HopalongPetrovski
I'm Spartacus!
I'm pretty sure he's saying your from the Bosphorus and I'm a moron.
BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
HopalongPetrovski
I'm Spartacus!
If you put more on you'll distort the performance characteristics.I'm pretty sure he's saying your from the Bosphorus and I'm a moron.
Pom down under
Top 20
I'm pretty sure he's saying your from the Bosphorus and I'm a moron.
Space Cadet
Regular
I'm waiting for the Pico to come out. Should be much more comfortableWell. If they are going to probe me I hope they use a well lubed AKD1500 and not an edge box!
SC
Jacob Kiraz
Regular
It Doesnt matter,
AGM Is coming up and the year is nearly into Its 3rd Month, Lers see how deep Brainchip goes inro the year.
Alot of products on offer
AGM Is coming up and the year is nearly into Its 3rd Month, Lers see how deep Brainchip goes inro the year.
Alot of products on offer
Bravo
Meow Meow š¾
Hi Pom or anyone.
@Bravo old girl take pity on a befuddled pensioner.
Sorry if I am going over old ground but was there an explanation arrived at regarding the variable cost per chip $2.94 amount quoted?
Either the chips cost $2.94 each, flat, regardless of the quantity we order or, given that word "variable", that is a minimum or maximum cost per chip depending on how many we order at a time.
Am I overthinking or under thinking this?
Hi Hoppy,
Unfortunately, I only know as much as everyone else here, which seems to be...not bloody much.
I feel like a bit of a goose because I seem to have started this whole debate of biblical proportions because I was trying to make a much simpler point, namely that none of us know what RTX (or anyone else for that matter) would actually be paying per chip. Although we've seen the slide that references a $4ā$50 price per chip (volume dependent) range, we donāt actually know what āvolumeā means in BrainChip world or where on that sliding scale a customer would sit.
So, my comment was about revenue uncertainty. But somehow, I know not how, it has apparently become incumbent upon me to explain the history of semiconductor costings and the empirical and ethical virtues of variable cost versus fixed cost allocations and how this does or doesn't tie in with the huge sum of money Jeff Bezos recently spent on the Melanoma documentary. I have no insight into any of those matters, sadly.
The truth is, I donāt know anything worth reporting other than the fact Iām about to top up another glass of vino before Iām asked to produce a wafer yield sensitivity spreadsheet.
Cheers!
Hi @Bravo ,
The CVD process for making ICs usually involves a carousel with a number of wafer stations, eg 6:
https://patents.google.com/patent/US11798825B2/en
Apparatus and methods of processing a substrate in a carousel processing chamber are described. A wafer pedestal has a support surface with a support shaft extending below the wafer pedestal. A roller pinion wheel is below the wafer support around the support shaft. The roller pinion wheel has a plurality of spokes in contact with the support shaft and a wheel with a plurality of roller pinions spaced around the outer periphery of the wheel. Processing chambers incorporating the wafer pedestal and processing methods using the wafer pedestal for in-situ rotation are also described.
An optimum production run would consist of multiples of 6 wafers.
There is CVD chamber set up and preparation time to account for. With electronic "photomasks" replacing tapeout, the process is much more efficient than in the olden days, but the equipment is much more expensive, which affects depreciation, but no doubt there are generous tax writeoffs/incentives.
Then there is the yield per wafer, determined by the number of chips per wafer minus the number of defective chips per wafer. With the very small sizes, you get a lot of chips per wafer, but the defect rate can be very high, particularly in the initial stages.
FG has chosen 28 nm for their radhard GR801, so it should have low defects, but lower chips per wafer, and of course, each chip includes the RISC-V processor, so there will be fewer Akidas per wafer than for a straight Akida wafer. This is probably true for most IP licences, as the licencee will probably wish to build an entire SoC with Akida as a coprocessor to a CPU/GPU/MCU.
The later versions of Akida will have a larger chip footprint as they use more bits, eg, Akida 2 is 8-bit, Akida 3 is INT16/FP32-bit, so fewer chips per wafer.
So really, it's your estimate of the length of the string, but I'd say somewhere between $3.99and $50.01 per chip.
The CVD process for making ICs usually involves a carousel with a number of wafer stations, eg 6:
https://patents.google.com/patent/US11798825B2/en
Apparatus and methods of processing a substrate in a carousel processing chamber are described. A wafer pedestal has a support surface with a support shaft extending below the wafer pedestal. A roller pinion wheel is below the wafer support around the support shaft. The roller pinion wheel has a plurality of spokes in contact with the support shaft and a wheel with a plurality of roller pinions spaced around the outer periphery of the wheel. Processing chambers incorporating the wafer pedestal and processing methods using the wafer pedestal for in-situ rotation are also described.
An optimum production run would consist of multiples of 6 wafers.
There is CVD chamber set up and preparation time to account for. With electronic "photomasks" replacing tapeout, the process is much more efficient than in the olden days, but the equipment is much more expensive, which affects depreciation, but no doubt there are generous tax writeoffs/incentives.
Then there is the yield per wafer, determined by the number of chips per wafer minus the number of defective chips per wafer. With the very small sizes, you get a lot of chips per wafer, but the defect rate can be very high, particularly in the initial stages.
FG has chosen 28 nm for their radhard GR801, so it should have low defects, but lower chips per wafer, and of course, each chip includes the RISC-V processor, so there will be fewer Akidas per wafer than for a straight Akida wafer. This is probably true for most IP licences, as the licencee will probably wish to build an entire SoC with Akida as a coprocessor to a CPU/GPU/MCU.
The later versions of Akida will have a larger chip footprint as they use more bits, eg, Akida 2 is 8-bit, Akida 3 is INT16/FP32-bit, so fewer chips per wafer.
So really, it's your estimate of the length of the string, but I'd say somewhere between $3.99and $50.01 per chip.
HopalongPetrovski
I'm Spartacus!
Thatās all right Miss pixie boots.Hi Hoppy,
Unfortunately, I only know as much as everyone else here, which seems to be...not bloody much.
I feel like a bit of a goose because I seem to have started this whole debate of biblical proportions because I was trying to make a much simpler point, namely that none of us know what RTX (or anyone else for that matter) would actually be paying per chip. Although we've seen the slide that references a $4ā$50 price per chip (volume dependent) range, we donāt actually know what āvolumeā means in BrainChip world or where on that sliding scale a customer would sit.
So, my comment was about revenue uncertainty. But somehow, I know not how, it has apparently become incumbent upon me to explain the history of semiconductor costings and the empirical and ethical virtues of variable cost versus fixed cost allocations and how this does or doesn't tie in with the huge sum of money Jeff Bezos recently spent on the Melanoma documentary. I have no insight into any of those matters, sadly.
The truth is, I donāt know anything worth reporting other than the fact Iām about to top up another glass of vino before Iām asked to produce a wafer yield sensitivity spreadsheet.
Cheers!
And thanks for the explanation but it seems that Mr P has taken a powder and scarpered.
I guess betwixt us weāve managed to quiet his fevered brow, allayed his concerns and put all his doubts to rest.
So, having done our good deed for the day, Iāll say good night sweet princess and gawd bless all who sail in your sauna.
Similar threads
