I was not aware, that there is a vendor for configurable embedded FPGA fabrics.
I think this is great, for several reasions.
For academic use where all this "what if" games are great to play with.
We've seen the Virtex2pro and now the Zinq idea of a FPGA with extensible processor (adding special purpose instructions etc.) which gives great flexiblity and potential performance boost. But it is expensive and a lot of people don't need gib FPGAs, they would prefere a great and fast microprocessor with some FPGA part.
In a lot of current designs there are only a few big digital chips in it (the rest is mixed signal or only analog). But there is still need for some glue logic and special requirements. Would be nice to include that part also to the main processor (where all the data ends anyways).
For us FPGA designers, there will be more work and fun for us :-)
@ChristophZ: In a lot of current designs there are only a few big digital chips in it (the rest is mixed signal or only analog). But there is still need for some glue logic and special requirements.
Very (very) often, an SoC is accompanied by an FPGA that can be used to implement changes or problems or things that were left out of the original spec -- it really makes sense to me to include one or more chunks of FPGA fabric in the middle of the SoC.
ChristophZ wrote: We've seen the Virtex-II Pro and now Zinq ... which gives great flexiblity and potential performance boost. But it is expensive...
We took a look at Virtex-II Pro back when it was fairly new. Technologically it was a good fit for our product line, because we were using PowerPC CPUs. However, Virtex-II Pro was way too expensive, with the cheapest chips over US$100 IIRC. So instead we went with an IBM (now AMCC / Applied Micro) PPC405EP for about US$10 plus a Spartan-IIE for about US$15, using a 32-bit PCI bus to communicate between them. Very nice solution, since both chips used +3.3V and +1.8V power.
I've been watching Zynq for a while now, still waiting for the price to come down. The Xcell Journal article from 2Q2011 says "With a starting point below $15..." I'm waiting for reality to catch up with marketing. Meanwhile, Spartan-3A and Spartan-6 are nice architectures.
ChristophZ also wrote: ... and a lot of people don't need big FPGAs...
I hear you. I would like a Spartan-6 LXT9 or LXT16 with PCI express, but the smallest Spartan-6 with PCIe is the LXT25.
Hi Max. In order to make the diagram more complete (last branch to the right), check us out at www.adicsys.com. ADICSYS has been designing and distributing embedded FPGAs and customized solutions implying generic RTL programmability for a few years now, not to mention our past FPGA activities. It is interesting to hear that Menta seems successful. Someone once told me that when there is competition, there are opportunities. So I take it as a good sign! Our approach is rather different and I do have a few questions about the pertinence of the described process, but this is probably not the place to start a strategic discussion ;-) . In the end, users will see more FPGA features of one kind or another on a wider variety of chips.
@Peer: In order to make the diagram more complete (last branch to the right), check us out at www.adicsys.com. ADICSYS has been designing and distributing embedded FPGAs and customized solutions implying generic RTL programmability for a few years now, not to mention our past FPGA activities.
Thanks for this info -- why haven't you contacted me before? LOL
Email me at email@example.com and maybe we can set up a briefing where to bring me up to date with everything you are doing so I can write a column and spread the word :-)
I agree in that embedding FPGA as a companion peripheral inside a SoC is going to be one of the future trends to follow.
As a quick example, I can imagine the use of this technology in order to implement flexible industrial field bus interfaces. TI is already embedding a kind of programmable state machine real-time units is some of its SoCs (e.g. the Sitara AM335X family) that can be used to implement PROFIBUS, EtherCAT, PROFINET... I'm sure that a little FPGA core could be a killer-app in this arena!
What looks scary to to me is that Xilinx is using a different parser for synthesis (Verific) and simulation (old-Modelsim), or am I wrong again so that I can sleep well at night, again ;-)
Certainly you have to simplify the diagram, but what I miss is a third branch of programmability, which is the multicore (or multiprocessor) domain. You can run complex algorithms down to timing critical virtual peripherals on such a processor arrays. I was wondering how the market will evolve for IP providers (such as Menta) for such a domain.