>>2597987Thought experiment: Could we make a large-format digital sensor of comparable size?
Current silicon wafers are 300mm in diameter. If we exactly fit the diagonal of a square sensor into a 300mm circle, we get something 300/sqrt(2) mm on a side, or about 212. This is just over 8 inches square. I'd say that that's comparable to large format.
How many pixels could we fit on it? Lets round the size to 212mm on a side, square. That gives us an area of 44944 square millimeters. Lets say we want our pixels to be comparable in size to those use on modern full-frame sensors. A Nikon D4 has pixels 7.3 microns, or .0073mm, on a side. Therefore each pixel has an area of 0.00005329 mm square. 44944/0.00005329 gives 843,385,250 pixels on our sensor. 843 pickles ain't too shabby.
Image sensors have a varying but low-single-digit number of transistors per pixel, lets be conservative and say we need five, which is probably an overestimate. Maybe we're compensating for defects so that a pixel will still work if one transistor is faulty or something. That means we need to put 4.2 billion transistors on the wafer. Considering Intel can put dozens of half-billion-transistor CPUs on a wafer, I'd say that's possible. With 843 pickles, and assuming two bytes per pixel (14-bit plus some file-format overhead and such) our hypothetical large-format digital would generate RAW files of about 1.7 gigabytes each.
How much would it cost? Intel can put 148 20x20mm dies on a 300mm wafer. Lets assume they have a perfect yield and sell all of them - Celerons sell for barely a hundred bucks, so lets assume that $100 per die is their cost of production. The whole wafer is then worth $14,800. Of course that's a big underestimate, because of retooling, business risk, etc. And more than that goes into a digital camera.
So counting the sensor, production, and everything else, I'd guess that, theoretically, an 8x8 large-format digital camera could be built for somewhere between $100 and $200k.
