TELKOM
NIKA
, Vol.11, No
.1, Janua
ry 2013, pp. 1~8
ISSN: 2302-4
046
1
Re
cei
v
ed Se
ptem
ber 20, 2012; Revi
se
d No
vem
ber
10, 2012; Accepted Decem
ber 3, 201
2
A Joint-Coding Scheme with Crosstalk Avoidance in
Network on Chip
Lei Zhou*
1,2
, Ning Wu
1
, Fe
n Ge
1
1
Colle
ge of Info
rmation Sci
enc
e and T
e
chno
l
o
g
y
, Nan
jin
g U
n
iversit
y
of Aer
o
spac
e an
d Astronautics,
Nanj
in
g, Chin
a
2
Colle
ge of Info
rmation En
gin
e
e
rin
g
, Yangz
ho
u Univ
ersit
y
, Y
angz
ho
u, Chin
a
*corres
pon
di
ng
author, e-mai
l
: tomcat8006
07
@12
6
.com
A
b
st
r
a
ct
T
he reli
ab
le tr
ansfer i
n
netw
o
rk on c
h
ip c
a
n be
guar
ante
ed by cr
osstal
k avoi
danc
e a
nd err
o
r
detectio
n
co
de.
In this
pap
er,
w
e
prop
ose
a j
o
int co
di
ng sc
h
e
me co
mbi
ned
w
i
th crosstalk
avoi
danc
e co
di
n
g
with error contr
o
l coding. The
Fibon
acci
numeral syst
em is
applied to s
a
tis
f
y the requir
e
m
ent of cr
osstalk
avoi
danc
e c
odi
ng, a
n
d
the
err
o
r d
e
tectio
n is
achi
eved
by
a
d
d
in
g p
a
rity
bits. W
e
a
l
so
i
m
p
l
e
m
e
n
t the
co
de
c
in reg
i
ster tran
sfer level. F
u
rt
her
mor
e
, the sche
m
es
of
cod
e
c ap
plyi
ng to f
ault-tol
e
rant ro
uter are a
n
a
l
y
z
ed.
T
he exper
i
m
e
n
t
al result sh
ow
s that "once e
n
c
ode,
mult
i
p
l
e
deco
de" sc
he
me
outperf
o
rms other sche
m
es i
n
trade-off del
ay,
area an
d pow
er.
Ke
y
w
ords
: cro
sstalk avoi
dan
ce, CODEC, fault tolera
nt, netw
o
rk on chip
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The ra
pid scalin
g of techn
o
logy int
o
the deep
sub-micron
regime h
a
s
bee
n
accomp
anie
d
by a d
r
am
atic in
crea
se i
n
transi
s
to
r
de
ns
itie
s
.
Ac
co
rdin
g
to
ITR
S
’s p
r
ed
ic
tion
, up
to 4 billion transi
s
tors
will be integrate
d
into one
ch
ip sin
c
e 20
10
[1]. However, the incre
a
si
ng
den
sities al
so lead to in
crea
sing
po
ssi
bility of
instantaneo
us fau
l
t becau
se
of more
crosst
alk
noises
and leak
age curr
ent. To inc
r
ease the r
e
liability of s
y
s
t
em, the c
r
oss
t
alk
avoidanc
e
and
error dete
c
tio
n
scheme h
a
s
be
come the
critical i
s
sue
s
in Net
w
ork
on Chi
p
(NO
C
) de
sig
n
.
In recent ye
ars, the
r
e h
a
s
bee
n an
e
v
olving effort in error d
e
tection a
nd
correctio
n
mech
ani
sm
s in the com
m
unication
subsy
s
tem,
a
nd cro
sstal
k
avoidan
ce
code
s (CACs) are
con
s
id
ere
d
a
s
effe
ctive scheme to
redu
ce th
e m
u
tual
inter-wi
re
co
upling
capa
cit
ance a
nd
hen
ce
the ene
rgy di
ssi
pation
of wire
segme
n
ts [2]. Yu
et
al [3]. prop
osed an
ada
ptive error
co
ntro
l
method
for
swit
ch-to
-
swit
ch li
nks in
a va
riabl
e
noi
se
envi
r
onm
ent, to
meet
reli
a
b
ility
requi
rem
ents and
achie
v
e ene
rgy-ef
ficien
cy. Sr
inivas [4] et
al propo
se
d bu
s-en
cod
i
ng
techni
que
s that decrea
s
e cro
s
stalk b
e
twee
n wires a
nd avoid adv
ersari
al swit
ching patterns on
the data b
u
s.
