Performance Variation in Digital Systems : Workload Dependent Modeling and Mitigation

Shrinking of device dimensions has undoubtedly enabled the very large scale integration of transistors on electronic chips. However, it has also brought to surface variation phenomena, that degrade system’s performance and threaten functional operation. Hence, the need to capture and describe these mechanisms, as well as effectively mitigate their impact is crucial. To this extent, we will focus on capturing, describing and characterizing variation threats in an efficient and accurate manner. In addition, we will move forward by proposing and developing a novel mitigation scheme that could tackle with performance variability and ensure dependability. First, we will study basic reliability phenomena that cause parametric and functional failures in digital systems. Bias Temperature Instability (BTI), Hot Carrier Injection (HCI), Random Dopand Fluctuation (RDF) are some of the dominant variation threats. After understanding the physics of these phenomena, we will proceed with following (or even developing) accurate and efficient models to describe them. This step is necessary in order to effectively mitigate performance variability later on. After having such models in our disposal, the main target of our research will be to study state-of-the-art (SoTA) mitigation mechanisms of performance variability. Finally, our aim is to propose a novel mitigation scheme that introduces an accurate yet energy-efficient approach to manage performance dependability.