission-of-data bottleneck. The latter is of course relieved by on-chip signal and image processing. For photography, who could possibly need more than a million pixels? (old joke).

At the very recent Int. Image Sensor Workshop in Crieff Scotland last week, it appears that sub-0.5um pixels will be coming in the next 5 years perhaps. Pixel pitch of 0.56um seems well in hand. Interestingly, the early jot patents specify pixel pitch of less than 0.5um and they will expire at about the same time as sub 0.5um pixels may appear. Sometimes it is better to not be too early!

 

[Thom Hogan]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction). What I tend to think is that the smaller size we target, the less we know today, but we may know something different tomorrow.

I'm with Dr. Fossum on this one: the limiting factor for 2D today has to only be the underlying structure size. Is that a hard limit? See above.

Finally, I'm of the belief that "more sampling" is technically always better, even if there is only an infinitesimal gain. That's one of the reasons why I love the jot idea.

 

[D Cox]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction). What I tend to think is that the smaller size we target, the less we know today, but we may know something different tomorrow.

I'm with Dr. Fossum on this one: the limiting factor for 2D today has to only be the underlying structure size. Is that a hard limit? See above.

Finally, I'm of the belief that "more sampling" is technically always better, even if there is only an infinitesimal gain. That's one of the reasons why I love the jot idea.

I think the cones in the human retina go down to about half a micron.

Smaller in birds of prey.

Loading…

oxfordre.com

Don

 

[hjulenissen]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction). What I tend to think is that the smaller size we target, the less we know today, but we may know something different tomorrow.

I'm with Dr. Fossum on this one: the limiting factor for 2D today has to only be the underlying structure size. Is that a hard limit? See above.

Finally, I'm of the belief that "more sampling" is technically always better, even if there is only an infinitesimal gain. That's one of the reasons why I love the jot idea.

I wonder if there are specialized "near-field" camera applications where diffraction plays out differently?

I imagine silicon inspection "cameras" that essentially let a sensor touch the surface of a cpu die. No lens involved.

-h

 

[alanr0]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction). What I tend to think is that the smaller size we target, the less we know today, but we may know something different tomorrow.

I'm with Dr. Fossum on this one: the limiting factor for 2D today has to only be the underlying structure size. Is that a hard limit? See above.

Finally, I'm of the belief that "more sampling" is technically always better, even if there is only an infinitesimal gain. That's one of the reasons why I love the jot idea.

I think the cones in the human retina go down to about half a micron.

Smaller in birds of prey.

https://oxfordre.com/neuroscience/d...01780F8607127EF4?rskey=Zxtqid&result=19&print

Don

I am not an expert in this field, but 0.5 µm sounds rather small. However I may be confusing cone size and cone spacing.

https://en.wikipedia.org/wiki/Cone_cell#Shape_and_arrangement quotes cone sizes between 0.5 and 4.0 µm, with smallest size and tightest packing in the fovea.

The Colour Vision Research Laboratory (CVRL) database reports peak foveal cone densities with a spacing around 2µm.

Loading…

www.cvrl.org

Loading…

www.cvrl.org

There is significant variation between publications and across the retina http://www.cvrl.org/database/text/density/cones.htm

Your linked paper has some fascinating background, including:

Neuroscience/Raptor Vision/Anatomy and Optics/Anatomical Spatial Resolution

... due to the wave nature of light, photoreceptors cannot function optimally when their diameters approach the wavelength of light. Indeed, the narrowest photoreceptors in the animal kingdom are about one micrometre wide (Land & Nilsson, 2012) and cones in wedge-tailed eagles (1.6 μ‎m; Reymond, 1985), common buzzards, common kestrels (1.7 μ‎m; Oehme, 1964), and brown falcons (Falco berigora)(1.8 μ‎m; Reymond, 1987) are among the thinnest reported in birds to date.

These cone widths are indeed smaller than spacings in the CVRL database for human subjects, but not less than 0.5 µm.

