Is noise performance of a sensor directly proportional to its size?

[nickzou]

I've always wondered what the relationship between low-light performance and sensor size was. Is it a linear relationship?

 

[tirediron]

Way too many variables involved to make a broad statement like that. Sensor size definitely plays a role, and typically, all other things being equal between two sensors, the larger one will have better high-ISO performance, but processing technology, pixel density, physical size of each pixel, pixel arrangement and a bunch of other things that I can't even pretend to understand.

 

[Derrel]

Take a look...the top 16 or so are all 24x36mm in size, and mostly made by Sony Corp...these are the lowest-noise sensors, which perform very well at elevated ISO levels. DxOMark - Camera Sensor Ratings

Currently, the Sony-made CMOS sensors used in d-slr type cameras (and the new SONY FF sensor used in that company's compact mirrorless camera) are ahead of the sensors made by other companies, so that the smaller, APS-C sensors can perform fairly close to the levels we USED TO associate only with FF (24x36mm) sized d-slr sensors.

AS far as noise being "directly proportional" to the size of a sensor...well, yes and no...medium-format sized sensors are actually not performing as well in terms of noise as are the newer, and smaller, 24x36mm-sized sensors. But the sensors used in FF d-slrs are performing a bit ahead of those in APS-C cameras, of the same generation and maker.

Clear as mud, eh?

 

[BrianV]

CMOS sensors integrate a number of signal processing functions onboard the chip. Some of these functions include taking ore than one sample off the detector and averaging the value to reduce noise. The idea being noise is random, and averages out over time. Low-light performance is greatly improved using signal processing, that's where all of the big gains have come for CMOS.

"Charge-Well Capacity", the number of electrons that can be counted before saturating- tends to be a function of element size. A larger detector can store more electrons. The Sensor used in the full-frame Nikon D4 has a very high capacity, much more than the D800.

 

[KmH]

Pixel size has a direct relationship to image noise as a function of the signal-to-noise ratio (SNR).
The bigger the pixels the better the SNR. Bigger image sensors generally have bigger pixels.

a Nikon APS-C image sensor has an area of 366.6 square mm. A full frame image sensor has an area of 864 square mm.
Each of 24 MP on an APS-C image sensor will be considerably smaller than each of 24 MP on a full frame image sensor.

However, in recent years, more and more pixels have been crammed onto image sensors, which means smaller pixels. So something else must be going on.

That something else is mostly about software, but part of it is also from improvements to electronics on the image sensor that aren't part of the pixels. Like the amplifier electronics.
DxOMark - SNR evolution over time

There are several different causes (kinds) of image noise. Image noise - Wikipedia, the free encyclopedia

The bottom line is there are several factors that contribute to minimizing image noise.

http://wwwimages.adobe.com/www.adob...ly/prophotographer/pdfs/pscs3_renderprint.pdf
Another technical problem prevents us from seeing all the way into the darkest regions of the scene. The sensor is sensitive to a wide range of radiation, not just the visible kind. This other radiation is all around us and is also emitted by components of the sensor itself. In the illustration the extra radiation is represented by condensation and leaking pipes. Each time we measure how full the bucket is, part of the sampled data is extra. It’s mostly random, and we don’t know how much of our measurement is this extra radiation. To compensate, the chip has rows of sensors with opaque covers. It measures the signal in these “dark” buckets to determine the average and maximum amount of stray radiation during the capture. This “dark current” measurement defines a level beyond which we can’t be sure of our data. In the vocabulary of signal processing we call this the noise floor.

 

[Garbz]

All other things being equal the quantum efficiency of a sensor determines how many photons are converted to electrons. If you can capture more photons (bigger pixels) then you have more electrons, and as I said all other things being equal including the noise it means you have greater signal to noise.

A lot of the above talks about signal processing and software but the reality is that CMOS sensors have changed a lot in recent times. Different layouts for the transistors and electronics introduce noise in different ways. There's been a lot of work done in recent years from ensuring that the electronics which control each quantum well don't cause noise. Then there's the other side of the sensor. Microlenses and coatings ensure that more photons end up hitting the sensor at right angles and as such have less of a chance of being reflected. Pixel dimensions have also increased in usable size, that is a pixel on a 35mm 12megapixel has a bigger surface area that converts photons into electrons than it sensors of yesteryear.

All of this in the name of a common goal, more signal, and less noise.

 

[rexbobcat]

Part of it is the sensor billy it also has to do with the processor.

Cameras typically have native ISOs that actually depend on the sensor (typically 100, 200, 400, etc...) but the others in between are approximations from what I understand, made by the processor, and so the processor does play some part in digital noise.

But typically bigger sensors have less noise IF they have large pixel wells. Medium format digital backs don't have ISO performance that is very good because they have so much resolution and therefore have smaller pixels. This causes an increase in heat (which can cause noise) and affects other factors that I'm not well versed about.

 

[KmH]

Native ISO is done electronically with amplifiers on the image sensor chip. It is my understanding the amplifiers contain a resistor ladder that provides for each native ISO value step, not just the full stop steps.

I don't know how much voltage a fully saturated CMOS pixel develops, but I would expect it to be in the µV range.

The Lo and Hi non-native ISO setting a digital camera may offer are done using software. Because they are done with software, a price is paid in the form of some decrease in image quality, like a noticeable reduction in dynamic range.

 

[Garbz]

Actually in most cameras the ISO is controlled in a combination thereof the analogue amplifier and altering the electrical bias of the photosite on the sensor.