ISBN 9783843928403

978-3-8439-2840-3, Reihe Informatik

Artur Quiring
Routing-driven Multiobjective 3D Floorplanning

163 Seiten, Dissertation Universität Hannover (2016), Hardcover, A5

Zusammenfassung / Abstract

Continuous down-scaling of feature sizes has been the main driver for performance improvement and functionality enhancement of integrated circuits. However, since moving from one technology node to the next becomes more and more uneconomic due to disproportionately increasing cost. New approaches are necessary to keep the current pace of innovation. One approach is 3D integration which offers a significant improvement of performance and power consumption due to possibly shorter interconnects. The additional third dimension in 3D integration requires new 3D-capable physical design tools with data structures and optimization algorithms adapted. To this end, a 3D floorplanner (SMF-3D) is presented in this thesis. As 3D floorplan data structure for SMF-3D the Multi Layer B*-Tree (MLBT) is proposed. It accounts for important 3D geometrical constraints such as 3D blocks which allow to further reduce interconnect length, 3D block groups, and 3D boundary constraints which facilitate modeling and optimization of complex design goals such as power consumption using a multiple supply voltage scheme. Furthermore, the routing algorithm Fast-ST-Routing, used within SMF-3D, is proposed to carefully plan wide global interconnects (WGIs), instead of using the half-perimeter wirelength metric to estimate interconnect length. Fast-ST-Routing uses Hanan grids and the approach Fast-Place to rapidly determine a routing solution. Fast-ST-Routing is sufficiently fast to be used within the inner loop of a Simulated Annealing (SA) based optimization algorithm. Experimental results show the necessity of WGI routing on floorplanning-level, and demonstrate the efficiency of Fast-ST-Routing compared to other state-of-the-art routers. Apart from WGI routing, several other design goals such as temperature, number of TSVs, and the fixed-outline constraint have to be considered in 3D floorplanning. Solving the respective multiobjective optimization problem is not trivial. Therefore, SMF-3D uses an enhanced SA based optimization algorithm which integrates problem specific information into the optimization process. To this end, an analytical function is proposed for each design goal. A combination of all analytical functions is used to guide the optimization process towards a low cost solution. Experimental results demonstrate the overall performance of SMF-3D compared to classical SA based optimization approaches.