Open ARM-wrestling in FPGAs
Why has the embedded RISC market in FPGAs been limited to ARM cores? That question occupied me for several hours after hearing about Altera’s licensing of the MIPS core Oct. 6. After all, embedded RISC instruction sets have maintained a diversity over close to 20 years in the ASSP and standalone-processor sectors of the embedded market.
Yes, there have been notable dropouts in embedded systems. Sparc, with a few exceptions, never hit it big there. But PowerPC, ColdFire, and MIPS cores have kept up a strong front, despite the overarching presence of ARM in most vertical sectors.
In the past, ARM had a few legitimate points to make about code compactness and efficiency of gate utilization. But as we move to multicore versions of Cortex, that argument carries less weight. And FPGA architectures have gotten large enough to allow for more experimentation in control-plane architectures.
Let’s make a core assumption to begin with: Dylan McGrath of EE Times raised the question of whether Altera was developing its own core based on the MIPS ISA. What other purpose would the licensing deal have? Altera almost certainly would not enter the standalone CPU business. A hard macro remains a possibility, but the licensing pact virtually implied soft core development.
While many choices of RISC cores are made for arbitrary reasons that have little do with thread efficiency or compactness of code, there is a definite school of thought that says MIPS offers a more register-rich architecture that is preferable for applications with bigger limitations in memory bandwidth, while ARM offers a clever use of arithmetic instructions that could prove better in array and vector problems. Merely by making this observation, I have no doubt offended some advocates of either camp, but it’s interesting to note that ARM-based FPGAs have been displacing some DSP designs of late. Perhaps the addition of MIPS will aid Altera in certain fast datapath designs for networked storage, mobile devices, and the like.
Maybe this was just an accident of timing, but it’s interesting to note that ARM announced a deal with GlobalFoundries, a day after the MIPS-Altera announcement, giving ARM access to 28-nm processes. Oh boy, the race is on.
In any event, the licensing decision means diversity of instruction sets. It might behoove FPGA companies to examine ColdFire and PPC as well, though a portfolio that is too broad could be as vexing as an ARM-only world.
Buy Cialis commented:
desert rat commented:
Loring commented:
Anonymous Coward commented:
desert rat commented:
Loring Wirbel commented:
Andy T commented:
desert rat commented:
