In a recent article, I argued that Intel's (INTC 3.03%) chances of winning a contract to manufacture Apple's (AAPL 0.04%) leading-edge A-series processors are essentially nonexistent. That particular argument hinged on the fact that Intel CEO Brian Krzanich had told investors that foundry-related revenue wouldn't move the needle over the next few years.
Although I think that argument is quite solid, there's one that's even more compelling. Allow me to explain.
The difference between a CPU process and a system-on-chip process
For any chip manufacturing generation, Intel typically releases two "flavors" of process. The first flavor of a given process is the one targeted at high-performance CPU products and is used to build the vast majority of the company's microprocessors.
The process flavor that follows is the company's system-on-chip process. This has been described by Intel executives and technologists as a "superset" of the high-performance CPU process. That is, a chip that can be built in the CPU-only process flavor can be built in the system-on-chip version of the process.
However, true system-on-chip designs typically require features that the CPU-only process flavor doesn't have. For example, the system-on-chip processes offer a wider variety of transistor performance options as well as greater flexibility to optimize the interconnect stack for different priorities (i.e., density, cost, or performance), as well as high-voltage transistors for I/O.
At any rate, what you need to know is that Intel would need to use a system-on-chip process to build something like an Apple A-series processor.
When will Intel go into production on its 10-nanometer system-on-chip process?
Intel has said that the first products based on its 10-nanometer manufacturing process will hit the market during the second half of 2017. Technically, if Intel is really aggressive about getting its 10-nanometer technology into production, it could hypothetically start producing 10-nanometer A-series processors in February/March 2017 in support of an iPhone 7s launch in the second half of 2017.
However, it's worth pointing out that there is usually a multiquarter lag between when Intel first begins producing products on its CPU-only process and its system-on-chip process. This delta has shrunken in recent years, but the gap was around two quarters at the 14-nanometer node.
If we assume a similar gap at the 10-nanometer node, and if we are very generous in assuming that Intel begins production of its first 10-nanometer CPUs in January 2017, then this would still put the start of production on the 10-nanometer system-on-chip process in July 2017.
That's far too late to intercept the iPhone 7s, which will almost certainly contain a processor manufactured in a foundry 10-nanometer technology.
10-nanometer should be ready for the iPhone 8, but will Apple want it?
The iPhone that would naturally follow the iPhone 7s would be the iPhone 8, which should launch in September 2018. Intel's 10-nanometer system-on-chip technology should be good to go for a chip launch in 2018, but at that point I would argue that Apple will have a better process option available from TSMC (TSM 1.26%).
Indeed, TSMC has said that it will go into "risk production" on its 7-nanometer technology in the first quarter of 2017. This implies volume production should begin in early 2018, allowing Apple to get a 7-nanometer chip inside of the iPhone 8.
Now, one could argue that the naming of these manufacturing technologies is misleading and that not all technologies with the same labels are created equal; this is a valid point. However, I believe that, for mobile chip designs, TSMC's 7-nanometer should be a better choice than Intel's 10-nanometer.
In a nutshell, it comes down to the fact that TSMC's 7-nanometer process should offer better chip density than what Intel's 10-nanometer process will deliver.
Proving that density claim
A rough metric that is widely used to get a sense of the relative maximum theoretical chip densities of a given process is minimum metal pitch multiplied by gate pitch. In the table below, based on public statements from TSMC and Intel, I have provided data on what Intel and TSMC have in production at 14/16nm and what the 10nm and 7nm processes from both companies should offer:
|
Intel |
TSMC | |
|---|---|---|
|
14/16nm |
3640 |
5760 |
|
10nm* |
1842 |
2995 |
|
7nm* |
932 |
1647 |
Source: Intel, TSMC, EETimes, author estimates.
The Intel 10nm and 7nm estimates come from applying a 0.506 shrink factor. Intel has publicly said that it intends to apply similar shrink factors in future processes, so that is where those estimates come from.
TSMC has publicly said that its 10nm process offers 0.52 times scaling over its 16-nanometer process, and it has also said that its 7-nanometer process delivers a 40-45% area scaling benefit over its 10-nanometer process. In the table above, I am giving TSMC the benefit of the doubt and assuming a scaling factor of 0.55x.
Intel should have a density lead "10nm to 10nm," but TSMC should move to what it dubs "7nm" before Intel does. At that point, TSMC should have the slightly denser process for mobile applications processors.
Given that TSMC should have a process that's a bit denser than Intel's, coupled with the fact that TSMC is quite experienced at high volume foundry work, it's hard to see -- barring some major materials innovation from Intel at 10nm that leads to a substantial performance boost from its 14-nanometer node -- why Apple would use Intel for a 2018 iPhone.