 

[MCLV]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction).

Diffraction is a very relevant effect in this context. One needs a pixel pitch that is about 3.5x smaller (when expressed in microns) than f-number to extract resolution provided by diffraction limited optics (i.e. 2 um pixel pitch is sufficient for f/7 or smaller apertures). It also means that pixel pitch derived from diffraction is DOF and format dependent.

One can have very small 1 um pixel pitch on a FF sensor but it would make sense only when shooting at f/2.8 or larger apertures. Also, the subject has to be very flat or quite far away. Otherwise, the resolution potential would be used only in a very small fraction of the image and the rest would consist of blurred out of focus areas anyway.

 

[MCLV]

is there a practical limit to the size of a pixel in a camera sensor?

Three practical limitations that come to mind are

1. node size vs expected wavelength
2. size of PSF vs size of pixel
3. number of expected photons per pixel at expected exposures

I think currently that boils down to about 0.5um pixels, though they tend to be used in binned configurations, so they are effectively 1 or 2 um pixels with the smaller actual size being used for additional benefits like OSPDAF and HDR.

In the limit there are jots.

Jack

I like to think of a continuum (in 2-D) silicon surface where photons are realized and converted to photoelectrons (and photoholes). Pixels are the sampling sites of those photocarriers (which could be 2-D or 3-D). In this sense, there is no real limit down to near crystal lattice dimensions.

You asked about practical limits so I think it depends on what you want to achieve with your samples. In addition to what Jack mentioned there is the practicality of readout electronics pitch, and the off-chip-transmission-of-data bottleneck. The latter is of course relieved by on-chip signal and image processing. For photography, who could possibly need more than a million pixels? (old joke).

At the very recent Int. Image Sensor Workshop in Crieff Scotland last week, it appears that sub-0.5um pixels will be coming in the next 5 years perhaps. Pixel pitch of 0.56um seems well in hand. Interestingly, the early jot patents specify pixel pitch of less than 0.5um and they will expire at about the same time as sub 0.5um pixels may appear. Sometimes it is better to not be too early!

With cameras, one usually has the same number of pixels on the sensor and in the resulting image (although there are exceptions). However, with jots and very small pixels, number of pixels/jots on the sensor likely gets decoupled from pixel count of final image. So it might be meaningful to separate discussion about pixel count of the output image and about the pixel pitch of the sensor.

 

[Eric Fossum]

is there a practical limit to the size of a pixel in a camera sensor?

Three practical limitations that come to mind are

1. node size vs expected wavelength
2. size of PSF vs size of pixel
3. number of expected photons per pixel at expected exposures

I think currently that boils down to about 0.5um pixels, though they tend to be used in binned configurations, so they are effectively 1 or 2 um pixels with the smaller actual size being used for additional benefits like OSPDAF and HDR.

In the limit there are jots.

Jack

I like to think of a continuum (in 2-D) silicon surface where photons are realized and converted to photoelectrons (and photoholes). Pixels are the sampling sites of those photocarriers (which could be 2-D or 3-D). In this sense, there is no real limit down to near crystal lattice dimensions.

You asked about practical limits so I think it depends on what you want to achieve with your samples. In addition to what Jack mentioned there is the practicality of readout electronics pitch, and the off-chip-transmission-of-data bottleneck. The latter is of course relieved by on-chip signal and image processing. For photography, who could possibly need more than a million pixels? (old joke).

At the very recent Int. Image Sensor Workshop in Crieff Scotland last week, it appears that sub-0.5um pixels will be coming in the next 5 years perhaps. Pixel pitch of 0.56um seems well in hand. Interestingly, the early jot patents specify pixel pitch of less than 0.5um and they will expire at about the same time as sub 0.5um pixels may appear. Sometimes it is better to not be too early!

With cameras, one usually has the same number of pixels on the sensor and in the resulting image (although there are exceptions). However, with jots and very small pixels, number of pixels/jots on the sensor likely gets decoupled from pixel count of final image. So it might be meaningful to separate discussion about pixel count of the output image and about the pixel pitch of the sensor.

I assume the OP was talking about physical size of the pixel on the sensor - the sampling pitch, and not the number of output pixels in a final fully-processed photograph. But sure, we could talk about both, although scaling of pixel count up or down seems a matter of computation and not really about technology or device physics.

 

[MCLV]

is there a practical limit to the size of a pixel in a camera sensor?

Three practical limitations that come to mind are

1. node size vs expected wavelength
2. size of PSF vs size of pixel
3. number of expected photons per pixel at expected exposures

I think currently that boils down to about 0.5um pixels, though they tend to be used in binned configurations, so they are effectively 1 or 2 um pixels with the smaller actual size being used for additional benefits like OSPDAF and HDR.

In the limit there are jots.

Jack

I like to think of a continuum (in 2-D) silicon surface where photons are realized and converted to photoelectrons (and photoholes). Pixels are the sampling sites of those photocarriers (which could be 2-D or 3-D). In this sense, there is no real limit down to near crystal lattice dimensions.

You asked about practical limits so I think it depends on what you want to achieve with your samples. In addition to what Jack mentioned there is the practicality of readout electronics pitch, and the off-chip-transmission-of-data bottleneck. The latter is of course relieved by on-chip signal and image processing. For photography, who could possibly need more than a million pixels? (old joke).

At the very recent Int. Image Sensor Workshop in Crieff Scotland last week, it appears that sub-0.5um pixels will be coming in the next 5 years perhaps. Pixel pitch of 0.56um seems well in hand. Interestingly, the early jot patents specify pixel pitch of less than 0.5um and they will expire at about the same time as sub 0.5um pixels may appear. Sometimes it is better to not be too early!

With cameras, one usually has the same number of pixels on the sensor and in the resulting image (although there are exceptions). However, with jots and very small pixels, number of pixels/jots on the sensor likely gets decoupled from pixel count of final image. So it might be meaningful to separate discussion about pixel count of the output image and about the pixel pitch of the sensor.

I assume the OP was talking about physical size of the pixel on the sensor - the sampling pitch, and not the number of output pixels in a final fully-processed photograph.

Yes, OP asked about sensor but I wanted to point out this difference since I have learned that what people mean and what they write are two things that are not necessarily the same. Let's see if OP clarifies it.

But sure, we could talk about both, although scaling of pixel count up or down seems a matter of computation and not really about technology or device physics.

Sure, you can adjust the pixel count by resampling afterwards but once you sample PSF of optics properly, there is no point in further increasing the pixel count of the output. Or is it? So this might result in a practical "limit" of pixel pitch, especially for large sensors. It seems to me that pixel pitches used in mobile phones are already beyond this practical limit for FF sensors.

 

[OpticsEngineer]

" ...Let's see if OP clarifies it."

Thanks for asking. I was interested in the pixel size and sampling pitch in my question. But I am happy when people extend things and add in related things as well.

 

[SrMi]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction).

Diffraction is a very relevant effect in this context. One needs a pixel pitch that is about 3.5x smaller (when expressed in microns) than f-number to extract resolution provided by diffraction limited optics (i.e. 2 um pixel pitch is sufficient for f/7 or smaller apertures). It also means that pixel pitch derived from diffraction is DOF and format dependent.

One can have very small 1 um pixel pitch on a FF sensor but it would make sense only when shooting at f/2.8 or larger apertures. Also, the subject has to be very flat or quite far away. Otherwise, the resolution potential would be used only in a very small fraction of the image and the rest would consist of blurred out of focus areas anyway.

Isn't diffraction independent from pixel pitch?

Diffraction limited pixels? Really? In Support of Depth of Field

Diffraction limited images, Really? In Support of Depth of Field

www.scantips.com

 

[Jack Hogan]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction).

Diffraction is a very relevant effect in this context. One needs a pixel pitch that is about 3.5x smaller (when expressed in microns) than f-number to extract resolution provided by diffraction limited optics (i.e. 2 um pixel pitch is sufficient for f/7 or smaller apertures). It also means that pixel pitch derived from diffraction is DOF and format dependent.

One can have very small 1 um pixel pitch on a FF sensor but it would make sense only when shooting at f/2.8 or larger apertures. Also, the subject has to be very flat or quite far away. Otherwise, the resolution potential would be used only in a very small fraction of the image and the rest would consist of blurred out of focus areas anyway.

Isn't diffraction independent from pixel pitch?

https://www.scantips.com/lights/diffraction.html

Yes, nonetheless there is an argument to be made that it is wasteful/unnecessary to sample an image on the sensing plane at a higher rate than some multiple of the highest spatial frequency present in it. And since today diffraction provides an upper physical limit to the highest spatial frequency acceptable for the setup, the two can become related that way.

Jack

 

[MCLV]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction).

Diffraction is a very relevant effect in this context. One needs a pixel pitch that is about 3.5x smaller (when expressed in microns) than f-number to extract resolution provided by diffraction limited optics (i.e. 2 um pixel pitch is sufficient for f/7 or smaller apertures). It also means that pixel pitch derived from diffraction is DOF and format dependent.

One can have very small 1 um pixel pitch on a FF sensor but it would make sense only when shooting at f/2.8 or larger apertures. Also, the subject has to be very flat or quite far away. Otherwise, the resolution potential would be used only in a very small fraction of the image and the rest would consist of blurred out of focus areas anyway.

Isn't diffraction independent from pixel pitch?

https://www.scantips.com/lights/diffraction.html

Yes, nonetheless there is an argument to be made that it is wasteful/unnecessary to sample an image on the sensing plane at a higher rate than some multiple of the highest spatial frequency present in it. And since today diffraction provides an upper physical limit to the highest spatial frequency acceptable for the setup, the two can become related that way.

Jack

Exactly. If 1 um pixels are used at f/7 instead of 2 um ones, one needs to readout four times more pixels and then to process them and store them on a card. This makes the camera slower, more energy hungry and uses a lot more storage. But the resulting image will be virtually identical in sharpness/amount of detail and lack of aliasing to one captured by 2 um pixels. So you get some significant disadvantages but no real benefit (I assume here that a conventional pixel is used, not jots or similar).

 

[SrMi]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction).

Diffraction is a very relevant effect in this context. One needs a pixel pitch that is about 3.5x smaller (when expressed in microns) than f-number to extract resolution provided by diffraction limited optics (i.e. 2 um pixel pitch is sufficient for f/7 or smaller apertures). It also means that pixel pitch derived from diffraction is DOF and format dependent.

One can have very small 1 um pixel pitch on a FF sensor but it would make sense only when shooting at f/2.8 or larger apertures. Also, the subject has to be very flat or quite far away. Otherwise, the resolution potential would be used only in a very small fraction of the image and the rest would consist of blurred out of focus areas anyway.

Isn't diffraction independent from pixel pitch?

https://www.scantips.com/lights/diffraction.html

Yes, nonetheless there is an argument to be made that it is wasteful/unnecessary to sample an image on the sensing plane at a higher rate than some multiple of the highest spatial frequency present in it. And since today diffraction provides an upper physical limit to the highest spatial frequency acceptable for the setup, the two can become related that way.

Jack

Exactly. If 1 um pixels are used at f/7 instead of 2 um ones, one needs to readout four times more pixels and then to process them and store them on a card. This makes the camera slower, more energy hungry and uses a lot more storage. But the resulting image will be virtually identical in sharpness/amount of detail and lack of aliasing to one captured by 2 um pixels. So you get some significant disadvantages but no real benefit (I assume here that a conventional pixel is used, not jots or similar).

Wouldn't the 1um image have less aliasing and better post-processing capabilities (NR, transformations, cropping)?

