On-chip NBTI and Gate-Oxide-Degradation Sensing and Dynamic Management in VLSI Circuits.

Abstract

The VLSI industry has achieved advancement in technology by continuous process scaling which has resulted in large scale integration. However, scaling also poses new reliability challenges. Currently the industry ensures the reliability of chips by limiting the supply voltage and temperature, but these constraints limit the benefits that are obtained from new process nodes. This method of managing reliability during design time is called Static Reliability Management (SRM). While SRM ensures that all the chips meet the reliability specifications, it introduces extreme pessimism in the chips as it margins for worst process, voltage, temperature and circuit state (PVTS), which will not be required for the majority of chips. To reduce the pessimism of SRM, the system needs to be made aware of its reliability by employing degradation sensors or degradation detection techniques. Using the degradation measurements, the system can estimate its lifetime and can adjust its operating points (supply voltage and temperature limits) dynamically and trade excess reliability slack with performance. This method of reliability management is called Dynamic Reliability Management (DRM). In this work we investigate different methods of DRM. We focus on two critical degradation mechanisms: Negative Bias Temperature Instability (NBTI) and Gate-oxide degradation. We propose NBTI and Gate-oxide degradation sensors with low area and power overhead, which allows them to be deployed in large numbers on the chip enabling collection of degradation statistics. The sensors were designed in 130nm and 45nm process nodes and tested on two test-chips. We then used the sensors to perform DRM in a silicon test for the first time. We demonstrate that DRM eliminates excess reliability slack which allows for a boost in supply voltage and performance. We then propose in situ Bias Temperature Instability (BTI) and Gate-oxide wear-out detection techniques. The in situ technique measures the degradation in the actual devices in the core and removes all the layers of uncertainty which arise because of the statistical nature of degradation and its dependence on PVTS. We implemented and tested these techniques on two test chips in a 65nm process node. We then use the BTI sensing technique to perform DRM.