Myth busted: small pixels have worse performance than large pixels

[Daniel Browning]

[Part 1 out of 3]

It is often stated that an image sensor with small pixels will create digital images that have worse performance characteristics (more noise, less dynamic range, lower sensitivity, worse color depth, more lens aberrations, worse diffraction, and more motion blur.) than those created by a sensor with large pixels. I disagree.

One line of reasoning used by proponents of that position is that a single pixel, in isolation, when reduced in size, has lower performance; therefore, a sensor full of small pixels creates images that have worse performance than the same sensor full of large pixels. But that's not seeing the forest for the trees: the reality is that the resulting images are generally the same, as indicated in a paper by G. Agranov at 2007 International Image Sensor Workshop.

Again, it is possible for small pixel sensors to have worse performance per pixel, but the same performance when actually displayed or used for the same purpose as a large pixel sensor. This fact may be unbelievable or at least counter-intuitive to many people who work with digital images, but I believe that is only because of the following five types of mistakes that are frequently made in image analysis:

• Unequal spatial frequencies
• Unequal sensor sizes.
• Unequal processing.
• Unequal expectations.
• Unequal technology.

Spatial Frequency
The first category, spatial frequency, is the most important and fundamental element of image analysis as it pertains to pixel size. This aspect of an image indicates the level(s) of detail under analysis: whether fine details (high frequencies) or or more coarse information (low spatial frequency). This is often ignored completely, other times poorly understood, but it always has a tremendous impact on the result of any comparison or performance analysis.

The great majority of image analysis is fundamentally based on the performance of a single pixel, so having worse performance per pixel and the same performance in the actual image, where it matters, would seem a contradiction. It isn't.

Performance scales with spatial frequency. In other words, the many important performance characteristics of a digital image are all a function of spatial frequency, including noise, dynamic range, color depth, diffraction, aberrations, and motion blur. Therefore, for any given sensor, analysis of higher spatial frequencies will never show better performance than analysis of lower spatial frequencies.

Every image sensor has a sampling rate, or Nyquist. This is the spatial frequency at which the image sensor samples information. But every resulting digital image also contains information at all other lower spatial frequencies. For example, Pixel A may have a native sampling rate of 30 lp/mm. But the resulting digital image also contains information corresponding to 20 lp/mm and 10 lp/mm, which are larger, coarser details. Pixel B may be much smaller, and may natively sample at 60 lp/mm, but the resulting image still contains all the information of Pixel A, it only has additional information.

100% crop is the most common way to compare image sensors, but it is very misleading when the sensors have different pixel sizes. The reason is that 100% means the maximum spatial frequency. But different pixel sizes sample different spatial frequencies. So 100% crop means higher spatial frequencies for small pixel sensors than it does for big pixel sensors. This results in comparisons of completely different portions of the image. A 100% crop of a small pixel image would show a single leaf, whereas a 100% crop in a large pixel image would show the entire shrub. It's a nonsensical comparison. Failing to account for that important and fundamental difference is one of the most common flaws in such comparisons.

This type of flaw is much more rare in optics analysis. There it is widely understood that standard optical measurements, such as MTF, are naturally and fundamentally a function of spatial frequency. If the MTF of lens A is 30% at 10 lp/mm, and the MTF of lens B is 20% at 100 lp/mm, that does not mean lens A is superior; in fact, the opposite is more likely true. It's necessary to measure the lenses at the *same* frequency, either 30 lp/mm or 100 lp/mm, before drawing conclusions. It's very likely that lens B has a much higher MTF at 10 lp/mm. Of course, comparing MTF without regard for spatial frequency is so obviously wrong that very few people ever make that mistake. However, those same people do not realize they are making the exact same error when they compare image sensors with 100% crops. They are comparing at their respective Nyquist frequencies, but
they have different Nyquists, so they are not the same spatial frequency.

Take a 100% crop comparison of a high resolution image (e.g. 15 MP) with a low resolution image (e.g. 6 MP) for example. The high resolution image contains details at a very high spatial frequency (fine details), whereas the low-res image is at a lower spatial frequency (larger details). Higher spatial frequencies have higher noise power than low spatial frequencies. But at the *same* spatial frequency, noise too is the same.

