In the previous three blogs (Parts 1, 2 and 3) I've outlined the background and key decisions involved in the development and implementation of the Elba testchip. Now we'll look at the final steps taken to bring Elba to life.
As our understanding of the various components broadened, the actual SoC architecture design activity then knew the details it needed in order finalize the design. We knew we had two Cortex-A9 macros, a Mali GPU and the various other components, but not too much about how we would best plug them together. Since we also wanted to investigate system level power management, many of the large system components were also placed into their own independent power domains. The layout of the design also became rather interesting. Unlike most of our testchips, this design was area-limited and we needed to make simplifications/modifications in order to fit the target size without breaking the initial design goals and device characterization. This last step proved to be one of the most time-consuming, and brought about various new design requirements to future projects within ARM.
What we ended up with was an SoC that had multiple synchronous and asynchronous clock domains, with 40 definitions (84 across CTS), with 26 asynchronous groups; four independent islands of voltage supply (and measurement) in addition to the various domains setup within the processor for power management; plus various temperature and silicon process monitors -- and one very complex job of ensuring appropriate SoC power management. Ironically, the solution to this power management task was one area in which we simplified the Elba design. Rather than place yet another processor on the SoC, we decided that an external Cortex-M3 processor-based microcontroller was by far the easiest option. It would also allow the software folk to start their task before the SoC was available since we could simulate the various inputs to the microcontroller externally.Maybe a future article that describes the various aspects of the Elba power management might be interesting -- let us know. ConclusionSo, what has been the conclusion of the Elba Program? I guess one that could easily be forgotten is that ARM has rather a lot of big-brain engineer types that work best when they are given rather big problems. Elba has become in ARM one of those programs that has not only changed both the way folk in the ARM world think, but also how the bigger world sees ARM. We are no longer simply that company that no one has heard of doing something to do with mobile phones, but a company that has a place to supply the necessary technology across the full richness of computing.The Cortex-A9 hard macro has been delivered to various companies, with the first public 2GHz laptops having been demonstrated by NuFront, who have big goals to supply low power, high performance personal computers. The PoP and multichannel libraries have also allowed other design configurations, and together, you, the consumer are starting to see multiple-GHz, high performance ARM processor-based compute platforms that can still operate all day on a small mobile battery.Links to Parts 1-3 of this blog series: