Hi,

I usually don't really ever comment on rumours, as I usually have nothing to say about it. But the rumours of the 5D has me wondering.

I've been reading some articles lately and learned that Canon can fit 20 full Frame sensor on each of their wafers with a 25% yield. So, you'll have about 5 good ones, and 15 bad ones.

Okay, I know that Intel and AMD puts however much CPUS on each wafer, but when the cpu doesn't make it to spec, they just downclock it and sell it as a slower CPU. I'm not an engineer or anything like that, so my facts may be blurred.

Anyhow, would this be possible with Sensors? Such as, Canon would get the 20 full frame sensors on their wafer. Only 4 will make it to a 1DS II, but instead of throwing out the other sensors, they'd just see if they meet the specs for the 5D sensor and use them for that? Giving a much higher yield for useable sensors per wafer.

Of course, I could just be wrong about the whole thing, and may have misunderstood the exact science about wafer processing.

take care
roger

The first thing you need to know is that CPU chips are very much smaller than sensors. The biggest mass market chips are in the order of 150-200 square millimeters. A full frame 35mm sensor is in the order of 860 square millimeters. You actually get more than 4 times the number of CPU off of a wafer because there is less waste space.

Next the CPU designer can design multiple redundant parts in to the chip and disable bits that don't work (for example a 1GB cache can be downgraded to 512MB if part of it doesn't work) so this increases the yield of good parts. ALL of an image sensor (except maybe a handful of pixels) has to work.

After you've done all of this you still get a good proportion of CPU chips that fail and go directly to the bin. The remaining chips are performance tested and split out by speed grade. THIS is what you're thinking of. The chips in slower CPUs aren't failures, just they haven't been found to be reliable at the maximum speed of the design.

I understand the failures within a CPU and test selection process... but for a camera senor would they be:-

a) single dead pixel(s) - as TFT screens --- this could be compensated for in post process by the camera or convesion process - not photographer
b) clusers --- this may be compensated, but may be seen on large prints
c) loss of large areas/section of the sensor --- therefore unusable
d) or/and mix

I think I would buy a 1DsMk2 or anything of that quality if there where a few compensated dead pixels.... because I have never had a perfectly clear photo anyway....and would you really notice a loss of multiple pixels

Also remember that an imaging sensor is both an analogue as well as a digital device. The actual image is not a digital image until after the ADC. First a signal must go from the photodiode, through an amplifier, then to the ADC. So the sensor design and manufacturing has to take into account how well the chip can be made, right down to how well each little photosite works.

In Canon's CMOS sensors, according to their lit, there are a couple of transistors around the photodiode, to allow both reading and refreshing. As I understand it, the difficult part in getting a sensor with a good dynamic range has to do with both extremes: (1) what happens when the photodiode is saturated (can't store anymore energy), and (2) how small a charge (from the photoelectric effect) at the photodiode can be reliably read. In manufacturing a sensor, techniques are used to tackle these problems which are not normally used in everyday CMOS chips, such as the common ASIC (such as DIGIC.)

Furthermore, on their writeups about the 1Ds sensor, Canon mentions that each sensor has to be made in twice as many steps as usual, as their actual stepper/photolithography devices cannot make chips that large (I assume they mean the image circle for photolithography part)! The largest chips they could make are the APS-C size.

Canon just recently opened a massive new chip facilty (I think their financial reports mentioned it was circa $500 million facility) in Japan. Part of the purpose of this facility is to make sensors for cameras, of course. Perhaps they now have solved the above size limitation they had in making the 1Ds sensor. This new facility very well may be the key reason why they now can offer more affordable FF sensors.

Dan9

There are many many many things that can affect the die on a wafer during the build process.

Creating die ( sensors or CPU's ) is just like creating an old photo. You take a bare
silicon wafer and you coat it with different chemicals/metals layer by layer until you build
up the electrical stackup that creates the components.

So there are hundreds of steps required during the stacking of the layers. Each layer
of chemical or metal deposition has its own inherent potential problems that can affect
the performance and affect the yield.

For sensor die one major killer is a defect called "PC" ( partical count ) and is nothing more than contamination in the form of small particles.... and even here there are hundreds of sources that contribute to un desired PC counts... some are stainless steel ( from wafer handlers ) some are caused from miss-process ( peeling tungsten, silane powder, mis-struck plasma, dirty deposition chamber, scratched wafers, broken wafers ) the list goes on and on for each and every step in the wafer build cycle. The idea is to
track down the source of PC contamination and fix it. This is one area that has great potential to get the yield up.

Those examples are only the PC defects... there's much more than that even, some are over etch, under etch, pealing photo resist, shifted photo pelical, too much implant,
too thick metal dep, too thin metal dep, too thin oxide, too thin SOG...

I would be very suprised that Canon would have yeild down at 25% , in our fabs we get that kind of yield during the startup of a device and the yeild numbers just climb from there... most if not all fab's run around ~90 + % after all of the processes have been tweeked.

Just lets say that there would be alot of process engineers fired at our fab's if they couldn't do better than 25%

I don't know for sure but I would also expect to see Canon producing some of the sensors in 300mm Fabs now and that will bring up the number of sensors to the market.

The trend for the last few years has been to " foundery " components where larger numbers of devices are required. The way this works is a Semi-conductor company designs the devices, works out the design/process issues in its own fab ( close to the designers ) and then sends the recipes and in some cases the equipment to a seperate
company that only has fab's and doesn't design ( Asia Fabs ) and the " Foundery Fab "
actually creates the products. Its much cheaper that way. And of course the yields have to be very good or there are other foundery fabs that compete at that level.

Sensor manufacturing versus CPU manufacturing are very different.... when a sensor is
bad, its just bad... no one wants a sensor with lots of dead pixels.

With CPU's... if some of the on-board memory is determined compromised then it can be mapped out, just like standard memory chips.. same thing.

There are several things that can affect the CPU's speed ratings... the races may have ended up too thick, too thin, resistance of components not at optimal...lots of other issues can have effect also... but the CPU is speed tested and of course they are bin'd
accordingly and sold at their respective performance level.... but like all semi-conductors, once the processes have settled in and all the pokes and tweeks completed the performance variances dwindle to nothing.

Also, As far as I know at this time... the largest CPU as in silicone real-estate is the
SUN Ultra Sparc... Its as large if not larger than any of the camera sensors.

dan9 wrote:


Furthermore, on their writeups about the 1Ds sensor, Canon mentions that each sensor has to be made in twice as many steps as usual, as their actual stepper/photolithography devices cannot make chips that large (I assume they mean the image circle for photolithography part)! The largest chips they could make are the APS-C size.

Dan9

Dan, The Canon Stepper/photolithography tool is actually just one huge Camera.

Its about the size of a small bedroom... A silicone wafer is coated with Photo Resist and then it enters the photolith portion of the tool, there in the tool is the actual negative ( pelical ) with the pattern that is to be projected onto the wafer. A single die is projected thru the photolith lens, then the wafer is " stepped " side to side, top to bottom ( the stepper portion of the tool ) and at each step another die pattern is projected onto the wafer until the entire wafer has been populated. After
that the photo resist is now developed and stripped off of the wafer. This leaves the pattern exposed for the deposition process... metal coatings and other required levels.

The limiting factor to size constraints is just like normal photo issues... the lens
can only project at that size before the actual lens performance deteriorates the image at the edges to the point it no longer projects a pattern that is usable in a manufacturing process to create semi-conductors.....

The 1D series sensors are actually two sensors side by side, they are easily seen
on the MKI camera's if you perform the heat saturation test as the two different sensors will show in the exposure.

So, with the size constraints presented by the limiting photolith lens, the stepper
has to make twice the amount of steps to create a die. The first step would be the
left side of the sensor and then projection and the next step would be the right side of the sensor and then a second projection... and then of course the next step would be to step over to the next die and the stepping/projection process starts over again... this being in contrast to a normal stepping process of just a single projection and then a single step over.

They create the die like this to eliminate " die parralelism"
and this method ensures that the left and right portions of the sensor are perfectly
parralel to each other in realtionship to the Z axis and the die attach process only has to worry about global parralelism during the glueing of the die to the carrier.

The number of the remaining steps to produce the sensor should be the same.....

dan9 wrote:

Canon just recently opened a massive new chip facilty (I think their financial reports mentioned it was circa $500 million facility) in Japan. Part of the purpose of this facility is to make sensors for cameras, of course. Perhaps they now have solved the above size limitation they had in making the 1Ds sensor. This new facility very well may be the key reason why they now can offer more affordable FF sensors.

Dan9

I'm wondering if the $500 million figure was just to upgrade the stepper/photolith tool and supporting equipment...

Almost every semi-conductor manufacturer in the world is in the process of building
at least one new Fab and many are building more ( Intel )... A new Fab is presently costing $2 Billion from a scratch investment. I have no idea how much the Mega Fabs in Korea cost, but its got to be a bunch.

The new stepper/photo tools are being priced at around $300 million per tool at this time and are the single most expensive semi-conductor tool in a fab...

Thats a lot of moola...

Thanks kites4 for all the info!

The new Canon fabrication facility actually is located in an already existing, large manufacturing complex Canon has in Northern Japan. Plus, Canon as one of the makers of steppers/photo tools (but I do believe they are not the largest here in Japan - I think that might be Nikon ) as well as making many other tools, hardware, and software - possibly combining all the above they could build a smaller facility at less cost than the electronics giants in other parts of asia.

Anyway, when the 1Ds came out they went through great pains to convince everyone how difficult it was to make the sensor, in order to explain why the camera cost so much. Now, at least in their corporate lit, they are trumpeting their new facilities and how it will be used for camera sensors. Bottom line for us, we hope, is that now at least there can be a FF camera for the rest of us.

Dan9

Dan, I have to agree... FF cameras are coming down in price and there will be several
models to choose from in the future...

If I remember correctly canon only made around 30,000 of the 1Ds bodies, + whatever the newer 1DsMKII adds up to be...
so the ROI ( return on investment ) just wouldn't add up unless canon goes ahead and starts pumping out FF sensor bodies... I think the future for Canon is FF sensors and
it can only get better for us the consumers.

This is also the reason that Canon has stated that the 1D and 1Ds bodies will merge into
one single body wtih FF sensor. Based on the current capabilities of the Canon Photolith
tools we should see a merge of these bodies with a 21mp sensor...
They have the capability now to mass produce such a sensor...

I think its just a matter of creating the next gen of the
DIGIC processor to handle the sensor off-loading so much data. I'm sure there are working copies in the design labs now.