Although it's not necessary or desirable, it is always possible to resample any two images (e.g. large pixel and small pixel images) to the same resolution for comparison. This would make it possible to compare 100% crops and draw conclusions about the spatial frequencies under analysis. However, sometimes the ability to resample is called into question, such as a blog post by Phil Askey at DPReview.com; however, it was thoroughly debunked:


http://forums.dpreview.com/forums/read.asp?forum=1018&message=30190836


http://forums.dpreview.com/forums/read.asp?forum=1000&message=30176643


http://forums.dpreview.com/forums/read.asp?forum=1031&message=31560647


There is ample proof that resampling works in practice as well as in theory. Given that fact, as long as small pixel sensors have proportionately higher noise and higher spatial frequencies, it will always be possible to resample the image and get lower noise power at lower spatial frequencies, so that the image is the same as that created by large pixel sensors. For example:


http://forums.dpreview.com/forums/read.asp?forum=1018&message=30211624


http://forums.dpreview.com/forums/read.asp?forum=1018&message=30190836

Again, in cases where it's not necessary to resample the image, it is best not to, since the resolution often has a highly beneficial impact on the image.

Another way to think about it is performance per detail. Say one small but important detail in an image is an eye. A large pixel sensor has a certain performance ”per eye”, so that over the area of the eye there is a certain noise power, dynamic range, etc. A small pixel sensor, too, has a certian performance per eye, again over the same area, only there are many more pixels. The noise power per pixel is higher, but since each pixel contributes a smaller portion of the eye, the noise power per eye is the same.

Here are some very good references about noise and spatial frequency:


http://luminous-landscape.com/forum/index.php?showtopic=29801&st=20&p=241562&#entry241562


http://forums.dpreview.com/forums/read.asp?forum=1000&message=30394220


http://forums.dpreview.com/forums/read.asp?forum=1034&message=31584345


http://forums.dpreview.com/forums/read.asp?forum=1000&message=30394220

Here are some images comparing small pixels and large pixels at low analog gain and high analog gain:


http://www.pbase.com/jps_photo/image/100092629/original.jpg




http://forums.dpreview.com/forums/read.asp?forum=1019&message=31512159

This error may have roots in the fact that the standard engineering measurements for sensor characteristics such as noise is necessarily at the level of the pixel. Sensitivity is measured in photoelectrons per lux second per pixel. Read noise is measured in RMS e- or ADU per pixel. Dynamic range is measured in stops or dB per pixel. There is nothing wrong with per-pixel measurements per se, but they cannot be compared with different pixel sizes without understanding the difference in spatial frequency.

Image sensor performance, like MTF, cannot be compared without understanding differences in spatial frequency.

[Continued in part 2.]

 

[Daniel Browning]

[Part 2 out of 3].

Unequal sensor sizes

Sensor size is separate from pixel size. Some assume that the two are always correlated, so that larger sensors have larger pixels, but that is not the case. Sensor size is generally the single most important factor in image sensor performance; therefore, it's always necessary to consider its impact on a comparison of pixel size. The most common form of this mistake goes like this:

• Compacts have more noise than DSLR cameras.
• Compacts have smaller pixels than DSLR cameras.
• Therefore smaller pixels cause more noise.

The logical error is that correlation is not causation. The reality is that it is not the small pixels that cause the noise, but small sensors. A digicam-sized sensor (5.6x4.15mm) sensor with super-large pixels (0.048 MP) will not have superior performance to a 56x41.5mm sensor with super-tiny pixels (48 MP). Even the size of the lens points to this fact: the large sensor will require a lens that is perhaps 50 times larger and heavier for the same f-number and angle of view, and that lens will focus a far greater quantity of light than the very tiny lens on a digicam. When they are both displayed or used in the same way, the large sensor will have far less noise, despite the smaller pixels.

Unequal processing.

Unequal processing is essentially all about the handling of the raw data from the ADC to the output of the converter. The best method to draw conclusions about pixel size is to analyze just the raw data itself, before a raw converter can introduce inequalities, bias, and increase experimental error; however, it is possible to get useful information after a raw conversion in certain conditions. The types of errors in this category include the following.

