Late on Sept. 24, chip giant Intel (NASDAQ:INTC) announced a new processor family, known as Coffee Lake, for the desktop personal computer market. These chips are notable because they offer significantly more processor cores than their predecessors, called Kaby Lake, for roughly the same price.
Now, while offering more cores to customers for the same price clearly improves the value proposition of these chips to potential buyers, the additional cores translate into larger chips. And, all else being equal, larger chips are more expensive to build than smaller ones.
In this column, I'd like to explore the potential cost structure increase that the new Coffee Lake chips bring for Intel compared to their predecessors.
Some estimates required
There are many unknowns involved in trying to estimate the manufacturing costs of a chip. These include the costs of processing the silicon wafers as well as the yield rates of the chips. No chip manufacturer publicly discloses that information, so the best that we can do is rely on our own, as well as third-party, estimates for such data.
Even with intelligently chosen estimates, the results we come up with could still be wildly off from what the true figures are.
With that major caveat in mind, here's what we know to be true:
- The size of the hex-core Coffee Lake chip comes in at 149 square millimeters, per a leak from the generally reliable BenchLife.info.
- The size of the quad-core Kaby Lake chip should come in at 123 square millimeters, per a measurement of the quad-core Skylake (architecturally virtually identical to the Kaby Lake chip) done by AnandTech.
Let's also assume that the wafers used to produce them cost the same (since they're both built using very similar manufacturing technologies) and that the defect densities (that is, the average number of defects in the wafers) are the same.
Crunching the numbers
Using this die-per-wafer calculator, I get about 472 dies per wafer for the quad-core Kaby Lake part and 393 dies per wafer for the Coffee Lake part.
Now, assuming a constant defect density of 0.2 defects per square millimeter, that same calculator tells us that of the 472 total chips produced per wafer, 372 of them will be usable.
Applying this to the larger hex-core Coffee Lake die, I get 286 good die per wafer for a yield rate of 74.8%
The trickiest part of this analysis is trying to properly estimate the costs of a wafer of each chip. One estimate pegs the manufacturing cost of a 14/16nm wafer (assuming 80% factory utilization) at about $4,800. I think Intel's costs will be higher than a typical contract chip manufacturer's costs because it manufacturers more complex, high-frequency processors, but for the sake of this analysis, I'll use the $4,800 figure.
The die cost of the quad-core Kaby Lake chip under these assumptions would come out to around $13, while the die cost of the hex-core Coffee Lake would be $16.78.
That's just under a $4 difference in raw die cost.
Yet there's more to the manufacturing costs of a chip than just die cost. Intel doesn't sell dies on its own -- those dies need to be mounted onto silicon packages (there is yield loss associated with this process, as well, which can lead to good dies being tossed out due to unsuccessful packaging attempts), tested, boxed up for shipment, and then ultimately shipped to distributors, who then sell these products to customers for an additional market.
Moreover, even though Intel is selling larger chips, the transition from smaller chips to larger chips could lead to an increase in factory utilization rate, which may lead to a reduction in effective wafer costs.
It seems reasonable to think Intel's overall manufacturing costs, on an apples-to-apples basis, will go up in the transition from Kaby Lake to Coffee Lake. However, Intel appears to have raised prices on some models -- the ones that'll likely be most popular with the desktop enthusiast crowd -- generation over generation to try to offset those increased manufacturing costs.
The key figure to watch in the quarters ahead to try to gauge the impact of this product transition will be Intel's gross profit margin percentage. We'll know soon enough if these new Coffee Lake chips, as well as the recently released Kaby Lake Refresh chips (these increased core counts in laptop computers generation over generation), will negatively impact Intel's gross margins.