See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328244357

VLSI Cell Partitioning Using Data Mining Approaches

Article in International Journal of Computer Sciences and Engineering · August 2018

DOI: 10.26438/ijcse/v6i8.10191027

CITATIONS READS
0 389

2 authors:

Suryakanta Nayak Mrutyunjaya Panda

4 PUBLICATIONS 8 CITATIONS
Utkal University
150 PUBLICATIONS 1,797 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Mrutyunjaya Panda on 18 March 2019.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Sciences and Engineering Open Access
Research Paper Vol.-6, Issue-8, Aug 2018 E-ISSN: 2347-2693

VLSI Cell Partitioning Using Data Mining Approaches

Suryakanta Nayak1*, Mrutyunjaya Panda2

1
Ph. D Scholar, P.G. Dep. of CSA, Utkal University, Vani Vihar, Odisha, India
2
P.G. Dep. of CSA, Utkal University, Vani Vihar, Odisha, India
*
Corresponding Author: [email protected]

Available online at: www.ijcseonline.org

Accepted: 15/Aug/2018, Published: 31/Aug/2018

Abstract- Theoretical studies on various cell partitioning methods are lucidly presented in the current research pertaining to
design and development of VLSI circuits. Owing to the difficulties in designing complex VLSI systems, it is extremely crucial
to partition the large circuit into tiny logic blocks to reduce time complexity, space complexity and power consumption. To
envisage the same, this communication scrutinizes a heuristic technique by using various data mining algorithms such as K-
means algorithms, K-Nearest Neighbor (K-NN), Fuzzy c-means and Support Vector Machine (SVM) for resolution of
complexity in VLSI circuits, where K-NN and SVM are employed for classification purpose and Fuzzy c-means and K-means
methodologies are deployed for clustering purpose. The upshot of the research revealed that K-NN and Fuzzy c-means methods
bestow optimum result pertaining to VLSI cell partitioning.

Keywords: K-means algorithms, K-nearest neighbor, Fuzzy c-means, The Support Vector Machines, Partitioning, and Data
mining

I. INTRODUCTION

The main challenges in development of a VLSI circuit lies in design of efficient cell partitioning into circuit components. Towards
achieving this objective, several parameters such as: (a) fabrication of the circuit with low-power and (b) high-speed computing
structures allotted for different applications needs to be addressed. The highly designed VLSI circuit must satisfy several
constraints that include: circuit area, no of pins, power consumption, noise, and throughput [1]. The physical architecture of
VLSI comprises of three prime prospects: Placement, Routing, and Partitioning. Placement comprises the process of appointing
locations to the constituents within the area. The method of interconnecting the placed constituents, subject to the restraint of
less wiring, is called as Routing. Separation of a complex system structure into the sub- components, to reduce the complexity,
is described as partitioning.

The objective of the partitioning is to separate the complex circuit into numerous components such that designing these
components individually and then combining these separately, minimizing the complication level of the whole design to a much
larger extent. The algorithms of Data mining present a means to investigate the data and extricate information from them.
Incorporating data mining methodologies into the VLSI partitioning gives a new dimension to the research domain. The Data
mining methodologies generally falls into two categories: Clustering algorithms and Classification algorithms. Both these kinds
find their usefulness in recognizing the hidden forms in data. This arrangement necessitates in forecasting a firm conclusion
based on the given input. In order to foresee the result, the algorithm exercises a training set that carries a group of attributes and
their corresponding result, usually known as prediction or goal attribute. The algorithm endeavors to govern relations among
the attributes that makes it practicable to forecast the result.

Cluster study is distinguished as a process of generating batch of clusters, or objects in a way that identical objects are located
in one cluster and the objects in disparate clusters are divergent [2].The prime rewards of cluster methodologies is that it may be
applied to a defined objective standard regularly to form categories. The consistency and speed of the clustering mechanism in
arranging the data together form an alluring reason to utilize clustering in the partitioning.
The rest of the paper is organized as follows. Section 2 elaborates the four algorithms of data mining in detail. Section 3
describes the implementation of different data mining algorithms in VLSI cell segregating. The outcomes are narrated in
Section 4. Finally, Section 5 concludes with future scope of research.

© 2018, IJCSE All Rights Reserved 1019

 International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

II. RELATED WORK

The authors [3] discusses about the available optimization algorithm addressing to VLSI large-scale partitioning, floor planning
and mixed-size placement problem by adopting multilevel hyper graph, so that the complexity is reduced to a great extent.
Darwin’s theory of survival of the fittest principle is used by the authors [4] for circuit partitioning, to have number of sub-
circuits, calculates the fitness value and discard the solutions that results in low fitness value. The authors [5] discusses about
the hold and jump process to improve the throughput in VLSI testing. The paper [6] presents a firefly algorithm for the partition
driven global cell placement in an efficient way and compared with its genetic counterpart. The authors [7] introduces hybrid
PSO and SA algorithm - a new circuit bi- partitioning algorithm for efficient reduction in the number of interconnections
among the VLSI circuit elements. A new interconnection oriented clustering algorithm is proposed by the authors [8] of
combinational VLSI circuit partitioning, by capturing clusters considering highly interconnected ones, in a circuits.

ALGORITHM DISCUSSION
This section briefs about the categorization algorithms: K-means algorithms, K-nearest neighbor, Fuzzy c-means and Support
Vector Machines. Though categorization and Clustering algorithms are believed to be cryptic, their differences are very
incomparable. There are previously defined set of classes in the categorization process and decision has to be taken on allotting
the already existent class to the newly data set. The underlying methodology in clustering technology is to find out some
relation among the data sets and grouping them based on their respective relation.

II.1 CLASSIFICATION ALGORITHMS

The central idea behind classification algorithm isthe selection of a hypothetical set from the existent
sets that best suitable for the given input data set. This segment involves detailed analysis of the identified
data mining categorization algorithms: Support Vector Machine Algorithm and K-Nearest Neighbor
Algorithm.

II.1.1 K NEAREST NEIGHBOUR ALGORITHM

K Nearest Neighbor (KNN) algorithm [9] is an indolent research algorithm. It is also a non-parametric by nature. Suppositions
are not made by taking the underlying data
as the actual data does not pursue any hypothetical suppositions. There is no definite generalization phase or it is very
insignificant. On the basis of the below assumptions, KNN algorithm is accomplished. The algorithm pretends that records are
in cadenced space described based on the distance (The distance is determined by any one of the followingThe data may be in
scalar form or multidimensional form. Each and every data point have notion of distance as they are in feature space. The
training data set holds a vector set and a class tag interrelated to the vector. However KNN works fine with arbitrary number of
classes. A number "k" which signifies the number of adjacent that are methods: Mahalanobis Distance, Murkowski Distance
and Euclidean Distance. The algorithm is simply known as the nearest neighbor algorithm if k=1.

II.1.2 SUPPORT VECTOR MACHINE ALGORITHM

A Support Vector Machine algorithm(SVM) [10] builds a model that authorizes stated data set inside one group or the other
group, from that the training data sets are belonged. So we may call that SVM algorithm is a bilinear classifier having non-
probabilistic property. Data are produced in spatial domain, thus the training data set of various categories are partitioned by a
distinct distance. Consequently the possibility of mismatching the stated data setis mostly reduced. But the

3.2.1 FUZZY C-MEANS Algorithm

In the fuzzy clustering, every data stage in the data set is associated with every other cluster by using a relationship function.
The algorithm of c-means permits overlapping of data set, i.e. the stated data set may come from more than two clusters with the
same duration. LetY=
{y1,y2, . . . , } be a set ofnumerical datain Let j be an integer, 1 < j<M. Given X, we say that j fuzzysubsets
{ :Y→ [0, 1]} are a j-partition of Y if the following conditions are satisfied[]:

0≤ for all k andl ...........................(1)

foralll ............................................................. (2)
0 < n forallk ............................ (3)

© 2018, IJCSE All Rights Reserved 1020

 International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

where = 1 ≤ k ≤ j and 1≤l≤M. Let the j M values filling the above

condition be arrayed as j×Mmatrix A= [ ].Then the set of all such matrices are the non-degenerate fuzzy j-partitions
of Y:
SVM algorithm apprehends more CPU memory and time than the KNN algorithm. The SVM algorithm deteriorates a serious
blow as it is directly relevant only in two classes and there is no probability of calibrated class relationship.

3.2 CLUSTERING ALGORITHMS

The objectives behind the clustering algorithms are to categorize objects into clusters or various subsets based on their defined
relationship with each other. Data in the same cluster will have similar attributes whereas those in separate clusters will have
dissimilar attributes. The algorithms explained under this segment are Fuzzy C means algorithm along with K means algorithm.

If all are either 0 or 1,we have the subset of hard j partition of Y

= =0or1
This Fuzzy j-segregations is treated as Fuzzy C-segregation. Fuzzy clustering is utilized in the mining complex along with
multi- dimensional datasets. The algorithm Fuzzy- C-means (FCM) is widespread, in that a chunk of data has limited relation
with each of the pre-specified cluster centers [5].

This Fuzzy j-segregation is treated as Fuzzy C-segregation. Fuzzy clustering is utilized in the mining complex along with multi-
dimensional datasets. The algorithm Fuzzy-C- means (FCM) is widespread, in that a chunk of data has limited relation with
each of the pre- specified cluster centers [11].

3.2.2 K- MEANS ALGORITHM

The K means algorithm is devised to cluster various data on the basis of the cluster centre known as mean. The summation of
clusters kis pretended to be static. The algorithm starts for primary k number of clusters, by designating the residual data to the
adjacent clusters and continuously modifying the relationship of various clusters between each other till the membership
becomes stable. It has two phases: Initialization and Iteration phases. During the former stage, the data is allocated to k clusters.
In the latter stage, the length between the data set along with each cluster is determined and data set is allocated
to the adjacent cluster. But, K means algorithm never employ overlapping of the data sets.

III. IMPLEMENTATIONAND EXPERIMENTALRESULTS

All the experiments are conducted using Xilinx 9.1i, FPGA Board and MATLAB 15 environment, in a Intel Core-2 processor
with 3GB RAM and 80GB HDD. The categorization and Clustering algorithms arecarried out for the beneath
circuit. The doorway at each nodule is characterized by the GM, gate matrix, and the assembly of gates is characterized by an
input matrix, IM. The gate matrix is outlined one column at unit time, with every consequent column holding the gates. The
input of the gates can be from the gates of prior column or wires; so the total number of columns in CGM, GM, is equal to the
utmost number of gates on any path from the circuit input to output [12]. A simple representation of Gate matrix is shown in
Table1.
Table1
NUMBER TYPE
1 AND
2 OR
3 XOR
4 INV
5 WIRE
6 NAND
7 NOR
8 XNOR
9 BUFFER

© 2018, IJCSE All Rights Reserved 1021

 International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

he count of rows in RGM, GM, signifies the cut size of the circuit. The gate in every node is denoted by a number in gate
matrix. The analogous numbers for every gate are mentioned in Table 1.The relation among gates in the gate matrix is
mentioned in input matrix, IM, where the count of columns in the input matrix and the gate matrix are same[12]. Finally, the
count of rows in the input matrix, RIM, is the multiplication of RGM and the utmost fan-in of each gate in the circuit. So, the
encouraging values for the column X are {1+RGM× [X–1], RGM×X} [12].
The quartetalgorithms discussed are enforced to the example circuit. But the matrix is stated as input to the categorization and
clustering algorithms.

System specification

Segmentation
Architectural Design

Planning

Logical Design and

Functional Design Placement

Circuit Design and

analysis Synthesis

Structural Design

Signal Routing

Structural verification

Time closure
Fabrication

Testing

Chip

Figure 1

© 2018, IJCSE All Rights Reserved 1022

 International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

The flow chart of the proposed implementation is shown in Figure1

The input circuit and the corresponding output in terms of gate level view with its look up table are shown in Figure 2, Figure 3
and Figure 4 respectively. Further, a sub-circuit of the input circuit is shown in Figure 5 with its truth table in Figure 6.logical
design placement

Figure 2: Input Circuit for VLSI cell partitioning

Circuit Design and

Structural Design

Structural verification

Figure 3: output Gate level view of the input circuit

© 2018, IJCSE All Rights Reserved 1023

International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

Figure 4: Look up Table for the input circuit

Figure 5: Sub-circuit view of the input circuit

© 2018, IJCSE All Rights Reserved 1024

 International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

Figure 6: Truth Table of the input circuit

IV. RESULT ANDDISCUSSION

The results obtained with the use of K-nearest neighbor and support vector machine, are displayed in Figure 7 and Figure 8
respectively. The K Nearest neighbor shows better categorization result than SVM algorithm as the VLSI cell partitioning
employs numerous classes. The SVM algorithm disappoints to produce efficient outcome while operating for different classes.
Further, the results obtained after using K- means and Fuzzy C-means Clustering algorithms are shown in Figure 9 and Figure
10 respectively, when enforced on the circuit. The objective function is considerably minimized in Fuzzy C-means algorithm
with very less count of repetitions whenever compared to the K-means algorithm. Furthermore, permitting overlapping of the
data sets makes the fuzzy C-means to be more competent than K- means algorithm.

Figure 7: K-NN Algorithm

© 2018, IJCSE All Rights Reserved 1025

International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

Figure 8: SVM Algorithm

Figure 9: K- Means Algorithm

Figure10: FCM Algorithm

© 2018, IJCSE All Rights Reserved 1026

 International Journal of Computer Sciences and Engineering Vol.6(8), Aug 2018, E-ISSN: 2347-2693

V. CONCLUSION

In this research, the VLSI cell partitioning using classification (K-NN and SVM) and clustering (K-Means and FCM) are
examined. It is observed that K -Nearest neighbor provides the optimum result with contemplate

to the VLSI Cell segregating. Further, from the clustering algorithms, it is identified that Fuzzy C-means gives optimum result
to VLSI cell partitioning. Finally, while comparing K-NN and FCM, FCM wins the race being the most efficient ones. In
future, we shall focus on the other efficient algorithms with their hybrids for best VLSI cell partitioning optimization.

REFERENCES

[1] Brian Von Herzen,VLSI Partitioning of a 2-Gs/s Digital Spectrometer, IEEE journal of solid-state circuits, VOL. 26, NO. 5, MAY1991.
[2] Gan, Guojun, Chaoqun Ma, and Jianhong Wu, Data Clustering: Theory, Algorithms, and Applications, ASA-SIAM Series on Statistics and
AppliedProbability, SIAM, Philadelphia, ASA, Alexandria, VA,2007.
[3] RajineSwetha R, B. ShekarBabu, SumithraDevi K.A, A Survey of Various Algorithms for Vlsi Physical Design, World Academy of Science,
Engineering and Technology International Journal of Electronics and Communication Engineering Vol:5, No:3,2011
[4] Sharadindu Roy, Samar SenSarma, improvement of the quality of VLSI circuit partitioning problem using genetic algorithm, Journal of Global
Research in Computer Science, , Volume 3, No. 12, pp. 18-22,2012.
[5] Ramalakshmi and S. Saravanan, Multiple Scan Base Partitioning Technique to Increase the Throughput in VLSI Testing, Indian Journal of
Science and Technology, Vol9(29), DOI: 10.17485/ijst/2016/v9i29/90862, August2016
[6] Maninder Kaur , Pradip Kumar Sharma, On solving partition driven standard cell placement problem using firefly-based metaheuristic
approach, International Journal of Bio-Inspired Computation,9(2), pp.121–127
[7] Shikha Arora, Neerumalhotra,Rajan Vohra, RajdeepSingh, Hybrid Algorithm PSO and SA in Achieving Partitioning Optimization for VLSI
Applications , International Journal of P2P Network Trends and Technology (IJPTT) - Volume 2 Issue 1,pp.1-4,2012
[8] MilonMahapatra, M Malathi and B Srinath. An Interconnectivity based Efficient Partitioning Algorithm of Combinational CMOS Circuits.
IJCA Proceedings on National Conference on VLSI and Embedded Systems NCVES(1):18-21, March2013
[9] Indu Saini, Dilbag Singh, ArunKhosla. Detection of QRS-complex using K-nearest neighbor algorithm, Int. J. Medical Engineering and
Informatics, 5(1),81-101,2013.
[10] Platt J. Fast Training of Support Vector Machines using Sequential Minimal Optimization. In: Scholkopf B, Burges C, Smola A, editors.
Advances in Kernel Methods: Support Vector Learning. Cambridge, MA: MIT Pressn,pp.185–208,1999.
[11] SubhagataChattopadhyay, Dilip Kumar Pratihar, SanjibChandra De Sarkar, A Comparative study of fuzzy c-means algorithm and entropy-
based fuzzy clustering algorithms, Computing and Informatics,vol30,pp. 701–720,2011.
[12] Ganesh K. Venayagamoorthyet al., Particle swarm-based optimal partitioning algorithm for combinational CMOS circuits,Engineering
Applications of Artificial Intelligence (2006), doi: 10.1016/j.engappai.2006.06.011.
[13] Ratnesh Kumar Shukla,An Introduction of Face Recognition and Face Detection for Blurred and Noisy Images ,International Journal of
Scientific Research in Research Paper . Computer Science and Engineering Vol.6, Issue.3, pp.39-43 , June 2018.
[14] T. Dheepak, Low Power Distributed MAC Protocol Against Various Kinds Of Attacks By Using Traffic Analysis Methodology, International
Journal of Scientific Research ,Computer Science and Engineering Vol.6, Issue.3, pp. 1-7 , June 2018

Authors Profile
Suryakanta Nayak has completed M.Tech from Biju Patnaik University of Technology University in 2012.
He is presently pursuing Ph. D. in Utkal University. His areas of interest include Data mining, VLSI Cell
partitioning, He has more than 6 years of teaching experience.

Mrutyunjaya Panda working as Reader in the Computer Science and Applications department of Utkal
University. He has 18 years of teaching and research experience. His teaching and research domains include
Data mining, Soft computing, Digital signal processing, Mobile communication, Social networking, Image
processing, Wireless sensor network, Text mining etc.. He has published 70 research papers in different
reputed journals and conferences. He has also written 2 text books, 3 edited books, 7 book chapters in
different fields.

© 2018, IJCSE All Rights Reserved 1027

View publication stats