Out-of-camera images. Processed formats that come directly out of the camera, such as JPEG (and including video such as HDV, HDCAM-SR, etc.) are good for drawing conclusions about the utility of that processed format for whatever purpose is needed, but it does not accurately reflect on the sensor itself. Furthermore, any conclusions, which are necessarily subjective, cannot be generalized to pixel sizes in all cameras. Too much processing has already been applied to the raw data, including noise reduction, saturation, black point, tone curve, and much more, which all have an affect on apparent noise, sensitivity, color, and dynamic range.

Unequal raw preconditioning. Most cameras apply a certain amount of processing before the raw file is written. Typically it includes at least hot/dead pixel remapping, but sometimes also other things. Many apply a black clip at or near the mean read noise level. Some remove the masked pixels. Nikon performs a slight white balance. Canon compensates for the sensor's poor angle of response by increasing brightness for f-numbers wider than f/2.8. Pentax applies slight median-filter based noise reduction on raw files when set to ISO 1600. Etc. Much of this pre-processing of the raw file can be factored into the comparison; the chief thing is to be aware of its occurance.

Unequal raw converters. One raw converter may use totally different processing than another, even if the settings look the same. For example, setting noise reduction to "off" in most converters does not actually turn off all noise reduction; it just reduces noise reduction to the lowest level the creator is willing to allow.

Another common mistake is to think that a given raw converter will process two different cameras equally. None of the popular raw converters do. There are some converters that do, e.g. dcraw, IRIS, Rawnalyze, etc. One of the most popular converters, ACR, for example, varies greatly in its treatment of cameras, with different styles for different manufacturers, models from the same manufacturer, and even from one minor version of ACR to the next.

Furthermore, even if a raw converter is used that can be proven to be totally equal (e.g. dcraw), the method it uses might be better suited to one type of sensor (e.g. strong OLPF, less aliases) more than another (e.g. weak OLPF, more aliases). Even this can lead to incorrect conclusions because some cameras combined with certain scenes may benefit more from certain styles of conversion. For example, one demosiac algorithm may give the best results for a sensor with slightly higher noise at Nyquist. One way to workaround this type of inequality is to examine and measure the raw data itself before conversion, such as with IRIS, Rawnalyze, dcraw, etc. The important thing to be aware of is the possibility for inequalities to arise from processing.

Unequal expectations.

This is the type of inequality that stems from having a different standard, expectation, or purpose for different pixel sizes. It unlevels the playing field. For example, there are some who claim in order for small pixels to be equal to large pixels, they must have the same performance (e.g. noise power) when displayed at much larger sizes (finer scales). But in reality, to be equal, they only need to have the same noise power when displayed at the same size. One might expect that a camera with 50% higher megapixels could be displayed 50% larger without any change in the visibility of noise, despite the same low scene luminance. That is an unequal expectation.

This can also be manifested in the idea that in order to be equal, one should be able to crop a smaller portion of the image from a small pixel sensor and still have the same performance (e.g. noise). The correct method is to crop the same portion out of both sensor for comparison. If one is cropping the center 10% from the large pixel sensor (e.g. 500x500), then one should crop the same 10% from the small pixel sensor (1000x1000).

If one only expects to it be at least display the same size, or the same portion of the image, and have the same performance (noise) for the same light, then that would be equal expectations.

Unequal technology.

This type of mistake is almost never made in support of large pixels, since large pixels are almost invariably the older technology. However, it is worth pointing out anyway. In one sense, it will never be possible to compare any two cameras with completely equal technology, because even unit-to-unit manufacturing tolerances of the same model will cause there to be inequalities.

[Continued in part 3.]

 

[Daniel Browning]

[Part 3 out of 3.]

Having now discussed the many factors that cause flaws in image analysis as it pertains to pixel size, let's look at some specific topics.

Fill factor

One topic that often comes up in a discussion of pixel size is fill factor, which is the relative area of the photoreceptive portion of the pixel, e.g. photodiode. It is commonly asserted that fill factor gets worse as pixel size shrinks (”smaller buckets means more space between the buckets”). In fact, this has not occurred. ”Fill factor pretty much has scaled with technology, ...” (CMOS sensor architect). Comparing the Sony ICX495AQN to ICX624, for example, pixel area shrunk from 4.84µm to 4.12µm, a decrease of 15%. But instead of losing 15% of the photoreceptive area, it actually increased by 7% (22% total):

Another assertion is that full well capacity decreases with pixel size. In fact, sensor designers say the reverse is true: "smaller pixels have greater depth (per unit area) and saturate 'later in time'".

