Is FF sensors going to slowly phase out?

Started Apr 11, 2013 | Discussions thread
blue_skies
Senior MemberPosts: 7,091Gear list
Like?
Re: Those are APS compacts with slow lenses
In reply to tomtom50, Apr 14, 2013

tomtom50 wrote:

Erik,

Your argument about Moore's Law not applying well is correct, but the reticle argument cuts the other way. FF sensors do not intrinsically require more steps to make; they are made in more steps because they are made in small quantities and dedicated tooling is not wothwhile for small volumes.

If FF volumes rise dedicated tooling will be developed, FF sensors will be made in fewer steps, design will be amortized over more units, and cost will drop.

Will price ever reach $100? Forever is a long time but it seems unlikely to me as well. The advantages of FF just do not seem that meaningful to the mass market. Only a small percentage photographers will ever understand, much less care about, depth of field. Very high IQ and low light capability can be achieved with 1", m43, and APS-C.

Falk Lumo goes into this really well in http://www.falklumo.com/lumolabs/articles/equivalence/ff.html

He has the following graphic, backed up well by theory:

http://www.falklumo.com/lumolabs/articles/equivalence/ff.html

All said, I can imagine FF costs dropping to $200 - $300 in the next five years if the entry-level FF cameras take off and volumes rise greatly. At that point a FF camera selling for $1000 on discount becomes imaginable.

I am sorry, but the referenced article, and the therein referenced article (by same author) has been heavily criticized on both this and on the m43 forum. The main points were that the author does not have a good comprehension of the topics discussed and writes opinions and conclusions based on nonsense analysis. Besides, if his theories hold any merit, we'd all be using iPhones right now

As to your argument about Moore's law, and cost, R&D amortization is actually rather low, as the sensor's cost is mainly stemming from production costs (and margins) and not R&D recovery. The R&D recovery cost is a small percentage of the total production cost.

The cost of ICs typically goes down by generations based on newer technologies which allow smaller ICs to be created that take up less geometry to perform the same functionality, and thereby fitting more on a single wafer. Camera sensors are an anomaly, because their size is fixed, regardless of the process node that it is targeted to. The camera sensor quality keeps increasing on every next process node, that is a fact. However, see graph, newer process nodes are much more expensive.

Wafer price per process node - note this does not take into account the wafer size

To bring the cost per sensor down, larger volumes indeed do help. This allows migration to larger wafers, meaning that proportionally more (large) sensors can be carved out of a single wafer (edge effect).

The process node advancement also allows for yield recovery such as bad-pixel remapping & masking. Simply, the more advanced process nodes allow for more advanced solutions. Some of this does not come from the sensor ICs, but from the processor ICs that process the sensor data in-camera.

A 12" wafer can hold about 50 FF die, a 16" wafer can hold about 125 FF die. But the 16" wafer will be proportionally more expensive. The price difference roughly holds up to the ratio of the radius-square, so a factor of 64/36 = 1.8x. But you see that the number of FF die increases by a factor of 2.5x, so there is an incentive to move to larger wafers, given large enough volumes.

Here is a simple #die/wafer estimater (but notice that it does not allow an FF-die - as the max reticle opening only goes up to 32mm):

It is hard to get today's exact numbers, but here is an historic article that give a good breakdown of wafer cost and breakdown thereof:

The big unknowns are exact wafer cost and die yield. A cost of around $25k per wafer is probably not realistic, but I would not go too far below that. Yield, even with the pixel recovery algorithms, is a big problem. It was once said that FF sensors could never be made because of their size/yield trade-off (meaning it was yielding less than one successful FF die per wafer).

But even with a 100% yield, and a wafer (structured) cost of $15k, assuming a 12" wafer at 50 die, the best possible cost is $300. Nowhere near your $100 price target. Up the actual wafer cost, and lower the yield to well below 100%, and you can see FF sensors costing in the range of $500 - $1000 (and probably even higher).

It has been said that FF cameras spend more than 60% of their cost on their sensors, which makes sense if you consider a $2000 FF camera body price as today's lowest price. Even with distributor's discounts, it is likely that today's FF sensor costs are still around $1k per die, and definitely not $100 per die.

