Opinion: A Material Delay – A Likely Reason for Camera Delays and Production Issues

[Richard CN]

Sony has done the most bragging about stacked sensors and one of the items that is clearly included in the stack is RAM and there is zero likelihood that those chips are made by either Sony or Canon. The biggest challenges in stacked sensors are a) the fabrication of the BSI sensor itself, since it has to be back lapped to extreme thinness and b ) the stack assembly process which includes very precise alignment and through-silicon vias in the rear layers.

a) that's just a fabrication issue. once you figure it out and can do so reliably it doesn't get any harder. Canon could do the ram chips btw - easily. Sony? not sure. also how much memory? given the area of a full frame sensor just because it's memory doesn't mean it's using small design rules past what sony or canon can achieve. it's not as if we didnt' have memory when 45 or even 90nm was cutting edge. 1Gbit SDRAM used anywhere from 100 to 60nm processes. and those chips were a lot smaller than the surface area of a sensor. 20 shot on sensor buffer of a 50Mp sensor would be 14Gbits. possible? it's not exactly out of the possibility.. So no, it's more than zero.

b) Canon bought a company to literally make that easier for them to do. it also depends on how they do the passthrough and the pads between substrates as well. edge placement is much easier than under each pixel as an example.

but do keep in mind that it costs a lot of money just to set up the sensor design. if you are making a million sensors, that's one thing, if you are making 10,000 - that does tend to be quite expensive. So while it may be easier for Canon to do right now, the design costs of the R3 sensor will still be pretty significant.

but once you figure out the manufacturing processes it doesn't get any more difficult with a new sensor, the processes and yields get better as time goes on. so the idea that the new sensors are holding up things? yeah no, i don't buy it.

the process and manufacturing of sub 5nm that TSMC is dealing with is entirely different than what we are discussing here. Sony uses 45nm from what I last recall seeing. Canon uses 90nm the last I've seen (a year ago).

 

[Dragon]

a) that's just a fabrication issue. once you figure it out and can do so reliably it doesn't get any harder. Canon could do the ram chips btw - easily. Sony? not sure. also how much memory? given the area of a full frame sensor just because it's memory doesn't mean it's using small design rules past what sony or canon can achieve. it's not as if we didnt' have memory when 45 or even 90nm was cutting edge. 1Gbit SDRAM used anywhere from 100 to 60nm processes. and those chips were a lot smaller than the surface area of a sensor. 20 shot on sensor buffer of a 50Mp sensor would be 14Gbits. possible? it's not exactly out of the possibility.. So no, it's more than zero.

b) Canon bought a company to literally make that easier for them to do. it also depends on how they do the passthrough and the pads between substrates as well. edge placement is much easier than under each pixel as an example.

but do keep in mind that it costs a lot of money just to set up the sensor design. if you are making a million sensors, that's one thing, if you are making 10,000 - that does tend to be quite expensive. So while it may be easier for Canon to do right now, the design costs of the R3 sensor will still be pretty significant.

but once you figure out the manufacturing processes it doesn't get any more difficult with a new sensor, the processes and yields get better as time goes on. so the idea that the new sensors are holding up things? yeah no, i don't buy it.

the process and manufacturing of sub 5nm that TSMC is dealing with is entirely different than what we are discussing here. Sony uses 45nm from what I last recall seeing. Canon uses 90nm the last I've seen (a year ago).

Yes, you can build 90 nm RAM, but it is not power efficient in today's world and heat is one of the biggest enemies of the modern camera, so I still say zero probability that the memory is fabbed in-house. BTW, TSMCs CoWos challenges have little to do with the small geometries they are using to fab chips. CoWos is the multi-chip packaging process that allows AI GPUs to exist and the scale and connection density is pretty similar to a stacked sensor. Very precise dedicated machines are required and , no, just because Canon was successful with the R3 does not mean they automatically have enough capacity for an R5 or that a possible 45 MP stacked QPAF sensor is not more challenging to assemble. I am not saying that Sensor difficulties are the delay issue, but rather questioning the Magnesium theory. Magnesium is currently $2,700 per metric ton which translates to about 50 cents per camera when you take into account that the Magnesium will be alloyed with Aluminum, etc. before it is cast into a camera body. The price peaked in November and both the price and overall usage of Magnesium have been dropping recently, so shortage of supply is not likely that big an issue (if so, the price would be higher). The most likely reason for the delay is that current inventory levels are too high and that theory is reinforced by the rather continuous discounts we have seen for the last 6 months. The current level on inflation (at least in the US) has had a damping effect on discretionary capital purchases in general.