Tuesday, June 24, 2008

The Nikon D700, D3x and the end of Moore's law

The folk at the Photography Bay have leaked PR materials for the forthcoming Nikon D700. It is in essence a Nikon D300 with the full frame FX sensor of the D3. Meanwhile the D3 itself is to be replaced by the 22MP D3x.

People like myself who recently bought a D300 might be wishing they had held off except for the fact that:

  • The D700 is rumored to cost $3,000-$3,200 - almost twice the going price of the D300
  • There is no advantage to buying an FX sensor camera unless you are also going to invest in the full frame wide angle lenses such as the $1,500 14-24mm zoom.
  • On past experience, the earliest deliveries of the D700 can be expected is November/December and they are likely to be on back-order for a year or more.

The D3x meanwhile is rumored to cost $6,500, an increase of $1.5K over the already pricey D3.

The digital camera market might well be mistaken for the PC market. But there is an important difference, in the digital camera world the end of Moore's law is already in sight. Cameras work in the visible light spectrum and physics is already limiting designs. Nobody knows exactly where the practical limit will be reached, but the physical limit is set by the wavelength of visible light.

Red light has a wavelength of 740nm, the FX sensor is 36mm wide, or 47,000 or so wavelengths, the DX sensor is 25mm wide, or 34,000 wavelengths. The 12 Megapixel sensors deliver an image 4256 pixels wide which means that each sensor is 11 wavelengths wide in FX format, 8 in DX format. If the physical limit is 2 wavelengths, Moore's law predicts that we will reach it in only 6 years, at which point the DX format will deliver 160MP and the FX format 320MP.

In practice, we are already near to the limit of the optics and the sensors. Nikon moved back to the FX format because the DX format was already starting to appear 'noisy' at high ISO values. Admittedly the noise is nothing like what we used to see on what passed for 'high ISO' in the film days. But expect noise on the 22MP D3x to degrade in much the same way as the D300 when used in hi-res mode rather than the D3. And expect each doubling of megapixels from here on to cost an ISO stop.

All of which means that a sports photographer is unlikely to feel the need to trade in their D3 for quite some time. Any further improvements in resolution will likely come at the cost of speed. If a 160MP DSLR ever arrives it is likely to be of real interest only to the likes of landscape photographers and folk who have to have the whizzyest gadget to brag about.

The current DSLR shootout between Canon and Nikon is then a last rush to sew up market share before the pace of technological advancement starts to slow. Whoever sells the most DSLR bodies in the next few years will be in pole position to sell lenses for them for decades to come. It took Nikon 45 years to roll out its first six generations of professional film body but less than a decade to deliver nine professional DSLR bodies.

Will we see something similar when Mr Moore stops delivering ever faster CPUs?

Welcome Nikon Rumors readers. Some folk might wonder why I am not talking about diffraction as the limiting factor here. The answer is, yes it is a limit but one that can be avoided by using a bigger lens. The other issue that I did not consider when writing the above but came to me later is that noise reduction post-processing can be used to push ISO response albeit at the cost of resolution. Big pixels appear to be less noisy only because the light sample is being averaged over a larger area.


Christophe Muller said...

Very interesting post.

Michael Ryan said...

Kind of reassuring as i've got my name down for a D700. then comes the investment in new glass.... or is that the next big leap?

Justin said...

Great point about pixel pitch in terms of wavelengths. Clearly the Bayer pattern has to go.

Moore's "law" was based on a simple heuristic for investment in lithographic scaling, which ultimately became a self fulfilling prophecy. It has no fundamental bearing to image sensors.

A lot can still be done to improve efficiency and pixel density to increase pitch, for example using amorphous silicon deposited on top of CMOS as your photosensor sites, including carefully controlled layers which are sensitive to individual RGB bands.

Sadly though, I fear the biggest driver for improved image sensors may be the cell phone / p&s market, solely due to volume. (The same reason why "crappy" CMOS has won out.) DSLR users may just have to wait for these novel technologies to make it up the food chain.

Todd said...

"On past experience, the earliest deliveries of the D700 can be expected is November/December and they are likely to be on back-order for a year or more. "

If past, as in recent experience proves, Nikon will release the D700 near the end of July.

Nikon definitely learned from past experiences that bringing a product to market soon after announcement (look at Canon) is what the consumer/professional photographer wants/needs.