DSLR Sensor Economics

By Gary Tomkins

We look at a lot of image sensors at Chipworks. It’s an interesting area, the state of the art is constantly evolving, and every sensor has unique features. The market is primarily driven by the mobile phone demands, but the digital camera is an attractive market on its own with over 100 million cameras shipped in 2007 and has been experiencing 20%+ annual growth. The DSLR (Digital Single Lens Reflex) camera is the fastest growing segment with an expected 9.1 million units shipping in 2008. (These figures come from the Camera and Imaging Products Association, CIPA here.)

DSLR cameras are now dominated by CMOS image sensors, this was initiated by Canon in 2003 when they launched the first DSLR to use a CIS, the Canon EOS 10D a 6.3 megapixel camera containing the Canon 704F CIS, demonstrating that CMOS image sensors are capable of producing the same quality images as CCD based cameras.

Now virtually all DSLR cameras contain CIS devices, driven by the need to get large pixel count data streamed off the chip quickly, and the lower power consumption of a CIS. The latest DSLR sensors that we are currently analyzing are 35mm full frame sensors from the top of line professional DSLR’s the Canon EOS 1Ds mark III, and the Nikon D3. These professional DSLR cameras contain some serious processing horsepower to get the signals out of the sensor. On the Nikon D3 we found a whopping 6 total Analog Devices AD9974 signal processors – presumably two for each color.

PCB in Nikon D3 DSLR

For the Canon – they are employing two of their Digic III Image Processors to handle the massive volume of data that a 21 megapixel camera can be imaging at five frames-per-second.

PCB in Canon EOS 1D Mark III DSLR

Interestingly Canon and Nikon, whilst both settling on CMOS sensors are diverging in their application. Canon with a whopping 21.1 megapixel sensor are driving resolution, whilst Nikon have chosen to use a 12.1 megapixel sensor, and have traded the bigger pixel area for increased sensitivity, (The Nikon D3 has an amazing ISO200-6400 sensitivity range in normal settings.)

What makes these sensors very interesting is the 36mm x 24mm full frame format. The majority of DSLR cameras use a 22mm x 15mm APS-C size sensor. To the photography enthusiasts the 1.6X multiplier that is required to image down to the APS-C format is a significant limiter to picture quality. Many expect the APS-C format to be a temporary stage, before all DSLR cameras are manufactured with full frame sensors. An excellent article on this, written by Bob Aitkins can be found here. Although for performance the full frame is clearly desirable, I think it will be a huge competitive challenge to provide this at the consumer or “prosumer” or hobbyist level.

These full frame sensors with ~6-9µm pixel sizes are fabricated with fairly relaxed design rules (~0.35-0.5µm ) compared to the current state of the art mobile phone sensors (Which are fabricated with 130 nm or even tighter 90 nm design rules). However there are challenges with sophisticated backend processing, and stringent junction leakage requirements and the low defect densities required to yield the required pixel performance.

Full frame sensors also present a unique challenge due to the die size being larger than what a normal lithography stepper is capable of printing. Current steppers from ASML, and notably Nikon and Canon typically have a 26 mm x 33mm maximum field size, large but still smaller than the full frame 24mm x 36mm requirement. To overcome this requires ‘stitching lithography’ where separate reticles (Stepper masks) each containing a portion of the die are exposed in serial fashion. This not only increases lithography costs due to the multiple steppings it is inherently difficult to ensure the on-die alignment needed.

Artifacts of stitching seen on Nikon D3 Sensor

Canon appears to be the only lithography equipment manufacturer to offer an i-line stepper configured for 200/300mm wafers with a 50mm x 50mm field. It is only capable of ~0.5µm design rules, but that is consistent with the geometries seen on the Canon CIS we are analyzing. We have not seen any evidence of stitching in the Canon device, thus if they are using this stepper in their CIS manufacturing line, they likely have some cost advantages over the D3 foundry.

The foundry for the Nikon D3 sensor is an interesting sidebar topic. Nikon have announced that they designed the sensor. Nikon has no wafer fabrication capability so they outsource the sensor production, but they are keeping the foundry close to their chests, so close that we must speculate to identify the source.

Die Markings on the image sensor found in the Nikon D3 DSLR

The obvious choice would be Sony, who build the sensor used in the Nikon D2X, however there are no Sony markings on the device, and the device structure is markedly different from the other Sony CIS we have analyzed. We considered Matsushita/Panasonic the device structure has similarities to the Panasonic CIS we have seen, but it is sufficiently different that we have doubts that it is theirs. Thus we are speculating who else could be the manufacturing source. I believe Nikon would stick with a Japanese foundry. An interesting possibility is Renesas, they have close ties with Nikon, supplying several imager processor chipsets, they have a patent portfolio in image sensors indicating they have active r+d in this field, and they have the fab capabilities. If indeed they are the foundry source for the D3, it makes them an interesting new entrant into the CIS manufacturing world.

 The consumer/prosumer cameras now span the $500-$1500 price range, the latest full-frame sensor DSLR’s span the $5000-8000 range. For the full frame sensors I believe achieving a average 50% yield would be impressive (One thing for sure the semiconductor manufacturers will not be disclosing what their die yields are, that is always the most closely guarded secret.) Thus I speculate with probing and packaging (These die justify a very expensive ceramic package) the cost to manufacture these sensors to be ~$300-400 each, compared to ~$70-80 for an APS-C sensor. Without the luxury of die shrinks available to reduce costs I believe it will be a long time before we see the full frame sensors on sub $1500 cameras. (But I would like to be proved wrong.)