Abstract
dc:descriptionProcess and environmental variations, as well as aging degradation, constitute major threats to circuit and system reliability, notably affecting speed and power dissipation. Although the relevant physical variation mechanisms do not differ to a great extent as scaling proceeds, their impact is no longer negligible due to the advancement of Moore’s law in technologies of a few nanometers. This increased loss of predictability in semiconductor devices poses a major challenge for the classical Static Timing Analysis (STA) algorithms and timing closure in general. Statistical methodologies are gaining significant attention due to their ability to quantify delay via statistical metrics. These methods take the form of Statistical Static Timing Analysis (SSTA) algorithms, Monte-Carlo (MC) simulations or probabilistic models. They handle cell and interconnection delays as random variables (RVs), following a particular Probability Density Function (PDF). Their purpose is to describe the maximum/minimum-delay PDFs at the outputs of the timing graph. This is achieved by either performing statistical max/min and add operations, and propagating the PDFs of constituent nodes along the timing graph, or by iteratively performing regular STA under process variations. Therefore, the maximum delay and the performance yield are obtained as a metric of the mean maximum delay and the related standard deviation. At a lower abstraction level, statistical models aim to probabilistically quantify circuit delay considering the relevant physical mechanisms, and define the mean delay and the standard deviation of delay as a function of transistor-level parameters, subjected to variations. Addition circuits are of particular importance due to their widespread usage, serving, among others, hardware implementations for neural networks, digital signal processing, and communication algorithms. Furthermore, they constitute building blocks for the implementation of more involved operations, such as multi-operand addition, multiplication, and division, and usually reside in timing critical paths. The literature concerning the delay assessment and modeling of adder circuits and architectures in the presence of process, voltage, and temperature variations is extensive. This thesis focuses on the timing variation analysis for certain adder architectures and more elaborate derived structures. Specifically, the focus is on the delay assessment of ripple-carry adders, borrow-save adders and parallel-prefix adders at technology nodes of a few nanometers in the presence of process variations. Following, closed-form models are presented for these architectures that interpret the delay variability measurements. The introduced models demonstrate sufficient accuracy compared to transistor-level simulations, and achieve substantial runtime acceleration with respect to MC Spice simulations. The purpose of this thesis is the development of detailed delay models for primary arithmetic structures, aiming to bridge the gap between statistical delay evaluation methodologies and circuit/block design. The use of the proposed modeling practices is motivated by the lack of detailed cell-level models for an early delay assessment of adder structures to drive logic synthesis, mapping, placement, and optimization processes. The introduced models find practical applicability within a great range of bit-lengths, circuit designs, and process technologies. Provided that the design of long bit length adders is a non-trivial task, as the related design space cannot be fully explored, the proposed models can accelerate variation-aware design decisions.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papachatzopoulos, Kleanthis
- Παπαχατζόπουλος, Κλεάνθης
Subjects
dc:subject × 8- Ολοκληρωμένα Κυκλώματα και Συστήματα
- Integrated Circuits and Systems
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Υπολογιστές, Υλικό (hardware) και Αρχιτεκτονική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Computer science, Hardware and Architecture
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58513