April 15, 2010 -- Within communications and networking chips, most of the general-purpose processors still operate in single or dual-core architectures, but across the semiconductor industry there is growing evidence that the number of processor cores operating in many-core architectures is duplicating every three years.

These cores, increasingly embedded in networking and multimedia products, are very often application-specific designs, tailored to the characteristics of their selected domains.

The principle behind the development of these application-specific processor cores has been that a simple architecture with a specialized instruction set can be largely more efficient than a complex architecture with a general-purpose instruction set. While this comparison can hold true on an instruction basis in isolation, it does not hold true within systems where the memories are shared among multiple clients and the communication is affected by long latencies. Under such limiting environments, more complex processor architectures are capable of delivering more performance.

In this article we argue that the evolution toward multicore systems requires some fundamental changes in the development of application-specific processor cores and we present the work that we have carried out on one application-specific processor to cope with these changes.

By Lorenzo Di Gregorio. (Di Gregorio is with Lantiq.)