I want medium format!

Started Aug 30, 2013 | Discussions thread
Joseph S Wisniewski Forum Pro • Posts: 34,136
Large sensors are not "easier to make"

stevo23 wrote:

Right - the digital isn't as good, I'd rather have a D800 right now. But truly - how hard does it have to be? Larger sensors should be easier to make...

Don't use the "s word", it's very rude, LOL.

Seriously, there's several reasons why larger sensors are much harder to make, and no reasons, at all, why they "should be" easier. Earlier, you drew parallels to large LCD panels. Those have such huge, simple pixels that they can use "amorphous" or "polycrystaline" semiconductors, technology where the transistors in the pixel are "printed" on the big glass sheets with a process that's a lot like a high-tech version of making posters in a printing press.

Sensors are much more complex, they have more transistors packed into a much smaller pixel, and they need to control them with literally 64-256 times more greater precision than an LCD (14-16 bit vs. 8 bit). To get that tiny size and high precision, they need fully "crystalline" silicon, just like computer chips. That means two things...

First, the "yield" sucks. Pure crystalline silicon chips are made from big round silicon "wafers", which are sawn from huge, laboratory grown "logs" of the purest silicon. There's no way to do that without doing a certain amount of damage to the wafer, creating microscopic flaws that will keep chips from working. This isn't much of a problem for small chips. Say you're using 200mm wafers. That's 31400 mm2. You can "dice" that wafer into:

  • About 120 chips the size of an Intel core i3 (181mm2) out of it. So, if you've got an average of 20 chip killing flaws on the wafer, the odds are they'll all fall on different chips, so you lose 20 out of the 120. That's not bad, it's an 82% yield.
  • About 50 chips the size of an APS sensor. They're 512mm2, and there's only so many rectangles you can cut out of a circle. There's still 20 flaws, and most of them end up on separate chips, so your yield is now 60%.
  • 24 FF chips, about 1053mm2. Fortunately, some of the chips manage to catch 2 or more flaws, and some of the flaws fall on the 20% of the wafer that doesn't accommodate rectangles, so only 16 of your chips are bad, and you get to keep 8, a yield of 33%. That means that the FF sensor costs about 4x what the APS sensor costs. 2x because it's twice the area, 2x because you get half as many keepers. 
  • 10 medium format chips at 1989mm2. Scatter 20 flaws randomly on that die, and you're doing great if 1 or 2 chips work. Yeah, just 10-20% yield. So, again, 4x the cost of FF, 2x because you doubled area, 2x because you halved the yield. That's why aerospace sensors cost so much. They might order a sensor that's so big you get just one per wafer, and you have to run 25 wafers before the odds let you have a single sensor where all the flaws fell outside the sensor part of the wafer.
-- hide signature --

Rahon Klavanian 1912-2008.
Armenian genocide survivor, amazing cook, scrabble master, and loving grandmother. You will be missed.
Ciao! Joseph

 Joseph S Wisniewski's gear list:Joseph S Wisniewski's gear list
Canon EOS 5D Mark II Nikon D90 Nikon D2X Nikon D3 Nikon D100 +43 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow