Citrine: A Methodology for Application-Specific Network-on-Chips Design

This paper presents a novel and efficient mapping algorithm, called Citrine. It generates a set of mappings which have nearly minimum possible distance between related cores results in reduction of communication cost in Network-on-Chip (NoC), while improving other performance metrics. Comparison of the communication cost results makes it obvious that final solutions found by our proposed approach outperform the results of other methods which proposed in literature. Vulnerability index is introduced and defined as a criterion for evaluating routing fault tolerance. Reducing the vulnerability index optimizes fault tolerant properties of NoC. Then by using the Citrine solutions, a total cost function is used to customize and prioritize the impact of metrics, i.e. communication cost and vulnerability index. The experimental results reveal the power of Citrine in generating different solutions with low bandwidth requirements that spread through the explored design space.