Apple-Samsung rivalry expands to 20nm chips

Editor's note: The rivalry between Apple and Samsung is said to have defined the global mobile market. Today, the companies are expanding the battlefield to a new front—the 20nm node. With their respective A8 and Exynos 5430 processors powering the soon-to-be-shipped iPhone 6 and Alpha smartphones, Tirias Research's Kevin Krewell looks at how Apple and Samsung both applied the silicon technology.

Samsung was expected to be the first supplier to ship a 20nm smartphone SoC with its Exynos 5430, but Apple's just-announced iPhone 6 and 6 Plus run on their own 20nm A8 processor and will be available for sale September 19. The Exynos 5430 will first appear in the Samsung Alpha smartphone that is also expected to ship in September. Which phone ships first is superfluous, but what is most interesting is that these companies used the new process node in different ways.

Samsung is using the 20nm shrink to reduce power and extend battery life, in part to compensate for the additional power required to support the LTE-Advanced modem in the Alpha handset. In Intel's "tick-tock" parlance, the 5430 would be a tick—a die shrink with no major architecture changes.

The 5430 design doesn't stray very far from previous 28nm parts such as the Exynos 5420 and 5422. The parts all offer the same Big.Little combination of four ARM Cortex-A15 and four Cortex-A7 CPUs, and the GPU appears to be similar. There are some new features in the 5430, such as a new audio processing section and an HEVC (H.265) decoder. However, most of Samsung's efforts seem to be focused on lowering power, rather than adding significantly more logic.

The Apple A8 appears to be the more ambitious chip. Though we haven't seen a chip teardown yet (that will come after the iPhone 6 ships on Friday), Apple says the chip contains 2 billion transistors—double the number in the A7 chip. Even with the doubling of the transistor count, the A8 die is 13 per cent smaller than the A7. The A7 die measures 102mm², so a 13 per cent shrink would put the A8 die area at 89mm².

The shrink from 28nm to 20nm obviously reduces the transistor size, but with double the transistor count, the A8 could have come closer to the same size as the A7. Apple likely improved the packing of its 64bit Cyclone CPU core, and some of the additional transistors could have come in the form of more cache memory, which is denser than logic used for CPUs and GPUs.

Apple devoted those extra billion transistors to improving CPU performance by up to 25 per cent and speeding up GPU performance by up to 50 per cent. At the same time, Apple is claiming power savings that allow it to sustain performance, while other processors start to thermally throttle performance.

Apple claims the battery life of the iPhone 6 is equal to or better than the A7-based iPhone 5S. The faster CPU performance is likely from a combination of improved CPU design, larger caches, and a slightly higher peak clock speed.

The GPU performance boost almost certainly comes from 50 per cent more GPU clusters. The A7 had an Imagination PowerVR Rogue GPU with four shader clusters. The 50 per cent performance increase can be directly attributed to a 50 per cent increase in the number of shader clusters. Apple also made some improvements to its image processing capability in the A8.

The availability of 20nm wafers is still spotty. Samsung has said its 20nm production is reserved for internal use, not foundry. It's not clear if the Apple A8 is manufactured by Samsung, TSMC, or both. Though both Apple and Samsung are using 20nm for their newest smartphone SoC, Qualcomm's own 20nm Snapdragon 810 and 808 chips won't ship until the first half of 2015.

Apple has chosen to use the extra billion transistors in the 20nm process to make significant improvements in the capabilities of its iPhone products, but Samsung seems focused on lower power and a smaller die—even skipping the transition to 64bit ARMv8 CPUs.

Because Samsung has a larger worldwide product footprint, it may be more concerned with reducing costs to compete with companies such as MediaTek and Qualcomm. Apple was more aggressive in adding performance and features to its die, but neither company made significant architectural changes to previous-generation products. More significant changes may have to wait until the 16nm/14nm FinFET node.

- Kevin Krewell
??Principal Analyst, Tirias Research