Processor Selection for SoC

Started by moogyd September 17, 2009
>moogyd wrote: >> Hi, >> >> I am looking at selecting a processor for our SoC platform. Currently >> we use an embedded 8051 core (small, low power, low performance, >> cheap). >> >> For our next project, we need more performance, and we are also trying >> to create a platform suitable for all future projects. >> >> The CPU must be available as RTL (VHDL or Verilog) source. >> >> Obviously, there are lots of options >> - Faster 8051 >> - 16-bit ? >> - 32-bit RISC (ARM Cortext M0, ARC 6XXX, OpenRISC) >> >> There are also lots of issues to consider >> - Power uW/MHz >> - Area >> - Performance >> - Cost/Licensing >> - Support and tools >> - >> Can anyone point to any data (comparisons) that would be useful in >> making a decision. >> >> Thanks, >> >> Steven >> >> >> > >Hi Steven, > >I work for ARM so my choice is certainly bias towards ARM cores. >But the following information might be useful for you. > >Faster 8051 >8051 IP cores are cheap, but as you can see the performance is limiting. >The fastest 8051 IP claimed their Dhrystone DMIPS is 0.1/MHz (but many >others are much slower). But you need to be careful as this often >require special compiler support or C libraries that make use of the >extra features. If using standard 8051 instruction set and features you >would get a lower performance. > >Due to the lower performance you might end up clocking the core faster >and hence getting higher power consumption. And in highend 8051 cores, >the register banks are implemented as D-flip-flop rather than SRAM (32 >8-bit registers, to allows single cycle execution of instructions). As a >result the gate count of the high-end 8051 can be quite large. > > > >16-bit >There are not too many commerical 16-bit processors. Most of them are >proprietary architecture and therefor the choice of compiler tools are >limited. The performance is several times better than 8051 but mostly >still less than half of ARM processors. Most of the 16-bit processors >has a Dhrystone DMIPS from 0.3/MHz to 0.5/MHz. > >The biggest problem you will find with 16-bit core is the 64kbyte memory >limitation. Some 16-bit architectures has workaround for this by allow >paging or segmentation of memory map but it will reduce the efficiency. > If your application requires more than 64kbytes or will expand in the >future, switching to ARM would be a better choice. > > >ARM Cortex-M0 >High performance at small size : Dhrystone DMIPS 0.9/MHz. Smaller than >most 16-bit cores and very good code density (smaller code size than >16-bit cores and 8-bit cores). >http://www.electronicsweekly.com/Articles/2009/06/10/46252/cortex-m0-used-in-low-power-touch-controller.htm > > >There are large number of choices for C compilers and debug tools, and >certainly future proof (e.g. tools, memory expansion). The processor >comes with debug features and an integration kit is included to help
design.
> >More details of Cortex-M0 can be found here: >http://www.arm.com/products/CPUs/ARM-Cortex-M0.html >http://www.arm.com/miscPDFs/24481.pdf >http://www.techonline.com/learning/webinar/215801022 > >If you want more technical details feel free to contact us directly or >let me know. You can try out Cortex-M0 using our Microcontroller >Prototyping System (MPS) without any NDA (details on >http://www.keil.com/mps/) > >regards, >Joseph >
And don't count the ARC 600 family out as well. - Higher performance than an an MO at the same same size (1.2 DMIPS/MHz) - as low as 12,800 gates - very low power - 0.0012 mW/Mhz on TSMC 90G - Robust and mature tools support including an overlay manager to take advantage of limited local memory - Code density on par with an 8051 due to it's 16/32 bit ISA Drop me a line if you'd like more information. Andy --------------------------------------- This message was sent using the comp.arch.embedded web interface on http://www.EmbeddedRelated.com