Fpga based highly reliable fault tolerant approach for network on chip (noc)

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Special Issue: 02 | Mar-2014 | ETCAN-2-14, Available @ http://www.ijret.org 21
FPGA BASED HIGHLY RELIABLE FAULT TOLERANT APPROACH
FOR NETWORK ON CHIP (NOC)
Jehosheba Margaret.M1
, Mary Susanna. M2
, Rajapirian.P3
1
Student, VLSI Design, Kings college of Engineering, Thanjavur, Tamilnadu, India. E-mail:jehosh17@gmail.com
2
Faculty, Department of Electronics and Communication Engineering, PRIST University, Kumbakonam, Tamilnadu, India
3
Faculty, Department of Electronics and Communication Engineering, Kings College of Engineering, Thanjavur,
Tamilnadu, India
Abstract
Network on chip is an interconnection between several processing elements and routers. There are several possibilities for the
occurrence of faults within the network. These faults degrade the performance of the network. In order to increase the
performance several fault tolerant methods has been used. They involve themselves in rerouting and hence take longer paths. To
make the path shorter, the router architecture has to be modified. The introduction of minimal routing algorithm for faults in the
network increases the overall performance of the network. When we use the algorithm, it takes shortest path regardless of the
presence of faults. The proposed algorithm is much simpler than the previous existing algorithm. It provides link among the
surviving routers in the network. It proves to be more efficient even in the presence of multiple faults. It has the ability to connect
the routers both horizontally and orthogonally even in the presence of faults. It has reduced delay over the network. It has proved to
be more reliable of 99.5% when multiple faults are found in the network. It can be used in 4x4 mesh topology with six faulty
routers to analyse the reliability of network. It also helps to estimate the various parameters such as reliability, latency, speed,
area and power.
Keywords- Network on Chip, faulty router, fault tolerant algorithm, chip network, faulty link.
-----------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
The recent technology in VLSI has made the chip to be
much smaller in size. To make it possible, the inner
modules and their interconnections are brought into small
size. Networks-on-Chip (NoC) has emerged as a promising
solution for on-chip interconnection in Multi Core NoCs
due to its scalability, reusability, flexibility, and parallelism
[1] [2] [3]. NoC consists of Network Interfaces, Routers,
set of links interconnecting the Routers and a defined
communication protocol for IP core interaction [4]. NoC
architecture design follows the computation architecture
design, which partitions a behaviour model and maps it
onto an computation architecture model [5]. The first step
is communication modelling and analysis. The second step
is topology and protocol design based on the
communication requirements. The aim of this step is to use
the least network resource to fulfil the communication
requirements showed by the communication pattern. To
estimate the delay on each interconnection in term of clock
cycles and analyse power and area of a NoC design, the
chip floor plan should be estimated in the third step to get
some design requirements for interconnections. The fourth
step is the performance analysis which simulates the NoC
architecture. In the fifth step, logic design can help increase
the accuracy of power and area analysis, but the
interconnection design is the centre of the analysis.
Faults refer to the failure of the system. Transient and
permanent faults are two different types of faults that can
occur in on-chip networks [6]. Transient faults are
temporary and unpredictable. They are often difficult to be
detected and corrected. Permanent faults are caused by
physical damages such as manufacturing defects and device
wear-out. These faults should be recovered or tolerated in a
way that the network continues functioning. A
deterministic routing algorithm uses a fixed path for each
pair of nodes resulting in increased packet latency
especially in congested networks [7]. In adaptive routing
algorithms, packets are not restricted to a single path when
traversing from a source to a destination router [8].
Moreover, non-minimal methods are usually more complex
with a larger number of virtual channels. These algorithms
are used in order to reroute packets around faults [9].
Some fault-tolerant algorithms are proposed to support
special cases of faults, such as one-faulty routers, convex or
concave regions. These algorithms either disable the
healthy components or require a large number of virtual
channels to avoid deadlock. There are other fault-tolerant
approaches [10] [11] [12] which do not require any virtual
channels. A common behaviour in fault-tolerant approaches

__________________________________________________________________________________________
is that packets are routed normally in the network until they
face a fault. At this time, turn models or other techniques
are used to reroute packets around the faults such that no
cycles be formed in the network.
In this paper, we present a new approach to tolerate
multiple faults in the network. This approach is termed as
Connection Retaining Fault tolerant approach which works
more effectively under faults. When a router becomes
faulty, the router is considered as a link to send the data
from source router to destination router. If a link is faulty, it
changes its direction so that the data reaches its destination
router.
The rest of this paper is organized as follows: Section II
reviews the related work. Basic details of the router are
given in Section III. The proposed fault-tolerant routing
algorithm is presented in Section IV. Multiple faulty cases
are discussed in Section V. The results are investigated in
Section VI. Finally the paper is concluded with a summary
in the last section.
2. RELATED WORK
There are few papers discussed to deal with faults in links
[12][13] and routers[11][14]. Few methods never use
virtual channels[9][10][11][15]. Some methods like
FTDR[17] and HARAQ[18] use routing table. Hence they
require large area [16][17]. Few approaches do not use
routing table [18] but they take non minimal path. The
routing algorithm scans the position in descending order
and selects its position. BFT-NoC[2] tolerate faulty links by
dynamic sharing of surviving channels. The algorithm [12]
tolerate single fault. Another algorithm [14] tolerates two
faulty links in the network. DBP [19] approach uses a
lightweight approach to maintain the network connectivity
among non-faulty routers. Here the connections are
embedded inside the chip. The turn based model is used to
avoid deadlock conditions [12][15][16]. The turn-based
routing methods have lower hardware costs but also lower
performance because of having less adaptivity.
The connection retaining algorithm has the following
advantages:(i) It provides shortest path in the network. (ii)
It requires knowledge about neighbouring router. (iii) It
consists few lines of codes. (iv) requires only two virtual
channels. (v) When router is faulty, it is considered as link.
(vi) High reliability. (vii) Does not convey extra
information about the routing process. Reachability is
another important factor as it consists of reachable faulty
neighbours.
2.1 Basic Router
Routers are considered to be the most important blocks in
network on chip. Routers consist of few main components
like buffers, ports, crossbar switch, flow control and arbiter.
There are four ports indicating the four directions namely
North, South, East and West. A Local port is always
connected to the processing element. In a XY network,
each input pair is connected to the output pair.
Fig.1. Basic Router
The type of router used in this method is double-Y network.
It provides adaptiveness with minimum number of virtual
channels. The network is partitioned into two sub networks
namely +X and -X. Eastward channel route through the +X
part in Y-direction whereas Westward channel route
through the -X part in Y-direction.
Fig 2. Router in Double-Y network
Flow control is a mechanism decides how the network
resources are allocated in the network. Buffers hold the
data in the router which are stored temporarily. Buffers are
provided for the four different directions namely North,
South, West and East.
3. ARCHITECTURE
3.1 Router Architecture
Faults are considered to be failure of a system. There are
several possibilities for the occurrence of faults in the
network. Since NoC consist of link, core and router, it may
lead to link fault, core fault and router fault respectively.
The network continues to work in case of link fault and
core fault but fail to work when the router fails. Faults in
the link can be rectified by the dynamic sharing of the

