During Intel's (INTC 1.94%) most recent investor meeting, Bill Holt -- the individual who heads up the company's manufacturing technology development group -- admitted that manufacturing yields on its 14-nanometer manufacturing process weren't in the shape the company had hoped they would be in by now.
This is interesting in its own right, but what I think investors really care about is the impact this has had, and will have, on the company's product costs and, ultimately, financial results. Let's take a closer look, shall we?
Here's what Intel had originally expected
Last year, Intel showed the following chart:
Source: Intel.
The company had projected that the manufacturing costs of its 14-nanometer Broadwell product would cross below the costs of its 22-nanometer Ivy Bridge/Haswell products (at the same point in their respective ramps) by the third quarter of 2015.
In light of the company's projections for the yields on its 14-nanometer process, the company put out the following forecast for product costs in each of its major product price segments:
Source: Intel.
As you can see, Intel had predicted modest bumps in cost across the board, no doubt due to the fact that it didn't expect 14-nanometer product costs to come below 22-nanometer product costs until the end of 2015 (meaning the average for the full year would still be higher).
That's not how things played out in reality, though.
Here's what Intel actually saw
As a result of the fact that Intel did not improve yields on its 14-nanometer technology at the rate it had hoped to (Intel is at least a year behind its original yield projections from what I can tell), here's how Intel's product cost structure by segment actually turned out for the year:
Swing and a miss. Source: Intel.
Note that in the mainstream and value segments -- where Intel largely transitioned to shipping products based on its 14-nanometer technology -- costs were not only up relative to 2014 levels, but they were up by far more than the company had originally expected.
In contrast, Intel did a little better than originally expected in the performance segment as it did not really begin transitioning these products to 14-nanometers until late in 2015, so costs for the year were only up slightly relative to 2014 levels.
What does the future hold?
Intel went ahead and gave cost projections by segment for 2016 as well, as you can see in the slide below:
Source: Intel.
Intel is expecting costs in the performance segment to move up significantly as it transitions its high-end products from 22-nanometer chips to 14-nanometer chips.
Given how big of a cost increase Intel is projecting for this segment in 2016, one has to wonder how bad the company's costs in this segment would have been if it had transitioned such products to 14-nanometer in 2015, particularly since products in this segment are probably much harder to build/more sensitive to yields than those in the mainstream and value segments.
In the mainstream and value segments, though, costs should come down from 2015 levels if the yield improvements Intel is projecting materialize.
Intel also included product cost projections by segment for the final quarter of 2016, since it expects its manufacturing yields to improve over the course of the year. If Intel is able to improve manufacturing yields as it hopes, it should see a nice reduction in cost by segment. In both the performance and value segments, it should still wind up with higher costs than in 2015, but they should at least be much better.
Some thoughts on the 10-nanometer transition
Perhaps the most troubling thing here is that Intel went into high volume production on its 14-nanometer technology in early 2014 and the company still needs to make substantial improvements in yields on this technology into 2016 in order to bring its costs down.
Intel says it will launch its first 10-nanometer product, known as Cannonlake, in late 2017. Given the issues Intel is having with its 14-nanometer process, can investors really be confident that by the time the second half of 2017 rolls around Intel will be able to manufacture 10-nanometer chips in volume at reasonable costs?
The 10-nanometer technology, per Intel, is supposed to bring another big jump in transistor density. However, from what I have been told by a source familiar with the situation, Intel's yield difficulties on 14-nanometers has been largely due to the aggressive density scaling that it went for.
With Intel planning a similarly large leap in transistor density at the 10-nanometer node, I worry the company will once again face significant yield issues that hurt product costs and, ultimately, gross profit margins in the years ahead.
