Photosite sizes on DSLR Sensors

So, as people who have been reading my Facebook posts would know – I recently became a Nikon convert – with a D3200 providing lots of megapixels for cheap. Of course, some of that is marketing gimmick, but when you want to crop things out, more pixels can work to your advantage. Unfortunately, it’s well known that having more pixels beyond certain points is detrimental to the quality of your photographs.

The reason is quite simple – the sensors themselves come in a fixed size. If you cram more photosites into the same area, each photosite has to be smaller. This has several problems – as each photosite is smaller, it collects less photons, and therefore is less sensitive – so you need to amplify it more. The signal also gets smaller, which means that noise from readback amplifiers and leakage becomes more prominent compared to your signal, reducing the signal to noise ratio. So the photosite area often correlates well with low-light performance.

This is an oversimplification of course, there are more variables – some of which include the noise reduction algorithms (getting better in more modern cameras, however, it does sacrifice detail to make a more pleasing image), microlenses (which improve the light collection ability of the pixels themselves), and lower noise electronics.

Having used my D3200 for a bit of a test – I noticed its RAWs were slightly noisy for even daylight shots, and fairly grainy for indoor shots. This was nothing like the Canon 400D (10MP) camera which I had used before which was pretty much clean in the daylight, and smeary-noisy in indoor shots. Of course, the onboard JPEGs from the Nikon looked quite good, even at high ISOs – but the detail was being trampled on by the inbuilt noise reduction.

So, I thought, how does the size of photosites on crop-factor (DX/APS-C) cameras compare to full-frame (FX/APS) cameras? If we can look at this, we can infer roughly what the comparative low-light performance would be.

I got out my trusty spreadsheet, and catalogued all of the current products and some of the past products from the two majors – Nikon and Canon. I took the given sensor size and output resolution to derive the photosite dimensions, which I multiplied to get area. This methodology isn’t perfect – as some of the sensor area may contain special sensors (not contributing to ouput image) whose area would be attributed to the image itself, but it gives a good indication nonetheless. What resulted from my graphing is intriguing. Click on the graph for Full Size –

So what did I learn/discover?

• The D3200 (along with the new D5200) which crams 24MP into a DX format sensor has the smallest pixel site area – an “obvious” duh moment. It’s likely to have the worst low-light performance of all.
• Canon has stuck to an 18MP APS-C sensor for most of their recent products and their photosite area is in the “gap” between the competing Nikon D5100/D7000 and D5200/D3200. In fact, I shoved in the 400D/450D/500D out of interest – they’re not actively being produced like all the other cameras on the chart.
• Almost all FX/APS format cameras (with exception of the D800) have larger sensor sites than the 400D (10MP) camera I started with! This means that almost any upgrade to full frame will produce cleaner high-ISO shots in low light – and might mean it’s actually quite worthwhile to go full-frame after all.
• The D800 photosite area is roughly equivalent to the D3100 14MP DX format camera and slightly worse than the 450D – so it’s not too bad, but it’s not too great either.
• If they adopted the photosite area of the D3200 to make a “new” D800 – their FX sensor would have 58MP!!!
• The Canon APS format cameras only span a small range in photosite areas – inbetween the D4 and the D600 of the Nikon range – whereas the Nikon leaves you wide-open for choice in terms of megapixels-to-noise tradeoff.
• There is a gap between the Nikon crop and Canon crop lines because of the slight dimensional differences between DX format sensors and APS-C sensors.
• The largest photosite (D700/D3S) is the same area as 4.8 D3200 photosites!
• The area of the photosites on my old 400D is almost 2.2x the size of the photosites on the D3200.

Maybe I should do less calculating, and more photographing – most people tended to say the DX sweet spot is about 12-15MP – if that’s true, the D800 sits right in there for FX, so … will 36MP be the FX sweet spot? We shall see.