A few weeks back, I wrote about the Apple A5 processor [link] introduced along with the iPad2. (See “Peeling back the layers of the onion that’s Apple’s A5 processor for the iPad2”) At that time, there wasn’t much hard information about the detailed design of the processor. Now the Microprocessor Report’s Linley Gwennap has written a long article (subscription required) giving away a few more hardware details. Gwennap’s article also contains some interesting analysis. One key theme in Gwennap’s article is that Apple designs its own processors as a cost saving measure for its higher-priced consumer gear. He notes that application processor vendors need to charge higher margins than foundries to cover development, marketing, and sales costs so that system vendors like Apple can actually save money making high-volume consumer products with custom-designed application processors.
Gwennap bolsters that premise by pointing out that the hardware design of the Apple application processor series isn’t all that different—not significantly—from commercial offerings like the Tegra and OMAP application processors offered by Nvidia and TI. Gwennap’s position coincides well with the EDA360 perspective, which says that software is increasingly the key differentiator in the design of such products. Software represents roughly half of the development effort; software development consumes roughly half of the design budget; and complex, application-based software now defines far more of the user experience than does the hardware. The hardware must supply some basic abilities, such as hardware decoding of H.264 video, but then the software is what wraps the user interface like a skin around the bones of the hardware.
Even so, Gwennap points out, there are real benefits to designing your own silicon if you have the projected sales volumes to justify the design cost. You can better optimize the design to get exactly the balance between die area and performance you want. You can also select precisely the design IP you want to go with the exact software you have chosen or developed. But even taking all of that into consideration, writes Gwennap, certain consumer products such as the Apple iTouch media player do not require cutting-edge silicon design and existing application processors and chipsets, once they’ve ridden the price curve, may well be the best choice in such cases.
Coincidentally, Gwennap’s colleague Jag Bolaria has written an editorial appearing simultaneously with Gwennap’s Apple A5 article suggesting that Intel consider spinning off its fab capabilities into a separate company so that the new Intel fab behemoth can fully participate in the foundry business, which Gwennap writes will grow to $50 billion by 2015. Interesting idea. Didn’t AMD just sort of do that?