Many-Core: Finding the Best Multi-Processing Tile

Omar Hammami, Associate Professor, ENSTA ParisTech & Ludovic Larzul, Vice President of R&D, EVE
EETimes (8/29/2011 4:01 PM EDT)

Very few people design with transistors these days. Everyone creates systems that contain transistors, but your average designer, while obviously being aware of that fact, wouldn’t know much about the intimate details of the transistors being used, aside from sizing or ratios.

And that’s intentional. It’s the result of the inexorable wave of abstraction. Transistors became gates became cells became IP. And one particularly important IP block is the microprocessor.

Today you can purchase a microprocessor IP block, configure it, and commit it to silicon without having to worry about the underlying logic, not to mention the transistors implementing that logic. Depending on the processor you choose, you will automatically inherit an entire ecosystem of companion IP as well as software infrastructure for implementing your application. All thanks to abstraction.

But it’s got further to go. We’re already in the multicore world, and, following on its heels, is the “many-core” world, which is the same as multicore, but dives right in with lots and lots of cores instead of timidly sticking its toes in with just two or four.

Click here to read more ...