In early August, Intel (INTC 0.64%) announced key details of its 14-nanometer chip manufacturing technology. The company said one primary feature of its 14-nanometer process was a significant lead over its peers in density, or how many transistors (the building blocks of a chip) could be crammed into a given area.
However, as Daniel Nenni from SemiWiki pointed out, the Apple (AAPL 1.27%) A8 -- which is built on Taiwan Semiconductor's (TSM -0.34%) 20-nanometer manufacturing process -- actually offers greater transistor density than the recently announced 14-nanometer Core M processor from Intel.
This led Robert Maire from Semiconductor Advisors to raise the possibility that Intel's "basic design and architectural capability" might be lacking.
In this article, I offer a different, less sensationalist, explanation of what might be going on.
The funny thing about density
According to Apple, the A8 packs in about 2 billion transistors in an area of roughly 89 square millimeters. The Core M, according to Intel, packs 1.3 billion transistors into 82 square millimeters.
It doesn't take a mathematics degree to work out that the Apple chip is, indeed, denser.
However, the funny thing about trying to compare the capabilities of a given manufacturing process using two (entirely different) chips as proxies is that this is not an apples-to-apples comparison.
Let's go into some more detail.
Taking a closer look
At Intel's investor meeting in 2013, it presented a slide discussing how chip designers can use different "metal stacks" in order to achieve certain performance and density targets. The idea is that a chip designer can, on a given process, optimize for high density at the expense of performance, high performance at the expense of density, or meet at some point in the middle.
The Intel Core M is a fairly high performance design, with two "big" CPU cores that clock at up to 2.6 GHz, and includes a sizable graphics processor that clocks up to 850 MHz. The A8, on the other hand, includes two low-clocked CPUs, likely a low-clocked graphics processor (if the A7's graphics processor is anything to go by), and a bunch of other blocks that probably don't require super high speed transistors.
Also keep in mind that Core M is likely to command a high selling price (Intel's official pricing is $281 in quantities of 1,000), while the A8 needs to be inexpensive, which likely drives many of the trade-offs with respect to performance, power, and density.
In other words, comparing the transistor counts of the A8 and the Core M and looking at the density irrespective of performance, clock speed targets, and the other blocks found on the chip is not likely to yield a very useful comparison of the underlying processes.
What about all of those density claims?
Realistically, when both Intel and the foundries compare "densities," the contrast is a bit misleading because, as discussed above, it depends on the kind of chip a company is designing.
For a better comparison of what Intel's "densest" chip looks like and what a foundry 20-nanometer/16-nanometer/14-nanometer chip can do, it is probably best to wait for Intel to launch a smartphone or tablet oriented low-power processor. Intel is expected to launch its 14-nanometer Atom processor known as "Cherry Trail" late this year or early next year; if the company discloses a die-size and transistor count for that chip, this will probably more accurately reflect any "density comparison."
But, even then, without comparing the exact same chips built on the respective processes, it's hard to come to a conclusion about the densities of the underlying processes or the relative skills of the design teams at Apple and Intel.
