You are on page 1of 6

Applied Soft Computing 21 (2014) 376381

Contents lists available at ScienceDirect

Applied Soft Computing


journal homepage: www.elsevier.com/locate/asoc

Optimal approach on net routing for VLSI physical design based on


Tabu-ant colonies modeling
Ping Yang a, , Haiying Yang a , Wei Qiu a , Shuting Wang b, , Chuanquan Li c
a
Laboratory of Advanced Design, Manufacturing and Reliability for MEMS/NEMS/OEDS, School of Mechanical Engineering, Jiangsu University, Zhenjiang
212013, PR China
b
School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan, Hubei 430074, PR China
c
School of Mechanical & Electrical Engineering, Guilin University of Electronic Technology, PR China

a r t i c l e i n f o a b s t r a c t

Article history: To get a more efcient program for net routing design in VLSI physical design, a new mixed algorithm
Received 1 December 2011 is presented by combining ant colonies algorithm and Tabu search algorithm for improving net routing
Received in revised form 29 January 2014 design scheme in VLSI physical design. The models by considering different structure property such as
Accepted 22 March 2014
two-terminal, multiple-terminal, multi-layers and gridless net routing are developed with introducing
Available online 1 April 2014
the proper parameters matching which can be obtained by computer experiments. The results show that
the new algorithm can avoid the low convergence rate in the initial stage of basic ant colonies system.
Keywords:
The efciency of the Tabu-ant colonies is improved about 16.667%; meantime, the Tabu-ant colonies
Tabu-ant colonies hybrid modeling
Net routing
system can also avoid the local optimal solution effectively. It builds a basis for future work in solving
Optimal path, gridless net routing multiple-terminal, multiple-layers and gridless net routing problems with high efciency.
VLSI physical design 2014 Elsevier B.V. All rights reserved.

1. Introduction estimation technique that generates routing solution without any


prior knowledge of placement, and pin locations. They compared
Net routing is a NP-complete (non-deterministic polynomial our experimental results with an exact global router values, and
time) problem in very large-scale integration (VLSI) physical design the accuracy of their result is quite encouraging. Zhang and Chu [3]
[115]. As the electric circuit integration increasing day by day, investigated an efcient detailed router applying regular routing
the accuracy of the net routing becomes more and more accurate, patterns. In this paper, Regular Route is proposed for potentially
the efciency of the algorithm becomes faster and faster, how- better design rule satisfaction based on a correct-by-construction
ever, the structure is more and more complex, the net ports are methodology. Regular Route generates a detailed routing solution
also increasing. The traditional algorithm cannot meet the design that satises the basic spacing-based design rules. To satisfy all the
requirements any more. So, we should propose some new algo- design rules, they propose an abstract idea of local optimization
rithms continuously. In recent years, many intelligent algorithms, based on local shift and rip-up-and-reroute, assuming that most
such as genetic algorithm, immune algorithm and ant colonies sys- design rules are complex functions of local and neighboring geome-
tem, have appeared and applied to many important elds gradually. tries. Their experimental results demonstrate the effectiveness and
These algorithms have a great application prospect in VLSI design efciency of Regular Route. Wu et al. [4] study global routing via
[1,57]. integer programming. This paper introduces GRIP, a global routing
Kahng [1] proposed a parallel, analog approach for the construc- technique via integer programming. GRIP optimizes wirelength and
tion of a minimum rectilinear Steiner tree (MRST), which intuitively via cost directly without going through a traditional layer assign-
shrinks a bubble around the pins of the signal net until a Steiner tree ment phase. By using integer programming in an effective manner,
topology is induced. Sardar et al. [2] developed a geometry inde- GRIP obtains high-quality solutions. Specically, for the ISPD 2007
pendent wirelength estimation method in VLSI routing for VLSI and 2008 benchmarks, GRIP obtains an average improvement in
Design. In this paper, they propose a partition driven wirelength wirelength and via cost of 9.23% and 5.24%, respectively, when com-
pared to the best result in the open literature. Hadlock [5] discussed
a shortest path algorithm for grid graphs. Huijbregts and Jess [6]
Corresponding author. Tel.: +86 511 88790779.
investigated general gate array routing by using a k-terminal net
Co-corresponding author. routing algorithm with failure prediction. A general approach to
E-mail addresses: yangpingdm@ujs.edu.cn, yangping1964@163.com (P. Yang). gate array routing is presented, based on an abstract routing space

http://dx.doi.org/10.1016/j.asoc.2014.03.033
1568-4946/ 2014 Elsevier B.V. All rights reserved.
P. Yang et al. / Applied Soft Computing 21 (2014) 376381 377

model. Kohira et al. [7] developed a fast longer path algorithm for
routing grid with obstacles using biconnectivity based length upper
bound. A fast longer path algorithm that generates a path of a net in
routing grid so that the length increased as much as possible is pro-
posed. Liao et al. [8] developed three hybrid ant colony optimization
algorithms for mixed discrete-continuous optimization problems.
The rst two hybrids, labeled ACOR-HJ and ACOR-DE, differs in
philosophy with the former integrating ACOR with the effective
Hooke and Jeeves local search method and the latter a cooperative
hybrid between ACOR and differentia evolution. The third hybrid,
labeled ACOR-DE-HJ, is the second cooperative hybrid enhanced
with the Hooke and Jeeves local search. It was showed that all
three algorithms greatly outperform the original ACOR in nd-
ing the exact or near global optima. Dogan and Korrek [9] study
a new ECG beat clustering method based on kernelized fuzzy c-
means and hybrid ant colony optimization for continuous domains.
The proposed method is applied to a dataset which is obtained
from MIT-BIH arrhythmia database. After several experiments it Fig. 1. Non-uniform grid graph.
is observed that the proposed method outperforms the traditional
fuzzy c-means and kernelized fuzzy c-means algorithms. Lin and El
The generation of the gridless graph is familiar with that of the
Gamal [12] discussed a low-power eld-programmable gate array
non-uniform gird graph. We need only notice that when we make
routing fabric. The results show that an FPGA using this fabric can
the scanning beam, we should not let them penetrate the obstacles.
achieve 1.57 times lower dynamic power consumption and 1.35
Then, the terminals of the scanning beam and the points of inter-
times lower average net delays with only 9% reduction in logic den-
section can constitute a lattice, which is named as routing model
sity over a baseline island-style FPGA implemented in the same
of the gridless graph. It can be shown in Fig. 2. The information in
65 nm CMOS technology. The new routing fabric is also well-suited
the graph can also be recorded by a matrix.
to monolithically stacked 3D IC implementation. Chaudhry et al.
We can see that the number of the nodes in Fig. 1 is much less
[13] discussed some efcient rerouting algorithms for congestion
than that in Fig. 2, this means that the storage space and search
mitigation. Borah et al. [14] developed a fast and simple Steiner
space in non-uniform grid graph model are larger than these in
routing heuristic. This paper presents a simple yet efcient heuris-
gridless graph model. But the later is more complex to implement.
tic for rectilinear Steiner routing. The basic heuristic introduces a
In this paper, we study with emphases on non-uniform grid graph
new edge into the existing tree for another, costlier edge such that
model. We make some explorations about gridless graph at the rear
the resulting graph remains a tree. The heuristic has been success-
of the paper.
fully applied to the problem of minimum-length Steiner routing
and minimizing critical-sink Elmore delay. Yang et al. [1517] dis-
3. Ant colonies system and Tabu search algorithm
cussed a mixed isomorphism approach for kinematic structure
enumeration graphs based on GA and Ant algorithm respectively.
3.1. Ant colonies system
Rescki [18] gave an introduction for interested mathematicians,
concentrating on some simple special subproblems which can
Ant colonies system is a new simulating evolution algorithm
be used as illustrations of the applicability of combinatorial
which was rst presented by Italian scholar Dorigo et al. [19,20].
optimization.
It is used to solve combinatorial optimization problem by studying
In this paper, in order to improve the efciency of the net rout-
the ability for nding the best path during the course of search-
ing in the IC net design, a Tabu-ant colonies modeling method is
ing for food. The algorithm has been used to solve the traveling
developed by combining ant colonies algorithm with Tabu search
salesman problem, assignment problem, job scheduling problem,
algorithm for VLSI physical design. A non-uniform grid graph and a
etc. successfully. The application of the algorithm of Ant colonies
gridless graph is designed as the routing models. The related param-
system to the two-terminal net routing can be described as follows.
eters are discussed and suitable values are taken. The computer
experiments can illustrate the validity of the new mixed algorithm
for solving the shortest path in VLSI routing design.

2. Description about routing modeling

The non-uniform gird graph and the gridless graph are two com-
mon models for net routing. The shadows in Fig. 1 present obstacles.
They can be either electronic component or routed net. The points
S and T in Fig. 1 are terminals for routing. We make vertical and
horizontal line passing S and T. We make the scanning-beam along
the boundaries of all the obstacles. So, the non-uniform gird graph
is generated. The information in the graph includes: (1) the pos-
itions of the terminals; (2) the positions of the obstacles; (3) the
positions and status of the nodes. Here, all the intersection points
of the lines are nodes. The status of the nodes means whether the
nodes are inside the obstacles or not and whether the nodes have
been passed by in the course of the ants searching for path. The
information can be recorded by a matrix. Fig. 2. Gridless graph.
378 P. Yang et al. / Applied Soft Computing 21 (2014) 376381

Step1 Initialize the routing graph and give the initial value to the makes the solution vary at some degree. The length of the Tabu
pheromone of the node in the routing graph; object is decided according to the situation.
Step2 Place all the ants at the start terminal S; The algorithm is described as follows:
Step3 One ant chooses the next node following the rule in For-
mula (1). We must choose the next node from the set of Step1 Set (Tabu list) H = , initial solution: x0 , present solution:
optional nodes. Then make the local updating of the value x cur = x0 ;
of the pheromone of the chosen node by Formula (3). This Step2 Check if it satises the condition for ending. If it does, output
is done repeatedly until the ant reaches the end point or the the results, end; otherwise, choose the candidate set able
set of optional nodes is null; (x cur), which is not tabued;
Step4 Check if all the ants have set out. If it has not, return to Step3; Step3 Choose the solution which has the optimal evaluation from
otherwise, record the shortest path; able (x cur); x cur = x best;
Step5 Make the global updating of the value of the pheromone by Step4 Update the information in H; return to Step2.
Formula (4);
Step6 Check if it satises the condition for ending. If it does not,
4. Tabu-ant colonies system
return to Step2; otherwise, end;
Step7 Paint the shortest path in the graph.
4.1. Disadvantages of ant colonies system

Supposing the position of one ant in the colony is r, this ant Though the study of the ant colonies system has shown its supe-
chooses the next nodes s by the following rules. riority in solving the complex problems, as a probability algorithm,
 it has some disadvantages for application.

arg max{[ph(s)] [dis(r, s)] } if q < q0
s= (1)
S else (1) Long searching time. The ant colonies system gets the opti-
mal solution instructed by the pheromones feedback. At the
We suppose the set of optional nodes is allowed(s). Where, s primary phase of the searching, because the pheromones dis-
allowed (s). ph(s) is the value of the pheromone on node s and dis(r, tribution is uniform, it is very hard to nd a shorter path
s) presents the distance from r to s. and are parameters. They immediately. This is even more obvious when the solution
show the comparative importance of the pheromone and the path. space is large. Ants move on the path and pheromone is left. As
Where, q is a random gure conned in [0,1] and q0 is a parameter the lapse of time, the path which has passed more ants than oth-
(0 q0 1). S is a ratio rule which chooses the next s by probability ers will have more pheromone than the others. But, this course
p(r, s). needs a very long time.
(2) The stagnation behavior of the algorithm. As the pheromones

[ph(s)] [dis(r, s)]

accumulation on the path, the solutions will be same or similar.
p(r, s) =

if S allowed(s)
(2)
{[ph(s)] [dis(r, s)] } That is to say, the algorithm cannot make a further search for the
optimal solution. The search falls into a local optimal solution.
0 else

Local updating of the value of the pheromone: 4.2. Improvement of ant colonies system

ph(s) = (1 )ph(s) + ph0 (3)


According to these disadvantages of ant colonies system, we can
Global updating of the value of the pheromone: improve it by using following measures.

ph(s) = (1 )ph(s) + ph(s) (4) (1) Adjust the initial distribution of the pheromone. The uniform
distribution of the pheromone is the main reason of the slow
where ph0 is the initial value of the pheromone and 1  is the convergence rate in the primary phase. So, we regard the area in
attenuation coefcient of the value of the pheromone. which appears the optimal path in a higher rate as the effective
area for routing. Experiences show that, in many cases, it is
C if s the global shortest path possible to nd the shortest path in the minimum rectangular
ph(s) = Lgb (5)
0 area which covers all the terminals. Therefore, we regard this
else as the main area for searching and strengthen the pheromone
of this area.
where Lgb is the length of the present global shortest path and C is (2) Improve the rules for moving. Based on the moving rules in Eq.
a constant. (1), we know that a heuristic rule is used in ant colonies system.
The pheromone of the node and the distance from the node to
3.2. Tabu search algorithm the next nodes are the main factors to be considered. But the
searching direction is not taken into consideration at all. Actu-
Tabu search algorithm is an extension of the local search algo- ally, because the existence of the end point, the nodes which are
rithms. It is the application of articial intelligence to combinatorial near to the end point must be taken into prior consideration.
optimization algorithms. Its presentation is mainly for avoiding the By this way, the moving direction can be controlled to avoid
local optimization in the searching course. A Tabu list is set up its getting away from the end point. It makes the tendency to
which can record the optimal path in the algorithm. In the later search for the end point more obviously.
steps, the solutions in the Tabu list are partly adopted or abandon (3) The ant colonies system can avoid the local optimal more effec-
absolutely. We can avoid the local optimal solution by this way and tively by combining with Tabu search algorithm. As the lapse
search in a wider space. The length of the object, the candidate set, of time, the path which has passed more ants than others will
evaluation function and some computing information consist the have more pheromone than the others. To avoid the early con-
main characters of the algorithm. The most important two indexes vergence to the local optimal solution, we can resort to the
are Tabu object and the length of the object. Tabu object always divergent mechanism of Tabu search algorithm. When the value
P. Yang et al. / Applied Soft Computing 21 (2014) 376381 379

Table 1
Results between the ant colonies and improved ant colonies.

Algorithms The shortest path length (mm) Number of successful runs Convergent time (s) Improvement
3
Ant colonies 425 300 times 180 10
16.6667%
Improved ant colonies 425 220 times 150 103

