Swapping a defective core on a reliable network-on-chip (NoC) system requires swift access to an extra core. But how should an efficient fault-tolerant NoC algorithm for flexibly placing a spare core in a mesh network be designed? Bhanu et al. present two alternative techniques for flexibly positioning a replacement core in a mesh-based NoC system.

The authors present integer linear programming (ILP) and particle swarm optimization (PSO) strategies for accurately placing a spare core in a NoC system. Specifically, a directed graph is used to explicate the standby core location problem. In the directed core graph of an application, a vertex contains a core, an edge indicates the communication between a pair of cores, and the communication cost between each pair of cores is the bandwidth requirement. The ILP algorithm is proposed for pinpointing the unsurpassed location for the fallback core in a mesh network. Assuming the availability of a core for each router, a PSO technique is also proposed for positioning the extra core in the mesh network.

In experiments, the ILP algorithm was used to minimize the communication cost among pairs of cores, contingent upon the communication paths between active/failed cores and the routers. Unfortunately, the exact simulation results were derived for small NoCs due to the limited memory and central processing unit (CPU) of the system used to investigate the spare core placement issues. Consequently, PSO was explored to locate the positions of fault-tolerant cores in large applications. The experimental results reveal considerable reductions in communication costs, network access delays, and power consumption.

Clearly, the authors articulate and present unique perspectives into discrete PSO applications for designing fault-tolerant NoC systems. The experimental results compare favorably with the evidential results of NoC studies in the literature. However, the system with limited CPU and memory used for the experimental simulations might limit the generality of the results. I call on network practitioners to read this paper and recommend adequate platforms for future comprehensive NoC system studies.