Ho
wever, th
e data
sho
u
ld
be divide
d
in
to gro
ups a
c
cordin
g to the
width. Ga
ngu
ly
[5] et al prop
ose
d
joint
cro
sstal
k
avoida
nce
and
triple
-erro
r
-co
rre
cti
on/qua
dru
p
le
-erro
r-d
etecti
on
cod
e
s, an
d their pe
rform
a
n
c
e was evalu
a
ted in
different NO
C fabrics.
Neverthel
ess this codi
ng
scheme
can
applie
d to an
y width of data, the c
ode
re
dund
an
cy rate is larg
er tha
n
others.
In this pape
r,
we propo
se
a joint codi
ng
sch
eme
whi
c
h combin
es
cro
s
stalk
avoidan
ce
codi
ng with e
rro
r control
coding. The m
a
in idea of
th
is sche
me is
to represent dataword
s
int
o
Fibona
cci nu
meral
syste
m
and ad
d parity bits in
t
o
codi
ng, for providing th
e fault detecting
cap
ability a
nd avoi
ding
cro
sstal
k
noise
si
mult
aneo
usly. F
u
rthe
rmo
r
e,
the RTL l
e
vel
impleme
n
tation of
CODE
Cs is offered
.
The
co
de
c
is a
pplie
d to
fault-tolerant
route
r
b
a
sed
on
End-to-E
nd p
r
otocol and P
o
int-to-Poi
nt proto
c
ol,
and
the perform
a
n
ce of these error controlling
scheme
s
is al
so an
alyze
d
.
2. Error Control in NoC Links
The p
r
op
ose
d
co
ding
sch
e
me is
ba
sed
on the comm
only used int
e
rconn
ect a
r
chitecture
Mesh,
a
s
sho
w
n i
n
Fi
gure.1. Each route
r
con
nec
t
s
ne
ighbo
rs in f
o
u
r
di
re
ction
s
.
Data
exch
an
ge
betwe
en the f
unctio
nal bl
o
c
ks ta
ke
s pla
c
e in th
e
form of pa
ckets.
This
schem
e
divides
pa
ckets
into fixed-length flow c
ont
rol units
(flits),as sho
w
n in Figure 2, with
buffers sto
r
in
g only a few flits.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 1 – 8
2
At most 8 flits co
mpo
s
e a
packet, inclu
d
ing on
e hea
der flit and 7
payload
s. Heade
r flit, which
contai
ned
ro
uting info
rma
t
ion, like
sou
r
ce
and
de
stin
ation a
ddress, packe
t l
engt
h, etc, e
nabl
es
the switche
s
to est
ablish
a path
an
d
su
bseque
nt
flits sim
p
ly follow thi
s
pa
th in a
pipeli
ned
fas
h
ion.
Figure 1.
Th
e
Topology of 2D-me
s
h
Figure 2. Structure of Pa
cket
The purpo
se
of the erro
r control me
cha
n
ism
is to de
liver the datawords ove
r
chann
el
reliably. The
mech
ani
sm can be cla
s
sified into two ways: End-to-E
nd proto
c
ol a
nd Point-to-P
oint
proto
c
ol. End
-
to-End
proto
c
ol m
ean
s th
e erro
r c
ontrol only exe
c
u
t
es on
ce
in t
he data
tran
sfer
betwe
en fu
nctional bl
ocks. In Point-to
-P
oint protocol, the e
r
ror co
ntrol
sho
u
ld
be p
e
rfo
r
med
in
every ro
uter
the data
w
ord
s
pa
ss thro
u
gh. In
gene
ral, the ham
ming, pa
rity cod
e
or
CRC is
applie
d to d
e
tect a
nd th
e data
retran
smissio
n
i
s
applie
d to
co
rre
ct the
error. Note that
the
increasing possibility of i
n
stant
aneous fault leads
to the l
o
cal congestion because
of
the
increa
sing n
u
m
ber of pa
cket retran
smi
s
sion.
Cro
s
stalk is t
he mai
n
course of i
n
sta
n
ta
neou
s fault, t
herefo
r
e
co
di
ng data
w
o
r
d
s
by CAC
based on e
r
ror co
ntrol
co
de ca
n red
u
ce the possibil
i
ty
of error effectively. A fe
w of CACs were
prop
osed i
n
literature. He
re
we
con
s
id
er F
o
rbidde
n
Pattern
Co
n
d
ition
(FPC)
cod
e
s a
s
th
e
Cro
s
stalk Avoidan
ce
sch
e
m
e. It wa
s first p
r
opo
se
d i
n
[6]. The
forbidde
n p
a
tterns
are
defin
e
d
a
s
3-bit patterns “101
” and “0
10”. For exa
m
ple, 1100
1
10 ha
s no forbidde
n pattern for there i
s
no
three
co
nse
c
utive bits. It has b
een
sh
o
w
n in [7]
th
at a co
de
whi
c
h
contai
ns
no f
o
rbid
den
pattern
experie
nces
maximum cro
sstal
k of no g
r
eate
r
than 2
0
C.
3. The Joint
Code
Although CA
C
a
nd
E
C
C address dela
y
and relia
bi
l
i
ty individuall
y
, the co
mbi
nation
of
CAC an
d ECC sh
ould
sati
sfy the followi
ng co
nditi
on
s [4]. Firstly, CAC need
s to
be perfo
rme
d
in
first ste
p
be
cau
s
e it i
n
volves n
onlin
ear
and
disruptive map
p
ing from
dat
a to code
word;
Secon
d
ly, ECC nee
ds to b
e
system
atic
to ensu
r
e th
a
t
the redu
ctio
n in tran
sition
activity and the
pea
k couplin
g tran
sition
constraint a
r
e
maintaine
d
.
And la
stly the addition
al pa
rity bits gene
rated
by ECC shou
ld be
en
cod
e
d
by a li
nea
r
CAC to
en
su
re they do
not
suffer from
crosstal
k d
e
la
y.
Acco
rdi
ng to the co
nstraint
s of
above, the con
s
tru
c
tio
n
of joint cod
e
is sh
own as Figure 3.
Nonli
nea
r
CAC is u
s
ed
prio
r to
othe
r en
co
ding
s,
k
bits
data
is
en
cod
e
d
to l bit
s
cod
e
word. Af
ter that the a
dditional m
p
a
rity bits
a
r
e
adde
d to the
cod
e
word to
contai
n the e
rro
r
detectio
n
abil
i
ty,
then the m bits are further en
cod
e
d
by linear CA
C for cro
sstal
k avoidan
ce t
o
obtain mc bit
s
. Total l+m
c
bits are
sent
over the bu
s lastly.
Mutyam [8]
prop
osed
a
bus en
co
din
g
techniq
ue
usin
g a
varia
n
t of bin
a
ry
Fibona
cci
rep
r
e
s
entatio
n as CAC
scheme, whi
c
h
indicate
s
that any n bits vector ca
n be expre
s
se
d by
Fibona
cci ele
m
ents:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Joint-Codi
n
g
Schem
e wit
h
Cro
s
stalk A
v
oid
a
n
c
e in Network on
Chi
p
(Lei Zh
ou)
3
}
1
,
0
{
1
0
k
m
k
k
k
d
f
d
v
(1)
whe
r
e
d
k
i
s
the
k-
th
bit in vector
while f
is the Fibo
n
a
cci elem
ent. Here we
defi
ne the Fibo
n
a
cci
Sequen
ce
s a
s
follow:
)
2
(
2
1
)
1
(
1
)
0
(
0
m
m
f
m
f
m
m
m
f
(2)
The lite
r
ature
[8] ha
s
prov
ed the
data
e
n
co
ded
by bi
nary Fi
bon
acci
rep
r
e
s
entat
ion
can
prevent
crosstalk delay; th
erefo
r
e
we u
s
e bina
ry
Fibo
nacci represe
n
tation a
s
CA
C. Du
e to CA
C
avoid multiple
bits e
rro
r effi
ciently; the E
CC
only
ne
ed
s to dete
c
t o
n
e
bit erro
r by
parity bits.
Wi
th
the con
s
tru
c
ti
on the en
codi
ng algo
rithm i
s
expre
s
sed
as Figu
re 4.
The
crosstalk avoidance
ability of the whol
e codeword has
been proved i
n
[7] for the
entire
cod
e
word
satisfy F
P
C. He
re we
only need to
prove the o
n
e
bit error d
e
t
ection ability
o
f
addition
al parity bits.
Theorem
1.
The
ad
ditio
nal p
a
rity bit
s
of
code
wo
rd po
sse
s
s o
ne bit
erro
r
detectio
n
ability.
Proof.
ECC
shiel
d
s
data
w
ord
s
by g
ene
rating t
w
o a
d
d
itional p
a
rity
bits, which refers to
the p
r
in
ciple
of even
or od
d
pa
rity. Wh
e
n
ECC
gets
code
word
d
m
,…d
k+1
d
k
ge
ne
r
a
te
d b
y
C
A
C
,
it
cre
a
ted
pa
rity bit value
p
throug
h
bitwi
s
e X
O
R op
erator. Th
en
th
e pa
rity bit v
a
lue i
s
exten
d
to
two pa
rity bits
d
m+
1
d
m+
2
t
o
s
a
tis
f
y FPC. Ac
c
o
rding to FPC, the firs
t bit
d
m+
1
should e
qual to
the
last bit of co
d
e
wo
rd of
CAC,
so the
parity ability is guara
n
teed
by
d
m+
2
. For example, if we
use
even pa
rity, the pa
rity bits
sho
u
ld e
qual
to 00 o
r
11
when p
=
0. Oth
e
rwi
s
e t
he p
a
r
ity bits is
01
or
10 whe
n
p
=1. The truth table of relationship betwe
en
p, d
m+1
and parity bits
d
m+
1
d
m+
2
is shown in
Table 1.
Table 1. Truth table of pari
t
y bits
p d
m
d
m+
2
d
m+
1
0 0
0
0
0 1
1
1
1 0
1
0
1 1
0
1
From T
able I
we
can d
edu
ce that
d
m+
1
=d
m
,
d
m+2
=p
d
m
=p
d
m+
1
=d
1
d
2
…
d
m+1
; it is
equivalent to
that the la
st b
i
t d
m+2
is eve
n
pa
rity bit of
the w
hole co
deword. The derived
proce
s
s
is co
nsi
s
tent
whe
n
usi
ng o
dd parity pri
n
ciple.
Figure 3 co
nstruction of joi
n
t code
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 1 – 8
4
4. Implementation of
CO
DEC
Acco
rdi
ng to the algo
rithm of joint code, encode
r tran
sforms the
n
-bi
t
binary data into
m
-
bit Fibon
acci
cod
e
, an
d ad
ds t
w
o
additi
onal
parity bit
s
to th
e tail o
f
it.
N
an
d
m
sat
i
sf
y
2
n
<f
m+
2
,
therefo
r
e the
origin
al 32 bit
s
bina
ry data
is m
appe
d to 46 bits Fib
ona
cci
code.
Adding 2 pa
rity
bits, the total width of cod
e
wo
rd is 4
8
. The en
cod
e
r base
d
on al
gorithm
can
be implem
ent
ed
using the struct
ure illustrated in Figure 5.
The origi
nal 3
2
bits data is trans
fe
rred in
to encod
er, compa
r
ed with
f
47
to determine the
value of
d
46
and
d
47
.
