Bitcoin Forum

It's been selling for $10k. How much does it cost to produce those?

a million

My guess is they're extremely overpriced to compensate how much the manufacturers would profit by just keeping the stuff and mining themselves.

Does ASICMiner have publicly available financial statements? If so, you can look there.

500k or less if you can clone one from avalon or asicminer

Marginal cost. The cost to produce one if you already have the factory set up.

When Avalon posted their TMSC contract, they didn't fully black out the prices and someone saw they pay around $4k per wafer. They get ~4k chips out of the wafer, and packaging would add less than $1 per chip, so their cost is likely in the $1-2 range. That might have gone down with their recent increase in volume though, who knows.

But how much was the mask cost?

OP says MARGINAL cost.

The mask cost is a major cost which you have to divide by the total volume. For low volumes it will outnumber the wafer cost. The Avalons are using very old technology where the mask cost is pretty low compared to a more recent technology. The software cost is also typically in the multi 100k$ range.

http://en.wikipedia.org/wiki/Marginal_cost
They've already paid for the mask. While the number of wafers produced will affect the average cost, it won't affect the marginal cost unless for some reason another set of masks needs to be made.

They still have to make up the cost of the mask.

It would cost tens of thousands to millions to produce TONS of asics.


Avalon uses TSMC to make the chips it seems.   ALTERA and XILINX use TSMC to make their asics and FPGAs as well.

a 130nm ASIC seems very old school technology, so I am guessing the cost was in the thousands to Hundreds of thousands of dollar range.

When you get to 13-28nm asic...then we are talking millions.

Again this is all speculation from myself working with Engineers, Managers, Directors, VPs and CEOs  in the Semiconductor industry.

"total cost" includes the mask.  Includes input costs, labor, etc.  i agree with kingkoin. source: wife is a cpa.

13-28nm is billions.  That's basically only Intel CPUs (and AMD @ 28nm)

Altera and Xilinx both have 28nm FPGA's. Altera has shown a demo of a 20nm FPGA and has announced that they are working with Intel on their next generation 14nm FPGA's
http://newsroom.altera.com/press-releases/nr-20nm-device-altera.htm
http://newsroom.altera.com/press-releases/nr-altera-intel-agreement.htm

There's also 28nm ARM's found in many cellphones and tablets.

The price the cpu manus pay has to do with complexity as well.
They need a huge die (new desktop cpu's have more than a billion transistors per die) to work perfectly so the quality of the process needs to be very high.

I know this is semantics, but "total cost" does include the mask. "Marginal cost," however, does not include the mask. The marginal cost is $1-2 per chip. Break-even cost (marginal cost plus sunk costs, loan interest) is about $8-15 (from how much they are selling the chips for in batches of 10,000 chips)

Edit: this is very basic accounting/economics. Marginal cost ignores all sunk cost and only considers the cost to produce one additional unit. It's the "slope" of the supply curve, excluding any offset due to NRE/tooling costs - but it does include additional costs of labor for each unit. They do need to make the money back, but that is not included in the calculation or definition of marginal cost.

Listen to Maqi-something.

This is why electronics are so cheap a year or two after it's released. The real investment is in the development, and when a new generation comes out, that old development is "worthless".

Some years ago you would pay $100 for a 128MB usb-stick. Now you get a 2GB for less than $5.

This is correct. I caught myself up in the term total cost. Marginal would just be simply that what you said.

The real question here is why do you care about the marginal cost. I assume its to gauge the markup on the product. This is more for the firm to decide if its worth producing more runs past the initial production run that would include mask costs. This info is pretty useless without knowing the initial costs and debt the company already has from r&d expense. The marginal costs maybe low but its possible it may take the firm some time to become profitable, so judging a product by marginal costs does not say much as to whether or not the product is priced fairly in regards to the firms initial sunk in investment.

NRE, development const, tool cost, mask production const, etc. should not be included in the marginal cost per the definition of the term.

However, I wanted to know the mask cost as for a miner these costs will outnumber the unit production cost by large. The volumes for miners, at least at present, is so low in an ASIC context where one usually expect to sell hundreds of thousands or hopefully millions of devices.

The fab cost is tremendous. Samsung has recently started production of 10nm NAND flash. They're building (it might even be ready by now) a new fab for mass production of 10nm NAND flash which costs $7 billion.

The 128MB usb-stick you paid $100 for several years ago was made using an old technology in a fab which might even have been closed down by now.

The $5 2GB stick could not even be made using the old technology (unless you used lots of dies or chips making them bulky and power hungry). The feature size have decreased so the density have increased which will fit more devices on a wafer, so each device gets cheaper, if the defects are fairly constant across the wafer the yield in terms of number of devices will go up.