University of Thessaly (UTH)
3DIC CAD placemenent flows and algorithms yielding improved power, performance, and area
Abstract
dc:descriptionIn recent years, the needs for more processing power, faster circuits, and less power consumptionlead the semiconductor industry to investigate innovating methodologies andtechnologies to continue the chip scaling process.Moore’s law, indicating that every 18 months the number of transistors is duplicated,comes to an end. The size of the transistors will not be able to be further reduced, whilethe manufacturing cost will continue to increase for each new technology node. In parallel,wires’ delay cannot be scaled as transistors can, making the adoption of new technologynodes more challenging.A solution allowing higher density is to stack transistors in multiple tiers, by utilizing athird dimension. Three-Dimensional (3D) integrated circuits (ICs) allow chip scaling, justlike skyscrapers can accommodate more and more people at the same building block. Bydesigning 3D ICs, there is neither need for new material usage nor is it necessary to spendlarge amounts of money on investigating smaller technological nodes. So, mature technologynodes can be used for 3D ICs, as the benefits of stacking transistors in multiple tiers maysurpass the total benefits of utilizing new technology nodes.Although the 3D ICs design has been around for decades, it is not being employed inmass manufacturing. Therefore, the semiconductor industry will have to deal with a numberof new challenges due to the introduction of the third dimension. One of these challengesis higher density which results in higher circuit temperature, while the efficient linking ofdifferent tiers is another challenge. The absence of Electronic Design Automation (EDA)tools is another impediment to the mass manufacture of 3D ICs. Most of the attempts todesign a 3D IC used traditional two-dimensional EDA tools and physical design flows, withmodifications. However, circuit quality is low when the convectional two-dimensional flowsand tools are used, as they do not inherently take the third dimension into consideration.Our objective in this study is to investigate and evaluate different methodologies thatprovide the ability to separate a design into tiers, i.e. the tier assignment problem. This is oneof the first steps for the 3D integration, while each methodology significantly impacts integrationquality. Our main target has been to improve power consumption, circuit performance,circuit size, and achieve a limited number of interconnections across several tiers.Firstly, we investigated several Block-Based techniques to divide and tackle the problemof assigning a large circuit to tiers. Since modern designs contain billions of standard cells,designing a chip that takes all of the components at once into account is quite challenging.To simplify the three-dimensional design, transitional two-dimensional divide-and-conquerapproaches are employed. So, we developed and consequently we analyze three-dimensionalfloorplanning and clustering algorithms for tier assignment in this study, to handle very largedesigns. This floorplanning algorithm aims to partition RTL functional modules into tiers ofequal size, while the clustering algorithm aims to generate well-formed (e.g. balanced, lowconnectivity) groups of standard cells that will be allocated to tiers.Despite the fact that Block-Based techniques can handle very large designs, their quality isrestricted due to their higher level of circuit abstraction. So, we also investigated Fine-Grainedtechniques as partitioning and three-dimensional legalization. Fine-grained approachesprovide greater freedom, which can lead to better solutions. However, these approaches mustbe able to handle today’s very large circuits, that contain billions of standard cells. But, theFine-Grained techniques that we propose to assign standard cells to tiers do handle thesevery large designs.Finally, we evaluate the merits and limitations of suggested techniques, algorithms, andflows by comparing the placed and routed designs. We also provide assistance on how toapply the methods, algorithms, and flows on three-dimensional circuits, to select the optimalone for the specified metrics.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Nikolaos, Sketopoulos
- Νικόλαος, Σκετόπουλος
Subjects
dc:subject × 10- Ηλεκτρονική Σχεδίαση Κυκλωμάτων
- 3Δ Ολοκληρωμένα Κυκλώματα
- Electronic design automation
- 3D Integrated Circuits
- 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/50170