cpopham, I think we don't disagree. I assumed an FF sensor would be three times the price for twice the surface. Pretty fair, I guess.
BTW, I changed my body building cost estimate above to 300$ as I confused $ for €. I also added an FF body building cost estimate.
Originally posted by cpopham With larger chips, you need to produce more wafers to produce the same number of chips, thus a lower yield rate.
Maybe you don't mean it the way you said it. However, the way you said it this statement is wrong. The yield rate
y is
not the number of chips per wafer. This is the yield. The yield
rate is the number of good chips per total number of chips produced. Typical yield rates range from 50% to 90%.
If you double the surface of a chip, the yield
rate y' for chips twice as large is as follows:
- y' = y^2 for chips w/o redundant structures, i.e., a single defect ruins the chip.
- y' = y for chips with full redundant structures, i.e., where only failure density must stay below a threshold.
CPU chips have a higher surface fraction belonging to the first category than CMOS sensor chips. That was my point. Nevertheless, I used the second formula with y=0.8.
Originally posted by cpopham Consider for a moment an LCD panel - a few defective pixels scattered on the screen are "acceptable", but a cluster of defective pixels are not.
Yes. For cluster defects one must apply the second formula. But
y will be close to 1 for cluster defects, so
y^2 will be too.
Originally posted by cpopham as the physical size of the circuit features decreases, the rate of defects also decreases
I assume APS-C and FF to have the same pixel density. IMHO, it otherwise makes no sense to produce FF sensors. Of course, the camera may store with less pixels. It has been shown many times that downsampled images from high pitch sensors outperform original low pitch sensor images (for same surface sensors). This is because of lower read-out noise and better bayer interpolation. Therefore, all my arguments assumed a constant pitch (pixel density). An FF sensor will have more pixels and more defects.
Originally posted by cpopham Also, the equation you've used for calculating the manufacturing cost of a Full Frame sensor has a minor flaw in it - a silicon wafer is circular in shape
Very minor, indeed. The exact mathematical formula is intricate, you basically have to draw the layout. The area too close to the border is called "waste". Typical waste ratios are 12-20%. With 36mm chips on 300mm wafers, it will be larger. The increase though, will only be about 5% absolute, like in 15 to 20%. I allowed to ignore this effect.
Originally posted by cpopham they *are* most definitely more expensive than APS-C sensors, and that I believe it is a mistake to trivialise the cost of the sensor. [...] It could be that this Sony subsidiary could actually make more money producing other chips
I didn't say otherwise. I guessed an FF sensor is 3x the price or ~150$. A Fab will obtain higher margins on FF chips than APS-C chips because price elasticity is smaller.
BTW, which is why DSLR is more profitable than P&S. Or why Pentax needs to withdraw from P&S and need an FF body: they would generate much higher margins from those (why Airbus had to make A380 when Boing made all its profit from 747).