In th
e next stage, the rest of the input r
46
is com
pared
with f
45
and f
46
to
gene
rate d
45
, and
then
th
e remaini
ng i
s
tran
sferred
to next
conti
nuou
sly u
n
til
d
1
is
left. Lastl
y
d
47
...d
2
d
1
is perform
ed by
XOR gate to
gene
rate d
48
. The co
mbin
ational logi
c
depth of en
coder
is too l
a
rg
e,
so
circuit i
s
d
i
vided into 1
6
stage
pip
e
lin
e. The p
r
o
c
e
ss
of ea
ch
st
age
con
s
u
m
e
s
one cy
cle; the encodin
g
proce
s
s is com
p
leted in 16
cycles totally.
Figure 6 depi
cts the struct
ure of de
cod
e
r.
Whe
n
re
ceived 48 bits code
wo
rd, d
e
co
der
firs
tly us
es
bitwis
e XOR to
d
47
...d
2
d
1
for recre
a
ting th
e parity value
p
. If
p
≠
d
48
, there
may o
c
cur
data
corrupti
on du
rin
g
da
ta tran
sfer, t
he e
rro
r fla
g
e will
be
m
a
rked
and
ro
uters will
re
q
uest
retra
n
smi
s
sio
n
. Othe
rwi
s
e
the ci
rcuit tra
n
sforms
the
Fibona
cci
co
de into
bin
a
ry data
acco
rding
to formula (1
) and tran
sfer
to the next router.
Figure 4 The
Joint Code Al
gorithm
P
r
o
ced
u
r
e
Jo
i
n
t
-
C
o
d
e
-
A
l
gor
i
t
hm
(
v
)
/*
v
i
s
t
h
e
or
i
g
i
n
a
l
da
t
a
w
o
r
d
,
t
h
e
l
e
ngt
h of
C
A
C
i
s
m
*/
if
v
>=
f
m+
1
th
e
n
//
C
A
C
e
n
c
o
d
e
d
m
=
1
r
m
=
v
-
f
m
e
l
s
e
d
m
=
0
r
m
=
v
e
n
d
i
f
f
o
r
k
=
m
-1
to
2
do
if
r
k+
1
>=
f
k+1
th
e
n
d
k
=
1
el
s
e
i
f
r
k+
1
<
f
k
d
k
=
0
el
s
e
d
k
=
d
k+
1
en
d
i
f
r
k
=
r
k+
1
-
f
k
·
d
k
e
n
d
f
o
r
d
1
=
r
2
p
=
0
/
/
g
e
t
p
a
r
i
t
y
c
o
d
e
fo
r
k
= 1
to
m
do
p=
p^
d
k
en
d
f
o
r
if
p
= 0
an
d
d
m
= 0
t
h
e
n
//
L
X
C
en
co
d
e
d
m+
2
d
m+
1
=
00
el
s
e
i
f
p
=
0
an
d
d
m
= 1
d
m+
2
d
m+
1
=
11
el
s
e
i
f
p
=
1
an
d
d
m
= 0
d
m+
2
d
m+
1
=
10
el
s
e
d
m+
2
d
m+
1
=
01
e
n
d
i
f
r
e
t
u
r
n
(
d
m+
2
d
m+
1
d
m
…
d
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Joint-Codi
n
g
Schem
e wit
h
Cro
s
stalk A
v
oid
a
n
c
e in Network on
Chi
p
(Lei Zh
ou)
5
Figure 5. The
implementati
on of encode
r
Figure 6. The
implementati
on of decode
r
4. Exeriment Result a
nd Analy
s
is
4.1 The Expe
riment Sche
me
To evaluate t
he complexit
y
of the CO
DECs,
we i
m
plemente
d
th
e fault-tole
ra
nt route
r
whi
c
h a
pplied
to the joint
code a
nd
con
s
tructed
4
×
4 2
D
-m
esh net
work. Pa
cket i
n
jectio
n follo
ws
a unifo
rm
distribution. A
c
cordin
g to
def
ault bit e
r
ror
rate
(BER),
we i
n
je
cted
e
rro
r
bits i
n
lin
ks
betwe
en ro
uters
to sim
u
la
te
the occu
rrence
of
in
sta
n
taneo
us fau
l
t. We
pro
p
o
s
e th
re
e typ
e
s
combi
nation
of CODE
C an
d route
r
:
(1).
