RED Pixel Size
Published : 4th April 2010
Pixel size does not necessarily translate directly to native sensitivity of a sensor. While it leans that way, there are a lot of other factors involved that can skew the rule. Is it a 3T, 4T or 5T pixel? The more electronics that live inside the pixel, the less light gathering area there is left and the less sensitive the pixel is. Is it a "back-lit" pixel design? Is it a new "MINA" pixel? The long and short of it is that it is dangerous to judge too much from the pixel size alone.
For example, Arri's new sensor program (which looks quite nice from what I could see at IBC) uses dual gain paths to create a great looking image at ISO 800. The Mysterium-X (in EPIC) uses a new pixel design and some new technology to get to the same place. The Mysterium-X has the same noise characteristics at ISO 800 of Mysterium (in a RED ONE) at ISO 250, even though the pixel size is exactly the same for both. Neither company's sensor program achieves this "clean at ISO 800" look with just the pixel size.
In the end, what matters is what the image looks like. What ISO can I shoot and get the noise results I desire? What is the latitude?
What is the resolution and frame rate I can shoot? These are truly exciting times. The idea that digital sensors can provide a great film alternative seems very realistic right now. And the progress has only begun.
WA, NV and CA
>>Pixel size does not necessarily translate directly to native sensitivity of a sensor.
Yes, quite so. As best I can tell, all else being equal, a large photosite and a small photosite will have the same sensitivity, if you base sensitivity on the amount of light it takes to saturate the sensor (hit the clipping limit).
If, OTOH, you base sensitivity on a certain level of signal above the noise floor (however defined), then the larger photosite will seem more sensitive, because it will be less noisy for any given light level.
Where it gets complex is that not everyone rates sensors the same way, and some of the newer, smaller CMOS sensors on some of the big-name cameras appear to be rated a stop or two faster than they natively appear to be (witness the prevalence of negative gain settings on some of the newer small-sensor prosumer cameras). Since sensitivity is a competitive item, it behoves vendors to push (pun intended) for higher ratings, as long as the sensor architecture and the preamps deliver low-enough noise to allow for it.
Adam Wilt / filmmaker, MeetsTheEye / writer, provideocoalition.com
>> Pixel size does not necessarily translate directly to native sensitivity of a sensor.
>>Yes, quite so. As best I can tell, all else being equal, a large photosite and a small photosite will >>have the same sensitivity, if you base sensitivity on the amount of light it takes to saturate the >>sensor (hit the clipping limit).
Exactly. I've been waiting to see who would bring this up. Should have known it would be Adam.
It's the reason I originally said "I'm not talking about quality here, just the amount of light it takes to hit peak white with the same stop on the lens and 0 db gain." In other words, what it takes to hit the saturation point, where an increase in light gathering no longer results in an increase in output.
If you have a bucket and a coffee cup, and you put them outside when it's raining, the bucket will certainly have the ability to gather more water, but the cup will hit "saturation" much faster.
If you replace the cup with a cylinder with half the diameter of the bucket, but tall enough so it has the same volume as the bucket, they will "saturate" at the same time.
You would have to make the cylinder opening a lot smaller than half the diameter of the bucket (and the cylinder a lot taller) before the bucket would "saturate" faster. And even then, if you put a funnel at the top of the cylinder (the way manufacturers put magnifying glass on the photosites), you could bring things back to being more or less equal in terms of hitting that "saturation" point.
Back to sensors, manufacturers have all sorts of tricks (mostly optical) they use to increase the sensitivity of smaller chips. Some use dichroic mirror assemblies instead of prisms (which are more lossy), other glue tiny magnifiers onto the sensor. And, in general, smaller chip cameras are noisier
than bigger chip ones, and that added noise is tolerated because smaller chip cameras are less expensive. But that also means they can have their preamps generate more gain and increase the apparent sensitivity of the camera further. And having a lens concentrate the same amount of incoming light onto a smaller area doesn't hurt, either.
Looking at raw sensor sensitivity, while interesting as an academic discussion, does not fully equate to the overall sensitivity of a camera. No one I know of shoots anything for money (or love) in a lab using a raw sensor.
>>Pixel size does not necessarily translate directly to native sensitivity of a sensor. [...] In the end, >>what matters is what the image looks like.
Agreed. But what happens when people don't know what the image looks like?
With film, people did it right: compare images from two different films at the same size. But with digital they use "100% crop" to see "what the image looks like", but that is highly flawed.
>> the larger photosite will seem more sensitive, because it will be less noisy for any given light >>level.
I kindly disagree. There are three common reasons why I find myself in disagreement with many others about pixel size:
1. Mixing up sensor size with pixel size
2. Image analysis flaws
3. Different assumptions/circumstances
Many discussions about image sensors take it for granted that any change in pixel size is always a change in overall sensor size, because reducing sensor size is vitally important in many lucrative image sensor applications. Within that context, it's correct to say that smaller pixels have worse performance, because what it really means is that smaller sensors have worse performance. But that's a totally different perspective than keeping sensor size the same and only changing pixel size, and the subject line would be "Sensor size..." instead of "Pixel size...".
As for the second reason, there are many common image analysis mistakes (explored below). If none of the analysis mistakes are made, then there are some circumstances where large pixels have slightly better performance. But it's a much smaller difference than most people think, because they are still making the many image analysis mistakes.
When image analysis is performed correctly, it becomes very clear to the viewer of an image that it is possible for small pixel sensors to have worse performance "per pixel", but the same performance "per image". That is, when images from both sensors actually displayed or used for the same purpose, the performance is equal, despite the fact that the performance per pixel is unequal. Furthermore, that very condition is common in real world image sensors. Such ideas may be very surprising or at least counter-intuitive to many people who work with digital images, but there is ample proof.
With correct image analysis, many statements about small pixels are only myth, thanks to the ingenuity of sensor designers.
Portland, OR (after a nice vacation in Whistler)