Article Preview
Top1. Introduction
To meet the power and performance requirements of single processor architectures, recent advances have moved towards the development of multi-core systems on chip. These systems, which are formed by integrating many intellectual property blocks on a single chip, have performance constraints in terms of on chip communication. However, some of the initial architecture of multi-processor systems on chip (MPSoC) include bus-based framework as on chip communication, but this is not the scalable solution if the number of cores on chip increases. This issue has been addressed by the Network-on-Chip, which not only resolves wiring scalability, but also deal with two other major performance bottleneck that is, communication latency and bandwidth. The networks on chip is a packet switched mechanism in which different cores communicate with the help of short interconnects between them as well as by using multiple packets flow over these interconnects. This type of wireless transmission in NoC infrastructure is possible by the use of routers at each node. For an on chip network the characteristics like topology, routing algorithm, flow control, and router micro architecture, are the key factors in identifying the overall performance of the network. The topology defines the connection between nodes and links so as to establish all possible paths between nodes. The routing algorithm then selects one particular path from source to destination. The flow control policy gives the efficient resource utilization by deciding how packets are transmitted in the network. Depending on routing and flow control policy, router micro architecture outlines its network implementation. For designing an on-chip networks, selecting an efficient and deadlock free routing policy has always been the major concern. The routing algorithm chooses the path that each packet will follow to reach the destination. It can be classified as deterministic, oblivious or adaptive, depending on path establish for packet transmission. In deterministic algorithm, the path is governed by the source and destination, irrespective of current network traffic. Although the algorithm is simple and less complicated, it does not keep evenness across the links when the traffic is not uniform or is bursty. In case of oblivious routing there are many routes possible between the source and destination and the state of the network is not considered while selecting the route to be followed. In the adaptive routing algorithm, the nodes determine the path between source and destination and route is selected by considering the congestion status of the network (Dally & Towles, 2004). As network status is involved in choosing a path, the possibility of getting into a faulty or a congested region is very less. This algorithm is further divided into two functions as shown in Figure 1.
Figure 1. Routing and selection function of a routing algorithm. The routing function evaluates possible output ports depending on destination address. Selection function based on network status selects a single output from the evaluated ports.
The routing function gives a set of possible output directions depending on the current node and the destination node. Based on the network status of these accepted output channels at the current node a selection is made by the selection function. This selection should be such so as to provide available free channel. In any case of availability of more than one output channel, a strategy is required to select one of them (Duato, 2002).