Are higher megapixel sensors harder to manufacture and therefore cost more?

[wallpaperviking]

As the title suggests, just wondering why higher megapixel sensors seem to cost so much more?  Are these harder to manufacture or is it simply that the manufacturers see an opportunity to make more money on these?

I guess the best example is the latest Phase One digital backs, that come in 50, 80 and 100 megapixel sizes but are all different prices?

I understand how a larger physical sensor size would influence this but am curious about the price difference between different megapixel sizes given that the physical dimensions are the same?

Hope that makes sense

[eronald]

As the title suggests, just wondering why higher megapixel sensors seem to cost so much more?  Are these harder to manufacture or is it simply that the manufacturers see an opportunity to make more money on these?

I guess the best example is the latest Phase One digital backs, that come in 50, 80 and 100 megapixel sizes but are all different prices?

I understand how a larger physical sensor size would influence this but am curious about the price difference between different megapixel sizes given that the physical dimensions are the same?

Hope that makes sense

i think silicon defects are dependent on surface, but CFA defects are dependent on the number of pixels.

just my intuition. yours may be closer to ground truth.

Edmund

[Joe Towner]

Hey Wallpaperviking,

In a word, yes.  Break this down a bit as there are a few differences between the sensors and their effective yields, age and volume.

The 50mp is 33x44mm in size, and a CMOS chip
The 80mp is a 53.7x40.4 mm size and a CCD chip
The 100mp is a 53.7 x 40.4mm size and a CMOS chip

So it's not a straight forward mapping of MP to price.  The 50mp is used by Phase, Hasselblad, Pentax and now Fuji.  The 80MP is exclusive to Phase, and the 100mp is Phase and Hassy.

[Graham Welland]

yes ...

[ErikKaffehr]

Hi,

I would guess it is about production process and sensor area. A larger sensor is much more expensive to make than a smaller one.

Would you make say a 100 MP sensor on 54 x 40 mm surface it would be much more expensive (several times) than a 50 MP sensor on 44x33 mm surface, if both used the same technology.

Would you make a hypothetical 100 MP sensor 44x33 mm surface, it would not be more expensive in production than the 50 MP sensor. But development costs are probably a significant part the real cost. So if you would sell 10000 100 MP sensors but only 5000 50 MP sensors using same chip size and technology the 100 MP sensor would be cheaper to produce.

On the other hand, technology changes. Just as an example, the sensor used in the A7rII is a much more complicated design than the older 36 MP sensor use in the A7r and the Nikon D800. The A7rII is back side illuminated, has dual gain conversion and copper wiring.

So, there is no definitive answer to your question. A good way to see it is that pixels are cheap and sensor surface is expensive.

But, it is not that simple. It seems that the 100 MP 54x40 mm Sony sensor has something like an EV advantage over the 50 MP 44x33 sensor. So, it doesn't just have more pixels but also better pixels, and that mean a new pixel design.

Best regards
Erik

i think silicon defects are dependent on surface, but CFA defects are dependent on the number of pixels.

just my intuition. yours may be closer to ground truth.

Edmund

[razrblck]

One thing often overlooked, that also plays a role in sensor pricing, is feature size (or process size, if you are familiar with CPU manufacturing). A higher resolution sensor would need smaller features (the pixel themselves and the corresponding electronics that go with it) if you keep sensor size as a constant.

The 50MP 44x33 sensor has about 34 pixels per square μm of surface (~0.17μm pixel side length). The 100MP 54x40 sensor has about 46 pixels per square μm (~0.14μm pixel side length). It's a lot denser so all the supporting on-chip electronics needs to be smaller as well.

While smaller production processes make it possible to get more electronics (pixels, transistors, whatever) per area, they are also more expensive. Pair this with the fact that the actual sensor is bigger as well than the 50MP and you get a much more expensive product in production costs alone.

[Chris Livsey]

Presumably, in other words a wild guess because I have no direct knowledge, the manufacturing costs will be higher with both a larger area chip and a higher density chip and compounded with both in play because defects in the manufacturing, if the process is constant across sizes and that may not be true, will be greater and the reject rate higher as area and density increases.

But if the Sony, I know "in house", SOTA chip in the A7rII is back side illuminated with dual gain conversion and copper wiring is, in the final product, considerably less expensive than the recently announced Leica M10 product. Of course bespoke tweaks and volume come into play but so they do in medium format, indeed on recent figures the M10 can be expected to outsell all the "traditional" medium format offerings grouped. But then Leica is Leica and play outside the conventional rules.

(Credit please for veering OT in the grand tradition)

[ErikKaffehr]

Hi,

I don't think that applies. The relevant sensors are all made by Sony and they probably use the same 0.18 micron design rules for all their large format sensors.

If you include older CCD and Leica it would be different technologies.

Best regards
Erik

One thing often overlooked, that also plays a role in sensor pricing, is feature size (or process size, if you are familiar with CPU manufacturing). A higher resolution sensor would need smaller features (the pixel themselves and the corresponding electronics that go with it) if you keep sensor size as a constant.

The 50MP 44x33 sensor has about 34 pixels per square μm of surface (~0.17μm pixel side length). The 100MP 54x40 sensor has about 46 pixels per square μm (~0.14μm pixel side length). It's a lot denser so all the supporting on-chip electronics needs to be smaller as well.

While smaller production processes make it possible to get more electronics (pixels, transistors, whatever) per area, they are also more expensive. Pair this with the fact that the actual sensor is bigger as well than the 50MP and you get a much more expensive product in production costs alone.

[dchew]

I’ve spent most of my career involved with costing and pricing low-volume products (1-100,000 units/yr). It is fraught with pitfalls. If pricing McDonald’s french-fries (sorry - chips), there are a host of things you can do to accurately price (and cost) the product. I was recently told the light bulb isle at Home Depot is a $4M isle per store per year. That’s roughly a customer per minute per store. All sorts of fun stuff you can learn about people’s utility for various things and the price elasticity of not only a product, but its individual features in a case like that.

When selling a few thousand of something per year it is very different. To some extent, the MF manufactures use all three types of price discrimination. Although not technically a two-part tariff, the fact that you can trade in an old back for a reasonable credit (especially with P1) creates a similar economic effect even if irrational: It’s expensive to get on the train and now that I’m on I don’t want to get off! There is also the obvious feature set difference between the IQ350 and 150. Combine all this with quantity discounts to high-volume institutional users and they have all the price discrimination boxes checked. The problem still is volumes are low, so it is very difficult for HB or P1 to determine the real price elasticity of a product (if they lower the price will the marginal revenue go up or down); there are just too many other factors going on with too little volume to perform a meaningful test. This is especially true for young products.

As a result, pricing often defaults to a cost-plus model, or at least a hybrid that includes some sort of desired gross margin and (un)educated guesses about market dynamics.

You would think at least costing should be an exact science but it often isn’t. All sorts of overhead costs can be added as a percentage of material cost, and those costs are a jumbled mix of fixed and variable that change depending on the time horizon and question being asked.

For example, P1 may assign warranty cost as a percentage of material cost (I have no idea if they do, but it is certainly done often). As one might guess, the warranty and other support costs for more complicated (higher-end) products are often higher than those of established, less “cutting-edge” products. I have no idea what sensors cost but it doesn’t really matter to illustrate the point of how costs can escalate when multipliers are applied to cover warranty support and other costs allocated on a material cost basis:

IQ180 Cost (sensor only): $3000 x 1.1 = $3,300
IQ150/350/50c Cost (sensor only): $4500 x 1.12 = $5,040
IQ3100/100c Cost (sensor only): $6000 x 1.15 = $6,900

Now pile on top of that the completely rational thought that newer products with less-established competitive offerings should generate higher gross margins. Again, difficult to test and verify. But any change in margin results in a big difference in price. Let’s say 40% for established products and 50% for new products:

IQ180 Price (sensor only): $3,300 @ 40% GM = $5,500
IQ150/350/50c Price (sensor only): $5,040 @ 45% GM = $9,164
IQ3100/100c Price (sensor only): $6,900 @ 50% GM = $13,800

So, a sensor that costs only $3,000 more ends up with a price adder of > $8,000 (again, sensor only). That doesn’t count the added memory, faster processing required, heat management, allocated sunk costs for developing feature sets, etc.

How much of the price difference is associated with cost vs. market pricing? Only HB and P1 know, and even they may not really know!

Dave

[kers]

I think this question can be applied to all chips and the efforts making them smaller.

So maybe of interest:
Just came across this recent interesting interview with Peter Wennink, head of ASML.
ASML stands at the bases of all chip production since they make the high end machines to make the chips.
At the moment they are using UV-light for making them even smaller.
The problem with UV-light is that the optical system is not made out of lenses but has to be made with mirrors...
it is in Dutch so Google translate is you partner.
https://tweakers.net/reviews/5155/asml-ceo-peter-wennink-over-de-race-naar-7nm-en-kleiner.html

[E.J. Peiker]

I spent 27 years of my life working in semiconductor production for Intel in various roles from process engineer, to product engineer, to engineering manager, to manufacturing manager and ultimately factory manager.  Your question is not quite as simple as you might think as there are a number of factors.  Note that this is looking at it from a cost perspective not a price perspective - all sorts of things can go into final pricing that have little to do with the actual cost of manufacturing as noted in the answer from dchew above.

If you are talking about two sensors that have the same size but different pixel counts such as a 24 megapixel and a 42 megapixel full frame sensor then the cost is largely similar as long as they are made on the same semiconductor technology (such as the Sony 0.18 micron process for example) but if the 24mp sensor is made on a more mature and older technology it is likely to be a bit cheaper to make.  I used the words "largely similar", not "the same" for a reason - there are often design related process sensitivities that may give one product a yield advantage over another.   Also, using the 24 vs 42mp example, the 42mp sensor has nearly twice as many circuit elements on the chip so the chance of a point defect impacting that chip is a bit higher but often defects are big enough that it would kill either but the sensitivity of the higher density sensor is a bit higher.

If you are talking about sensors that have different different physical sizes (and possibly different pixel counts), the larger size overwhelms any other cost differences as cost goes up exponentially with size.  I wrote an article on this many years ago that should explain that relatively well:
https://www.naturescapes.net/articles/techniques/the-economics-of-digital-photo-sensors/

Overall here is the general order of impact on sensor cost assuming they are made in the same plant by the same company:
1. physical sensor size (overwhelmingly the biggest factor)
2. process technology node employed (smaller minimum feature size is more expensive due to higher capital requirements - more expensive semiconductor processing equipment and since it is newer it is less likely to be fully depreciated, also often a less mature technology doesn't yet have the same yields as a more mature one)
3. design related process yield sensitivities
4. circuit density related yield sensitivities

[Paul2660]

Hi Dave,

I agree the costing has to be complicated but I still think that the actual costs are considerably lower.

I only have to ask, if the stated price point of the 50MP chip Sony is 9K with various costing, then how can Pentax afford to sell the 645Z at around 10K, originally, then now down to around 7.5K? and also if this same pricing would apply then the new 50Mp cameras by Fuji/Hasselblad would both be under water.  There is no doubt that the 50MP Sony has some of the best DR ever produced, as shown by the P1 cameras using it, the amazing fact that Pentax with the 645Z produce even more DR and the new Hasselblad's images seem to be in line with Pentax or possibly slightly cleaner.  The 50MP chip did for MF what the Sony 36MP chip did for 35mm cameras.

Actual costing by P1 IMO is less since they have almost zero marketing cost, as this is mainly done by their dealer network, unlike the other companies mentioned above.  P1's price does have to be inflated to cover the dealer margins but from some pricing I have seen recently there is a lot of room out there especially in the lenses. 

P1 has had the ability to keep the cost of their product line in the stratosphere for a long time, and with the 100MP back they can continue that as odds are the follow on to the 50MP 30% cropped chip will be 75 maybe 100MP but still not the size of the 100MP chips.  But just look what Fuji was able to do with the older 2014 50MP Sony, creation of ES and EFC, both of which have never been available before now, even with Pentax. So Fuji is giving you even more features out of the same 50MP piece of silicon. Not only vastly cheaper, but more value/features also.

Most don't seem to give Pentax the credit they deserve with the 645Z and price point they were able to bring such a powerful camera to market at.  Pentax made some in roads but not as much as the new Fuji and Hasselblad offerings were.  I also feel that the Fuji offering will end up having more impact due to the fact it has a focal plane shutter and thus can possibly operate with other manufacturers lenses unlike the Hasselblad.  Either way the 50MP market has been brought to a new price point and it will be difficult for other companies to continue the stratosphere price points.  If in the future an adapter is made so that the Fuji can be mounted to a Universalis or Aptus then things will get very interesting.  The only issue I see there is that LR has a terrible LCC process and C1 will continue to keep the blinders on and not support other MF products so movements with the Fuji will be difficult.  Unless the chips are close enough that a renaming of the exif to P1 IQ150 or 350 etc will allow C1 to open the files.  It has worked that way in the past.

Paul Caldwell

[Graham Welland]

I think that it's less to do with cost as it is to do with yield. As the process improves for a given sensor chip, yields improve and the per item cost thus falls. Likewise the number of chips per wafer is less as size increases, and also more elements in a larger chip, the greater the potential for errors and unusable chips in production.

[ErikKaffehr]

Hi,

Thanks a lot!

Best regards
Erik

I spent 27 years of my life working in semiconductor production for Intel in various roles from process engineer, to product engineer, to engineering manager, to manufacturing manager and ultimately factory manager.  Your question is not quite as simple as you might think as there are a number of factors.  Note that this is looking at it from a cost perspective not a price perspective - all sorts of things can go into final pricing that have little to do with the actual cost of manufacturing as noted in the answer from dchew above.

If you are talking about two sensors that have the same size but different pixel counts such as a 24 megapixel and a 42 megapixel full frame sensor then the cost is largely similar as long as they are made on the same semiconductor technology (such as the Sony 0.18 micron process for example) but if the 24mp sensor is made on a more mature and older technology it is likely to be a bit cheaper to make.  I used the words "largely similar", not "the same" for a reason - there are often design related process sensitivities that may give one product a yield advantage over another.   Also, using the 24 vs 42mp example, the 42mp sensor has nearly twice as many circuit elements on the chip so the chance of a point defect impacting that chip is a bit higher but often defects are big enough that it would kill either but the sensitivity of the higher density sensor is a bit higher.

If you are talking about sensors that have different different physical sizes (and possibly different pixel counts), the larger size overwhelms any other cost differences as cost goes up exponentially with size.  I wrote an article on this many years ago that should explain that relatively well:
https://www.naturescapes.net/articles/techniques/the-economics-of-digital-photo-sensors/

Overall here is the general order of impact on sensor cost assuming they are made in the same plant by the same company:
1. physical sensor size (overwhelmingly the biggest factor)
2. process technology node employed (smaller minimum feature size is more expensive due to higher capital requirements - more expensive semiconductor processing equipment and since it is newer it is less likely to be fully depreciated, also often a less mature technology doesn''t yet have the same yields as a mmore mature one)
3. design related process yield sensitivities
4. circuit density related yield sensitivities

[dchew]

Hi Dave,

I agree the costing has to be complicated but I still think that the actual costs are considerably lower...

Paul Caldwell

Paul,
My sensor cost guess was from the standpoint of the camera manufacturer, not the chip maker, but regardless I'm sure you are right: my sensor costs are too high. If we take your Pentax example and work backwards, assuming a 30-50% sensor cost vs total cost and a GM of 35-55%, then the sensor cost is between $1350 and $3250 using the original $10k camera price. Probably somewhere in that range. The real point I was trying to make is that regardless of sensor price/cost, I don't think there is a straightforward answer to the OP's question. I don't think camera pricing is as unjustified as some others do, but I certainly agree it is not only related to cost. Few things in the market are priced based on cost, other than government contracts.

Dave

[E.J. Peiker]

Paul,
My sensor cost guess was from the standpoint of the camera manufacturer, not the chip maker, but regardless I'm sure you are right: my sensor costs are too high. If we take your Pentax example and work backwards, assuming a 30-50% sensor cost vs total cost and a GM of 35-55%, then the sensor cost is between $1350 and $3250 using the original $10k camera price. Probably somewhere in that range. The real point I was trying to make is that regardless of sensor price/cost, I don't think there is a straightforward answer to the OP's question. I don't think camera pricing is as unjustified as some others do, but I certainly agree it is not only related to cost. Few things in the market are priced based on cost, other than government contracts.

Dave

It's been reported elsewhere that depending on quantities and customization (SKU) the 50mp 44x33 sensor is being sold by Sony from $3K to $4K per unit.  I can't vouch for the accuracy of those reports but that sounds reasonable accurate based on my experience in the industry. 

[Jim Kasson]

I spent 27 years of my life working in semiconductor production for Intel in various roles from process engineer, to product engineer, to engineering manager, to manufacturing manager and ultimately factory manager.  Your question is not quite as simple as you might think as there are a number of factors.  Note that this is looking at it from a cost perspective not a price perspective - all sorts of things can go into final pricing that have little to do with the actual cost of manufacturing as noted in the answer from dchew above.

If you are talking about two sensors that have the same size but different pixel counts such as a 24 megapixel and a 42 megapixel full frame sensor then the cost is largely similar as long as they are made on the same semiconductor technology (such as the Sony 0.18 micron process for example) but if the 24mp sensor is made on a more mature and older technology it is likely to be a bit cheaper to make.  I used the words "largely similar", not "the same" for a reason - there are often design related process sensitivities that may give one product a yield advantage over another.   Also, using the 24 vs 42mp example, the 42mp sensor has nearly twice as many circuit elements on the chip so the chance of a point defect impacting that chip is a bit higher but often defects are big enough that it would kill either but the sensitivity of the higher density sensor is a bit higher.

If you are talking about sensors that have different different physical sizes (and possibly different pixel counts), the larger size overwhelms any other cost differences as cost goes up exponentially with size.  I wrote an article on this many years ago that should explain that relatively well:
https://www.naturescapes.net/articles/techniques/the-economics-of-digital-photo-sensors/

Overall here is the general order of impact on sensor cost assuming they are made in the same plant by the same company:
1. physical sensor size (overwhelmingly the biggest factor)
2. process technology node employed (smaller minimum feature size is more expensive due to higher capital requirements - more expensive semiconductor processing equipment and since it is newer it is less likely to be fully depreciated, also often a less mature technology doesn't yet have the same yields as a more mature one)
3. design related process yield sensitivities
4. circuit density related yield sensitivities

Excellent summary. What would you say about techniques like those required for back side illumination? How about the precision assembly for stacked sensors?

Another issue: are there cost discontinuities with mask size? Are there cost issues with tiling masks? With 18 nm processes, are multiple displaced exposures per mask necessary? Is that a significant expense?

Jim

[Jim Kasson]

It's been reported elsewhere that depending on quantities and customization (SKU) the 50mp 44x33 sensor is being sold by Sony from $3K to $4K per unit.  I can't vouch for the accuracy of those reports but that sounds reasonable accurate based on my experience in the industry.

If Fuji is paying at least $3K for the sensor, how can they sell a complete camera for $6500, including dealer markup?

Forward pricing?

Jim

[E.J. Peiker]

Excellent summary. What would you say about techniques like those required for back side illumination? How about the precision assembly for stacked sensors?

Another issue: are there cost discontinuities with mask size? Are there cost issues with tiling masks? With 18 nm processes, are multiple displaced exposures per mask necessary? Is that a significant expense?

Jim

Masks for tighter geometries are more expensive but the minimum geometry in a photographic sensor is not dictated by the pixels which are enormous compared to the minimum feature size so that can pretty much be ignored if the two sensors are manufactured on the same technology node.  As for BSI, that likely requires at least a couple more mask layers so an equivalent BSI sensor would be a bit more expensive to one that uses traditional methods.  Stacked sensors would also add cost and complexity.  Steppers from ASML that employ phase shift technologies and or multiple displaced exposures are handled within the tool but of course those tools are more expensive.  The capital outlay for smaller and smaller geometries goes up exponentially.  When I started in the industry we were at the 2 micron node and the equipment for an entire factory could be purchased for the approximate price of a single stepper capable of 0.12 microns.  I started with Intel in 1983 and retired from there in 2010.

[E.J. Peiker]

If Fuji is paying at least $3K for the sensor, how can they sell a complete camera for $6500, including dealer markup?

Forward pricing?

Jim

A lot of the development cost is probably amortized among common things with other cameras like the X-T2 and X-Pro2 - there are many common things with those cameras (which is similar to how Pentax was able to offer the 645Z at such a competitive price - it basically used a lot of the "guts" of their K series DSLRs).  In larger sensor cameras by far the most expensive component is the sensor.  Image processors and memory for the buffer, etc are very cheap compared to those as they are small IC's.  Of course the body and its manufacturing costs are pretty expensive and probably cost quite a bit but it wouldn't be a ton more than the manufacturing cost of an X-T2 (sure it's smaller and doesn't have the removable EVF but then there is cost in it's sensor as well so I don't think it would be dramatically more and those cameras sell for about $1600 which means their cost is less than that).   I can easily see the total cost out the door being in the $5K range which leaves a 30% margin once development costs are paid for on the body that they can play with for things like their profit and dealer margin.  Of course this is pure speculation on my part as I only have insight into the IC costs.  I suspect the real profit margin is in lenses and accessories.