Internet in a chip

The MIT researchers' new 36-core chip is "tiled," meaning that it simply repeats the same circuit layout 36 times. Tiling makes multicore chips much easier to design. Credit: MIT

It is now at least ten years that chip manufacturers are trying to increase the processing power of chips by multiplying the cores, each one a processor in itself. This allows for parallel operations and a theoretical multiplication of processing capacity. It is a good approach since you can multiply the capacity without increasing hte clock speed and hence the power required...

The problem however is that in order to have parallel processing you need to have a problem that can be subdivided in parts that can be processed in parallel. And you need to have a practical way to program such parallelism in a manageable way.

There is also an additional issue, and that is to find a way to connect the various cores on the chip in such a way that they can operate independently of one another, when possibile and in cooperation when needed.

So far communications among the cores is ensured via a common bus connection all of them. The problem with this bus is that as the number of cores grow the contention on the bus grows too (since the bus has to be reserved by one core when it needs to communicate). 

Now researchers at MIT have revealed a chip with 36 cores interconnected via a miniaturised Internet that solve the contention issue as well as related cashing problems.

This is interesting also at a global architectural level since one could imagine that cores are interconnected among them in the chip through this mini internet and can be interconnected with other cores in other chips, in the same board or in boards on the other side of the Earth still using Internet and being programmed with the same software paradigm.

I can see in the future a super massive distributed processing architecture taking advantage of whatever processing capacity exist, transforming processing points into a processing fabric that will contain storage capacity and communications capacity, whre the local processing will be as relevant as the global processing. We have got a good example of this architecture and we take it around every day with us: our brain.

Author - Roberto Saracco

© 2010-2018 EIT Digital IVZW. All rights reserved. Legal notice. Privacy Policy.