Delay Measurements and Self Characterisation on FPGAs

Abstract

This thesis examines new timing measurement methods for self delay characterisation of Field-Programmable Gate Arrays (FPGAs) components and delay measurement of complex circuits on FPGAs. Two novel measurement techniques based on analysis of a circuit's output failure rate and transition probability is proposed for accurate, precise and efficient measurement of propagation delays. The transition probability based method is especially attractive, since it requires no modifications in the circuit-under-test and requires little hardware resources, making it an ideal method for physical delay analysis of FPGA circuits. The relentless advancements in process technology has led to smaller and denser transistors in integrated circuits. While FPGA users benefit from this in terms of increased hardware resources for more complex designs, the actual productivity with FPGA in terms of timing performance (operating frequency, latency and throughput) has lagged behind the potential improvements from the improved technology due to delay variability in FPGA components and the inaccuracy of timing models used in FPGA timing analysis. The ability to measure delay of any arbitrary circuit on FPGA offers many opportunities for on-chip characterisation and physical timing analysis, allowing delay variability to be accurately tracked and variation-aware optimisations to be developed, reducing the productivity gap observed in today's FPGA designs. The measurement techniques are developed into complete self measurement and characterisation platforms in this thesis, demonstrating their practical uses in actual FPGA hardware for cross-chip delay characterisation and accurate delay measurement of both complex combinatorial and sequential circuits, further reinforcing their positions in solving the delay variability problem in FPGAs.