Thursday 23 August 2007

Nikon's D3 and D300: Implications for Sony?

Nikon shooters have had some very good news today, the long awaited and rumoured D3 and D300 have been unveiled to a rapturous audience. They both boast impressive specifications and many other sites have covered these in depth so I won't reproduce them here. I will speculate and point out some features of the new Nikons which may have some bearing on Sony's DSLR efforts. Given the long history between Nikon cameras and Sony sensors, hopefully the latest information on the new models will provide some insight into the machinations behind Sony's closed doors.

We'll start with the D3. From Rob Galbraith, Nikon says that the sensor is a 36mm x 23.9mm CMOS sensor with 8.45 micron square pixels and a 12 channel, 14bit readout giving at least 9 fps in full readout mode and 11 fps in cropped mode. It is stated that the sensor is an original Nikon design, though it is unknown who actually fabricates it.

From the above, we note a few things. The 12 channel readout is different from the parallel column ADC system used in the IMX021. This is the conventional way of reading out a focal plane array and uses a few high speed analogue to digital converters to quantize the photo-electron signal in each pixel. The 14-bit ADC is interesting, I am not sure whether it truly represents a greater than 12 bit dynamic range or merely a marketing, "keeping up with the Jones'". We can estimate the full-well-capacity at about 55k electrons, similar to the Canon 1DII. The minimum sampling rate at 9fps is 9MHz per ADC. To exceed 12 bit dynamic range (DR), we require a read noise of better than 13.4e which is plausible, though whether this can be achieved at 9MHz is to be seen. The switchable 12/14 bit resolution may indicate that Nikon engineers do not consider the system having greater than 12 bit DR.

The D300 sensor is the most relevant to Sony's plans as on the surface, it looks like the IMX021 (left) recently announced by Sony. The D300 sensor is a Sony-made 12MP APS-C sized CMOS sensor with 5.49 micron square pixels and 14 bit readout at at least 8fps. The IMX021 only specifies 12bit output but at 10fps. It is unknown whether the D300 sensor employs parallel column ADC.

For such a smaller pixel (half the area and presumably FWC), the read noise would have to be half that of the D3 sensor to achieve the same DR. Hence, it is unlikely that the DR exceeds 12 bits, making the 14 bit ADCs superfluous, unless parallel column ADC and multi-sampling have been employed to significantly improve read-noise. A rough estimate gives a 36 multi-sampling with an ADC frequency of 1MHz at 10 fps leading to a reduction in read noise of 6.

It is this latter scenario which is promising for a future Sony DSLR. Though Nikon have a habit of ordering bespoke sensors from Sony (e.g. sensors for the D70s, D2x), the IMX021 seems to be a wholly Sony in-house development, hence the ability to use the parallel column ADC architecture for low noise, high DR imaging may not be proprietary to Nikon.

We will hopefully be able to infer more when Pentax makes its announcements soon, otherwise we will have to wait till Sony make their rumoured announcements early September.

PS A useful introduction to noise sources in CCD imaging is here. Much of the material on signal analysis applies equally well to CMOS sensors.

No comments: