Search

The conventional planar bulk MOSFET is difficult to scale down to sub-20nm gate length, due to the worsening performance variability and short channel effects. Thin body transistors, including Multiple-Gate (FinFET & Tri-Gate FET) and Fully Depleted SOI (FD-SOI) MOSFETs are anticipated to replace the current transistor architecture, and will be used in future CMOS technology nodes. Strained Silicon technology is widely used today to boost planar bulk transistor performance. Thus it's technically important to examine the strain-induced performance enhancement in these thin body transistors, for nanometer scale channel length. A comprehensive study on impact of channel stress on ultra-thin-body FD-SOI MOSFETs is presented. It's found that strain-induced mobility enhancement diminishes with Silicon body thickness scaling below 5nm for electrons, but not for holes. Strain-induced carrier transport enhancement is maintained with gate-length scaling. By applying forward back biasing (FBB) through the ultra-thin Buried Oxide layer, both carrier mobilities and their responses to strain get enhanced. For Multiple-gate FETs, the impact of performance enhancement through various types of stressors (including CESL, SiGe Source/Drain, Strained SOI and Metal Gate Last process) is studied, for different fin crystalline orientations and aspect ratios, to provide guidance for 3-D transistor design optimization.

Data-driven modeling is emerging as a new modeling paradigm necessitated by the ubiquitous practice of massive data collection. The high dimensionality of today’s dataset has become the major bottleneck of most modeling techniques. The considerable percentage of irrelevant, redundant or inaccurate data hidden underneath this high dimensionality drastically deteriorates the performance and comprehensibility of models built directly on these raw data. This dissertation proposes an Aggressive Dimensionality Reduction (ADR) methodology, aiming to tackle the “curse” of dimensionality and facilitate data modeling to truly utilize collected data for better decision making. Center to the proposed ADR are two new dependency tests, the empirical conditional expectation test between a continuous and a discrete variable and the orderized fractal dimension between two continuous variables. They constitute two powerful additions to the existing dependency test arsenal. Three case studies show the proposed ADR methodology is capable of effectively reducing dimensionality and improving model performance and comprehensibility.

Process planning has been imposed by the Computer Integrated Manufacturing (CIM) ideology a post-design nature, and researches in Computer Aided Process Planning (CAPP) have been mistakenly directed to automatic process planning. However, the nature of process planning dictates a successful CAPP system should be a decision support system rather than a decision generation system. Setup planning is an intermediate phase of process planning, and automatic setup planning constitutes a critical component of CAPP. The contradiction between the seemly productive research in the automatic setup planning literature and the reality of a near zero commercial application has to be ascribed to the same misconceived goal of freeing human being from every decision making process. In this thesis, a quantitative setup plan evaluation system driven by multiple attributes utility analysis coupling with manufacturing error simulation is proposed to serve three purposes, 1) to answer what is optimality of setup plans, 2) to build a framework of comparison for setup plans in practice and 3) to incorporate in existing automatic setup planning systems a human interface to fulfill their potential values.