Is there a theoretical limit?

[Loupey]

Just wondering if anyone has heard of any manufacturing limitations in regards to pixel density of imaging sensors.

I'm sure no manufacturer is willing to leak out information regarding what they will be capable of in the future, but I'm thinking that there must be some limitation of cramming so many pixels and microlenses in a given area.

The latest breeds of cameras now exceed 40,000 pixels per square millimeter.

I'm wondering if the pixel density curve is/will slow towards some asymptote. Having that knowledge may help people understand where they stand with their current cameras and how many upgrades cycles they expect to undergo before the number of MP becomes irrelevant (perhaps already?). A lot has to do with how large of an image or how extreme of a crop one wants to perform so the "need" will always be there for more I suppose. But can mass-production machines continue to make smaller and smaller pixels? And how much smaller?

Or, conversely, is there a theoretical density beyond which most lenses cannot produce an appreciable difference? So no matter how many more pixels we have, lenses cannot utilize them?


Will the pixel game follow Moore's Law so that we are hopelessly shackled forever to the "need" to upgrade?


Just thinking out loud.

[Skyman]

When I worked in a camera shop we sat down (the other salesman was a bio chemical phd student at the time) one slow day and based on current technology then (about 5 years ago) we calculated that it would be theoretically possible to build a sensor over something like 300,000 mega pixels on a standard ccd. connecting all those pixels up and stopping them from turning into a molten mess is another story. I am sure there will be something on the horizon that will render the ccd / cmos redundant. In an ideal world we could have cameras with interchangable ccd's, like the old days when upgrades were more about improvements in film rather than camera.

[another view]

I have no idea - however I will say that one thing that got my interest with my Lumix LX3 was that they increased the physical size of the sensor without increasing the number of pixels (10mp). It just came in, and is too new for a lot of reviews on it but image quality is supposed to be better than the LX2. I didn't have one of those but so far I'm pretty impressed (nothing worth sharing yet).

People's perception (in general, not everyone of course) is that more is better with mp. I don't think so - give me a good number of high quality pixels, however they have to do it.

[mikeb380]

Just wondering if anyone has heard of any manufacturing limitations in regards to pixel density of imaging sensors.

I'm sure no manufacturer is willing to leak out information regarding what they will be capable of in the future, but I'm thinking that there must be some limitation of cramming so many pixels and microlenses in a given area.

The latest breeds of cameras now exceed 40,000 pixels per square millimeter.

I'm wondering if the pixel density curve is/will slow towards some asymptote. Having that knowledge may help people understand where they stand with their current cameras and how many upgrades cycles they expect to undergo before the number of MP becomes irrelevant (perhaps already?). A lot has to do with how large of an image or how extreme of a crop one wants to perform so the "need" will always be there for more I suppose. But can mass-production machines continue to make smaller and smaller pixels? And how much smaller?

Or, conversely, is there a theoretical density beyond which most lenses cannot produce an appreciable difference? So no matter how many more pixels we have, lenses cannot utilize them?


Will the pixel game follow Moore's Law so that we are hopelessly shackled forever to the "need" to upgrade?


Just thinking out loud.

Loupey;
I think the only thing keeping the game going is all the many people who absolutely have to get "THE LATEST AND GREATEST!"

In my experience, most of the people in my camera club never print a photo, they just publish to the web at 72dpi. Yet all of them (well, most, anyway) insist they have to have a camera with the highest MP ratings and nothing but L series lenses ( for Canon users) and yet none of them print photos; at least not that I have seen. I have a Canon XT with 8MP as I gave my 300D to my son-in-law and I really can't tell the difference in photos taken with the two cameras, even cropped and printed at 11X16.5 inches. And I have only one L lens, a Russian lens for my FEDs and Zorkii RF cameras.

People keep trying to "Keep up with the Joneses" even when they, themselves, are the Joneses. All I can say is that all the manufacturers are delirious with joy as the know nothings keep buying at every new announcement.

As to your original question, I don't know if there is an answer, just that I don't feel it is important. Once they get all the MP crammed into a camera and there can't be any more, the manufactirers will find some other trigger which will keep those people coming back for more. Meanwhile, I'll keep printing 11X16.5(17) and 13X19 photos from my meager 8MP. (No, I haven't printed a 13X19 yet but am buying paper next payday. I'm looking forward to it.) Actually, your question is the same one asked years ago about microcircuits on computer chips. So far there is no limit there, so maybe no limit to MP in a camera¿

Cheers
Michael

[danic]

Loupey;

Once they get all the MP crammed into a camera and there can't be any more, the manufactirers will find some other trigger which will keep those people coming back for more.

Have a look at Canons new 5D. It has all the bells and whistle's. What a waste IMO

[drg]

Excellent question!!

1. There's different answers dependent upon the manufacturing technology. Fundamentally the number of ('transisitors' or circuits per unit area) divided by the (number of 'transistors' or circuits necessary to make the photos sensitve buckets plus the number of 't' or c's needed to transport the data off the chip(s).
I intentionally used common terminology to make this a bit more readable. The chip makers have been very guarded as to what tech they use and how far they will go. There's a doubling or two left easily I believe (~100 Mega-pixels) in what is currently employed for the current realm of DSLR format camera. Could you use the image?? I doubt it for many reasons.

100 Megapixels in a large chip (MF or sLF) is virtually available now. Slow, unwieldy and you need to mortage something to get one.

2. This does not factor in issues with noise, heat, and power that are all related by formulas available from the appropriate manufactuer via spec books. Different level of integration (UVLSI, SSIC, etc.) have distinct characteristics.

3. Many lenses were exceed in resolution by the large MF-style chips some time ago. I've got older lens that unique characteristic flaws that on film add character, with the bigger chips just are unusuable. Thus the redesign of all the Hasselblad gear over the past few years. Also I believe the same is applied to the new Leitz lens offering.

Finally as new sensor arrangements emerge the limiting factor also may become how much processing the market will bear in the camera and whether they will be full time video capable as well.

I'm waiting for something along the lines of the 3 chip 'still' camera. There have been experiments but they have been costly for the consumer, even professional due to product consideration.

I've got some more thoughts on this for later!

[Sushigaijin]

What about the foveon chips? Those effectively triple the usable space on a chip surface, which is why they didn't sell well - no one would buy a 3 mp camera in a 9mp market. However, I have a feeling they didn't deliver the promised resolution boost because they were stillborn; they sold poorly and that left no money for R&D. I'm sure with proper funding these could be very nice sensors.

A foveon would reduce a 40,000 p/mm2 to about 13,000. That's a grande pixel with an RGB value rather than three small pixels with only R, G, or B values...reduce the pixel size to 40,000 p/mm2 and you have a CCD equivalent of 120,000 p/mm2

Sounds like a bandwidth nightmare to me.

[mikeb380]

Have a look at Canons new 5D. It has all the bells and whistle's. What a waste IMO

Danic;
I agree with that. I never use the "bells & whistles" on my camera. I set everything flat and shoot RAW, if the menus went away I'd never miss them except to use mirror lockup and format the CF card and those could be replaced with a button or switch. To me, the best possible digital camera would be a Canon F1 with an 8MP digital back. Who needs AF and/or AE? I learned to use my camera to make exposures and do pretty well, thank you. I tried to teach classes on basics of exposure and focus to members of my camera club and most were too lazy to attempt. Just flash, bam, thank you maam and they were off to the next snapshot. :mad2:

LOL
Michael

[SmartWombat]

Bandwidth?
Chip has 4 edges, so use 4 image processors
Canon are using dual DIGIC now, I predict quad DIGIC in future.

Foveon is great, I really like the concept.
But the interconnections in each layer of the sensor will make a 3-dimensional puzzle that to some extent blocks the light to the lower layers of the sensor.
Or limits the space between the photosites so you can't get so many on the sensor.

[Axle]

To me, the best possible digital camera would be a Canon F1 with an 8MP digital back.

Ya know, when I still had by Minolta X-7A, I was thinking the same thing, why not working on adapting film cameras to use Digital mediums to record images, a Digital back as you said.

Them have them for Medium format cameras...why not 135 size?

Off topic, sorry.


But anyways, there is a theoretical limit I think, however with every day technology improves and that limit is pushed further. We are limited in our views by our level of technology and available materials.

[freygr]

There are some problems, 1) the smaller the sensor cell is the more amplification you need and amplification means more noise. 2) the more pixels the less room for pixels (more interconnects).

The pixel count will keep increasing as they improve the low noise amplifiers and making finer features. But for all piratical purposes unless your going to print large photographic prints it will be why.

[Franglais]

I have another take on this. There IS a theoretical limit to the number of pixels you put on a sensor due to light being made up of photons, which are like particles and indivisible. You can't keep on making the pixels smaller and smaller because there's going to come a time when there aren't enough photons to go around. The only thing you can do is to increase the exposure time to allow more pixels to arrive i.e. you drop the ISO sensitivity of the sensor.

[drg]

Another factor is chip/wafer yield. When IC's are fabricated there's always a critical/break even point as to how many you must manufacture to make money. The sensor chips in cameras have untill recenly been made in such small numbers as to only barely approach yield potential. Yes, even millions of sensor chips is barely enough and that's why so many cameras have at times 'shared' the basic sensor.

The small sensor (less than 1 Mpixel) are of sufficient heritage to make security, manufacturing, and remote sensing cameras quite cheap if CCD technology is employed. With today's processing environments, you can do almost anything in real time (or no noticeable lag) to this imagery.

One yield issue that is always interesting in wafer making is how many usable circuits there are. In the not too distant past less the 85% active links was the threshold. Now, how many extra pixels are really on a sensor is unknown. We are seeing chips now with some extras (the real number obviously isn't being reported) that can be cross mapped (like what used to be the first thing you did with a hard drive) to eliminate hot-pixels.

Then comes the second and third generations of CCD and CMOS chips. More connections, heat, production loss, quality, etc. that added to the cost. This becomes a serious expense with each new and larger chip. The 10 Mpixel range chips are reportedly the 'break' even chip size for all cameras. Six to 8 Mpixel were the first reliable quality units for production and by the third of fourth 'release' they were profitable. Look at the quarterly margin/profits and stock holder reports for Kodak, Fuji, and Sony. Compare this to what was being sold. Analytic reports from various firms are available, but the more easily available and essentially free public shareholder reports are just as telling with a little bit of deduction.

The FOVEON chip has a 'bigger' problem than just getting 14 million pieces of data (latest version) off the chip. That problem hopefully has been somewhat improved with the new imaging processing engine. FOVEON chips have a 'well' problem. The light has to get to the bottom of the well for all three colors to be registered! It isn't about circuitry blocking light as much as it is about the layers of circuitry alongside the sensor/light path limiting the number of light sensitive locations. These are limited by sensor density (p/mm2) due to the complex nature of the design. Remember these chips use the natural phenomenon of silicon to transmit the different wavelengths of light at different rates to acheive delivery to the three photo sites.

FOVEON equipped cameras perform very differently with various lens constructions. I've shot 25,000 shots probably on FOVEON equipped cameras and still get fooled even when I'm using one with some regularity when I switch lenses and don't check 'the notes' as to what will occur. This is as much due to the angle the light strikes the mirror/lenses into the photo sites as anything. Again, a real estate problem!

The more space that can be put between the sensor sites vs. the ability to control noise, vs chip manufacturing yield v how much I'm willing to pay for the d*** thing is going to be the real limit for a few more years.

[Loupey]

Thank you all for sharing your excellent insights, ideas, and comments.

I too think that manufacturing technology will continue to push limits of pixel density (and the associated processing of all that data and handling of heat and noise) on an on-going basis for quite some time.

But I would think that, at some point, the physical properties of the world is going to step in - like the wavelength of light (albeit the wavelength of red is still many orders of magnitude smaller than the current pixel spacing) or perhaps the air itself inside the camera body. Definitely the resolving power of the current lenses will be the first "roadblock" for many manufacturers. Once that limit is exceeded, a whole new line (as drg mentioned) would be necessary to reap the rewards of such an ultra-high MP camera. But will anyone/anything be able to hold a camera that still? Perhaps the minute seismic vibration of earth itself will prevent the need of such theoretically high resolution lenses/sensors?

Perhaps that roadblock will never be traversed. Perhaps by then, the average consumer will have had enough upgrades to resist starting all over with lenses.

Still, I don't want to dwell on the "why" aspect of more MP (that topic has been beaten to death). Instead, I was just wondering how far we can go.


Spy satellites went from being able to resolve something the size of cars to being able to read their license plates. And they have to do it while shooting through the entire atmospheric layer. Were the improvements just with the lenses used? Or was it the imaging sensor? Combination of many things I'm sure. But fun to think about

[Loupey]

I happened to catch a science documentary last night that was discussing chip technology and Moore's Law specifically.

Current theory is that Moore's Law, which states that computing power/capacity doubles every 18 months which has been in effect for the last 50 years, will soon be broken. The size of the transistors will reach a point no longer feasible by silicon.

When that happens, productivity will cease in increase, global economy will sputter, and doomsday will ensue.


Kinda like now.

[Speed]

Hey Loupey,

I was reading an article about this a couple of weeks ago. The march to more megapixels seems to have slowed down, though Sony may give that a renewed push...

The issue isn't as simple as how many megapixels can they cram onto a chip, but rather, at what point is pixel count irrelevant. As the count goes up, the pixels themselves become smaller. Trade offs are made, such as light sensitivity. While the manufactuers may be able to make a 100 megapixle sensor, if it's only usable in bright sunlight is it worth building?

Could Nikon have crammed 20 megapixels onto the D3's sensor? Of course. But they didn't. And the result is unbelievably high, and usable, ISO's that we didn't think was possible just a couple of years ago.

The D3 has 12 megapixels. The 1Ds has 21. I'll take the D3...