Do wafer costs decline as a node matures?

[Fred Chen]

"Cost" and "Price" are two entirely different things. Price is always based on the balance of supply and demand - those price hikes are largely due to demand being higher than supply. Cost, however, is much more predictable. Depreciation is the biggest factor for cost per wafer, but even as depreciation is being written off (typically in a straight line bases over useful life), as the process matures, yield increases (sometimes very significantly), so cost per KGD drops. However, as I noted above, if we take depreciation and yield improvements out of the equation (focus only on variable costs - same as marginal costs), I don't think those drop notably as the process matures.

Viewed from a different perspective, the costs for taking an IC into production are heavily weighted to fixed costs versus variable costs. Therefore, the goal for a company fabricating ICs is to keep capacity utilization high (divide fixed costs over the maximum number of wafers). That is why prices (and margins) decline sharply when we operate with excess capacity.

Agreed.

 

[Fred Chen]

https://www.sec.gov/Archives/edgar/data/937966/000093796621000007/asml-20201231.htm?s=03 p. 70

ASML's 2021 update of the litho energy consumption table, I'll also start separate thread:

Annual Report 2021 | ASML

ASML published its annual report, combining financial and non-financial results, on February 9, 2022.

www.asml.com

p. 57

 

[Eric Bouche]

To go back to the original question: of course they do! And it is quite dramatic. The key reasons are yield improvements, productivity improvements and depreciation of the tools

 

[jms_embedded]

To go back to the original question: of course they do! And it is quite dramatic. The key reasons are yield improvements, productivity improvements and depreciation of the tools

Several people have stated this in this thread and elsewhere, but there's little quantitative information to back it up. I realize that is next-to-impossible because of proprietary considerations, but without some kind of quantitative information, I don't think it helps much to assert this.

 

[Tanj]

To go back to the original question: of course they do! And it is quite dramatic. The key reasons are yield improvements, productivity improvements and depreciation of the tools

Thanks! PMCW was correct to highlight "price" vs. "cost", and the vendors like TSMC have discretion to flatten the price curve so it does not match the cost. At some date the process may be trailing edge so prices will fall with costs.

But, I was mostly curious about pricing.

 

[Daniel Nenni]

Thanks! PMCW was correct to highlight "price" vs. "cost", and the vendors like TSMC have discretion to flatten the price curve so it does not match the cost. At some date the process may be trailing edge so prices will fall with costs.

But, I was mostly curious about pricing.

You can see semiconductor manufacturing process improvements now with new node names. TSMC 16/12nm, 7/7+/6nm, 5/4nm and the different variations for the different markets to increase the life of a fab (N3, N3E, N3P, N3X). Given that there are still plenty of 200m fabs in production I see a very long life for the 300mm FinFET nodes. The wafer price of the nodes will continue to go up/down based on demand. A change in materials cost will mostly be due to inflation or supply constraints. TSMC definitely has the advantage there with economies of scale, large ecosystem, strong supply chain partnerships, etc...

 

[nghanayem]

To further add on to what Dan and others have said you can also look at what intel does with it's nodes over their lifetime. Initially high margin laptop chips are made. As the process matures server products can be made. When client and datacenter products move to the next node you see more price sensitive chips like chipsets, eDRAM caches, Atom processors, interposers, ect. be made. These parts are enabled by much of the tooling being partially or fully depreciated, recipes and process engineers becoming more efficient, improved maintenance quality/procedures, increasing yields, as well as any tool upgrades that allow the tools to run the with a higher quality or output.

 

[hist78]

To further add on to what Dan and others have said you can also look at what intel does with it's nodes over their lifetime. Initially high margin laptop chips are made. As the process matures server products can be made. When client and datacenter products move to the next node you see more price sensitive chips like chipsets, eDRAM caches, Atom processors, interposers, ect. be made. These parts are enabled by much of the tooling being partially or fully depreciated, recipes and process engineers becoming more efficient, improved maintenance quality/procedures, increasing yields, as well as any tool upgrades that allow the tools to run the with a higher quality or output.

For Intel the situation is getting more complicated after it adopted the so-called "Semiconductor co-investment program" with Brookfield. Much of this program's details are not public but I suspect it's a sale-and-leaseback contract in disguise.

While TSMC completes its 5-year equipment depreciation as required by tax law and accounting rules, Intel may still need to pay Brookfield decent amount of equipment and facility fees/rents/rights for many more years. It may reduce Capex Intel needed in the short term but increase Intel's cost in the long run.

During the recent quarterly earnings conference call, TSMC CFO stated that TSMC won't use such financial arrangement due to higher cost.

 

[nghanayem]

For Intel the situation is getting more complicated after it adopted the so-called "Semiconductor co-investment program" with Brookfield. Much of this program's details are not public but I suspect it's a sale-and-leaseback contract in disguise.

While TSMC completes its 5-year equipment depreciation as required by tax law and accounting rules, Intel may still need to pay Brookfield decent amount of equipment and facility fees/rents/rights for many more years. It may reduce Capex Intel needed in the short term but increase Intel's cost in the long run.

During the recent quarterly earnings conference call, TSMC CFO stated that TSMC won't use such financial arrangement due to higher cost.

Maybe. I was just giving evidence of how node costs decay over time, because otherwise you would have never seen things like 14nm chipsets.

 

[Scotten Jones]

For a given node the cost to run wafers goes down by more than a half over the first five years. Initially ramping up and improving yield, eventually equipment becomes fully depreciated and that is a huge cost drop. Labor costs generally drift up and most material costs drift down, although starting wafer are going up but depreciation swamps out everything else by far.