Read noise

Sensors vary greatly in read noise. There are some large pixel sensors that have a high gain mode with lower read noise than other large pixel sensors and small pixel sensors, but they invariably have much higher read noise at base gain. On the other hand, at base gain, small pixel sensors have much less read noise than any of the large pixel sensors, with or without variable gain amps. Furthermore, generally, as large pixels are made smaller, with or without variable gain amps, read noise is improved.

Voltage

Some assert that smaller pixels put out lower voltages, and the increased amplification necessary results in worse noise. However, Eric Fossum has stated that no sensor designer would allow this to be the case. The simplified description of how it works is given here by an electrical engineer:

The accumulated charge of photoelectrons collects on the gate of the source follower transistor which has a gain of something less than one. The output of this amplifier is the voltage output of the cell (actually transferred via the column amplifier). This voltage is determined by the voltage on the source follower gate, which in turn is given by the value of the accumulated charge divided by the capacitance of the gate (by the well known expression V = Q/C).

If you take a cell and scale it uniformly by a factor s. Now the cell is s^2 times smaller in terms of area, so the accumulated charge is s^2 times smaller. However, the gate capacitance is s^2 times smaller also. So the output voltage is now (Q/s^2)/(C/s^2) = Q/C = V. Scaling has not changed the output voltage of the cell, so no extra amplification is needed.

Optical and mechanical issues.

There are many things that can affect the resolution of an image, including diffraction, aberrations, manufacturing tolerances, and motion blur (from camera shake or subject movement). In the face of these issues, some will claim that small pixels are actually worse than large pixels. This is easily proven false. The reality is that all of these factors may cause diminishing returns, but returns never diminish below 0%.

Diffraction, in particular, is often misunderstood. As pixel size decreases there are two points: one at which diffraction is just barely beginning to have an affect on the image, and another where the resolution improvement is so small that it's immeasurable. The most common mistake is to think both are the same point, or that they are anywhere near eachother. In fact, they are very far apart.

Other considerations

There are at least three things to consider with regard to pixel size

• File size and workflow
• Magnification value
• In-camera processing (JPEG, HDCAM, etc.)

File size is an obvious one. Magnification is what causes telephoto (wildlife, sports, etc.) and macro shooters to often prefer high pixel density bodies.

Out-of-camera JPEGs are affected by pixel density because manufacturers have added stronger noise reduction which may not be desired and may be difficult to tune with in-camera settings.

Higher pixel densities may require bigger files, slower workflow, and longer processing times. Lower pixel densities may result in smaller files, faster workflow, and shorter processing times. This is an area where there are many possible software solutions for having most of the benefit of smaller pixels without the size/speed downsides. REDCODE is a good example.

Finally, here is the math on a comparison of dynamic range between the LX3 and 5D2.

Compare the 2-micron pixels of the LX3 (10.7 stops DR) with the immensely larger 6.4 micron pixels of the 5D2 (11.1 stops DR). Going by the per-pixel numbers, it seems that the smaller LX3 pixels have less dynamic range. But remember that the LX3-sized pixel samples a much, much higher spatial frequency.

At the same spatial frequency, the scaled LX3 pixels have 12.3 stops of dynamic range, 1.2 stops greater.

5D2 maximum signal: 52,300 e- LX3 maximum signal: 9,000 e-
5D2 read noise at base ISO: 23.5 e- LX3 read noise at base ISO: 5.6 e-
5D2 per-pixel DR at base ISO: 11.1 stops (log_2(52300/23.5))
LX3 per-pixel DR at base ISO: 10.7 stops (log_2(9000/5.6))
LX3 scaled maximum signal: 92200 (9000 e- * (6.4µm/2.0µm)^2)
LX3 scaled read noise at base ISO: 17.92 (sqrt(5.6 e-^2 * ((6.4µm/2.0µm)^2)))
LX3 scaled DR at base ISO: 12.3 stops (log_2(92200/17.92))

I hope that these posts were informative and helpful.

