My first reaction was, “It's about time.” My second reaction was, “I hope they did it right.” Let me explain. Xilinx, considered by many to be the market leader for FPGAs, had a hole in its lineup, at least in my eyes. For at least a couple of years, I asked the folks at Xilinx why they weren't making a serious run at ARM-based FPGAs.
I learned that it wasn't as simple as dropping the core into the company's library. There were It actually took some design issues that needed to be overcome to ensure that the ARM core could operate at its maximum efficiency. Those changes were put in place last fall, when Xilinx announced a technology agreement with ARM.
Essentially, the technology agreement revolved around changes made to the AMBA bus to keep the programmable logic tightly coupled with the processor core. Xilinx adopted ARM physical IP, and the two companies made a technical commitment to work together to define the AMBA 4 specification, which is the de-facto industry standard for on-chip communications on SoCs designed with an ARM core.
With that technology in place, it was clear where Xilinx was headed. However, there were a few details on which Xilinx remained mum. As of today at the Embedded Systems Conference Silicon Valley, that silence is broken and all questions are being answered. They're calling it their Extensible Processing Platform that takes advantage of ARM's dual-core Cortex-A9 MPCore processors, each running at up to 800 MHz. With the platform, designers can apply a combination of serial and parallel processing for applications that require high-speed access to real-time inputs, high-performance processing, and/or complex digital signal processing.
Thanks to the changes made in the architecture, a software-centric development flow is enabled by the processor-centric approach which presents a full processor system. This includes caches, memory controllers, and commonly used connectivity and I/O peripherals. It's built using Xilinx's high-performance, low-power 28-nanometer technology.
The high-bandwidth AMBA-AXI interconnects keep the programmable logic tightly coupled with the processor core. This architectural approach addresses common performance bottlenecks between these parallel and serial computing environments, memory, and I/O. It also gives the processor control of the programmable logic, including dynamic reconfiguration.
The architecture abstracts a lot of the hardware burden from software developers, who can now tap into the vast off-the-shelf open-source and commercially available software component libraries. Another key feature is the FPGA's ability to boot an operating system (OS) at reset.
Pricing and availability will be announced for products based on the Extensible Processing Platform architecture in early 2011.