__________________________________________________________________________________________
remaining channels. When a router fails, it cannot send or
receive the data from one location to another. It causes
entire system to fail leaving the core and link as faulty. The
core can perform its regular operation only when the fault
is recovered.
Fig.3. Modified router architecture
In this proposed method, the router architecture can still be
used even in the presence of faults. In the normal router
architecture, it consists of input buffers, routing unit, virtual
channel allocator, switch allocator and crossbar switch. In
this method, the architecture is modified with reference to
the normal architecture. It needs extra multiplexer,
demultiplexer, link checker and wiring connections. When
the router is faulty, the west input channel is directed
towards the east output channel while the east input channel
is directed towards the west output channel. In the same
way, the north input channel is directed towards the south
output channel while south input channel is directed
towards the north output channel. It needs extra wiring and
costs overhead.
The above architecture is used in 4x4 mesh topology. It can
tolerate multiple number of faults in the network by acting
as a link with its neighbouring routers.
Fig 4 4x4 Mesh topology network with fault and their
result after applying algorithm
3.2 Fault Tolerant Algorithm In Router Architecture
In this section, the fault tolerant algorithm is used to
tolerate the number of faults present in the entire network.
It supports the faulty router by taking non minimal paths.
Due to the characteristics of adaptiveness, the data chooses
different path by bypassing the faulty router to reach the
destination router from the source router. By default all the
data are routed in the Y-direction of the network. If the
neighbouring routers are faulty in the Y-direction then the
data is routed in the X-direction.
If there are three routers in the network with one as faulty
then the data is transferred by bypassing the faulty router.
In case of six routers in the network with one faulty router
in the Y-direction of the source router then the data is sent
in the X-direction. After selecting the path, it further sends
the data to the next router to reach destination. If the next
hop router is faulty, it is considered as a link. Finally, it
reaches its destination router. If the destination is two hops
away from the source then the data is routed to the non
faulty router first. It tries the best to avoid routing through
the faulty router.
Suppose, if both the direction are non faulty then it moves
through the less congested area. If the network is designed
to tolerate a few numbers of faults, packets can be routed
adaptively in the network as long as the remaining distance
along both directions is equal or greater than two hops. To
sum up, it is guaranteed that all situations of a single faulty
router are covered by the retaining algorithm taking only
the shortest paths between each pair of source and
destination routers. In addition, fully adaptiveness is
provided when the packet is not close to the destination
router (i.e. the packet is close to the destination when the
distance along one dimension is at most one hop).
However, for better reliability, the adaptiveness of
algorithm can be limited. If the distance is equal in both
directions then the data is sent in either way.
Fig 5 Tolerating one fault

__________________________________________________________________________________________
Moreover, this algorithm requires only the information
about the adjacent neighbouring routers. So, it requires low
area overhead. The presence of faults occurs mostly in
additional resources and less used resources. This makes
the reliability to be more efficient.
3.3 Link Checker
Link checker is used to check the connection (i.e., link)
between the two adjacent routers. In a network, all routers
are well connected to each other in a well established
manner. There are several possibilities for the failure of
link. Care must be taken so that these failures do not affect
network performance.
Fig 6 Link checker with router
If a link fails, the transfer of data is stopped. It is necessary
to have the information about the status of the links in order
to complete the data transfer from source router to
destination router. The usage of link checker prevents
performance degradation in the network.
4. RELIABILITY OF ROUTERS
In this section, it deals how the faulty routers are tolerated
in the presence of two or more faults. When the number of
faults is increased, it undergoes several problems in the
network. The solution can be brought by analysing the
probabilities of faults occurrence.
4.1 Reliability for Two Faulty Routers
Fig. 7.Tolerating two faulty routers
Reliability estimation is carried out under two possible
conditions in the network.
Reliabilty1:
By using the fault tolerant approach how the data is
successfully sent under the existence of faults in the network.
At first, we have to calculate the number of total
combinations of two faulty routers in the network. Then, we
have to measure the number of combinations in which two
faults occur in diagonal positions. By dividing these two
numbers, the reliability value is obtained. The number of
different combinations of two faulty switches in an nxn mesh
network can be measured by using the formula,
N(all combinations):
N=(n2
(n2
-1))/2 (1)
N(diagonal combinations):
N=2(n-1)2
(2)
Reliability1=1-4((n-1)2
/n2
(n2
-1)) (3)
Reliability2:
The estimation of reliability considered when the data is sent
over the faulty router. Only the healthy routers send the data
to another router. Any faulty router cannot send the data but
only can receive them. The total number of data per
combination is given by,
N(delivered per combination)=(n2
-2)(n2
-3) (4)
Total=N(deliveredpercombination)*N(all combinations).
Total number of data defeated,
Defeated=2X N(diagonal combinations)=4(n-1)2
(5)
Reliability2, R2=1-(Defeated/Total).
4.2 Reliability for Multiple Routers
Reliability check is made for multiple router faults. The
number of faults in the routers cannot be increased for
small size networks. Atleast one non faulty router must
present in the network. As the number faulty routers
increases, the performance decreases. Even under these
severe fault conditions, the rest of the packets can be routed
to their destinations through the shortest paths. The focus of
this work is to tolerate faults by using the shortest paths
without any performance loss. However, non minimal paths
or virtual channels can be used to support the remaining
cases.