 

[MCLV]

I've learned not to rely on anyone who says there are limits ;~). Amongst other examples I remember conversations with camera designers at the turn of the century where they said that sensors would never go below 2 microns (they were worried about diffraction).

Diffraction is a very relevant effect in this context. One needs a pixel pitch that is about 3.5x smaller (when expressed in microns) than f-number to extract resolution provided by diffraction limited optics (i.e. 2 um pixel pitch is sufficient for f/7 or smaller apertures). It also means that pixel pitch derived from diffraction is DOF and format dependent.

One can have very small 1 um pixel pitch on a FF sensor but it would make sense only when shooting at f/2.8 or larger apertures. Also, the subject has to be very flat or quite far away. Otherwise, the resolution potential would be used only in a very small fraction of the image and the rest would consist of blurred out of focus areas anyway.

Isn't diffraction independent from pixel pitch?

https://www.scantips.com/lights/diffraction.html

Yes, nonetheless there is an argument to be made that it is wasteful/unnecessary to sample an image on the sensing plane at a higher rate than some multiple of the highest spatial frequency present in it. And since today diffraction provides an upper physical limit to the highest spatial frequency acceptable for the setup, the two can become related that way.

Jack

Exactly. If 1 um pixels are used at f/7 instead of 2 um ones, one needs to readout four times more pixels and then to process them and store them on a card. This makes the camera slower, more energy hungry and uses a lot more storage. But the resulting image will be virtually identical in sharpness/amount of detail and lack of aliasing to one captured by 2 um pixels. So you get some significant disadvantages but no real benefit (I assume here that a conventional pixel is used, not jots or similar).

Wouldn't the 1um image have less aliasing and better post-processing capabilities (NR, transformations, cropping)?

No, because there would be already no aliasing with 2 um pixels (at f/7). That means that sensor was able to properly sample all available information/frequencies coming from the lens at such aperture.

Have a look at this link where Jim Kasson shows diffraction blur on Fuji GFX 100 with 3.76 um pixel pitch: https://blog.kasson.com/gfx-100/a-visual-look-at-gfx-100-diffraction-blur/

Above-mentioned ruled of thumb predicts that 3.76 um pixel pitch is sufficient for aperture of 3.76*3.5 = f/13. Jim's results show that there is a hint of aliasing at f/11 and none at f/16. That agrees quite well with a prediction of f/13.

Sensor with 1.88 um pixel pitch will show the same aliasing (which is almost none) at f/5.6 than 3.76 um sensor at f/11.

 

[ProfHankD]

At the very recent Int. Image Sensor Workshop in Crieff Scotland last week, it appears that sub-0.5um pixels will be coming in the next 5 years perhaps. Pixel pitch of 0.56um seems well in hand. Interestingly, the early jot patents specify pixel pitch of less than 0.5um and they will expire at about the same time as sub 0.5um pixels may appear.

At Electronic Imaging 2023 I heard of some new sensor tech that is explicitly using sub-wavelength sensel sizes that would be smaller than 0.5um, so probably not waiting long at all for that size in specialized configurations. (It was a multi-spectral sensor design.)

However, I personally don't get how sub-wavelength pixels manage to see much light. My understanding was that we quickly got into issues of needing EOT (Extraordinary Optical Transmission) with subwavelength structures, but I haven't heard EOT mentioned by any of the sensor folks. What really does happen when the pixel dimensions become smaller than the wavelength of the light?

 

[OpticsEngineer]

I wondered about that briefly too, but then recalled that a single atom can absorb a photon. and an atom is much smaller than a wavelength of light.

I guess we could go into overlap of wavefunctions, cross sections and transition rates, but I don't know if it is worth it for this discussion. but obviously more atoms means more absorption, along with line broadening.

 

[MCLV]

Some sensors used in mobile phones nowadays employ quad- and nona-Bayer CFA. Is there a limit how small can the individual elements of CFA be? And is it more manufacturing issue of CFA itself or a crosstalk issue due to absorption of light in silicon?