Once e
n
code
, once de
co
d
e
: The
co
mbi
nation
follo
ws End-to
-End
proto
c
ol, that
dataword
is
only
en
cod
e
d
in
the ro
uter whi
c
h con
necte
d
to so
urce
fun
c
tion
block and d
e
co
ded by
destin
a
tion ro
uter.
(2).
Multiple en
co
de, multiple
decode: Th
e
comb
i
nation
follows Poi
n
t-to-Point p
r
o
t
ocol, that
dataword i
s
e
n
co
ded a
nd d
e
co
ded by ev
ery route
r
the
datawo
r
d pa
ssed in the ro
uting path.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 1 – 8
6
(3).
Once en
co
de
, multiple de
code: data
w
o
r
d is
o
n
ly en
coded i
n
sou
r
ce
route
r
, the
route
r
s in
routing p
a
th decode a
nd parity the cod
e
wo
rd, and tr
ansfe
r the ori
g
inal code
wo
rd directly if
cod
e
word h
a
s
no bit error.
The
experi
m
ent impl
eme
n
ts the
three
erro
r-cont
rol
sche
me
s m
entione
d a
b
o
v
e usi
ng
Verilog
HDL,
and th
en
syn
t
hesi
z
e
s
the
m
in
De
si
g
n
Compli
er of
SynopsysTM.
Based
on it,
the
perfo
rman
ce
of average d
e
l
ay, power a
n
d
area i
s
discussed a
s
follo
ws.
4.2 Dela
y
The d
e
lay of
CO
DEC i
s
fix
ed,
therefore
the extra
del
ay su
ffered i
n
thre
e type
s ca
n b
e
cal
c
ulate
d
as
follow:
decoder
encoder
Delay
Delay
Delay
1
(3)
)
(
2
decoder
encoder
Delay
Delay
Delay
(4)
decoder
encoder
Delay
Delay
Delay
3
(5)
whe
r
e Delay
e
n
coder
and Del
a
y
decoder
are the del
ay of the en
co
de
r a
nd de
co
de
r resp
ectively,the
value are 16
cycle
s
an
d 1 cycle
re
sp
ectively
acco
rding to the i
m
pleme
n
tatio
n
in se
ction
4.
α
expre
s
ses th
e averag
e ho
ps of 2D-me
s
h netwo
rk, thi
s
implie
s that:
2
11
2
n
m
m
j
n
i
ij
C
hop
(6)
whe
r
e
m
a
nd
n
indi
cate
th
e num
ber of
node
s in
ho
ri
zon
and
verti
c
al di
re
ction
of the
netwo
rk, and
hop
ij
i
s
h
o
p
s
from n
ode
i
to
j
. We
use 4
×
4 2
D
-me
s
h
netwo
rk a
s
th
e expe
riment
al
netwo
rk,
m
=4,
n
=4. In ex
perim
ents th
e injectio
n rates
follo
w
uniform di
stri
bution, there
f
ore
α
=2.67,
∆
Del
a
y
1
=1
7,
∆
Del
a
y2=4
5.39,
∆
Delay3 =18.
67.
Except the fixed delay of CO
DEC, the
delay suffe
re
d from
erro
r
control al
so i
n
clu
d
e
s
the con
s
ump
t
ion of d
a
ta
retra
n
smi
s
sio
n
. In net
works we a
s
su
me the BE
R is
0, 1/10
0
000,
1/1000
0, 2/1
0000,
3/1000
0, 4/
100
00, 5
/
10000,
6/10
000, 7/1
0000
, 8/1000
0, 9/
1000
0 an
d 1/
1000
unde
r the
inj
e
ction
rate
of
0.04flit/cycle
/node, 0.0
8
flit/cycle/no
de
a
nd 0.1
0
flit/cycle/no
de, Fig
u
re
7 plots the av
erag
e delay versus BER.
(
a
)
(b)
(c
)
Figure 7. Del
a
y versu
s
BER
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Joint-Codi
n
g
Schem
e wit
h
Cro
s
stalk A
v
oid
a
n
c
e in Network on
Chi
p
(Lei Zh
ou)
7
The gap b
e
twee
n delays
of three types is sli
ght wh
en BER and i
n
jectio
n rate
s are low.
Whe
n
BER a
nd inj
e
ctio
n
rates i
n
cre
a
se
, the d
e
la
y of
first type
p
r
o
m
ote di
stin
ctly. Although t
he
delay of
co
d
e
cs in
first type i
s
small
e
st, this
type i
n
volves m
o
re
packet
retran
smissio
n
whi
c
h
must pa
ss throug
h ea
ch router in routi
ng pat
h. For second type
, the incre
a
si
ng of delay is
limited becau
se that the retran
smi
ssi
o
n
happ
ene
d wh
e
n
route
r
finds e
r
ror in
the routing
p
a
th,
this meth
od
shorten
s
th
e p
a
th of ret
r
an
smissi
on e
ffectively. Note that in
low BER and injecti
o
n
rates, th
e d
e
l
a
y in this type is l
a
rger th
an othe
rs
for this
type
s
u
ffers
more delay penalty in
encodin
g
an
d
de
codin
g
p
r
oce
s
s. The
r
e
f
ore thi
s
type
is n
o
t fitted to low loa
d
system. Th
e thi
r
d
type avoid
s
e
n
co
ding
du
rin
g
ro
uting
pro
c
e
s
s co
mpa
r
ed to th
e
se
cond type, th
e
r
efore the
del
ay
outperfo
rm
s than othe
rs.
4.3 Po
w
e
r an
d Area
The code
cs are
synthe
sized usi
ng a
SMIC 0.18-
μ
m CMOS st
anda
rd
cell li
bra
r
y in
De
sign
Comp
iler of Synopsys
TM
, The results of power
and area a
r
e
sho
w
n in Ta
b
l
e 2.
Table 2. Power and Area o
f
code
c
To simplify the co
mpa
r
ison, we o
n
ly ev
aluate the
con
s
um
ptio
n of power
and a
r
ea
cau
s
e
d
by co
decs. Figu
re
8 plots the co
ns
um
ption of power an
d area und
er thre
e types.
In first type,
encodin
g
a
n
d
de
co
ding
p
r
ocess
in so
urce route
r
a
nd
d
e
stin
atio
n
router
respe
c
tively, therefo
r
e
co
d
e
cs a
r
e
pla
c
ed in l
o
ca
l p
o
rt of e
a
ch router. In
se
cond first typ
e
,
encodin
g
a
n
d
de
codi
ng
ha
ppen
ed i
n
ev
ery h
op i
n
ro
uting, code
cs sh
ould
be
pl
ace
d
in
po
rts of
four
dire
ction
s
. In thi
r
d typ
e
, en
codi
ng i
s
p
r
o
c
e
s
sed
whe
n
d
a
tawo
rds e
n
ter th
e
netwo
rk a
nd t
he
route
r
s de
co
de in
ro
uting
path, so o
n
e
en
cod
e
r is pla
c
ed i
n
lo
cal
port
and
four d
e
code
s are
placed i
n
p
o
rt
s of
four di
re
ction
s
. It is o
b
vious
that th
e second
type con
s
ume
s
more
po
we
r
and
area tha
n
oth
e
rs. Note that the consump
t
ion of t
he third type is larg
er than the first type, but th
e
gap bet
wee
n
them is slig
ht becau
se the
con
s
um
ption
of deco
der i
s
far less than
encode
r.
Figure 8. The
compa
r
i
s
on
of powe
r
and
area
5. Conclusio
n
In this pape
r we propo
se
d
a joint codin
g
scheme
wh
ich combin
es cro
sstal
k av
oidan
ce
cod
e
with e
r
ror dete
c
tion
co
de, to
so
lute the
relia
ble p
r
obl
em
of No
C i
n
d
eep
su
b-mi
cron
regim
e
. We
mappe
d the dataword into
Fibona
cci n
u
m
eral
system
to avoid cro
s
stalk, an
d ad
ded
Power
Area
(
μ
m2
)
D
y
namic (mW)
Leakage (
μ
W)
Total
(mW
)
Combinational
Logic
Sequential
Logic
Total
Decoder
1.9959
32.1384
2.0280
365897.35
73483.50
439380.85
Encoder
0.8106
1.7212
0.8124
29069.40
2767.56
31836.96
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 1 – 8
8
parity ability based on it. The impl
ementation of co
dec in l
o
w
complexity is showed. Further we
analyzed th
ree
scheme
s
of co
de
c a
p
p
lying to
No
C,
the experi
m
ental re
sult
s sho
w
th
at the
“On
c
e en
co
d
e
, multiple de
cod
e
” type o
u
tperfo
rm
s th
an others fro
m
the view of delay. Although
the power an
d area of this
type is increa
sing
slight
ly compa
r
ed to the be
st
one, it still is the most
approp
riate schem
e in the three types
which
sati
sfie
s
the requi
rem
ent of erro
r control.
Referen
ces
[1]
MS W
u
, CL
Le
e. Usi
ng
a
peri
odic
squ
a
re
w
a
ve test s
i
g
nal
to d
e
tect cros
s talk fa
ults.
IEEE Des
i
gn
&
T
e
st of Comp
u
t
ers
. 2005; 22(
2): 160-1
69.
[2]
H Z
i
mmer, A Jantsch. A F
ault
Model N
o
tatio
n
and Er
r
o
r-Co
n
trol Schem
e for S
w
itch-to-S
w
itc
h
Buses
i
n
a Net
w
ork-
on-
Chip.
First IEEE/ACM/IFIP
International
Co
nferenc
e on Hardware/Software Codesign
and Syste
m
s S
y
nthesis
, Ne
w
York, NY, USA. ACM, 188-19
3
[3]
Q Yu, P Ampad
u. Ada
p
tiv
e
Error C
ontr
o
l for N
o
C S
w
itc
h
-to-S
w
itc
h
Links
in
a V
a
ria
b
le
No
is
e
Enviro
nment.
IEEE International Sy
m
p
osium
on Defect
and Fault Tolerance of VLSI System
s
. 20
09
[4]
SR Srid
hara,
NR Sh
anb
ha
n
.
Codi
ng for
S
y
st
em-o
n-C
h
i
p
Net
w
o
r
ks: A Unifie
d F
r
am
e
w
ork.
IEEE
TRANSACTIONS ON VLSI SYSTEMS
, 2005, 13(6): 65
5-6
67.
[5]
A Gangu
l
y
,
PP
Pand
e, B Belz
er. Crosstalk-A
w
a
r
e C
h
a
n
n
e
l
Codi
ng Sc
hem
es for Ener
g
y
Efficient an
d
Reli
ab
le No
C Intercon
nects.
IEEE Transaction on VLSI System
s
, 20
09
[6]
SR Sridh
a
ra a
nd NR S
han
bh
ag, “Cod
in
g for relia
ble
on-ch
i
p
buses: fun
d
a
m
ental l
i
mits a
nd practic
a
l
codes”,
Proceedings of IEEE Internatio
nal Conferenc
e on VLSI Design
, 20
0
5
: 417-4
2
2
[7]
C Dua
n
, VH C
o
rder
o Ca
lle,
SP Khatri. Effici
ent On-C
hip
Crosstalk Av
oi
danc
e CODEC
Desig
n
.
IE
EE
T
r
ansactio
n
s On VLSI Systems
, 2009, 17(
4): 551-
560.
[8]
M Mut
y
am, “Pr
e
venti
ng crosst
alk de
la
y
usin
g
F
i
bon
acci re
presentati
on,”
in
Proc. Int. Conf. VLSI Des
.,
200
4: 685
–6
88
.
Evaluation Warning : The document was created with Spire.PDF for Python.