Will XMOS displace low cost FPGAs?

[infiniteimprobability]

Building complete systems from software running on threads is certainly a new way of doing things. The ability to create not only processing and control blocks but interfaces too, means we have new type of programmable device to choose from. That’s something that doesn’t happen too often in our industry.

There is, of course, a technology that has been around for over 25 years that lets you put different blocks together and build a system – the FPGA. The introduction of the soft processor around 10 years ago allows software to be run these devices too.

So how do they compare? Clearly if you are designing a 160Gig router for core networks or a 3G cellular base station with massive custom DSP pipelines using programmable devices then the FPGA is the only show in town.

However if you are building more modest systems with microcontroller style peripherals, medium speed DSP ( < 100MMACs) and a few 100MIPS of processing, then an XMOS chip will do the job just as well. If power, cost and the requirement to design using only a high level language are important, then you could argue that XMOS is a better choice.

Well that’s what we think and we’ve released a white paper comparing the two. http://bit.ly/fpgavsxmos
It’s written by a hardware engineer who designs chips, from a hardware engineer’s viewpoint.

What do you think though? When does XMOS make sense and when do you dive into RTL?

[leon_heller]

I mentioned the paper on the comp.arch.fpga newsgroup. Only one response from someone, but he is thinking of using XMOS for a new project instead of an FPGA and had a couple of questions which I answered. I've mentioned XMOS there before, but no one seemed interested.

A friend of mine has ordered some XMOS kit where he works. He tried to get some colleagues interested, but they want to use some very expensive "real-time" Java boards ($700 each). The manufacturer boasts of sub-microsecond response times to events!

[shawn]

Find the best mix is where it is at, XMOS even as a softcore in a Fpga
would more likely be more effective than even the best extensible classic
Von Neumann type. Fpga's are 80 percent routers , they like links!

The price, scale, speed, in midsized Fpga's is the sweet spot.
XMOS technologies should handle all other tasks elegantly.
Then there are some nice arrays we may attach or exotic technologies we might attach.
Xmos provides a very predictable means to Manage all this abundant silicon. The PPC, the ARM
the two primary co-host for Fpga's technically are subservient to the events creating more and more heat, like the traveling salesman problem just to manage evermore silicon. XMOS is a SOC with out the help of anyone. If I took 9 L2's to 5 million system gates or 9 G4's to 10 million system gates, the links are proportional to the gates more so than a classic processors, Also you have a nice price parity.

All fpga are blanks with very little Hard-IP they differentiate them self's primarily with there licensee's and tools, that's their real revenue. XMOS is mostly Hard-IP great licences and tools a real symbiosis to Fpga's especially the exponentially expensive biggest fastest exotic Fpga's...
Being able to manage virtual silicon effectively will directly place XMOS in the consumer market displacing most contemporary design.


PS Simple answer " Absolutely!"