large chip cost

malcolm82

Forum Enthusiast
Messages
360
Reaction score
0
Location
UK
http://www.extremetech.com/article2/0,1697,2053781,00.asp

This is the chip of a new graphics card that is just out. It looks to be about 22x22mm wich is over half the size of a full frame sensor. These graphics cards are going for 450-650$ and include a lot of expensive memory and a board and cooler that also cost quite a lot i think so i suspect this chip isnt all that expensive. Also considering the full frame sensors are made on really old equipment how much can they really cost? Seems to me canon is lying.
 
A few things you're neglecting to consider.

1. The Canon sensor is almost twice the area of the chip you're referring to (864 vs. 484 sqmm) this makes a pretty significant difference in yield and gross die per wafer.

2. The graphics card will sell in far greater quantities than the Canon sensor, so they can amortize all the engineering costs over far more units.

3. Canon isn't lying, a full frame sensor is a formidable fabrication task and the costs are exponentially higher than smaller devices. Are they also padding the price a bit because they have the market all to themselves, of course they are, if they didn't they'd be stupid.

If you're trying to use this argument to claim Canon is publicly lying about the sensor cost just for the sake of gouging us on 1Ds pricing, I think your argument doesn't hold up.

You're always free to shop the competition and buy their full frame sensor camera at a lower price.
http://www.extremetech.com/article2/0,1697,2053781,00.asp
This is the chip of a new graphics card that is just out. It looks
to be about 22x22mm wich is over half the size of a full frame
sensor. These graphics cards are going for 450-650$ and include a
lot of expensive memory and a board and cooler that also cost quite
a lot i think so i suspect this chip isnt all that expensive. Also
considering the full frame sensors are made on really old equipment
how much can they really cost? Seems to me canon is lying.
 
I work as a photolithography engineer as my profession.. 90nm is not a cake walk and requires specilized equipment and chemicals (not old stuff) .. we use DUV wavelength resist that cost 4-5K a gallon...the best scanners or steppers cost $15M +....we have thin films equpiment that is larger then my livingroom..I can go on and on. Chip making is tough business.. memory making is worse.
--
JohnnyG
 
I read here before that canon is using something like 1micron process for their sensors, i didnt mean 90nm is old.
Interesting comments though :)
 
Apparently the 5D sensor costs around $600 to make which may not sound that much compared to the price of the camera but it soon adds up once you factor in the additional amount Canon needs to charge to cover R&D, setup costs, profit margin for their semi division etc.

It may be that the manufacturing method for the 5D sensor is different from that used with the 1DS2 which partly explains the lower cost but obviously there is some element of Canon making hay while the sun shines. Since no-one else is offering 35mm sensors in their cameras, why wouldn't they charge a premium?

At least it's not a large format sensor - last I heard, a 10x10cm chip will cost you ~$100,000!
 
It is obvious that Canon does not price its products like IKEA does - the price premium for being alone in the FF business is huge.

Comparing a $1495 30D with a $2795 5D shows almost a 90% increase in price with a sensor area increase of 150%

Comparing a $3460 1D2N with a $6820 1Ds2 shows almost a 100% increase in price with a sensor area increase of just 60%

A bit meaningless comparison as the sensor cost is just a piece of the body cost.

Seen a different way around, the cost increase absolute dollar terms is $1300 vs. $3360 for just a little larger piece of silicon. Given that the 5D body is a bit more sophisticated than the 30D body, a slight price premium can be expected there, independent of the sensor size. However, the 1D2 and 1Ds2 bodies are identical and the electronics excluding the sensor must be fairly equal. OTOH, the 1D2 shutter is probably more expensive.

Although I don't know what CMOS process Canon is using, assume a 300mm wafer with a theoretical area of some 70,686 sq.mm.

30D: 22.5x15 = 338 sq.mm = 200 sensors/wafer
1D2: 28.7x19.1 = 548 sq.mm = 120 sensors/wafer
FF: 36x24 = 864 sq.mm = 80 sensors/wafer

The real chip size will be larger, as there is some area surrounding the sensor. The practical output is lower than the figures above, as the dies cannot utilize 100% of the wafer area and larger dies gives more area loss, but it somewhat gives an idea about what output we're talking about.

A typical wafer cost is somewhere in the $1500-2500 range, although the technology used by Canon may be a bit more expensive. Further, the wafers may be smaller etc. OTOH, given the wafer volumes Canon buys, the wafer cost is most likely in the lower region anyway.

But independent how you calculate on yield and chip sizes, even with a $3000 wafer and a production yield as low as 20%, the silicon cost for a FF sensor must still be less than $200. Let's further say that a 1D2 sensor cost with its smaller size and hence larger yield is $50.

How can a $150 (which most likely is much smaller) increase in silicon cost be translated into $3360 increase in body price?

You can always argue that the economy of scale is different, but hey – the 1D2 is not a mass market camera nor is the 1Ds2.

... but when an APS-C sensor camera (30D) grows into a FF sensor camera (5D), the price "just" increase by $1300. Here, the production volumes of the APS-C sensors vs. 5D FF sensors must be huge.

Yes – we already know it – because Canon have decided how the market shall be segmented and after all it seems like we're all prepared to pay for it.

... Well, just wonder how much Hasselblad pays for their 39MP H3D sensors...
 
Did you measure the actual size of the chip, or just the plastic case it is contained in? Such chips are no more than about 4-5 mm on a side and are inside the plastic case which people think is the chip.
http://www.extremetech.com/article2/0,1697,2053781,00.asp
This is the chip of a new graphics card that is just out. It looks
to be about 22x22mm wich is over half the size of a full frame
sensor. These graphics cards are going for 450-650$ and include a
lot of expensive memory and a board and cooler that also cost quite
a lot i think so i suspect this chip isnt all that expensive. Also
considering the full frame sensors are made on really old equipment
how much can they really cost? Seems to me canon is lying.
--
So many lenses, so little time!
 
Oops, just now noticed the link with the pictures - so my previous comment was mistaken.
--
So many lenses, so little time!
 
http://www.extremetech.com/article2/0,1697,2053781,00.asp
This is the chip of a new graphics card that is just out. It looks
to be about 22x22mm wich is over half the size of a full frame
sensor. These graphics cards are going for 450-650$ and include a
lot of expensive memory and a board and cooler that also cost quite
a lot i think so i suspect this chip isnt all that expensive. Also
considering the full frame sensors are made on really old equipment
how much can they really cost? Seems to me canon is lying.
A sensor isn't just a silicon chip. Also needs microlenses, all the coatings and protective layers. Then there's the rest of the camera, controls, LCD, metering, processor, memory, CF interface, USB controller, etc. etc.

--
http://public.fotki.com/wibble/public_display/

 
Im comparing the 5D with the 30D wich also has all of that other stuff and the 5D with the 30D sensor would cost very little extra wich means they are charging 1000$ or more extra for the sensor.

I think those coatings and microlenses are a small part compared to the actual sensor.
 
I read theyve said things about it in interviews and there is also their full frame white paper right? I read that they were saying the cost of the 5D sensor is giving them problems as if they are selling the 5D for a loss and the sensor costs 1500-2000$.
 
JakobE wrote:
....
Although I don't know what CMOS process Canon is using, assume a
300mm wafer with a theoretical area of some 70,686 sq.mm.
Canon use 8inch (200mm) wafers to start with
30D: 22.5x15 = 338 sq.mm = 200 sensors/wafer
1D2: 28.7x19.1 = 548 sq.mm = 120 sensors/wafer
FF: 36x24 = 864 sq.mm = 80 sensors/wafer
Which assumes that there is no waste space on the wafer, patently incorrect. A 200mm wafer can actually hold 20 FF sensors as opposed to the theoretical 36 that your math would assume. See http://www.robgalbraith.com/public_files/Canon_Full-Frame_CMOS_White_Paper.pdf
...
A typical wafer cost is somewhere in the $1500-2500 range, although
the technology used by Canon may be a bit more expensive. Further,
the wafers may be smaller etc. OTOH, given the wafer volumes Canon
buys, the wafer cost is most likely in the lower region anyway.

But independent how you calculate on yield and chip sizes, even
with a $3000 wafer and a production yield as low as 20%, the
silicon cost for a FF sensor must still be less than $200.
Using those numbers you'd get 4 good FF sensors per wafer, each costing $750 to make. At that point you still have the costs of testing, adding the micro-lenses and CFA, packaging and the optical/AA filter.
 
Do you think the wafers for the new 90-65nm process are still only 3000$ or less? They could probably get 300 or more cpu's out of this wich makes them very cheap.

They have said that these fabs are getting so expensive that the next gen fabs will become too expensive to be economically feasible. I think the fabs cost about 2-3billion$ now? I read somewhere an amd fab with 20mm wafers made 30000wafers per month, do those 30mm wafer fabs make as many? If they would that would mean they can make about 100million cpu's per year wich brings the fab cost per cpu to 5$ in just a few years. Given that they are selling most of these cpu's for 100-300$ this is a negligible amount. It seems the industry is in a habit of exagerating.
 
It seems that soon the full frame sensors should be cheap anough for a 2000$ 5D replacement.
 
Im comparing the 5D with the 30D wich also has all of that other
stuff and the 5D with the 30D sensor would cost very little extra
wich means they are charging 1000$ or more extra for the sensor.
I think those coatings and microlenses are a small part compared to
the actual sensor.
But what you think isn't really all that important compared to the truth huh?

Heck, You know what - the D30 used to cost what the 5D costs now - does that make the D30 the biggest ripoff of all time? Maybe things move on, technology gets cheaper the more experience you have and there is a price for the best early on.

--
http://public.fotki.com/wibble/public_display/

 
The CMOS sensors used in cameras most likely do not require the most recent processes in terms of small geometries. Further, the state-of-the-art sub-micron processes used by high density digital, such as CPUs and memories are inferior when it comes to current leakage, which must be an absolutely no-no for imaging sensors.

Analog performance, low noise (low leakage) is the key, so something like a 0.35 or 0.5 micron process is probably likely. These processes are very mature and "old" equipment and fabs can probably be used.

Furter, given that Canon apparently uses 8" wafers, I strongly doubt that a wafer will cost over $1000, probably much less.
 

Keyboard shortcuts

Back
Top