Kind regards,

 

[Pawel Achtel]

Can we make this sticky, please?

This is priceless information, Daniel. Thank you! :thumbsup:

 

[David Mullen ASC]

Since a lot of that was over my head...

When does the law of diminishing returns kick in? Is there no reason to not use tinier photosites if there are no downsides involved? Why not put the smallest photosites possible in every sensor then? There is no relationship between light gathering / quantum efficiency, whatever, and photosite size?

 

[I Bloom]

Hmmm... one thing though is with CMOS, you can fit more complexity into a larger photosite, and get things like 16 bit ADC with current fabrication techniques. Hence the monstro will benefit from larger pixels because it simply has more to work with in the digital realm. I think this additional complexity might also help with read reset times, another important consideration for cinematography. Just speculation, it seems to be all a trade off, and I wonder if there is a long life ahead for the Mysterium X.

Great post by the way.

IBloom

 

[I Bloom]

Since a lot of that was over my head...

When does the law of diminishing returns kick in? Is there no reason to not use tinier photosites if there are no downsides involved? Why not put the smallest photosites possible in every sensor then? There is no relationship between light gathering / quantum efficiency, whatever, and photosite size?

Smaller pixels definitely perform worse but..I think what he's saying in part is that across the same surface area the gains you might make with larger photosites in terms of signal to noise, are balanced by the loss of resolution, across the same surface area.

So for example if I combine (say average) four neighboring noisy pixels, I can get one reduced noise pixel roughly equivalent to what I might have had with a photosite the size of four pixels.

IBloom

 

[David Mullen ASC]

Smaller pixels definitely perform worse but..I think what he's saying in part is that across the same surface area the gains you might make with larger photosites in terms of signal to noise, are balanced by the loss of resolution, across the same surface area.

So for example if I combine (say average) four neighboring noisy pixels, I can get one reduced noise pixel roughly equivalent to what I might have had with a photosite the size of four pixels.

IBloom

But that's a bit of a circular argument, like saying that larger sensors don't give you less depth of field because you can stop down the lens to compensate. You're saying that smaller pixels do have worse performance issues, just that there are mitigating and compensating factors around that. So the myth isn't busted.

 

[I Bloom]

But that's a bit of a circular argument, like saying that larger sensors don't give you less depth of field because you can stop down the lens to compensate. You're saying that smaller pixels do have worse performance issues, just that there are mitigating and compensating factors around that. So the myth isn't busted.

It's busted for sensors not pixels. The image you are getting from the same surface area on the sensor. If you take the entire image coming off the sensor into account then the myth is busted so to speak. So it's different from stopping down since the same lenses same conditions same surface area, only difference is frequency of photosites.

IBloom

 

[Daniel Browning]

When does the law of diminishing returns kick in?

Good question.

First, there's diffraction, lens aberrations, and mechanical issues such as collimation, alignment, backfocus, tilt, manufacturing tolerances, etc. Diminishing returns have definitely kicked in here already.

For diffraction, if anyone is shooting the RED ONE at f/32 because they need really extreme DOF, they are not going to get any benefit from pixels smaller than RED ONE: the returns will be close to 0%. Even f/11 will have returns diminished slightly.

Lens aberrations can be an issue too. Usually even the worst lenses will have pretty good performance in the center, stopped down. But their corners wide open will sometimes not benefit very much from smaller pixels, so the returns may be 0-5% due to aberrations.

And there's the mechanical issues. If your collimation is not perfect, but it's good enough for RED ONE, then it will have to be better to get the full return of even smaller pixels. This relates to manufacturing tolerances of everything in the image chain: the higher the resolution, the more difficult it is to get full return from that additional resolution. Even things like tripods have to be more steady to prevent diminishing returns.

So essentially the diminishing returns depend on the circumstances, but the higher the resolution, the more often the returns will be diminished. Some people don't need more resolution than they currently are getting, so as long as the improvement from smaller pixels is 0% or more, that's OK for them.

On the other hand we have noise, dynamic range, and QE. For these, G. Agranov's paper showed no difference from 5.6 microns to 1.7 microns in typical situations, which agrees with my own survey of available sensor data and images. Now data on 1.4 micron sensors is available and they too appear to have performance that scaled in proportion with their size. I chalk this up to the ingenuity of sensor designers. I think we're getting close, though, so perhaps 1.2 microns will finally be the point at which diminishing returns kick in for typical situations.

One important exception is performance in low light. There is a technology that reduces read noise at high gain, and it has been making its way from large pixels down to smaller and smaller pixels; currently 4.7 microns. If the technology only works at certain pixel sizes, then when they hit that size some day then we'll no longer benefit from it. But until there is any evidence of that I'm going to continue to assume that sensor designers will scale pixel performance with size so that the image-level performance remains the same.

Another exception is angle of response. Pixels of all sizes tend to have lower response from oblique angles, such as an ultra wide angle non-retro-focus f/1.4 lens, but smaller pixels have even more difficulty because it's hard to scaling the depth (z dimension) in proportion with the area (x and y, width and height).

Is there no reason to not use tinier photosites if there are no downsides involved? Why not put the smallest photosites possible in every sensor then?

Mostly because of the file size and processing speed problems I mentioned. It's difficult enough for post production, but even more difficult inside the camera. I understand that heat is a major issue.

There is no relationship between light gathering / quantum efficiency, whatever, and photosite size?

Aside from the above exceptions, I think that's correct. There is a strong correlation between all those things and *sensor* size, though; such that larger sensors (with a proportionately larger lens and thinner DOF) have much improved light gathering capability.

With CMOS, you can fit more complexity into a larger photosite, and get things like 16 bit ADC with current fabrication techniques.

That's a good point, although modern microlens design and shared transistor techniques can allow for quite a lot of complexity in small pixels. With two transistors shared among four pixels, one can get the benefit of a 6T squeezed into the space of a 4T. Backside illumination would help here, too.

I think this additional complexity might also help with read reset times, another important consideration for cinematography.

Yes.

But that's a bit of a circular argument, like saying that larger sensors don't give you less depth of field because you can stop down the lens to compensate.

Hmm, where have I heard that before.

You're saying that smaller pixels do have worse performance issues, just that there are mitigating and compensating factors around that. So the myth isn't busted.

It depends on how the thread title is interpreted. I was going for "sensor full of small pixels vs. sensor full of large pixels", and I tried to clarify that in the opening paragraph. But if it is interpreted as "small sensor full of small pixels vs. big sensors full of big pixels", then the myth isn't busted.

 

[David Mullen ASC]

It's busted for sensors not pixels. The image you are getting from the same surface area on the sensor. If you take the entire image coming off the sensor into account then the myth is busted so to speak.

So the title of the thread should be amended to "Myth busted: sensors with smaller photosites..."

 

[Jeff Kilgroe]

So the title of the thread should be amended to "Myth busted: sensors with smaller photosites..."

Yes it should be...

Although, as for actual size of photosites relating to performance, the whole is a sum of many parts. Sensors are like microprocessors in that they are continuously evolving with new designs, manufacturing methodologies, new material combinations, etc.. It's a technological progression and each sensor and its photosite designs are created with a set of performance parameters in mind.

A big reason for this "myth" is that for several years, both CCD and CMOS sensors suffered large performance impact in terms of sensitivity with smaller photosites. And truth be told, they still do to some extent. It's just a fact and a limitation of the technologies available. When we start getting into various micro-lens structures that can only absorb or register a percentage of the light that contacts them, many of these structures see decreased light-gathering ability on an individual basis as the size of their photon wells and micro-lenses are downsized.

Depending on photosite structure or design, photon absorption is not constant or even possible across the entire surface of the photosite. Often, the inactive areas may be restricted to a specific width due to manufacturing abilities and tolerances or even the materials used and technology available. Keeping this in mind, as photosites decrease in size, the size of the wells decreases at a higher rate, or at least this holds true with all current manufacturing processes that I'm aware of.

For example, let's assume we have a 6 micron photosite with an active well size of 5.5 microns. Assume it's built with a process that requires photosite structures to have a 0.25 micron inactive structure space on each side of the well. For simplicity, we'll just assume that the well has a 100% registration factor. Anyway, this means that for those 6 micron photosites, which are 0.0360mm^2 in area, there is an area of 0.0303mm^2 capable of photon absorption, or roughly 84% of the photosite surface.

