Some clarification: The ARM/Xilinx agreement isnt exclusive because that's against ARM's business model of being agnostic. Moreover, they've had a strategic relationship with Actel, who has a low power FPGA with a soft core M1 in the market (called Fusion) and more than a year ago, they announced an agreement for the Cortex M3, so you see where this is heading...... The ARM-embedded SOC market is getting more and more interesting. First Actel with Fusion, then Cypress with PSoC and now Xilinx. (Only Altera has decided to go with MIPS). All targeting different applications and markets, but it's definitely the new course for FPGAs looking for new growth areas.
Thanks for reminding me of Triscend (I wrote about them often enough, but how quickly we forget).
And given that ARM was interested in buying and Xilinx elephanted in and snapped Triscend up by trumping ARM's best offer, we can see that the processor-plus-FPGA debate has been rumbling on for quite awhile (see Xilinx sees the logic of Triscend acquisition published in March 2004).
Back then it was about building FPGA-enhanced flexible microcontrollers around a processor core and clearly at that time Xilinx did not want that perogative slipping out to ARM. It would have been a strange move by ARM putting it into potential competition with some of its licensees.
Now the mood seems more cooperative and about welcoming Xilinx into the ARM ecosystem while kicking the PowerPC to the curb.
Tom's comment brings another question to mind: what is the availability of IP for AMBA platforms versus IBM CoreConnect, or MIPS EC, etc.? (Bus bridges, or lack of, aside; that could be another issue!) Certainly for Xilinx, there is a substantial IP library for IBM CC in their EDK.
Xilinx has had Power PC hard-core processors for years in the Virtext II Pro, Virtex 4 and Virtex 5 families, they also use a PowerPC based bus standard to hook up IP. It looks like they have concluded that PowerPC is losing in the processor shake-out and ARM is now the partner of choice. The soft core processors are nice but they are not fast enough for a lot of applications, better memory architectures are also needed.
Xilinx is the largest player in FPGA so ARM has gone with them and left Altera with MIPS.
The interesting FPGA startup in this context is Triscend rather than Elixent or XMOS. Triscend was founded by people who came from Xilinx, it had an FPGA+ARM based microcontroller and it eventually got bought by Xilinx after a proposed sale to ARM failed.
You may well be right. And Altera has its Nios soft processor and Xilinx has its MicroBlaze soft processor. So there is plenty of choice about how software programmability gets added to FPGA fabric. But what the addition here is that Xilinx is taking physical IP to help it implement Cortex cores efficiently and ARM has agreed to co-develop an important piece of peripheral intellectual property (IP) - the next iteration of the AMBA on-chip bus.
Can we see a sea of ARM processor cores linked by a grid of AMBA highways surrounds by a halo of FPGA-fabric? And how do you make the software- and hardware programmability of the ensemble make sense?
By the way, doesn't Altera offer a line of FPGA with hard ARM processors in its fabric? And MIPS was featured in Quicklogic? If so, an interesting change or rather, addition of partners by the FPGA top two ...