Combining Sensors...is it possible?

[Wild]

I just read an article over at PopSci about a new camera being constructed that uses dozens of large format sensors in a large array to create images. It got me to thinking about using a similar technique in DSLRs. Why can't a camera have say four m4/3 sensors in a 2x2 pattern to make a similar sensor to 35mm? Anybody have any ideas?

Here's the article:
http://www.popsci.com/technology/article/2012-04/worlds-biggest-digital-camera-designed-observing-space-gets-closer-reality

 

[PeterJ]

Link seems to be broken, but do you think 4 smaller sensors would be significantly cheaper than a 35mm sensor?

Technically it may be a little cheaper for the sensors themselves, production costs and yields tend to go up a fair degree with increasing die size, however that would be offset by increased costs of the sensor read electronics and assembly costs. I'd also assume they'd be a gap between sensors which might not matter for that sort of application because maybe they always interpolate multiple frames. I suspect in that particular case it makes sense because they are dealing with a size well beyond any practical manufacturing techniques for making it on a single die.

 

[FunPhotons]

Link seems to be broken, but do you think 4 smaller sensors would be significantly cheaper than a 35mm sensor?

Technically it may be a little cheaper for the sensors themselves, production costs and yields tend to go up a fair degree with increasing die size, however that would be offset by increased costs of the sensor read electronics and assembly costs.

How do you know this? I work in the semiconductor test industry, and while I don't know for sure I believe that cost of producing larger IC's goes up way faster than producing more smaller. Consider a CPU board which has many small IC's that you can buy for $100. If all those were combined into one SOIC it would cost a fortune.

I believe this technique of combining smaller chips is used in very large sensors for scientific applications (astronomy mostly). Knowing nothing about it, I would guess the problem is that you would need to have the FOV's overlap by 'tipping' the sensors relative to each other, which would seem to work when you're always focused at infinity (astronomy) and not so well when you're taking a picture at near field. Don't know.

 

[jhpeterson]

Link seems to be broken, but do you think 4 smaller sensors would be significantly cheaper than a 35mm sensor?

Technically it may be a little cheaper for the sensors themselves, production costs and yields tend to go up a fair degree with increasing die size, however that would be offset by increased costs of the sensor read electronics and assembly costs.

How do you know this? I work in the semiconductor test industry, and while I don't know for sure I believe that cost of producing larger IC's goes up way faster than producing more smaller. Consider a CPU board which has many small IC's that you can buy for $100. If all those were combined into one SOIC it would cost a fortune.

I believe this technique of combining smaller chips is used in very large sensors for scientific applications (astronomy mostly). Knowing nothing about it, I would guess the problem is that you would need to have the FOV's overlap by 'tipping' the sensors relative to each other, which would seem to work when you're always focused at infinity (astronomy) and not so well when you're taking a picture at near field. Don't know.

Apparently the reason why the cost for larger sensors increases significantly with their size has more to do with the greater effect that a defect anywhere on the semiconductor wafer has upon a particular sensor. If a wafer yields only a very small number of sensors, a single defect has a much larger impact than when that same wafer would yield several hundrerd.

 

[PeterJ]

How do you know this? I work in the semiconductor test industry, and while I don't know for sure I believe that cost of producing larger IC's goes up way faster than producing more smaller.

Sure they do, they have lower yields, but unlike a PC motherboard for example you have the additional cost of trying to precisely match and align things. Not to mention anyone can buy say an ARM processor for under $10, but you won't see an ARM core systems on the top 500 supercomputer list.

At some point / volume it makes sense to put everything on one piece of silicon, unless you're at the cutting edge when you need a 1000 image sensors or CPUs for your cutting edge design.

 

[YellowJersey]

The Gaia spacecraft will use 106 CCD sensors to survey the stars and will have a combined resolution of about 938MP.

http://en.wikipedia.org/wiki/Gaia_%28spacecraft%29

 

[elflord]

I just read an article over at PopSci about a new camera being constructed that uses dozens of large format sensors in a large array to create images. It got me to thinking about using a similar technique in DSLRs. Why can't a camera have say four m4/3 sensors in a 2x2 pattern to make a similar sensor to 35mm? Anybody have any ideas?

Here's the article:
http://www.popsci.com/technology/article/2012-04/worlds-biggest-digital-camera-designed-observing-space-gets-closer-reality

I read somewhere that they used this technique to make medium format sensors. It's possible, but may not be the most cost effective approach.

 

[Wild]

Link seems to be broken, but do you think 4 smaller sensors would be significantly cheaper than a 35mm sensor?

Technically it may be a little cheaper for the sensors themselves, production costs and yields tend to go up a fair degree with increasing die size, however that would be offset by increased costs of the sensor read electronics and assembly costs. I'd also assume they'd be a gap between sensors which might not matter for that sort of application because maybe they always interpolate multiple frames. I suspect in that particular case it makes sense because they are dealing with a size well beyond any practical manufacturing techniques for making it on a single die.

Sorry the link didn't work for you, here's a different one about the same story:
http://www.photographybay.com/2012/04/30/3-2-gigapixel-camera-has-189-image-sensors/

And ya I was specifically referring to the cost of the overall sensor. I'm not sure, but I figure since the cost of sensors relative to their size is exponential that combining smaller sensors would be more cost effective. Oh and I was also just curious :

 

[Wild]

How do you know this? I work in the semiconductor test industry, and while I don't know for sure I believe that cost of producing larger IC's goes up way faster than producing more smaller.

Sure they do, they have lower yields, but unlike a PC motherboard for example you have the additional cost of trying to precisely match and align things. Not to mention anyone can buy say an ARM processor for under $10, but you won't see an ARM core systems on the top 500 supercomputer list.

At some point / volume it makes sense to put everything on one piece of silicon, unless you're at the cutting edge when you need a 1000 image sensors or CPUs for your cutting edge design.

I figured that there's probably some obvious costs or difficulties involved with this sort of technology which is why I've never heard of anything like this before. I think it's impossible that nobody has ever looked into this as a possible way to lower large-format digital sensor costs.

 

[Wild]

The Gaia spacecraft will use 106 CCD sensors to survey the stars and will have a combined resolution of about 938MP.

http://en.wikipedia.org/wiki/Gaia_%28spacecraft%29

That was an interesting read. Same principle here I guess, just trying to figure out if it'd be worthwhile to do something like that in a consumer DSLR.

 

[Mt Spokane Photography]

Developmental costs would be pretty high. Digic processors are only setup to handle one sensor. Balancing color and brightness between sensors, combining images, synchronizing them, and getting rid of 4X the heat would probably push the cost out of sight, and even then, a large sensor would likely outperform the combination.

I'd never buy one over a single sensor, too many compromises that are not necessary. For a multimillion dollar aerospace uses where a multi million dollar supercomputer reassembles the images over a period of days with a staff of scientists to manage it is one thing, a consumer camera is another.

 

[Wild]

Developmental costs would be pretty high. Digic processors are only setup to handle one sensor. Balancing color and brightness between sensors, combining images, synchronizing them, and getting rid of 4X the heat would probably push the cost out of sight, and even then, a large sensor would likely outperform the combination.

I'd never buy one over a single sensor, too many compromises that are not necessary. For a multimillion dollar aerospace uses where a multi million dollar supercomputer reassembles the images over a period of days with a staff of scientists to manage it is one thing, a consumer camera is another.

Your reasoning makes a lot of sense. It seems like it would be pretty impractical to use an arrayed sensor setup like the ones they're using for telescopes. I just thought it would be pretty cool if it actually was cheaper than the equivalent sized sensor and could offer the same image quality. Thanks for the response!

 

[3kramd5]

For a multimillion dollar aerospace uses where a multi million dollar supercomputer reassembles the images over a period of days with a staff of scientists to manage it is one thing, a consumer camera is another.

The alignment/reassembly shouldn't be that difficult. It's that essentially what Hasselblad's H4D-200M does (albiet by moving the sensor rather than combining sensors, thus eliminating your first set of concerns re. color, brightness and heat).

 

[!Xabbu]

Very interesting concept and I don't see why this shouldn't be a possible way technology will go. Once you set up your assembly line to calibrate everything it should be business as usual from that point on. However, perhaps I'm also way underestimating the difficulties.

 

[bycostello]

Photoshop and panning, you can have unlimited sized sensor....