Mfcttrf

You will occasionally encounter articles about how awesome full-frame cameras are. Lot of that is probably over-enthusiasm over a new piece of equipment or simple marketing, but it seems to me that at least these things are true:

• Sensor with a large area captures more light


• Sensor with large individual pixels would have less noise


• Larger sensor can fir much more pixels

The full-frame cameras are much more expensive. This is weird to me, since I had the impression that making electronics smaller is always harder, since you need more precise equipment.

That must've been even more important in the dawn of digital single lens cameras, many years ago.

So why was it chosen to make sensors smaller than is the film originally used in the cameras? AFAIK some lenses made for film cameras still work with some DSLRs, so why make the sensor different from the film?

Note that I'm interested more about history of the initial decision (since film frame size was the status quo, and DLSRs were expensive anyway), than the price difference.

Making large semiconductor devices with no, or only a very small number, of defects is very hard. Smaller ones are much less demanding to make.

In particular the yield – the proportion of the ones you make which are usable – for semiconductors drops as you try and make them larger. To understand this properly needs statistics which I don't want to try and do on the fly, but it's easy to see why this happens from a simple example.

Let's say that you can make 24x36mm devices (ie, full frame), and you have a 90% chance of any device you make having a serious enough defect that it is scrap. Your yield of 24x36mm devices is 10%: you need to make 10 devices for every 1 you can sell. (manufacturers are shy about what their yields are, but this is not stupidly low).

So instead you use the same technology, with the same defect rate, to make 10x15mm sensors, of which you can fit four on each wafer you were using for the 24x36 ones, with room for cutting the substrate up after making them. 90% of the wafers are defective, but the defects are localised (this is what defects are like in reality), so you get 3 good sensors from the defective wafers. So your yield rate is now (1 * 4 + 9 * 3)/40 (10 wafers of which 9 are defective), which is 31/40 or about 77%. In fact it is a little lower as I've assumed wafers only have 0 or 1 defect, while in fact they have some probability of a defect per unit area (or something) and thus there will be wafers which yield less than 3 good sensors.

But the basic point is clear: by making a sensor which is smaller you have increased your yield by a factor of seven. This means your cost to make these smaller sensors is seven times lower.

(You can make an even more extreme case: let's assume that every 24x36mm wafer is defective, but none have more than one defect (again this is simplification). The yield for 24x36mm sensors is 0: you can't make them. But the yield for 10x15mm sensors is (somewhat less than, because of the multiple-defect case) 75%: you get 3 good ones from each wafer.)

The first mainstream applications for electronic image sensors (be it Image-Orthicons, Vidicons, Plumbicons, or CCDs, or CMOS active pixel sensors, be it analog-electronic or digital workflows) were in video, not in still images.

Video followed form factors similar to movie film. In movie film, 35mm (equivalent to full frame still) or even 70mm were exotically large formats only used for actual (cinematic) movie production due to significant costs.

Also, the resolution demands for most video applications used to be much smaller - if pre-HD home televisions (maximum resolution 625 lines of maybe a 1000 pixels each) were the major target, high resolution capabilities were not necessary.

Also, in the non-cinema moving image world the demands on lenses appear to be different - much more expectations on lens speed and zoom range, much less on image quality. This can be done far more cost effective with lens designs that only have to service a small image circle.

Digital still cameras existed several years before interchangeable lens cameras became plausible, and these used tiny sensors first that were very likely designed for or based on designs for video.

APS-C sized sensors were HUGE compared to a normal digital camera sensor when early DSLRs were introduced; the few early full frame DSLRs (think Kodak DCS) and their sensors were extremely expensive, probably because there was very little design experience in making economical sensors in that size.

Image sensors are very coarse in actual structure compared to what CPUs or memory chips even in the 1990s used - for example, a common CPU for late 1990s desktop computers used 250nm feature size, which is quite smaller than what would even be physically useful on a visible-light imaging sensor. Today, 14nm (!!) is about state of the art.

The necessity to avoid large die sizes per part, regardless of the structure sizes, as already explained in other posts, has not changed much.

Big sensors cost more than small sensors for more-or-less the same reason that big TVs cost more than small TVs. Compare a 30-inch TV and a 60-inch TV (about 75cm and 150cm, if you prefer). Miniaturization is no problem — we could make all of the parts of the 30-inch TV way smaller without running into any difficulty. The 30-inch TV costs less to make than the 60-inch TV because it uses less materials and requires less work to finish. And the 60-inch TV will have a higher defect rate — 4 times the area means much higher the chances that something goes wrong somewhere on the screen, creating a dead pixel. Because customers hate dead pixels, a panel that has more than one or two (or maybe even more than zero) gets scrapped, or sold as part of a lower-cost product. The production costs for defective units get rolled into the price of the acceptable units that are sold, so the bigger you go, the more expensive things get.

The same considerations apply to image sensors. Even the smallest sensors on prosumer cameras have features that are huge compared to what semiconductor technology is capable of, so the cost of miniaturization isn't a major factor. Compact cameras and cell phones normally use far smaller sensors, and even budget phones normally have two cameras, with fancier ones having three or four! For reasonable sizes, smaller costs less, not more. The defect issue also comes into play. The bigger you make the sensor, the more likely you will have a defect that requires you to scrap the whole thing, and the more money (in materials) you will lose when you do scrap it. That drives cost up with size, dramatically beyond a certain point.

The largest-format digital camera you can get as of this writing has a whopping 9"x11" sensor (that's more than 8 times the diagonal of a "full frame" sensor, or more than 64 times the area), and it only has 12 megapixels so obviously miniaturization isn't an issue — those pixels are huge. It retails for over $100,000.

Smaller sensors have higher production yields, and the electronics to process are lower cost.

Double the sensor, and roughly square the processing power needed.

The reality is that DX sensors are often higher resolution and greater dynamic range than films they are replacing.

Because you specifically asked about history...

I'd suggest: size, weight, & cost.

All those considerations were equally true in the pre-digital (ie film) days. A popular film format was the 110 size. See:
https://en.wikipedia.org/wiki/110_film

The 110 film was cheaper, the cameras were cheaper, and many of the cameras were a lot smaller and lighter than the smallest 35mm film compacts. They could fit very easily in a small pocket. Of course those same constraints exist today with digital cameras, as others have pointed out. So it's not just small and big image sensors today; it was also small and big film formats back then as well.