## First Steps Towards Experimental Design in Evaluating Layout Layout Algorithms: Wire Length versus Wire Crossing in Linear Placement Optimization (1998)

Venue: | October), CBL, CS Dept., NCSU, Box 7550 |

Citations: | 2 - 2 self |

### BibTeX

@TECHREPORT{Ghosh98firststeps,

author = {Debabrata Ghosh and Franc Brglez and Matthias Stallmann},

title = {First Steps Towards Experimental Design in Evaluating Layout Layout Algorithms: Wire Length versus Wire Crossing in Linear Placement Optimization},

institution = {October), CBL, CS Dept., NCSU, Box 7550},

year = {1998}

}

### OpenURL

### Abstract

. This report introduces an experimental design that discovers a new relationship between a cell placement for minimized wire crossing in bipartite (two-layer) graphs and a cell placement in linear arrangement, optimized for minimum total wire length as measured after rectilinear routing in a single channel. We introduce hypercrossing, a new crossing model for graphs, and demonstrate that the total wire length in the channel, as measured after routing, has a positive linear correlation with the total hypercrossing, and a coefficient of correlation of near 100%. While heuristics to directly minimize total hypercrossing are yet to be invented, our experiments show that the minimization of total (regular) wire crossing leads to placements with significantly reduced wire length and wiring area. A methodology and new datasets are now available to determine, with statistical significance, which of the two cost function classes will yield a better placement in general: costs based on the mode...

### Citations

347 |
Graph Drawing: Algorithms for the Visualization of Graphs
- Battista, Eades, et al.
- 1999
(Show Context)
Citation Context ...inimum crossing number is NP-complete, even for 2-layer graphs [19]. Graph drawing packages can now be obtained from the Web [20], [21]. A textbook, bringing this research up-todate, is now available =-=[22]-=-. This paper introduces hypercrossing, a new crossing model for graphs, and a series of experimental designs that demonstrate, with statistical significance, important relationships between the cell p... |

290 | Logic synthesis and optimization benchmarks user guide version 3.0
- Yang
- 1991
(Show Context)
Citation Context ...ll treatments are applied to each instance of a netlist in the class. The netlists of 2-layer graphs in our experiments are representative of connected components extracted from netlists available in =-=[30]-=-. Only seldom would such components have a very large number of nodes -- most have but a few nodes, particularly in arithmetic circuits. We find it important to conduct the first set of experiments wi... |

233 | A technique for drawing directed graphs
- Gansner, Koutsofios, et al.
- 1993
(Show Context)
Citation Context ...lity of hierarchical structures [18]. The problem of placing vertices for minimum crossing number is NP-complete, even for 2-layer graphs [19]. Graph drawing packages can now be obtained from the Web =-=[20]-=-, [21]. A textbook, bringing this research up-todate, is now available [22]. This paper introduces hypercrossing, a new crossing model for graphs, and a series of experimental designs that demonstrate... |

214 |
Crossing number is NP-complete
- Garey, Johnson
- 1983
(Show Context)
Citation Context ...ars. Crossing theory has been developed to improve the readability of hierarchical structures [18]. The problem of placing vertices for minimum crossing number is NP-complete, even for 2-layer graphs =-=[19]-=-. Graph drawing packages can now be obtained from the Web [20], [21]. A textbook, bringing this research up-todate, is now available [22]. This paper introduces hypercrossing, a new crossing model for... |

200 | GORDIAN: VLSI placement by quadratic programming and slicing optimization
- Kleinhans, Sigl, et al.
- 1991
(Show Context)
Citation Context ...mostly to total wire length. Algorithms that use wire length as a cost function include approaches as diverse as simulated annealing [1, 2], quadrisection [3], and recursive and quadratic programming =-=[4, 5, 6, 7, 8, 9, 10]-=-. Comparisons between different wire length objectives are presented in [11, 12]. Models and algorithms for accurately estimating the average interconnect length for both random and optimized placemen... |

108 |
Efficient and effective placement for very large circuits
- Sun, Sechen
- 1995
(Show Context)
Citation Context ... the cost function that continues to dominate the research relates mostly to total wire length. Algorithms that use wire length as a cost function include approaches as diverse as simulated annealing =-=[1, 2]-=-, quadrisection [3], and recursive and quadratic programming [4, 5, 6, 7, 8, 9, 10]. Comparisons between different wire length objectives are presented in [11, 12]. Models and algorithms for accuratel... |

87 |
VLSI Placement and Global Routing Using Simulated Annealing
- Sechen
- 1988
(Show Context)
Citation Context ... the cost function that continues to dominate the research relates mostly to total wire length. Algorithms that use wire length as a cost function include approaches as diverse as simulated annealing =-=[1, 2]-=-, quadrisection [3], and recursive and quadratic programming [4, 5, 6, 7, 8, 9, 10]. Comparisons between different wire length objectives are presented in [11, 12]. Models and algorithms for accuratel... |

84 |
New lower bound techniques for VLSI
- Leighton
- 1981
(Show Context)
Citation Context ...on the wire crossing cost function. The crossing number and wire area have been investigated to find lower bounds on the layout area and the maximum edge length of a variety of computational networks =-=[15]-=-. Given a global routing, the problem of `properly' distributing the set of crossings among the regions has been studied in [16]. Crossing minimization as a post-optimization problem after global rout... |

82 | Analytical placement: A linear or a quadratic objective function - Sigl, Doll, et al. - 1991 |

65 |
Statistical Theory and Methodology
- Brownlee
- 1965
(Show Context)
Citation Context ... where the parameter description is the same as already provided for (1). Notably, the wire length statistics for the two treatments, tr20 in (1) and tr03 in (2) appear distinct even without a t-test =-=[28]-=-. Indeed, for a two-tailed test at the ff = 0:05 significance level we have t0:975 = \Sigma1:972 while t-test based on statistics in (1) and (2) shows that t = 28:38, clearly outside the range of t0:9... |

52 |
Module Placement based on Resistive Network Optimization
- Cheng, Kuh
- 1984
(Show Context)
Citation Context ...mostly to total wire length. Algorithms that use wire length as a cost function include approaches as diverse as simulated annealing [1, 2], quadrisection [3], and recursive and quadratic programming =-=[4, 5, 6, 7, 8, 9, 10]-=-. Comparisons between different wire length objectives are presented in [11, 12]. Models and algorithms for accurately estimating the average interconnect length for both random and optimized placemen... |

51 |
Crossing theory and hierarchy mapping
- Warfield
- 1977
(Show Context)
Citation Context ... and computer science, research on crossing number minimization in graphs has been carried out for many years. Crossing theory has been developed to improve the readability of hierarchical structures =-=[18]-=-. The problem of placing vertices for minimum crossing number is NP-complete, even for 2-layer graphs [19]. Graph drawing packages can now be obtained from the Web [20], [21]. A textbook, bringing thi... |

45 |
RITUAL: A performance driven placement algorithm
- Srinivasan, Chaudhary, et al.
- 1992
(Show Context)
Citation Context ...mostly to total wire length. Algorithms that use wire length as a cost function include approaches as diverse as simulated annealing [1, 2], quadrisection [3], and recursive and quadratic programming =-=[4, 5, 6, 7, 8, 9, 10]-=-. Comparisons between different wire length objectives are presented in [11, 12]. Models and algorithms for accurately estimating the average interconnect length for both random and optimized placemen... |

37 | Partitioning very large circuits using analytical placement techniques
- Riess, Doll, et al.
- 1994
(Show Context)
Citation Context ...ches as diverse as simulated annealing [1, 2], quadrisection [3], and recursive and quadratic programming [4, 5, 6, 7, 8, 9, 10]. Comparisons between different wire length objectives are presented in =-=[11, 12]-=-. Models and algorithms for accurately estimating the average interconnect length for both random and optimized placements have been presented in [13]. Abstract placement models, such as presented in ... |

33 | Faster Minimization of Linear Wirelength for Global Placement
- Alpert, Chan, et al.
- 1998
(Show Context)
Citation Context |

28 | Synthesis of wiring signature-invariant equivalence class circuit mutants and applications to benchmarking
- GHOSH, KAPUR, et al.
- 1998
(Show Context)
Citation Context ...to belong to the mutation class, each perturbation graph must also satisfy the signature oe mut in (3) -- a non-trivial procedure which has been generalized for arbitrary netlists and is described in =-=[25]-=-. Both the isomorphism class G iso and the mutation class Gmut satisfy the signature oe mut in (3): both have the same number of nodes, edges, and connected components -- as well as the distribution o... |

25 |
A unified approach to partitioning and placement
- Kuh
- 1991
(Show Context)
Citation Context |

22 |
Analytical placement: A linear or quadratic objective function
- Doll, Johannes
- 1991
(Show Context)
Citation Context |

19 | On crossing minimization problem
- Chen, Lee
- 1998
(Show Context)
Citation Context ...uting, the problem of `properly' distributing the set of crossings among the regions has been studied in [16]. Crossing minimization as a post-optimization problem after global routing is reported in =-=[17]-=-. However, in fields as diverse as economics, social sciences, mathematics and computer science, research on crossing number minimization in graphs has been carried out for many years. Crossing theory... |

18 | On bipartite drawings and the linear arrangement problem
- Shahrokhi, Sýkora, et al.
(Show Context)
Citation Context ...ing number in the bigraph, is the number of line intersections induced by f . Similarly, a bigraph model is used to formulate a linear arrangement problem and the associated wire length cost function =-=[26]-=-. The work in [26] also brings out important relationships between the wire crossing and the linear arrangement problem in graphs. Note that while the wire length 1 We leverage the notion of a netlist... |

17 |
Average interconnection length estimation for random and optimized placements
- Sechen
- 1987
(Show Context)
Citation Context ...erent wire length objectives are presented in [11, 12]. Models and algorithms for accurately estimating the average interconnect length for both random and optimized placements have been presented in =-=[13]-=-. Abstract placement models, such as presented in [14], also aim * F. Brglez and D. Ghosh have been supported by contracts from the Semiconductor Research Corporation (94--DJ--553), SEMATECH (94--DJ--... |

16 | Design of experiments in BDD variable ordering: Lessons learned
- Harlow, Brglez
- 1998
(Show Context)
Citation Context ... rely on hashing the input data may unknowingly undo the randomization of input presentations and confound the experiments. An important lesson on this subject has been learned and reported elsewhere =-=[29]-=-. A number of experiments, reported in the next section, demonstrate the utility and application of both the isomorphism and the mutation classes. V. Experimental Results The experiments are based on ... |

14 | Heuristics and experimental design for bigraph crossing number minimization, in
- Stallmann, Brglez, et al.
- 1999
(Show Context)
Citation Context ...l hypercrossing are yet to be invented, 18 out of 19 placement algorithms whose objective is to minimize (regular) wire crossing achieve a better minimum on average (of wire crossing and wire length) =-=[23]-=- than a stateof -the-art placement algorithm whose objective is to minimize the total wire length [24]. ffl An equivalence class of two-layer graphs that are all isomorphic to each other (but presente... |

13 |
A Quadrisoction-Based Combined Place and Route Scheme for Standard Cells
- Suaris, Kedem
- 1989
(Show Context)
Citation Context ...t continues to dominate the research relates mostly to total wire length. Algorithms that use wire length as a cost function include approaches as diverse as simulated annealing [1, 2], quadrisection =-=[3]-=-, and recursive and quadratic programming [4, 5, 6, 7, 8, 9, 10]. Comparisons between different wire length objectives are presented in [11, 12]. Models and algorithms for accurately estimating the av... |

13 |
The Crossing Distribution Problem
- Marek-Sadowska, Sarrafzadeh
- 1995
(Show Context)
Citation Context ...area and the maximum edge length of a variety of computational networks [15]. Given a global routing, the problem of `properly' distributing the set of crossings among the regions has been studied in =-=[16]-=-. Crossing minimization as a post-optimization problem after global routing is reported in [17]. However, in fields as diverse as economics, social sciences, mathematics and computer science, research... |

13 |
OASIS 2.0 User's Guide
- Kozminski
- 1992
(Show Context)
Citation Context ...ze (regular) wire crossing achieve a better minimum on average (of wire crossing and wire length) [23] than a stateof -the-art placement algorithm whose objective is to minimize the total wire length =-=[24]-=-. ffl An equivalence class of two-layer graphs that are all isomorphic to each other (but presented in different random orders) can induce nearly as much variance in the optimized wire length (and wir... |

12 |
Analysis of PlacementProcedures for VLSI Standard Cell Layout
- Hartoog
- 1986
(Show Context)
Citation Context ... Variations such as observed and reported for the example of an isomorphism class in (1), are not unusual -- they have been reported for a number of netlists and placement algorithms as early as 1986 =-=[27]-=-. Additional and more recent instances are reported elsewhere [25]. Not all instances of reference circuits produce an isomorphism class that induces as much variability as we report in (1). Study of ... |

10 |
Experimental Appraisal of Linear and Quadratic Objective Functions E ect on Force Directed Method for Analog Placement
- Mahmoud, Asakura, et al.
- 1994
(Show Context)
Citation Context ...ches as diverse as simulated annealing [1, 2], quadrisection [3], and recursive and quadratic programming [4, 5, 6, 7, 8, 9, 10]. Comparisons between different wire length objectives are presented in =-=[11, 12]-=-. Models and algorithms for accurately estimating the average interconnect length for both random and optimized placements have been presented in [13]. Abstract placement models, such as presented in ... |

9 | E cient and e ective placements for very large circuits - Sun, Sechen - 1993 |

7 |
Accurate Net Models for Placement Improvement by Network Flow Methods
- Doll, Johannes, et al.
- 1992
(Show Context)
Citation Context |

6 | A Uni ed Approach to Partitioning and Placement - Tsay, Kuh - 1991 |

6 | Crossing theory and hierarchy mapping - eld, J - 1977 |

3 |
On the Mincut Bipartite Arrangement Problem
- Bui, Lee
- 1987
(Show Context)
Citation Context .... Models and algorithms for accurately estimating the average interconnect length for both random and optimized placements have been presented in [13]. Abstract placement models, such as presented in =-=[14]-=-, also aim * F. Brglez and D. Ghosh have been supported by contracts from the Semiconductor Research Corporation (94--DJ--553), SEMATECH (94--DJ--800), DARPA/ARO (P--3316--EL/DAAH04--94--G--2080), and... |