Now, if that same process is used to manufacture a new sensor with photosites that are 5 microns in size, that yields a well size of 4.5 microns. So that means for any given sensor built of those photosites, it will have photosites that are 0.0250mm^2 in area and wells that are 0.0203^mm in area, or will have light absorption ability across 81% of it's total surface area. Or in other words, a reduction of total light absorbing ability by 3% in contrast to the 6 micron photosite sensor.

What Daniel has posted is correct, in several ways, but also makes a few assumptions that do not coincide with the technology that is available now. All things considered, once photosite manufacturing can be brought to a level where photon absorption can be held mostly constant across the entire photosite and the entire photosite structure can be scaled uniformly, then we'll be in business. But we'll also have to bow down to the physical properties of our universe and accept that at a certain point, photosites can become too small to individually receive entire wavelengths of light. But I suppose this is why cutting edge, theoretical sensor design is already heading toward designs that don't use individual photosites or "pixels".

The 3 micron photosites of the 2/3" Mysterium-X are already at a point where the red end of the spectrum could easily split photon registration 50/50 between two photosites and potentially lose 10~30% of a photon strike depending on the size of the inactive (or less active) area bordering the photon wells.

With all that said and the theoretical back-and-forth here, what people really should take away from all this is that there is no direct restriction on sensitivity based on photosite or pixel size. Case in point, today's 12MP DSLRs have more than double the sensitivity and 2 stops more range than the 8MP DSLRs from 3 years ago. Sensor design is still evolving and progressing. No sensor out there has come close to any theoretical maximum light gathering ability, at any photosite size. RED is going to show us this as well when they release the 2/3" Scarlet. It's already said to have more range and I'm betting it will have equal or greater sensitivity than the RED One, even though the photosites are 60% smaller!!! It will be interesting to see what the differences are between the 5 micron Mysterium-X (S35 sensor) and the 3 micron Mysterium-X (2/3" sensor).

 

[David Rasberry]

Thanks for the posts Daniel. Great explanations.

 

[sander kamp]

Can somebody please make pictures with this article? I have a hard time understanding what exactly the point is...

 

[Daniel Browning]

Thanks for the response, Jeff.

A big reason for this "myth" is that for several years, both CCD and CMOS sensors suffered large performance impact in terms of sensitivity with smaller photosites.

I kindly disagree. In still cameras that was not the case. Smaller pixels, such as the FZ50, always had higher sensitivity than large pixels; it's only recently that large pixels have finally started to catch up, thanks to gapless microlenses many years after digicams had them.

And truth be told, they still do to some extent.

It's true that there are some exceptions, which I mentioned, but I don't think that's a good generalization.

Depending on photosite structure or design, photon absorption is not constant or even possible across the entire surface of the photosite. Often, the inactive areas may be restricted to a specific width due to manufacturing abilities and tolerances or even the materials used and technology available. Keeping this in mind, as photosites decrease in size, the size of the wells decreases at a higher rate, or at least this holds true with all current manufacturing processes that I'm aware of.

Unlike statements about noise, QE, FWC, DR, etc., I cannot refute statements about fill factor with my own measurements and data because I cannot measure fill factor directly. However, I can measure QE and FWC; they are greatly affected by fill factor, but they have not gotten worse with smaller pixel sizes; therefore they provide no indication that fill factor has gotten worse.

There is one other resource I can utilize to analyze fill factor, which is statements made by actual sensor designers. In my second post above, I quoted the person who invented CMOS APS: "Fill factor pretty much has scaled with technology, ...”. Above you may also notice one specific example I showed where fill factor remained the same (actually, it improved).

Let me know if you think the resource I quoted is incorrect and example I gave was a fluke and I'll come up with some additional data for you.

With all that said and the theoretical back-and-forth here, what people really should take away from all this is that there is no direct restriction on sensitivity based on photosite or pixel size.

Agreed.

Case in point, today's 12MP DSLRs have more than double the sensitivity and 2 stops more range than the 8MP DSLRs from 3 years ago.

