Chinese high end CPUs are now in the game – details: Part 3 – ARM and others
In this final part of the Chinese CPU development coverage, we look at the local ARM processor flavours, as well as China's own instruction set attempts aimed at the general market.
While MIPS and Alpha were at the forefront of RISC high end architecture development, the sole Europe-developed surviving instruction set architecture, ARM, was from the very start in 1985 aimed at the entry level – whether it was the BBC micro home computer successor then, or the myriad of smartphones and netbooks today. The Chinese have embraced ARM architecture as well for this part of the market, with several licenses up to now. These cover the full spectrum of consumer devices, from smartphones and tablets to netbooks, DTV settop boxes and car gadgets.
The Fuzhou-based RockChip offers Cortex A8-based custom ARM CPUs and SoC chips for personal entertainment devices. Their newest RK29xx is the first chip to decode Google's WebM VP8 in hardware. The 1.2 GHz CPU with 512 KB L2 cache also has an integrated 60 million polygons/s GPU as well as DSP-accelerated 1080p playback and encoding in most formats. It supports tablets and smartphones with up to 1280×800 displays. A dual-core version is supposedly under development as well.
The Hangzhou-based NationalChip licensed the ARM over 3 years ago, with specific focus on derivatives for digital entertainment, mainly digital TV sets and set-top boxes. Considered as one of top ten Chinese IC design companies by EETimes China, the company offers GX1100, 1200, 1500, and 3000 families of integrated SoC-approach components for digital entertainment.
Then, the Shanghai-based Leadcore Technology, the chip design arm of Chinese communications equipment company Datang Group, is working on custom ARM processors based on the Cortex-A9 MPCore, the ARM Mali-400 MP graphics core and Cortex-A9 optimization pack for the TSMC 40 nm low power process technology. Their focus is putting together uni and dual-core versions of such chips with its own baseband chip to target high-end smartphones based on the China's 3G standard, TD-SCDMA.
Another Shanghai company, Brite Semiconductor Corp., a fabless startup founded in 2008, has licensed most major ARM processor cores, including Cortex, ARM9, ARM11 and Mali on a long term arrangement. The license also covers Coresight debug and trace technology and peripherals that are compliant with the AMBA on-chip bus. Brite provides design services to electronics companies and works with SMIC, the local foundry, on the manufacturing side. They have already successfully output 40 nm chips from this foundry earlier this year.
Yet another company from 'New New York' of Asia, Shanghai InfoTM Micro-electronics, has licences the ARM11 processor core, Cortex-A5 and Cortex-A9 processor cores and the Mali300 and Mali400 GPUs for 3-D enabled mobile computing devices to be manufactured by Shuoying Digital Science & Technology (China) Co. Ltd. which is its both owner and main customer. They also have multicore ICs ready as of now.
In Zhuhai, AllWinner, focusing on HD media semiconductors, took the ARM Cortex-A8 processor and the Mali-400 MP GPU for their own HD-enabed procssors to be used with a range of Android OS-based tablets, smart TVs, personal media players, eBooks, smart media boxes, IP cameras and automotive multimedia gadgets. The Allwinner Technology SoC designs are available since this past summer.
Finally, we look at the ultimate approach – designing your own instruction set from ground up, a venture few dare to try, especially these days since X86 is pre dominant for the past decade across the board. ICube, a Shenzhen company, created the Harmony Unified Processor Technology, which is supposed to tightly integrate two different processor types, CPU and GPU, into one unified core – sounds somewhat like AMD Fusion approach, but with a fresh instruction set optimised from scratch for the purpose. This technology consists of the Multi-Thread Virtual Pipeline parallel computing core (MVP), an independent instruction set architecture (ISA), an optimizing compiler and the Agile Switch dynamic load balancer.
Even though these are big-named things reminiscent of what you see in servers, ICube's technology is actually used in small SoC solutions for the hand-held computing and communication market, with a focus on the Android OS. The initial product, ICube IC1, is a 600 MHz dual core 32-bit SoC with 8 threads (4 per core) in parallel and 5160 DMIPs declared throughput, a 70 million polygon/s, 600 Mpixel/s GPU, and a host of integrated features such as FullHD display driver up to 1920×1200 with HDMI/DVI, a camera interface, 720p video acceleration, 5.1 audio, memory card, USB, 3G and Wifi connectivity.
What's interesting here is not only the fine grained CPU multithreading with OpenMP and Pthread (both used in HPC and general SMP apps a lot) support , but also the GPU support for Data parallel, Task parallel, and Function parallel computing with minimised interrupt and context switch overhead due to multithreading, and heterogeneous GPGPU applications with both OpenGL ES2.0 and OpenCL support. Each core has 64KB I-Cache , 64KB D-Cache, 64KB SRAM and 32-bit GPR file, 8-channel DMA and 16-source interrupt controller. Each core only takes 3.0 mm2, including memory, with operating power of about 300mw.
The built in support both homogeneous (OpenMP and such) and heterogeneous (OpenCL and such) parallel programming APIs through native compiler and MVP drivers is quite a good news here, as a new ISA needs the easiest possible programming enablement to ensure software support.
In summary, China is covering the ground well at the mainstream level as well, ensuring a well varied supply of CPUs for all classes of consumer devices, having ARM compatibility yet local cost, design and manufacturing control. At the same time, going for its own instruction sets is the next frontier.
Images Credits to Products of Respective Vendors