Incremental placement for field-programmable gate arrays

As the logic capacity of FPGAs continues to increase with deep submicron technology, performing a full recompilation for small iterative changes in a large design is an extremely time-consuming and costly process. To address this issue, this thesis presents a new incremental placement algorithm for FPGAs named "iPlace" that significantly reduces the time required for recompilation. The iPlace algorithm is based on shifting, compaction, and annealing. Key ideas from the algorithm include a placement super-grid that is larger than the physical size of the FPGA. The super-grid allows insertion of additional CLBs into areas with no free locations by CPU-efficient shifting. This is followed by a compaction scheme to re-legalize CLBs that are shifted to illegal locations outside of the physical size of the FPGA. The algorithm ends with a low-temperature anneal to improve quality. This algorithm is capable of handling multiple design changes across large regions of a FPGA. This is especially useful for hierarchical designs where sub-circuits are re-used multiple times. If one such sub-circuit is modified, iPlace can quickly produce a high quality incremental placement solution. For a single region of design change, we found that iPlace is 34 to 260 times faster than the academic tool Versatile Place and Route (VPR) in default mode. Compared to VPR's reduced-quality "-fast" placement option, iPlace is 3 to 28 times faster with equivalent quality. For multiple regions of design changes, iPlace is still 50-70 times faster compared to VPR in default mode when up to 2/3 of the CLBs are modified; Compared to the "-fast" placement option, iPlace is still 5-8 times faster. We believe that iPlace is the first academically available incremental placement algorithm capable of handling significant changes to a netlist for very large circuits.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat