Affiliations: [a] Department of Electrical and Computer Engineering, Concordia University, 1455 de Maisonneuve Blvd. West, Montreal, Canada | [b] Department of Electrical and Computer Engineering, University of Nevada, 4505 Maryland Parkway, Las Vegas, NV, USA | [c] Department of Electrical and Information Engineering, Otto-von-Guericke University of Magdeburg, PO Box 4120, D-39016 Magdeburg, Germany | [d] Department of Computer Science, University of Nevada, 4505 Maryland Parkway, Las Vegas, NV, USA
Correspondence:
[*]
Corresponding author: Department of Electrical and Computer Engineering, 1360 Barrington Street, P.O. Box 1000, Room C 367, Dalhousie University, Halifax, NS, B3J 1Z1, Canada. Tel.: +1 902 494 3993; Fax: +1 902 422 7535; E-mail: [email protected].
Abstract: Practical analog layout synthesis techniques have been the subject of active research for the past two decades to address the growing gap between the increasing chip functionality and the design productivity. In this paper, we present a novel macro-cell placement approach following the optimization flow of a genetic algorithm controlled by the methodology of simulated annealing. A process of cell slide is adopted to drastically reduce the configuration space without degrading search opportunities. In addition, this cell-slide process is used to satisfy the symmetry constraints essential for analog layouts. Furthermore, the dedicated cost function captures subtle electrical and geometrical constraints, such as area, net length, aspect ratio, proximity, parasitic effects, etc. required for analog layout and subsequent intellectual property reuse. To study the algorithm parameters, fractional factorial experiments and a meta-GA approach are employed. The proposed algorithm has been tested using several analog circuits. Compared to the simulated-annealing approach, the dominant one currently used for the analog placement problem, the proposed algorithm requires less computation time while generating higher quality layouts, comparable to expert manual placements. Furthermore, our hybrid algorithm and the method of parameter optimization can be readily adapted to different optimization problems across disciplines.