__________________________________________________________________________________________
Based on the above two methods the reliability of the router
is estimated.
Fig. 8 Tolerating multiple faulty router
5. EXPERIMENTAL RESULTS
To evaluate the performance of the network, an NoC
simulator is used along with the VHDL codes is applied
over the entire components of the Network on Chip. The
data width is set to 32 bits for all routers. Each input buffer
can accommodate 8 flits in each virtual channel. As a
performance metric, we use latency defined as the number
of cycles between the initiation of a message issued by a
Processing Element (PE) and the time when the message is
completely delivered to the destination PE.
The performance estimation of this connection retaining
approach is compared with two methods, called
Reconfigurable Switch algorithm [13] and Reconfigurable
Links algorithm [14]. In the other methods, packets may
take unnecessary longer paths to reach destinations.
Reconfigurable Switch algorithm requires one virtual
channel along each dimension and is able to tolerate all
single faulty routers. Reconfigurable Links utilizes two
virtual channels to support all two faulty routers.
5.1 Reliability for Uniform Traffic
In the uniform traffic profile, each processing element (PE)
generates data packets and sends them to another PE using
a uniform distribution [20]. The mesh size is considered to
be 4×4. To evaluate the reliability of connection retaining
algorithm, the number of faulty routers increases from one
to six. All faulty routers are selected using a random
function and its performance is evaluated.
Fig 9 Reliability Analysis
Reliability is based on the above two performance metric
measurement. When the faults are considered to be
minimum, the reliability is good for all the three methods.
As the number of faults increases, the other two methods
degrade in their performance. Unlike the other methods
connection retaining method holds good under multiple
faults. It is proved to be moreefficient with the reliability
estimation of about 99.5% under multiple faults in the
network.
5.2 Performance Analysis for Uniform Traffic
The average communication latency of Reconfigurable
Switch algorithm and Reconfigurable Links are obtained
under the cases of single and two faulty routers as more
faulty routers are not well supported. The average
communication latency of Connection retaining fault
tolerant algorithm is measured under one to six faulty
routers in the network. To have a fair performance
comparison we use two virtual channels in all three
methods. The extra virtual channels are used for the
performance purposes. In a fault-free network, the
performance of all methods is comparable while
Reconfigurable Links outperforms others because of its
better adaptiveness. As the number of faults increases, the
performance of Reconfigurable Switch algorithm and
Reconfigurable Links significantly decreases. We increase
the number of faulty routers to six faults and measure the
performance of connection retaining method. Surprisingly,
the performance gradually starts growing under the same
traffic load.

__________________________________________________________________________________________
Fig 10 Performance Analysis under uniform traffic
5.3 Performance Analysis for Hotspot Traffic
Under the hotspot traffic pattern, one or more routers are
chosen as hotspots receiving an extra portion of the traffic
in addition to the regular uniform traffic. The performance
Analysis is for Connection retaining algorithm,
Reconfigurable Link and Reconfigurable Switch.
Connection retaining algorithm proves to more promising
under hot spot condition.
Fig 11 Performance Analysis under hotspot traffic
6. CONCLUSIONS
In this paper, we proposed a fault-tolerant approach to
retain the connectivity of the network. In the presented
approach, data are routed through the shortest paths,
maintaining the performance of NoC even in the presence
of multiple faults. The router architecture is slightly
modified to route the data safely between the routers. The
main purpose of the modification in the router architecture
is to maintain the connectivity among the remaining
routers. For this to happen, when a router becomes faulty,
the router links are connected to each other along the
horizontal and orthogonal directions. This method is a very
simple, lightweight, and adaptive approach. It takes only
one andtwo virtual channels along the X and Y dimensions
respectively. Obtaining high reliability by a simple
approach is the conclusion of this work.
REFERENCES
[1] Xu, Jiang, W. Wolf, J. Hankel, S. Charkdhar, “A
Methodology for design, modeling and analysis for
networks-on-Chip,” in Proc. of IEEE International
Symposium on Circuits and Systems, pp. 1778-1781,
2005.
[2] W. Tsai, D. Zheng, S. Chen, and Y.H. Hu, “A fault-
tolerant NoC scheme using bidirectional channel”,
in Proc. DAC, pp.918-923, 2011.
[3] E. Rijpkema et al., “Tradeoffs in the design of a
router with both guaranteed and best-effort services
for networks on chip,” in Proc . Of DATE’03, pp. 350-
355, 2003.
[4] Kangmin Lee, Se-Joong Lee, Hoi-Jun Yoo, “Low-
power network-on- chip for high-performance SoC
design”,press, 2008.
[5] L. Benini, G. De Micheli, “Networks on chip: a new
paradigm for systems on chip design”, Design,
Automation and Test in Europe Conference and
Exhibition, 2002.
[6] M. Ali, M. Welzl, S.Hessler, “A Fault tolerant
mechanism for handling Permanent and Transient
Failures in a Network on Chip,” in Proc. Of
international conference on Information Technology,
pp.1027-1032, 2007.
[7] J. Duato, S. Yalamanchili, L. Ni, “Interconnection
networks: an engineering approach”, Morgan
Kaufmann Publishers, 2003.
[8] M. Daneshtalab, M. Kamali, M. Ebrahimi, S.
Mohammadi, A. Afzali-Kusha, and J. Plosila,
"Adaptive Input-output Selection Based On-Chip
Router Architecture," Journal of Low Power
Electronics (JOLPE), Vol.8, No. 1, pp. 11-29, 2012.
[9] F. Chaix, et al., “A fault-tolerant deadlock-free
adaptive routing for On Chip interconnects,” in Proc.
of DATE, pp. 1-4, 2011.
[10] S. Jovanovic, C. Tanougast, et al., “A new
deadlock-free fault- tolerant routing algorithm for
NoC interconnections”, in Proc. of FPL, pp.326-331,
2009.
[11] J. Wu and D. Wang, “Fault-tolerant and deadlock-free
routing in 2- Dmeshes using rectilinear-monotone
polygonal fault blocks”, in Proc. Of Parallel
Algorithms., pp.99-111, 2005.
[12] Z. Zhen, A. Greiner, S. Taktak, “A reconfigurable
routing algorithm for a fault-tolerant 2D-Mesh
Network-on-Chip”, DAC, pp. 441-446, 2008.
[13] M. Ebrahimi et al., “MAFA: Adaptive Fault-Tolerant
Routing Algorithm for Networks-on-Chip,” in Proc.

__________________________________________________________________________________________
of DSD, pp. 201-206, 2012.
[14] M. Valinataja, S. Mohammadi, J. Plosila, P. Liljeberg,
H. Tenhunen, “A reconfigurable and adaptive
routing method for fault-tolerant meshbased
networks-on-chip,” International Journal of
Electronics and Communications (AEU), v. 65, I.7, pp.
630-640, 2011.
[15] D.Fick, A. DeOrio, G. Chen, V. Bertacco, D.
Sylvester and D. Blaauw,“A highly resilient routing
algorithm for fault-tolerant NoCs,” Design,Automation
and Test in Europe (DATE), pp. 21-26, 2009.
[16] M. Valinataj, S. Mohammadi, S. Safari and J. Plosila,
“A link failure aware routing algorithm for
Networks-on-Chip in nano technologies,”IEEE
Conference on Nanotechnology, pp. 841-844, 2009.
[17] Ch. Feng, Zh. L, A. Jantsch, J. Li, M. Zhang, “A
Reconfigurable Fault tolerant Deflection Routing
Algorithm Based on Reinforcement Learning for
Network-on-Chip”, in Proc. of NoCArc, 2010.
[18] M. Ebrahimi, M. Daneshtalab, F. Farahnakian, P.
Liljeberg, J. Plosila, M. Palesi, and H. Tenhunen,
"HARAQ: Congestion-Aware Learning Model for
Highly Adaptive Routing Algorithm in On-Chip
Networks," in Proc of NOCS, pp. 19-26, 2012.
[19] M. Koibuchi, H. Matsutani, H. Amano, T.M.
Pinkston, “A Lightweight Fault-Tolerant Mechanism
for Network-on-Chip”, in Proc. of NoCS, pp.13-22,
2008.
[20] C.J. Glass et al., “The Turn Model for Adaptive
Routing”, in Proc. Of 19th Int'l Symp. Computer
Architecture, pp. 278-287, 1992

Fpga based highly reliable fault tolerant approach for network on chip (noc)

Recommended

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Fpga based highly reliable fault tolerant approach for network on chip (noc) (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Fpga based highly reliable fault tolerant approach for network on chip (noc)