ARM Community: Think Globally, Act Locally: Why coherency matters to Smart Mobile Devices - ARM Community

What does this well known phrase have to do with chip design for Smart Mobile Devices? In the future, everything – let me explain. There was certainly a lot of press when we launched our Cortex-A15 processor and in my recent blog I highlighted what that device will mean for mobile devices of the future. But last week at ARM Technology Conference (ARM Techcon) we had another big technology announcement from our graphics processor family with the launch of the ARM Mali-T604 graphics processor, which delivers high-end performance and power efficiency for mobile graphics and visual computing.

So what do these two technologies have in common besides being both low-power and multi-core scalable? They play well together because they speak the same language and this is very important to designing complex SoCs.

Importance of power and interactions on ARM-based solutions
The power advantage of ARM-based solutions is not only low-power CPUs, is also due to the rest of the key components including graphics, video, audio, DSP, security and more being integrated on the same silicon die, working in harmony at close quarters. Counter-intuitive as it may seem, the ARM processor is usually only a small portion of the real-estate on a mobile phone chip. Thus how efficiently the rest of the system interacts with the ARM processor is a very important portion of overall battery life and user experience (see Jeff Chu’s blog from last year).

How these components interact is very important especially when they operate on common data which is often the case for mobile SoCs. Components need to know when to work on and when to wait for shared data – as it may be being worked on by another unit. In the real world this is not just speaking the same language, but actually understanding each other in the SoC world; this is considered “coherent” operation. Usually the point of coherency is established in external memory and is regulated by software. The introduction of hardware coherency is a big step to performance and power efficiency of next generation chips.

Analogy: compare travelling between cities to communication on a PC
Let us draw a geographical travel analogy for simplicity: A PCI card or motherboard being considered a continent and a chip being considered a city. With this basis, then in the PC world communication between units such as the Central Processing Unit (CPU) and the Graphics Processing Unit (GPU) actually happens across separate boards (each with its various chips), the communication between the two would be akin to trans-continental travel for each interaction. Imagine doing these by boat! Not only that, there would be multiple such travel transactions before delivering the final result. This is extremely energy and time inefficient.

Analogy: compare travelling between cities to communication on an ARM processor-based SoC
If instead you compare this to a current ARM processor-based systems-on-a-chip (SoC), the CPU and GPU would on the same chip and hence by analogy be in the same city, but the point of coherence – the memory on the same board (and in some cases soldered on top of the application processor itself) – is another city a few hundred miles away. The intra-city communication helps, but there are regular time-consuming inter-city trips (due to coherency requirements) before we arrive at the final result. The result is a great big improvement in time and energy efficiency over a PC system, but from our perspective there is still with some work to be done.

Mobile ARM system-based SoC: Improved efficiency and interaction
Now let’s look at a system where a Cortex-A15, a Mali-T604 graphics engine are connected together by a system-coherent AMBA 4 fabric component, such as CoreLink CCI-400 Cache Coherent Interconnect. This is analogous to having the two still in the same city but now connected by an efficient highway system and most interactions between them happen within “city limits” and there are few (if any) inter-city transactions before delivering the final result. The end result is a system that acts locally for improved efficiency and a much-reduced software overhead.



Not exactly your average mobile configuration… but just illustrating what could be possible

So as we look at the next generation of products that demand for higher-performance, we need to improve not just the efficiency of each component, but also the overall interaction across the SoC. So Smart Mobile Devices of the future will think globally but keep the critical actions within “city limits”.

Nandan Nayampally, Director of Apps Processor Products, ARM, He has long history with application processors having managed the ARM11 family, the Cortex-A8 before taking on the Cortex-A15 and is now looking at next-generation processors. Having been a keen observer of the accelerating trend towards connected and differentiated devices and how they affect our everyday lives, this reluctant blogger intends to pitch in with a processor-centric viewpoint now and then.