Sunday, February 10, 2013

Nikon D7000 Sensor: Cross Section Review and Implications for Nikon D5200

Sony IMX071 sensor, used in Nikon D7000 and D5100

In a previous post, I went into detail about the so-called "ISO-lessnes" of the Nikon D7000 sensor. A combination of very high quantum efficiency (the ability to convert light into electrical signal) and very low read noise means that for normal use, there is virtually now difference between raising the ISO in-camera or digitally amplifying an under exposed image in post-processing. The D7000 sensor is now superseded by the D5200 sensor, but it is interesting to see in what areas that the latest and greatest will have to beat what was previously a state-of the art sensor. The following electron micrographs come from Chipworks by way of The Landingfield, an astrophotography blog. This is a cross section of the D7000 sensor (Sony IMX071 Exmor):

Nikon D7000 sensor.

 And for comparison, here is the Canon 5D Mark I (Left) and Market II (Right):

Canon 5DmI and 5DmII sensors.

Just for reference, what you are seeing from top to bottom is:

  • Microlens
  • Colour filter array
  • Wiring
  • Light Gathering Substrate

A few things stand out. The 5D Mark I is the oldest camera of the bunch, and the cross section shows it; the wiring takes up a much larger proportion of the area under the colour filter array and microlens, meaning that the 5D is the least efficient of the three at gathering light. Compare that to the 5D Mark II, where the path from microlens to substrate is wider.

Compared to either of the the Canons, the Sony sensor in the D7000 takes things further. Notice how narrow the wiring interconnects are, they hardly impede light coming in from the microlens. Also notice that the micro lenses in the D7000 are gap-less. The ones on the Canon sensors are almost hemispherical, but do have gaps. The microlenses on the Sony sensor aren't perfect hemisphere's, but flattened out a bit, with each lens connected into the next. Gap-less microlenses are more efficient at directing light past the wiring interconnects into the photodiode, but they theoretically are less capable with light rays coming in at extreme angles. I'm not an expert, but my guess is that the very narrow wiring interconnects ameliorates this, so the D7000 sensor gets the double benefit of being able to gather a high quantity of light and handle extreme angles (important when capturing images with wide-open apertures) at the same time. What's striking is that you can see from the cross-section that the D7000 sensor wouldn't benefit that much from back-side illumination (BSI), as the path from the microlens into the photodiode is already very wide and clear. Here is what the tightly packed Bayer array looks like from above, courtesy of Chipworks by way of Foto Actualidad:

It was(is) a surprise to learn that the D7000 microlens arrays gap-less, since neither Sony nor Nikon made much mention of this at the time of launch. Digging around into the Chipworks literature, it also turns out that the the Sony-made D7000 sensor gets by with ultra-low read noise because of on-chip Analog-to-Digital Converters (ADC's) running at a very low rate, something on the order of 20kHZ. By comparison, the D3/D700/D3s and Canon full frame sensors use a different design, where the ADC is external and running at a speed of 10MHZ, which is 500 times faster.

Looking ahead and seeing what Toshiba has been able to accomplish with the Nikon D5200 sensor, it would be interesting to see what the copper wiring interconnects will look like under an electron microscope. Until now, Sony has been using aluminum fabs. However, aluminum has roughly two thirds the conductivity of copper, so I'm interested to see if Toshiba managed to further reduce the thickness of the wiring under the microlens. The problem with going to a higher pixel density is that the wiring takes up a larger proportion of the photodiode area as the megapixels increase. However, the Toshiba sensor doesn't show much, if any, degradation in per-pixel noise and dynamic range compared to the D7000, so my guess is that the thickness of the wiring interconnects have shrunk proportionally with the overall size of the photodiode.

No comments:

Post a Comment