For one, Canon's DSLR sensitivity (QE) improved only 5% in the last 1 year, 30% in the last 3 years, and 115% in the last 6 years. [10D: 0.042 photo-electrons per 12-bit ADU per square micron. 50D: 0.100 e-/adu/um^2.] The dynamic range improvement on Canon was a bit less a single stop in 4.5 years. [Area-referred base ISO read noise 20D: 1.81, 40D: 1.15, 50D: 0.97. Area referred full well stayed about the same.]

Sensor design is still evolving and progressing. No sensor out there has come close to any theoretical maximum light gathering ability, at any photosite size.

I kindly disagree. We've been very close to the theoretical maximum QE for a long time. Take off the color filters and most current desings, especially small pixels, hit 90% QE at some wavelengths. With color filters a lot of sensors still hit 50-70%. There's less than a full stop improvement even theoretically possible, nevermind what's practically possible.

Read noise and dynamic range are the areas where I think there is a lot of room for improvement, both theoretically and practically.

I have a hard time understanding what exactly the point is...

The point is: don't worry, be happy.

Can somebody please make pictures with this article?

That is something I intend to do someday, but haven't found the time. Here's a start: click the thumbnail to see a comparison of a large pixel sensor and a small pixel sensor (from one of the above links):



Kind regards,

 

[sander kamp]

So how does this all apply to RED's cameras: if you have the processing power then more pixels on the same chip size is better until you reach physical limit of the lens system?

 

[Sidney L. Plaut]

my head is spinning.

 

[Steven Caesare]

This is a really good post and subject treatment. Thanks Daniel.

Incidentally, it has never made sense for me why folks ignore the "per area unit" part of the pixel size equation. Just as I don't understand when they ignore that focal length AND sensor size determin a correct "DoF".

The parallel in film is sensitivity/grain size... when discussing that it's all relative to a full 35mm frame... you don't crop out a portion of a 35mm frame and enlarge it to another arbitrary size in order to compare...

 

[I Bloom]

Incidentally, it has never made sense for me why folks ignore the "per area unit" part of the pixel size equation. Just as I don't understand when they ignore that focal length AND sensor size determin a correct "DoF".

Marketing snake oil... been with photography for a long time I think.

IBloom

 

[Jeff Kilgroe]

There is one other resource I can utilize to analyze fill factor, which is statements made by actual sensor designers. In my second post above, I quoted the person who invented CMOS APS: "Fill factor pretty much has scaled with technology, ...”. Above you may also notice one specific example I showed where fill factor remained the same (actually, it improved).

Actually that seems pretty good from the glance I gave it. The big issue here is that fill factor or well size is continuously improving with newer designs and manufacturing processes.

Going back and looking at my post, I'm not too happy with it. Mostly I was just trying to say is that given equivalent designs and manufacturing abilities, photosite efficiency generally decreases as size decreases. Or the fill factor or surface to photon receptor area ratio diminishes with size. But the market itself doesn't work that way and typically sensor designs that have smaller photosites are most often constructed of a different design or are a generation or two newer than a larger-photosite predecessor. No one has achieved a 100% fill factor on any sensor yet. No one has yet achieved a photosite pitch gap smaller than 0.185 microns. But there have been ways of increasing sensitivity with smaller photon wells and it's actually easier to control the behavior of a photosite with a smaller well (as in noise control and read reliability).

I kindly disagree. We've been very close to the theoretical maximum QE for a long time. Take off the color filters and most current desings, especially small pixels, hit 90% QE at some wavelengths. With color filters a lot of sensors still hit 50-70%. There's less than a full stop improvement even theoretically possible, nevermind what's practically possible.

Close is subjective. To me, a 50-70% range still represents a long way to go, especially when that last 30% will be the most difficult. Also not sure how you determine there is less than a full stop of improvement.

Overall, I was agreeing with you. In a round-about sort of way. Just trying to point out that you weren't taking surface to receptor area efficiency into account or active area / fill factor. But I was also saying that in the end it doesn't matter because of the progression and evolution of the technology. While one photosite design and manufacture process may give diminishing returns if scaled to a smaller size, that does not necessarily hold true for the next design. So far, sensor manufacturers have largely continued to deliver improved performance along with higher photosite densities, just as microprocessor manufacturers continue to deliver smaller, and more powerful CPUs. This discussion becomes largely academic in nature because at this time, and for the foreseeable future, sensor makers are not going to break that trend.