Once you go to smaller sensors, the cost structure changes rather quickly, but not because of the total number of die per wafer (it roughly doubles going from FF to APS-C), but because the yield increases dramatically (smaller die is more likely to be functional - defect density area formulas).

To reach $250 per APS-C die, you need to yield twice as many good die as FF-die (there are 2x on a wafer, so $1k -> $500, and then you need to have twice the yield, so $500 -> $250). It is likely that the yield ratio is well above 2x, meaning that APS-C die can be made for $150-$200 or so.

Please do not take the above as science either - I am merely trying to illustrate as to why FF sensors remain much more expensive than their smaller counterparts.

Moore's law sounds great, but it was never a law, rather an observation, and it doesn't apply to ICs that do not change in size (sensor ICs). Keep this in mind.

I do not foresee a $250 cost per FF die to be realized within the next five years, and perhaps never. A $1000 FF camera within five years is a pipe-dream. We are barely at $400 cameras for APS-C today, and these are heavily de-featured (no EVF, etc). APS-C (and m43) cameras that are 'full-features' are selling well over $1k.

Sony's RX1 is probably as 'cheap' as an FF camera can get today. It lacks a number of features that FF cameras have (e.g. EVF), and is comprised out of fewer components. Depending on how much you want to allocate for the lens, I'd suggest that this camera is barely braking the $2000 FF camera body price barrier. If you add the (fixed) lens and the (removable) EVF, you are well over $3000 for this camera.

Your best bet for a $1000 FF camera is to get a used one. The 5DII is selling for about $1500 used today, so maybe in another five years you can snatch one of those.

Camera and sensor technology will keep improving, but we seem to have identified sweet spots (e.g. 12Mp for m43, 16Mp for APS-C, and 30Mp for FF). Anything outside of those parameters leads to rather poor high-ISO performance (e.g. 16Mp P&S), with excessive NR algorithms (smearing). Yet, customers want (and manufacturers build) the 16MP P&S, 16MP m43, 24Mp APS-C and 50Mp FF. Why? Well, because at ISO 100 it works, and more is always better ...

If Metabones' reduced the SpeedBooster from $600 to $400, it will be far more compelling to use an FF legacy lens on a Nex-7N with the SB than to step up to a FF Nex for another $1000 extra. But the SB has many drawbacks, one of them being focusing speed, and the RX1 sales into non-professionals shows an appetite for such higher priced products does indeed exist.

A FF Nex camera at $2500 may indeed be a reality - hopefully within a year.

For professionals, a $2500 FF camera expense is recovered within just a few assignments, and is just a 'tool-of-the-trade'. Most will upgrade (and sell their old one) their camera every one or two years anyways, whereas amateurs tend to keep their cameras much longer.

And yes, smaller sensor can deliver very high IQ, but they will never truly compete against the larger sensors - don't fool yourself here. Simplest way to convince this is to look at highest 'useable' (not max) ISO on a camera model - it will tell you the story. Today's P&S should really only be used below ISO 400, 1" below ISO 800, m43 below ISO 1600, APS-C below ISO 3200, and FF below ISO 6400. These are incredible numbers compared to only a few years ago (and even more when thinking about film), but my point is not the actual level, but the relative ratio - it is all related to sensor size in the end.

And, one last comment, under low light, f/2.8 and ISO 6.4k still works on an FF sensor, but f/2 and ISO 3200 already fails on an APS-C camera. The SB can help here and lower the f/2 to f/1.4, allowing for ISO 1600. This, imho, is a really BIG factor - APS-C can then be used with f/2 lenses at indoor settings (ie. 1/60th at ISO 1600).

-- hide signature --

Cheers,
Henry

 blue_skies's gear list:blue_skies's gear list
Canon PowerShot S95 Sony Alpha NEX-7 Sony Alpha NEX-6 Sony Alpha 7 Sony a6000 +30 more
Reply   Reply with quote   Complain
Post (hide subjects)Posted by
NoNew
NoNew
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow