Microsoft's Curved Sensor - Thinking Outside the Box

[Tim Tyler]

The Project Vermont Team set out to solve one of the most well-known optical issues in conventional photography: field curvature.

 

 

Field curvature exists because lenses don’t want to focus on a plane. A lens delivers sharpest focus along a curved path, rather than on a flat sensor. The result: objects appear out of focus across the image. By curving the focal surface, we eliminate the need for the lens to address field curvature and can use the resulting optical design freedom to improve other aspects of the system. Specifically, we can improve resolution, light-gathering capability, lateral chromatic aberration, distortion and illumination uniformity while also reducing system size, complexity and cost.

 

Project Vermont

https://www.microsoft.com/en-us/research/project/project-vermont/

 

Highly curved image sensors: a practical approach for improved optical performance

https://www.osapublishing.org/oe/abstract.cfm?uri=oe-25-12-13010

[Simon Wyss]

One good idea in the video realm for a change

[Mark Kenfield]

I assume such a sensor would require lenses redesigned so that the don't correct for field curvature?

[Dom Jaeger]

I assume such a sensor would require lenses redesigned so that the don't correct for field curvature?

 

Indeed, all previous lenses would be unsuitable. Bit of a fly in the ointment.

[Phil Rhodes]

I think this is mainly being considered for fixed-lens cameras, where it would make the lens design a lot easier.

[Tyler Purcell]

I don't know how you can make a pixel curve. That would introduce a whole new problem.

[Mark Kenfield]

I don't know how you can make a pixel curve. That would introduce a whole new problem.

 

What's that?

[Tyler Purcell]

What's that?

Umm, how can you "curve" a pixel and keep them the same shape across the entire imager?

[Gregg MacPherson]

In one of those articles they talk about how they deform the sensor into a spherical surface. I couldn't tell if each pixel took this deformation or if it's accommodated by the space between the pixels. If one had to deform the pixels themselves, they are so small relative to the surface curvature that the deformation for each pixel would be very small.

 

But it is puzzling, the motivation. It could make all existing lenses obsolete. Maybe its an idea that will be applied to mass volume stuff with tiny sensors like cell phones...?

[Simon Wyss]

Man, a little go-ogle and Wikipedia tells

 

https://www.dpreview.com/articles/2279255612/sony-s-curved-sensors-may-allow-for-simpler-lenses-and-better-images

 

it’s been around for some time.

[Robert Hart]

Nothing really new. The Kodak Brownie 127 snapshot camera of the mid 1950's ran roll film over a horizontally curved film plane.

It will be interesting to see where they go with a spherical sensor. There will be all manner of possibilities with VR image aquisition.

Edited June 21, 2017 by Robert Hart

[Tyler Purcell]

Nothing really new. The Kodak Brownie 127 snapshot camera of the mid 1950's ran roll film over a horizontally curved film plane.

Umm, it has a very slight curve, nothing like this new imager. Plus, going horizontal or vertical isn't a problem, especially at not much of an angle. When you do BOTH horizontal and vertical, now you add a whole new can of worms.

[George Ebersole]

The first thing that comes to mind is better miniature effects' shots, but since everyone uses CGI these days ...

[Phil Rhodes]

I think the intention is to flatten out distortions inherent to the lens, which would put everything back in square. That's sort of the point.

 

P

[Gregg MacPherson]

Phil, are you taking the piss? Flattening out distortions sounds like a crude description of what the best lens designers have achieved already (for flat film planes and sensors).

[Phil Rhodes]

I think the idea is that if we didn't have to build that sort of correction into the lens, the lens would end up being better, cheaper, or some combination of the two. There have been attempts at similar things as regards chromatic aberration, where some interchangeable-lens cameras can be aware of the lens and its settings, and automatically correct aberration digitally. Yes, it's possible to correct for chromatic aberration optically, but there are downsides to doing so.

 

Again, I think this tech is really intended for either simple interchangeable-lens cameras, or fixed-lens cameras, for things in the consumer world - not necessarily for high-end cinematography devices.

[Adrian Sierkowski]

This certainly; currently, isn't for anything at all "pro" it's the idea to simplify the design of lenses (and minimize the use off elements therein) by making the sensor itself essentially into a corrective optic by deforming it based on the design of the whole thing. This would be great in cell phones or gopros, or anything where you have a small sensor and a fixed lens. While they mention going up to APS-C size in their tests (in the paper provided) it's much more an idea to help get the most out of the tiny sensors in phones etc.

[George Ebersole]

So it's essentially marketing to draw more buyers to a product, and not really an improvement on existing industry grade tech. Go figure.

[Phil Rhodes]

Based on the theory, I don't think that's entirely fair - it could make things better without making them bulkier or more expensive, or it could make things less bulky and cheaper without sacrificing performance. Regardless of the market segment at which it's targeted, those are worthwhile things.

 

P

[Simon Wyss]

The direction is towards XL capture. When you have enough light and can afford an f/2.8 opening to begin with you don’t need that. Tens of thousands of small-gauge film cameras have been in use with triplet lenses like this. At f/1.4 and wider sharpness looks different.

[Keith Walters]

The problem with most lenses is the majority of the incoming photons wind up out of focus and far from the target surface. With a curved sensor a much greater percentage of them actually get used. The lens has to be a completely different design of course.
This is really only intended for improving cellphone cameras, though; it's not really applicable to normal program production.

Curving the sensor surface isn't quite as difficult as it sounds. Part of the process of making computer chips is that the silicon wafer is first criss-crossed with diamond scribers to create fracture lines (similar to what glass cutters do), the wafer is glued to a stiff rubber sheet, and then the sheet is deformed by placing it across the mouth of a cup-shaped container attached to a vacuum pump.

At a certain point, the wafer shatters into small square chips, which is where the term "chip" actually comes from. The chips are then mounted in their application packages and microscopic gold wires are welded between the chips' connection pads and the package's solder pins.

The thing is, it's quite astonishing how far the wafer can deform before it shatters.

Most modern CMOS sensors use so-called "backside illumination", where the light gathering photocells are on one side of the chip and the processing circuitry is on the other, which gives a CCD-like fill factor but with CMOS versatility.

Doing this requires the chips to be extremely thin, which means they can be deformed considerably without damage. The hard part is actually applying the Bayer Mask.

[George Ebersole]

Based on the theory, I don't think that's entirely fair - it could make things better without making them bulkier or more expensive, or it could make things less bulky and cheaper without sacrificing performance. Regardless of the market segment at which it's targeted, those are worthwhile things.

 

P

 

Maybe. I mean all tech, products and other progress is market driven, but it's like the major film companies improved their products because they wanted to, and this doesn't feel like it was done in that vein.

 

Just my cynical side.

 

But yeah, it could slim down tech. Yet it seems like digital image gathering tech is slimmed down enough for practical purposes. Still, it's interesting.

[Mei Lewis]

I'm interested in what the optical character of newly designed lenses for curved sensors would be. How would the bokeh and flare look? Would there be other, perhaps pleasant, aberrations like the non-circular bokeh of anamorphics?

[Simon Wyss]

A curved sensor isn’t a remedy for inherent lens flaws such as chromatic aberration or distortion. It helps in designing optical systems with less coma, astigmatism, and higher order defaults. Field flattening elements can be left out, so simpler and shorter rear groups are feasible. In short, triplets can have a renaissance. I could also imagine the Heliar design to find back to the show.