Impact of defect tolerance techniques on the criticality of a SRAM-based mesh of cluster FPGA

As device sizes shrink, circuits are increasingly prone to manufacturing defects. One of the future challenges is to find a way to use a maximum of defected manufactured circuits. One possible approach to this growing problem is to add redundancy to propose defect-tolerant architectures. But, hardware redundancy increases area. In this paper, we propose a method to determine the most critical elements in a SRAM-based Mesh of Clusters FPGA and different strategies to locally insert hardware redundancy. Depending on the criticality, using defect tolerance, area and timing metrics, five different strategies are evaluated on the Mesh of Clusters architecture. We show that using these techniques on a Mesh of Clusters architecture permits to tolerate 4 times more defects than classic hardware redundancy techniques applied on industrial mesh FPGA. With local strategies, we obtain a best trade off between the number of defects bypassed (37.95%), the FPGA area overhead (21.84%) and the critical path delay increase (9.65%).