So what has changed?
For one thing, the cost of SoC development has gone up exponentially and, therefore, so too has the volume threshold to justify an SoC. In response FPGAs are taking over the old semi-custom market and starting to be the deployment-of-choice in lower-volume, higher-value applications. But many of those applications require a processor or even multiprocessor and a power efficient one at that.
So it may be that ARM and Xilinx believe that the time of the multiprocessor-plus-FPGA is coming soon and they want to create standard ICs that are nonetheless software- and hardware-programmable and scalable.
It may be that this ARM-Xilinx announcement is a response to the stake thrust into the ground by XMOS Semiconductor Ltd. (Bristol, England) with its software-defined silicon?
Another way to think of this is that the Xilinx architecture has been based on the look-up-table (LUT) for its 25 years of existence and, to be frank, it is too fine-grained for some of the things system architects now want to do. It is the processor core that is on its way to becoming the NAND-gate or LUT of the next generation of field-programmable systems.
It is interesting that in some sense the ARM-Xilinx announcement parallels a recent one from processor IP licensor MIPS Technologies Inc. and programmable logic vendor Altera Corp. But what makes this latest deal look strategic is that both ARM and Xilinx are market leaders in their respective fields. Now if only we had a bit more detail.
Related links and articles:
ARM success could trigger takeover bid, says analyst
Commentary: ARM's Osprey broadens the battle front with Intel
CEO Interview: Moshe Gavrielov of Xilinx
Sizing up the FPGA contenders
FPGA debug and verification productivity