az-dave wrote:
FX is only currently significantly more expensive. Technology has historically gotten cheaper over time. Just look back at how expensive microwaves, VCRs, flat screen TVs and other items were at introduction compared to where they are priced now. I remember trying to decide between 40 and 70 MB hard drives for about 300 dollars now we buy 1 TB drives for around a 100 dollars.
Yes, because for those applications there are cost savings to be had by scaling the size of the transistors downward. There is no such cost savings with image sensors as the size of the sensor is fixed.
However, if and when the cost of DX vs FX becomes insignificant, then you will see DX disappear all together as DX crop mode will work quite well for those needing DX format. We are decades from that.
Aug 25, 2010 at 05:23 PM
Andre Labonte Offline • • • • • • Upload & Sell: Off
Andre Labonte wrote:
Yes, because for those applications there are cost savings to be had by scaling the size of the transistors downward. There is no such cost savings with image sensors as the size of the sensor is fixed.
However, if and when the cost of DX vs FX becomes insignificant, then you will see DX disappear all together as DX crop mode will work quite well for those needing DX format. We are decades from that.
The other way to get the cost down is to improve manufacturing and sensor yield per run. I don't think we are there yet but I do not believe we are decades away either. It is unlikely that DX will be totally abandoned. It will continue to live in the lower end bodies and smaller form factors. All I can say is these are exciting times and I can't wait to see what comes out over the next few years.
Aug 25, 2010 at 07:49 PM
Andre Labonte Offline • • • • • • Upload & Sell: Off
az-dave wrote:
The other way to get the cost down is to improve manufacturing and sensor yield per run. I don't think we are there yet but I do not believe we are decades away either. It is unlikely that DX will be totally abandoned. It will continue to live in the lower end bodies and smaller form factors. All I can say is these are exciting times and I can't wait to see what comes out over the next few years.
That's been done. 300mm class 1 FABs have been the norm for some time now. FAB technology is relatively mature and there are deminishing returns there. Increasing transistor density still is best way to drive functionality and cost, thus the reason for <45nm tech nodes.
There will be so many defects per square cm and with a fixed sensor size, going smaller on the transistors does nothing for you. Also, you will always get approximately 1/2 the number of FX sensors than DX sensor for the same wafer. Therefore an FX sensor will ALWAYS and FOREVER cost 2X a DX sensor.
Also, as long as the focus is on improved performance from the sensor, there is little chance to reduce prices. NEW computers for instance, really have not come down in price in the last 10 years, but they have imroved in performance. The camera industry is much the same. When an older tech falls below a price point, it is abandoned for the next node. Most FAB costs are fixed for a given loading and the equipment must be payed for. The new FABs cost billions of $$$ to build and increasing wafer size (the other way of reducing costs) is becoming harder to do in part due to size and handling, and more due to the up-front equipment costs. Also, larger wafers only reduce costs where there is volume. Camera sensors, of the 100s of varieties out there, are hardly volume in comparison say RAM or micorporcessors. P&S sensors maybe, but not SLR sensors.
I don't think FX is as expensive as people make it out to be. The difference between the D700 and D300 when I made my purchase was about $400-500. That is chump change in DSLR world unless you are just getting your first body. The one DX lens I used was sold for barely any loss. The best lenses Nikon offer are all FX lenses, so it isn't like you are making any kind of sacrifice there either.
As FX becomes more popular, it will continue to drive the price down. Sensor yields do improve over time if you are not shrinking the size and even though the size of a sensor never changes, the process becomes cheaper as other chip makers advance technology to shrink down die size. The idea that the cost will remain the same forever is silly.
I think hlavo is mostly correct in that DX will be relegated to the entry level bodies for a few years, then scrapped. From my experiences, I personally think the IQ of FX files is much better than DX and that is enough to make that extra reach a moot point for me. TC's aren't that expensive.
Aug 26, 2010 at 06:36 AM
Andre Labonte Offline • • • • • • Upload & Sell: Off
As FX becomes more popular, it will continue to drive the price down. Sensor yields do improve over time if you are not shrinking the size and even though the size of a sensor never changes, the process becomes cheaper as other chip makers advance technology to shrink down die size. The idea that the cost will remain the same forever is silly.
.....
You contradict yourself here. First you say that yields will improve over time even if not shrinking size and then you say the process becomes cheaper as the chip makers advance technology to shrink down die size!
That's my point, the sensor does not shrink in size so it does not become cheeper to build. The state of the art for FABs is class 1 and has been for over a decade with no work being done to improve it. The only other way to improve cost is if you have sufficient volume to move to a 300mm FAB ... we are there already, so next is 450mm but that won't come into being for another 10 years given the current state of the technology and when it does come come, it will be memory and microprocessors first, not image sensors. Also, you never get rid of the 2x cost difference. Remember that defect yeild is proportional to the square of the area. So the DX chips will have 1/4 the defect hits to yeild in addition to producing 2x the number of die per wafer.
The one place where there will be some cost improvement is in process yeild (as opposed to defect yeild). That's fine if there is capacity to run the old technology node over the current cutting edge node AND there is a market for the old node. This is why P&S have come down so much in price over the last 10 years and why the sensor is a small part of the cost. However, in the DSLR market, people want the latest tech node (i.e. highest performance) in the latest model. The result is that as the FABs finally catch up on the process curve and yeild well, they have to switch off to the next tech node for the latest and greatest. This keeps sensor cost higher for DSLRs. Also, the DSLRs have much larger sensors.
Cost for DSLR sensors are not going to come down anytime soon. They will some day if the process stabilizes out, but we have a long way for that to happen.
Aug 26, 2010 at 07:18 AM
Andre Labonte Offline • • • • • • Upload & Sell: Off
Steezus wrote:
I don't think FX is as expensive as people make it out to be. The difference between the D700 and D300 when I made my purchase was about $400-500. That is chump change in DSLR world unless you are just getting your first body. The one DX lens I used was sold for barely any loss. The best lenses Nikon offer are all FX lenses, so it isn't like you are making any kind of sacrifice there either.
.....
Another thing I thought of. You indicate $400-$500 differnece in your purchase price of the D700 and D300. This is not the norm and I don't consider $400 to $500 to be "chump change". Current new prices for these bodies are:
D700 $2400
D300s $1500
B&H prices and you have to "add to cart" to see it since it's below MSRP
A difference of $900. This represents a 60% preimium just to get FX and a slower frame rate. But for the sake of argument, let's use the average of your number and say FX sensor costs $450 more than the DX sensor. That means the FX sensor cost $900 total and the DX sensor cost $450 (the 2x factor I previously mentioned). That means that the sensor is 30% of the cost for a D300s and 38% of the cost for the D700. That is HUGE. And remember, I was being conservative in using your number which is half of the $900 actual idfference. I'm sure that the difference is not all sensor, but most of it is.
Andre Labonte wrote:
There will be so many defects per square cm and with a fixed sensor size, going smaller on the transistors does nothing for you. Also, you will always get approximately 1/2 the number of FX sensors than DX sensor for the same wafer. Therefore an FX sensor will ALWAYS and FOREVER cost 2X a DX sensor.
is not x4 the cost, fx is double the size and has half the yeild?
Aug 26, 2010 at 11:29 AM
Andre Labonte Offline • • • • • • Upload & Sell: Off
louis fusco wrote:
is not x4 the cost, fx is double the size and has half the yeild?
That depends on wafer size and the number of die per wafer.
Example:
Let's say you can get 20 FX chips on a wafer. You will be able to get about 45 DX chips on the same wafer. More than 2x because you are fitting rectagles into a circular wafer and smaller means more area covered. Now let's say that you get 12 killer defects on the wafer and assume a kill ratio of 0.8, meaning 10 dead die.
Yeild = 10 FX chips or 35 DX chips. If you reduce the defect denisity to 5 dead chips, it 15FX and 40 DX. So the actual number is somewhere between 2x to 4x the cost. The 2X number is the absolute best case which never happens.
p.7 #10 · what'll this mean for future FX cameras?
Andre Labonte wrote:
You contradict yourself here. First you say that yields will improve over time even if not shrinking size and then you say the process becomes cheaper as the chip makers advance technology to shrink down die size!
Do you really think that the term 'process' in regards to semi conductors means shrinking the size Process invokes all of the technology and parts that are used in manufacturing a semi conductor. It improves, it doesn't stay the same. According to your theory it would cost Intel the same amount of inflation adjusted dollars to produce a PIII Coppermine wafer today as it did almost 10 years ago. Also, I guess you think the yields would remain exactly the same? Do you really think the processes involved in making these chips are unchanged and only the microarchitectures have changed? Obviously that is not reality and many other factors other than unit volume will work to bring down the price of an FX chip. Of course that works for DX chips as well, but the FX capabilities are enough that it is a clear successor to DX and will all but replace it.
Aug 26, 2010 at 01:04 PM
Andre Labonte Offline • • • • • • Upload & Sell: Off
p.7 #11 · what'll this mean for future FX cameras?
Steezus wrote:
Do you really think that the term 'process' in regards to semi conductors means shrinking the size Process invokes all of the technology and parts that are used in manufacturing a semi conductor. It improves, it doesn't stay the same. According to your theory it would cost Intel the same amount of inflation adjusted dollars to produce a PIII Coppermine wafer today as it did almost 10 years ago. Also, I guess you think the yields would remain exactly the same? Do you really think the processes involved in making these chips are unchanged and only the microarchitectures have changed? Obviously that is not reality and many other factors other than unit volume will work to bring down the price of an FX chip. Of course that works for DX chips as well, but the FX capabilities are enough that it is a clear successor to DX and will all but replace it. ...Show more →
How does pointing out your inconsistency imply that I think that changing the semiconductor process = shrinking the size? I never said that.
I'm a process development engineer in a wafer FAB and have been for the last 10 years. I worked on the Foveon chip and a number of other imager processes. While the process do change (and not just transistor size reduction), the cost associated with the tool sets and their operation are fixed. Gains are made by improving yeilds and finding other efficiencies or increasing wafer size with volume.
Trust me, COST IS KING in a wafer FAB. It's a big part of what I do all day, every day.
And to use your example of a PIII ... well, a PIII is not exactly cutting edge now is it? Look at the price of the cutting edge today, or even a node down from that, and look at the cost of the cutting edge (or a node down) ten years ago and you will see the prices are/were about the same. That is a KEY point to my argument. If sensor technology stayed fixed, I'd agree that costs would go down as yeilds on the mature process improve. But, the processes are always changing and the current stuff today costs the same as the current stuff did 10 years ago (now the old stuff today which is cheep). But who wants the old stuff, even if cheep, in a new camera?
I'm not alone in this, the other engineers who work with me in the FAB agree with my point of view with respect to image sensors.
p.7 #12 · what'll this mean for future FX cameras?
I guess you lost me because this is what you said:
Andre Labonte wrote:
You contradict yourself here. First you say that yields will improve over time even if not shrinking size and then you say the process becomes cheaper as the chip makers advance technology to shrink down die size!
It doesn't make any sense because I am not changing the term process to mean always shrinking the die. While research goes into creating smaller chips, it has the effect of creating new and more efficient processes that old microarchitectures can benefit from. So it is kind of like setting up a straw man because you and I both know that if Coppermine chip could be built a lot cheaper and with much higher yields than was possible 10 years ago thanks to advances in the process as a result of engineers shrinking down size.
I also strongly disagree that prices have stayed the same. Everything inside of servers, desktops, and monitors have drastically dropped in price. A run of the mill desktop with crt monitor would set you back $1k back in 2000. Building it on your own was not much cheaper. Now you can get the modern version of a run of the mill desktop for about $5-600 with monitor. Bleeding edge systems are the same story. There is no possible way you are going to tell me I am wrong about that! Shit was expensive 10 years ago, now days is awesome because you purchase hardware that is one cycle behind for cheap.
Process invokes all of the technology and parts that are used in manufacturing a semi conductor. It improves, it doesn't stay the same. According to your theory it would cost Intel the same amount of inflation adjusted dollars to produce a PIII Coppermine wafer today as it did almost 10 years ago. Also, I guess you think the yields would remain exactly the same? Do you really think the processes involved in making these chips are unchanged and only the microarchitectures have changed? Obviously that is not reality and many other factors other than unit volume will work to bring down the price of an FX chip. Of course that works for DX chips as well, but the FX capabilities are enough that it is a clear successor to DX and will all but replace it.