Characterisation and mitigation of long-term degradation effects in programmable logic

Abstract

Reliability has always been an issue in silicon device engineering, but until now it has been managed by the carefully tuned fabrication process. In the future the underlying physical limitations of silicon-based electronics, plus the practical challenges of manufacturing with such complexity at such a small scale, will lead to a crunch point where transistor-level reliability must be forfeited to continue achieving better productivity. Field-programmable gate arrays (FPGAs) are built on state-of-the-art silicon processes, but it has been recognised for some time that their distinctive characteristics put them in a favourable position over application-specific integrated circuits in the face of the reliability challenge. The literature shows how a regular structure, interchangeable resources and an ability to reconfigure can all be exploited to detect, locate, and overcome degradation and keep an FPGA application running. To fully exploit these characteristics, a better understanding is needed of the behavioural changes that are seen in the resources that make up an FPGA under ageing. Modelling is an attractive approach to this and in this thesis the causes and effects are explored of three important degradation mechanisms. All are shown to have an adverse affect on FPGA operation, but their characteristics show novel opportunities for ageing mitigation. Any modelling exercise is built on assumptions and so an empirical method is developed for investigating ageing on hardware with an accelerated-life test. Here, experiments show that timing degradation due to negative-bias temperature instability is the dominant process in the technology considered. Building on simulated and experimental results, this work also demonstrates a variety of methods for increasing the lifetime of FPGA lookup tables. The pre-emptive measure of wear-levelling is investigated in particular detail, and it is shown by experiment how di fferent reconfiguration algorithms can result in a significant reduction to the rate of degradation.