of the pheromone of one node is higher than a certain con- The improved moving rules as following
stant, the node will be put into the Tabu list A. Then we decide 
1
weather put this node into the set of optional nodes according arg max{[ph(s)] [dis(r, s)] dis(start, end) } if q < q0
s= (6)
the information in list A. Besides, the optimal solution which is S else
obtained in one of the repeats is put into the Tabu list B. And
 1
we also decide weather implement the global updating of the
[ph(s)] [dis(r, s)] dis(start, end)
if S allowed(s)
value of the pheromone. By this way, we can effectively avoid p(r, s) =
{[ph(s)] [dis(r, s)] dis(start, end)
1
} (7)
the accumulation of the pheromone on some of the paths and 0 else
avoid lose the global optimal solution.
where start presents the start point, end presents the end point.
And dis(start, end) is the distance from the start point to end point.
According to these aforesaid measures, an improved algorithm The ant colonies algorithm and the improved ant colonies are
can be presented which is named as Tabu-ant colonies system. The used to search the shortest path for the net routing in two-
improved algorithm can be described as followings for the two terminals. The h results are shown in Table 1. It shows that all these
terminal net routing. two algorithms can nd the shortest the path, but there are differ-
ent between them. The Number of successful runs and time of the
Step1 Initialize the routing graph and give the initial pheromone improved ant colonies are improved. The efciency of it is improved
value to the net nodes; about 16.667%.
Step2 Give the start point and the end point. Strengthen the value
of the pheromone in the minimum rectangular area which 5. Experimental results and analysis
covers all the terminals;
Step3 All the ants are placed at the start point S; 5.1. Parameter selection
Step4 Each ant chooses the next nodes follow the rules in Eqs. (6)
and (7). Check if the node in the Tabu list A. If it does not, The two-terminal net routing in the non-uniform grid is investi-
make the local pheromone updating, then check the value gated rstly. In the ant colonies system, the parameters are selected
of the pheromone on this node. If it exceeds the upper limit according to the actual situation. Now, we can set
ph max, put the node into Tabu list. If this node has already 1
a = 1, = 1, ph0 = (n dis(start, end)) , n = 20,  = 0.1 q0 = 0.5.
been in the Tabu list, make sure weather put the node into
the set of optional nodes; The comparative importance of the experience and searching
Step5 Repeat, until the ant gets the end point or the set of optional for new paths can be decided by q0 . Its value can affect the experi-
nodes is null; mental results greatly.
Step6 Check if all the ants have started. If they do not, return to Fig. 3 shows the different searching characteristics when q0
Step4; otherwise, record the shortest path; chosen different values. The results show that when q0 = 0.1, the
Step7 Put the shortest path in this repeat into the Tabu list B. We pheromone cannot be accumulated and reserved effectively, so the
decide weather make the global updating according to the searching cannot get a convergence though the optimal solution
information in the Tabu list; can be obtained. And, when q0 is equal to 0.9, the searching char-
Step8 Check if it meets the condition for ending. If it does not, acteristics shows that the q0 is too large to get an effective search.
return to Step3; otherwise, end; In this situation, the pheromone is accumulated so quickly that the
Step9 Paint the shortest path in the graph. search converged to the local optimal solution in early stage. But

Fig. 3. Searching characteristics of the ant colonies searching system.


380 P. Yang et al. / Applied Soft Computing 21 (2014) 376381

Fig. 4. Shortest path searched by ant colonies system and Tabu-ant colonies.

when q0 is equal to 0.5, the searching characteristics shows that


a better searching results can be obtained. Hence, the parameter Fig. 6. Multiple-terminal net routing (six-terminal for example: A, B, C, D, E, F).
q0 is set as 0.5 [17]. Under these parameters, the shortest path of
two-terminal net routing is searched based on the ant colonies and 5.3. Examples about multi-pleterminal net routing, multilayers
Tabu-ant colonies shown in Fig. 4. net routing and gridless net routing

