The losers met various fates. The corpse of Advanced Switching Interconnect is left in public view, dead from a lack of silicon support. StarFabric is zombified, with prospects for new designs dim. InfiniBand and HyperTransport have been banished – although both are quite successful in other venues, they didn’t fit this space.
The three winners are Ethernet, PCI Express, and RapidIO.
Ethernet is ubiquitous, and continues to advance in speed as 10GigE develops. It’s difficult to bet against – in fact, the question was posed by an audience member that if Ethernet keeps winning, why are other interconnects still being considered? Ethernet is a great local-area and wide-area network, good for board-to-board, but not at all useful in chip-to-chip interconnect.
PCI Express is also enjoying new-found ubiquity with native Intel processor support. Its strength is as a chip-to-board I/O path, fanning out from a single processor to many I/O cards easily. PLX Technology conducted a workshop on PCI Express showing it as a versatile tool, with transparent and non-transparent bridging examples illustrating how embedded designers can create configurations to solve I/O problems. Cornet Technology was showing PCI Express on their boards. And P. A. Semi previewed support for PCI Express on their new PWRficient processor.
RapidIO has become the third clear option for board designers, with strength in chip-to-chip and board-to-board interconnects. Tom Cox, executive director of the RapidIO Trade Association, is signaling for RapidIO silicon support to double this year, with an already impressive chart showing devices from Freescale, Tundra, TI, IDT, PMC Sierra, Mercury Computer Systems, Xilinx, Altera, and Lattice. Curtiss Wright demonstrated their new CHAMP-AV6 and VPX6-185 boards with RapidIO paths transferring data.
Interestingly, each of these winning interconnects has formally recognized the others – they now show almost identical use case charts. With peace among them and critical mass for each established, maybe the industry can finally stop wasting energy on this debate … but probably not.