Study of Monitoring Circuitry for Ageing in FPGAs

Abstract: Along with the down-scaling of CMOS technology, ageing has become one of the most important reliability challenges in CMOS devices. Ageing is defined as degradation in certain device characteristics such as delay, which can result in failure. Field Programmable Gate Arrays (FPGAs) are typically the first among the CMOS devices to adopt the latest technology. It is, therefore, crucial to tackle ageing in FPGAs. As design-time-only techniques might prove insufficient in providing enough margins for future CMOS technology nodes, it becomes important to also monitor for the ageing degree to ensure the correct functionality. This thesis aims to review previous work to understand ageing and find effective ageing monitoring methods for FPGAs, in terms of the ability to detect degradation of the fabric resources with high accuracy and high precision. A comprehensive survey of previous methods is conducted, reporting a comparison of monitors in detail and comparing their pros and cons. Based on the survey, we perceived the process/performance variation mapping method by use of ring-oscillators (the so-called PV mapping) to have great potential and enhanced the existing PV mapping method by (1) introducing sensors based on new ring-oscillator types that cover significantly more hardware resources, and thus enhance the monitoring coverage, (2) pushing the number of uniform sensors (each sensor being comprised of a ring oscillator and a frequency counter) by the use of carefully developed placement constraints to almost 80% of the maximum theoretically possible number of sensors of the suggested type, and (3) designing extra ring-oscillator types and circuitry for gaining more insights into the precision and coverage of the proposed performance variation (PV) mapping method. The PV mapping method was applied to 20 Digilent Nexys4 boards, featuring a 28nm XILINX ARTIX 7 XC7A100T FPGAs to validate that the proposed method is capable of detecting delay differences among uniformly shaped sensors on the same device, and for each sensor among multiple boards. In addition, the precision and accuracy of the proposed method are reported.