Multiple-terminal net routing can be processed as a complex


5.2. Comparison between some searching results form of two-terminal net routing. Fig. 6 shows a six-terminal net
routing. The course can be described as follows:
The comparison between some searching results can be imple-
mented by using the depth-rst searching system, the ant colonies (1) All terminals (A, B, C, D, E, F) can be regarded as a set S.
system and Tabu-ant colonies system respectively for net rout- (2) We choose two terminals randomly from set S (for example B
ing problem. The results show that all these three methods can and C) and route them. So, it became a two-terminal net routing
complete and nd the shortest path, but they are different in the problem which we have discussed in previous paper. The path
convergence shown in Fig. 5. from B to C can be found. Then, terminals B and C are deleted
The convergence time of ant colonies is 180 s, the number of suc- from S.
cessful runs of ant colonies is 300 (times); while the convergence (3) All the nodes on the path from B to C (B, p1, p2, C) can be
time of Tabu-ant colonies is 150 s, the number of successful runs regarded as a set T.
of Tabu-ant colonies is 220 (times). The results show that all the (4) We can get a terminal randomly from the residual nodes of set
ant colonies system and Tabu-ant colonies system can nd out the S (for example node A). Route A to T. This can also be regarded
shortest path, but it reduces that the convergence time and number as a two-terminal net routing problem (A as start point and set
of successful runs of the Tabu-ant colonies. The comparison shows T as end point). Then put A into set T and delete A from set S.
that the Tabu-ant colonies system can get a more effectively search- (5) Repeat (4) until all the terminals are included in set T.
ing characteristics than other algorithms. Because the pheromone
in the effective area for routing is reinforced, the convergence rate The multi-layers net routing result by using the intelligent Prim
in the early stage is greatly improved. So it avoids the blindness of algorithm is shown in Fig. 7. Fig. 7(a) shows the rst layer result
ant colonies system in the early stage. In the meantime, it can avoid and Fig. 7(b) the second layer result. The results show that the
the local optimal solution effectively by combing with Tabu search Tabu-ant colonies method is an effective means for multi-layer
algorithm. So it can get a convergence to the global optimal solution net routing. Tabu-ant colonies system can be applied to the grid-
after 220 repeats. This is better than ant colonies system. Besides, less graph model. Fig. 8 shows two-terminal net routing in gridless
the time is also shorter than that of the ant colonies system. graph model. The results show that the Tabu-ant colonies method

Fig. 5. The convergence characteristics of different searching methods.


P. Yang et al. / Applied Soft Computing 21 (2014) 376381 381

Acknowledgments

The authors would like to acknowledge the support of National


Natural Science Foundation of China (61076098), the support
of National Natural Science Foundation of China (51165004,
51275182), the support of Natural Science Foundation of Gangxi
Advanced Manufacturing Key Laboratory (13-051-09-013K), the
Special Natural Science Foundation for Innovative Group of Jiangsu
University during the course of this work.

References

[1] A.B. Kahng, A Steiner tree construction for VLSI routing, in: Proceedings.
IJCNN-91-Seattle: International Joint Conference on Neural Networks, 1991,
Fig. 7. Multilayer net routing based on the Tabu-ant colonies system (two layers for
pp. 133139.
example). [2] R. Sardar, R. Mondal, T. Samanta, Geometry Independent Wire-
length. Estimation Method in VLSI Routing[C]//VLSI Design, 2013, pp.
257261.
[3] Y. Zhang, C. Chu, Regular route an efcient detailed router applying regular
routing patterns, IEEE Trans. VLSI Syst. 21 (9) (2013) 16551668.
[4] T.-H. Wu, A. Davoodi, J.T. Linderoth, GRIP: global routing via integer program-
ming, IEEE Trans. Comput. Aided Des. Integr. Circ. Syst. 30 (1) (2011) 7284.
[5] F.O. Hadlock, A shortest path algorithm for grid graphs, Networks 7 (1977)
323334.
[6] E.P. Huijbregts, J.A.G. Jess, General gate array routing using a k-terminal net
routing algorithm with failure prediction, IEEE Trans. VLSI Syst. 1 (4) (1993)
473481.
[7] Y. Kohira, S. Suehiro, A. Takahashi, A fast longer path algorithm for routing grid
with obstacles using biconnectivity based length upper bound, in: Proceedings
of the Asia and South Pacic Design Automation Conference, ASP-DAC, 2009,
pp. 600605.
[8] T.W. Liao, R.J. Kuo, J.T.L. Hu, Hybrid ant colony optimization algorithms for
mixed discrete-continuous optimization problems, Appl. Math. Comput. 219
(6) (2012) 32413252.
[9] B. Dogan, M. Korrek, A new ECG beat clustering method based on kernelized
fuzzy c-means and hybrid ant colony optimization for continuous domains,
Appl. Soft Comput. J. 12 (11) (2012) 34423451.
[10] A. Recski, D. Szeszlr, Routing vertex disjoint Steiner-trees in a cubic gird and
connections to VLSI, Discrete Appl. Math. 155 (1) (2007) 4452.
[11] S. Stepanenko, B. Engels, New Tabu Search based global optimization methods
outline of algorithms and study of efciency, J. Comput. Chem. 29 (5) (2008)
Fig. 8. Gridless net routing based on the Tabu-ant colonies system. 768780.
[12] M.J. Lin, A. El Gamal, A low-power eld-programmable gate array routing fabric,
IEEE Trans. VLSI Syst. 17 (10) (2009) 14811494.
can be used for the multi-terminals, multi-layers and gridless graph [13] M.A.R. Chaudhry, Z. Asad, A. Sprintson, J. Hu, Efcient rerouting algorithms
models in the net routing. for congestion mitigation, in: Proceedings of the 2009 IEEE Computer Society
Annual Symposium on VLSI, ISVLSI, 2009, pp. 4348.
[14] M.J. Borah, R.M. Owens, M.J. Irwin, A fast and simple Steiner routing heuristic,
6. Conclusions Discrete Appl. Math. 90 (13) (1999) 5167.
[15] P. Yang, K. Zeng, A high performance approach on mechanism isomorphism
identication based on an adaptive hybrid genetic algorithm for digital intelli-
The net routing design is an important technique to achieve an gent manufacturing, Eng. Comput. 25 (4) (2009) 397403.
effective conceptual design for VLSI physical design. During the [16] P. Yang, X. Qin, A hybrid optimization approach for chip placement of multi-chip
process of nding the optimal scheme by using some method, it module packaging, Microelectron. J. 40 (8) (2009) 12351243.
[17] W. Qiu, P. Yang, An intelligent multi-layer net routing method based
shows a NP-complete problem. The aim of this paper is to provide
on minimum spanning tree, Int. J. Mater. Struct. Integr. 2 (4) (2008)
a systematical method to intelligent net routing approach for very 332344.
large-scale integration (VLSI) physical design. Based on analyzing [18] A. Recski, Minimax results and polynomial algorithms in VLSI routing, Ann.
Discrete Math. 51 (1992) 261273.
the working characteristics of the VLSI, a Tabu-ant colonies sys-
[19] M. Dorigo, V. Maniezzo, A. Colorni, Ant system: optimization by a colony of
tem is developed by combining with the Tabu search algorithm for cooperating agents, IEEE Trans. Syst. Man Cybern. B 26 (1) (1996) 2941.
nding the optimal scheme of net routing in VLSI physical design. [20] M. Dorigo, L.M. Gambardella, Ant colony system: a cooperative learning
A non-uniform grid graph and a gridless graph are designed as approach to the traveling salesman problem, IEEE Trans. Evol. Comput. 1 (1)
(1997) 5356.
the routing modes. The related parameters are discussed and suit-
able values are found. The mixed algorithm shows it can get a Yang Ping is currently a professor in Jiangsu University in China, also is currently a
director of China Precision Machine Society and a senior member of Chinese Insti-
higher convergence rate than ant colonies algorithm or Tabu search tute of Electronics. He received his Ph.D. in mechanical engineering from Huazhong
algorithm. In addition, it can effectively avoid local optimum. Fur- University of Science & Technology (HUST) in 2001. He engaged in sciences research
thermore, the performances of the algorithm can be illustrated by in Concordia University. His research interests focus on the theoretical aspect and
CAD of mechanical system for the purposes of design and control. He has authored
comparing with other algorithm with some examples. Computer
over 70 professional and scholarly publications in famous international journal in
experiments show that the new algorithm is a valid algorithm to the very specialized eld of the theoretical aspect and CAD of mechanical system
solve the net routing problem. for the purposes of design and control.

You might also like