Canon Doesn't Think Consumer Cameras Should have Raw Video

Except there is. One of the biggest technical reasons is that you can only get a certain amount of sensors from one wafer.

Building costs for new fabrication processes and factories are high, for GPU/CPU's we are talking in the billions. CMOS plants maybe more cheaper but the fundamental fact is that the bigger the sensor (or cpu/gpu) the more space it takes from one wafer and thus more wafers have to be constructed. There is a huge difference between an APS-C sensor and fullframe on one wafer. Even bigger compared to 1/3" sensors.

CPU/GPU makers tend to bin big chips so they can sell defective ones too at lower prices but that's difficult to do with sensors.

The cost of the wafers we don't know so it's quite hard to predict prices. Also wafers keep improving and the amount of bad quality sensors thrown away is getting smaller all the time.

He made exactly those calculations: how many APS-C and FF sensors you can fit into a wafer, how many of them will be functional assuming standard fault distributions for that area, etc. And he concluded that nowadays a FF sensor costs about $100 more to manufacture than an APS-C sensor.

He's not completely open with his assumptions, but if you run the numbers it's difficult to find a difference of more than $300. You can fit 40-50 and 20 APS-C and FF sensors, respectively, on an 8" wafer that costs around $1500 at a 0.18um foundry. In order to get a $300 difference, you have to assume that most APS-C sensors in every wafer come out not working and get thrown away, which makes the probability of getting a working FF sensor really low (you get 25 working APS-C chips but only 3 or 4 FF working chips). If only 20% of the APS-C chips are faulty, then the difference is less than $100 (you get 40 working APS-C chips per wafer, but only 11 FF working chips).

Even with a $300 difference, it means that something like a NEX-6 but with a FF sensor could be sold for around $1K. And the proof is in the pudding: we are actually getting there!
(for a DSLR it's not so simple because the prism and mirror mechanisms are more expensive too, and whereas I understand CMOS economics, I have absolutely no idea about these)

edit: for 1/3" chips, you can get around 1000 working chips per wafer

edit2: that $1500 price is what I have found for small manufacturers with small runs. If you're a big manufacturer, it seems by 2009 you only had to pay around $850 per 8" wafer with a 0.11um CIS FSI process (here).
(0.18um and 0.11um is not great by today's standards, but big sensors are usually manufactured with even older processes, 0.50um for the Canons and 0.18 only for the D800, here)
 
Last edited:
Beyond the hard costs for the parts, don't forget that when you reduce the numbers of what you're making your per unit costs go way up. Not just for the common components, but when you spread out the tooling, packaging, research & development, marketing, support, etc per unit sold it makes a big difference. For a round number example, spread a million dollars of costs over a million units of a consumer product and it's only a buck each. Spread a million dollars over a thousand professional units and now you've got a thousand dollars you have to tack on per unit.
 
Yes, the tooling is very expensive, and even more so for a FF sensor (since standard masks don't cover the whole sensor). But it's no different from any other sensor; a $1K FF NEX would sell more units than many other cameras that are actually being produced.
 
Back
Top