Last week I published an article titled The day I designed my own mobile SoC. It was based on a presentation I had been working on for several months. The main goal was to map our IP portfolio onto next generation SoCs for a range of markets, including mobile.
Little did I know my article would enjoy such popularity and spark a heated debate. Among the feedback that I got was one particular question which got me back to the writing table. One of our readers asked what type of SoC I would put inside an affordable mobile phone.
Since a significant part of future growth in mobile is predicted to come from entry-level devices, I sat down and tried to imagine what I would include in this type of apps processor. After some discussions with my colleagues, I came up with the diagram below:
Like most people who work in PR, I regularly carry two phones with me: one for personal use and another for business use, one of which is usually an entry level device. I therefore tried to answer our reader’s question by thinking what latest generation, low cost technologies would replace what I already have in my pocket.
Selecting a CPU for general purpose processing
Choosing a CPU for an entry-level SoC can be a very challenging task. You want best in class performance efficiency but you also have to remember that cost plays a huge factor.
My preferred CPU configuration would include a cluster of four dual-threaded MIPS interAptiv processors. Since interAptiv is significantly more power- and area-efficient than competing designs, companies can deliver a significant performance boost to ultra-affordable devices in a similar power and area envelope to what is available today; you can read more about the latest interAptiv CPU we’ve recently announced here.
Additionally, each of the four CPUs supports hardware multithreading – a technology unique to the MIPS architecture among licensable CPUs which can significantly boost app performance. This means that four dual-threaded MIPS interAptiv CPUs are seen by Linux SMP-based operating systems (e.g. Android) as an octa-core apps processor.
Therefore, I am getting a better performing dual-threaded core in the same area and power as a competing single-core CPU. The diagrams below give you a numerical representation of the points made above. A single core, dual threaded interAptiv is only 0.08 larger than the competing single core CPU while a single core, single threaded version of interAptiv is 80% the size of CPU B:
Selecting a GPU for graphics and compute, VPUs for video decode and encode and an ISP for imaging
Providing the right mix of GPU performance and features is very important for entry-level devices. For my ultra-affordable mobile device, I’d have to go with a PowerVR G6110 GPU.
PowerVR G6110 packs all the latest features of the Rogue architecture, including 32 ALU cores and our area-optimized PVR3C triple compression technology. G6110 is a single-cluster GPU that includes our latest generation PVRIC2 technology; this creates an important competitive advantage for low-cost SoCs where memory bandwidth is a limited and valuable resource.
PowerVR G6110 is designed to support the latest graphics and compute API standards, including OpenGL ES 3.1, OpenCL 1.2 and DirectX 9_3, and delivers much better performance compared to a competing dual- or even quad-core solution.
A part of my work involves watching YouTube clips: interviews with key execs, the latest device reviews, live reports from tradeshows or conferences – and occasionally cats. Therefore having a low power video decoder inside my SoC is a top priority. Our PowerVR D5500 VPU can handle all of the commonly used online video standards (H.264, VP8 and MPEG-4) as well as the upcoming H.265 format. Most importantly, PowerVR D5500 is a highly- efficient, power sipping video decoder: it only needs 10mW to decode H.264HP 1080p at 30 fps.
When it comes to the camera, I am prepared to accept a 5 MPixel CMOS sensor. However, I still want those pictures or videos to look crisp and clear, albeit at a lower resolution. This is where integrating a PowerVR V2500 imaging processor can make all the difference. First of all, PowerVR V2500 supports 10-bit color depths so image quality is preserved between the camera and the PowerVR E4500 video encoder.
Secondly, PowerVR V2500 decreases power consumption and system costs since it is integrated on chip, at the same process node as the other engines.
Selecting a radio processing unit for combo connectivity
I cannot stress how important Wi-Fi access is to me; I always need my smartphone connected to the Internet. Adding a low area, power saving Ensigma C4521 combo RPU which supports high-speed 802.11ac, Bluetooth Classic, and digital/FM radio is the perfect way to cover my basic connectivity needs.
In this article I wanted to offer my perspective on what to expect from a next-generation, extremely power-efficient design. Even though making decisions for any mobile SoCs requires a deep understanding of processor design, optimal feature set and performance levels, I tried to make the selection process as transparent as possible and offer our readers the opportunity to get better acquainted with our products.
I hope you enjoyed this article and look forward to reading your thoughts on what you think of my entry-level mobile SoC. If you have any questions, please don’t hesitate to use the comment box below.