Int
ern
at
i
onal
Journ
al of
El
e
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
491
2
~
492
1
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
491
2
-
492
1
4912
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Detectio
n and M
onitorin
g I
nt
ra
/
I
nter C
rossta
lk in
Op
tical
Network
on Chip
Ah
med
J
e
didi
Coll
ege of
Enginee
ring
,
Ahl
ia Unive
rsit
y
,
Bah
rain
CES L
abor
at
or
y
,
Nati
on
al E
ngin
e
eri
ng
Schoo
l
of
Sfax,
Unive
rsit
y
of
Sfax,
Tuni
si
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r
30
, 201
8
Re
vised
Jun
18
, 201
8
Accepte
d
J
ul
29
, 2
01
8
Multi
proc
essor
s
y
stem
-
on
-
chi
p
(MP
SoC
)
has
b
ec
om
e
an
at
tr
active
solut
ion
for
improving
t
he
per
form
ance
of
single
chi
p
in
object
ive
to
sati
sf
y
th
e
per
form
anc
e
gr
owing
expone
n
ti
all
y
of
the
computer
app
lications
as
m
ult
imedia
app
l
ic
a
ti
ons.
How
ev
er,
th
e
comm
unic
ation
bet
we
en
t
he
diffe
r
ent
proc
ess
ors’
cor
e
s
pre
sents
the
f
i
rst
cha
l
le
ng
e
fro
nt
the
h
igh
per
f
orm
anc
e
of
MP
SoC
.
Beside
s,
Network
on
Chip
(NoC)
is
among
the
m
ost
prom
ine
n
t
soluti
on
for
handl
ing
the
on
-
chip
comm
unic
at
ion.
Beside
s,
NoC
pote
ntia
l
li
m
it
ed
b
y
ph
y
si
ca
l
l
imita
t
ion,
p
ower
consum
pti
o
n,
la
t
ency
and
bandwidt
h
in
the
bo
th
c
ase
:
in
cre
asing
data
ex
cha
nge
or
sc
al
ab
il
ity
of
Mult
ic
or
es.
Opti
cal
comm
unic
at
ion
offe
rs
a
wider
bandwidt
h
and
lower
power
co
nsum
pti
on,
base
d
on,
a
ne
w
te
chnol
og
y
n
amed
Optic
al
N
et
work
-
on
-
Chip
(ON
oC)
ha
s
bee
n
in
troduce
d
in
MP
SoC
.
How
eve
r,
ON
oC
components
induc
e
th
e
cro
ss
ta
lk
nois
e
in
the
ne
twork
on
both
form
s
int
ra
/i
nt
er
cro
ss
ta
lk.
This
serious
proble
m
det
er
iora
t
es
th
e
qualit
y
of
sig
nal
s
and
degr
ad
es
net
work
per
form
anc
e
.
As
a
result
,
detec
t
ion
and
m
onit
oring
th
e
i
m
pai
rm
ent
s
bec
om
ing
a
ch
allenge
to
k
ee
p
th
e
per
form
an
ce
i
n
the
ON
oC.
In
thi
s
ar
ti
c
le,
we
propose
a
ne
w
sy
st
em
to
det
e
ct
and
m
onit
or
t
he
cro
ss
ta
lk
nois
e
in
ON
oC.
Parti
cularl
y
,
we
pre
sent
an
anal
y
ti
c
m
odel
of
int
ra/int
er
c
ross
ta
lk
at
th
e
opti
c
al
device
s.
The
n,
we
ev
al
ua
te
the
se
impairm
ent
s
in
obje
c
ti
ve
to
pre
sent
the
m
oti
vat
ion
t
o
det
ect
and
m
o
nit
or
cro
ss
ta
lk
in
ON
oC,
in
whic
h
our
sy
ste
m
has
the
c
apa
bi
li
t
y
to
d
etec
t
,
to
lo
ca
l
iz
e
,
and
to
m
onit
or
the
cro
ss
t
al
k
noise
i
n
the
whole
netw
ork.
Thi
s
s
y
stem
offe
rs
high
rel
ia
b
il
i
t
y
,
sc
al
ab
il
ity
and
eff
iciency
with
t
ime
running ti
m
e
l
ess t
han
20
m
s.
Ke
yw
or
d:
In
te
r/
i
nt
ra
c
ro
s
sta
lk
MPSoC
ONoC
Op
ti
cal
co
m
m
un
ci
at
io
n
RTL desi
gn
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ah
m
ed
Je
did
i,
Dep
a
rtm
ent o
f C
om
pu
te
r
E
ng
i
neer
i
ng,
Ah
li
a
Un
i
ver
si
ty
,
Ba
hr
ai
n.
Em
a
il
:
ajed
idi
@ah
li
a.e
du.bh
1.
INTROD
U
CTION
The
high
integ
rati
on
rate
of
t
ran
sist
or
has
pushe
d
t
he
sem
ic
ondu
ct
or
i
ndus
try
to
a
s
hif
t
fr
om
the
sing
le
-
c
ore
to
a
m
ulti
-
cor
e
in
on
e
chi
p.
Be
sides,
a
ne
w
pa
rad
i
gm
is
intr
oduce
d
m
ulti
pr
oces
sor
syst
em
one
chip
(MPS
oC).
Howe
ver
,
M
PSo
C
has
to
fa
ce
serio
us
pro
blem
s
in
te
r
m
s
of
e
nergy
co
nsum
ption
,
e
xec
ution
tim
e,
heat
diss
ipati
on
an
d
da
ta
flow
.
Data
trans
fer
betwee
n
cor
e
s
in
MPSoC
becam
e
t
he
first
chall
e
ng
e
t
o
i
m
pr
ove
these
par
am
et
ers.
H
ence,
Netw
ork
on
Chi
p
(
NoC)
is
e
m
erg
in
g
as
a
prom
isin
g
s
olu
ti
on
pr
ov
i
ding
low
e
nergy
co
ns
um
ption
a
nd
high
ba
ndwidt
h
to
i
m
pr
ove
the
perf
or
m
ance
of
MPS
oC.
Ho
wever,
acc
ordi
ng
the
gro
wing
of
the
pe
rfo
rm
a
nce
of
com
pu
t
er
ap
plica
ti
on
No
C
beco
m
e
a
bo
tt
le
nec
k
f
or
the
scal
abili
ty
and
powe
r
dissi
pation
i
n
the
MP
S
oC.
I
ndeed
,
el
e
ct
rical
interco
nnect
s
are
not
able
to
bo
os
t
tra
ns
m
issi
on
rates
an
d
powe
r dissi
pation w
hich
it
m
a
kes hig
hly desi
rab
le
t
o
re
plac
e them
[
1].
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d Mo
nitorin
g In
t
ra
/I
nter Cr
os
st
alk in O
ptical
Ne
tw
or
k on C
hi
p
(
Ah
med Jedi
di)
4913
Op
ti
cal
com
m
un
ic
at
io
n
offe
r
s
a
high
ba
nd
width
with
lo
wer
powe
r
co
ns
um
ption.
I
n
fact,
it
can
achieve
band
w
idths
in
t
he
ord
er
of
te
ra
bits
pe
r
seco
nd
by
e
xp
l
oiti
ng
wav
e
le
ng
th
div
isi
on
m
ulti
plexing.
Sinc
e
photonic
s
yst
em
has
a
po
te
nt
ia
l
capaci
ty
an
d
bec
om
e
attracti
ve
to
us
e
it
to
i
m
pr
ove
co
m
m
un
ic
at
ion
in
the
chip [
2]. As a
r
esult,
op
ti
cal
net
work on c
hip “O
No
C”
is pr
esented
to repl
ace t
he
tra
diti
onal
NoC.
Mi
cro
res
onat
ors
(MRs)
a
nd
na
nophot
on
ic
waveg
uid
e
s
are
t
he
key
dev
ic
es
in
ONoC. Ho
wev
e
r,
th
ese
op
ti
cal
de
vices
and
wav
e
guides
pro
vide
trans
par
e
ncy
(
a
com
po
ne
nt
is
cal
le
d
X
-
tra
nspare
nt
if
it
fo
r
wards
incom
ing
signa
ls
fr
om
inp
ut
to
ou
t
pu
t
with
ou
t
exam
ining
the
X
aspect
of
the
sign
al
)
capab
il
it
ie
s
wh
ic
h
they
induce
opti
c
al
vulne
rab
il
it
ie
s
“cr
os
sta
lk”
in
O
N
oC
where
is
does
not
exist
i
n
a
co
nv
e
ntio
nal
N
oC.
Fu
rt
her
m
or
e,
one
of
the
seri
ous
pro
blem
s
with
trans
par
e
nc
y
is
the
fact
that
op
ti
cal
cro
ss
ta
lk
is
add
it
ive
,
an
d
thu
s
t
he
ag
gre
gate
eff
e
ct
of
cro
sstal
k
over
a
whole
O
N
oC
m
a
y
be
m
or
e
ne
far
i
ou
s
t
ha
n
a
sin
gle
po
int
of
cro
sstal
k
[3
]
.
Con
se
quently
,
op
ti
cal
cr
os
sta
lk
de
grade
s
th
e
qual
it
y
of
sign
al
s
an
d
i
ncrea
ses
their
BE
R
(Bi
t
Error
Ra
te
)
pe
rfor
m
ance
[4]
.
In
fact,
bo
t
h
f
or
m
s
of
op
ti
cal
cro
sstal
k
can
arise
in
ONoC
syst
em
:
inter
-
cro
sstal
k
a
nd intra
-
c
r
os
sta
lk.
In
te
r
-
cr
os
sta
lk
arises
w
he
n
the
cr
os
sta
lk
si
gn
al
is
at
a
w
avelen
gth
s
uffi
ci
ently
diff
ere
nt
from
the
aff
ect
ed
sig
nal
’s
wav
el
e
ng
t
h
that
the
dif
fe
ren
ce
is
la
rg
e
r
than
the
rece
iver’s
el
ect
rica
l
band
width.
I
nter
-
cro
sstal
k
ca
n
a
lso
oc
cu
r
t
hro
ugh
m
or
e
inter
act
ion
s
t
hat
a
r
e
in
direct.
For
exam
ple,
if
one
cha
nnel
af
fe
ct
s
the
gain
see
n
by
ano
t
her
c
ha
nnel
,
as
with
nonlinearit
ie
s.
I
n
the
seco
nd
pl
ace,
intra
-
cr
osst
al
k
arises
w
hen
t
he
cro
sstal
k
sign
a
l
is
a
t
the
sa
me
wav
el
en
gth
as
that
of
the
af
fected
si
gn
al
or
suffici
entl
y
cl
os
e
to
it
t
hat
the
diff
e
re
nce
in
wav
el
e
ng
t
hs
is
within
the
rec
ei
ver
’
s
el
ect
rical
ban
dwidt
h.
In
tra
-
cr
os
sta
lk
arises
in
trans
m
issi
on
li
nk
s
due to
r
e
f
le
ct
ion
s [5]
-
[6
]
.
Cros
sta
lk
is
one
of
th
e
m
ajo
r
pro
blem
arisi
n
g
in
ONoC
an
d
it
is
a
barrier
of
th
e
scal
abili
ty
and
th
e
perform
ance
f
or
t
he
ev
olu
ti
on
of
the
MP
So
C.
As
a
res
ult,
fin
d
a
m
eth
od
to
detect
,
local
iz
e
and
m
on
it
or
cro
sstal
k
will
be
essenti
al
.
I
n
this
pap
e
r,
we
pro
pose
a
new
syst
em
to
detect
and
m
on
it
or
the
intr
a/
inter
cro
s
sta
l
k
in
O
No
C.
I
nd
ee
d,
we
prese
nt
an
analy
ti
c
m
od
el
fo
r
t
he
both
form
s
of
cros
sta
lk
induce
d
by
the
ONoC
com
ponen
ts
.
Ba
sed
on
this
stud
y,
we
evalu
at
e
and
po
i
nt
out
the
i
m
po
rta
nce
to
detect
and
m
on
it
or
cro
sstal
k
noise
in
ONoC.
Furt
her
m
or
e,
we
de
sign
a
nd
im
pl
e
m
ent
a
per
for
m
ed
hardw
a
re
syst
e
m
to
detect
and
m
on
it
or
inter/i
ntra
c
rossta
lk i
n
the
wh
ole n
et
work.
The
rest
of
t
he
pap
e
r
is
orga
ni
zed
as
f
ollo
ws:
The
sec
ond
se
ct
ion
will
pres
ent
an
e
xtensi
ve
ov
e
rv
ie
w
of
the
relat
ed
work
i
n
the
li
te
ratur
e
.
Th
e
th
ird
sect
i
on
will
descr
i
be
the
inn
e
r
arc
hitec
tu
re
of
the
basic
dev
ic
es
in
O
No
C
a
nd
the
net
wor
k
m
od
el
us
e
d
in
t
his
wor
k.
T
he
f
ourt
h
sect
ion
wi
ll
dep
ic
it
the
analy
ti
c
m
od
el
of
t
he
cro
sstal
k
noise
ind
uce
d
in
op
ti
cal
com
ponen
ts
.
Ma
inly
,
we
will
descr
ibe
the
cr
os
st
al
k
pr
ogress
i
n
the
diff
e
re
nt
de
vices
an
d
in
t
he
whole
netw
ork.
T
he
fifth
sect
ion
will
descr
i
be
the
ha
rdwar
e
desi
gn
an
d
i
m
ple
m
entat
io
n
of
the
pro
posed
syst
em
of
detect
ion
a
nd
m
on
it
or
the
cr
os
sta
lk
in
O
N
oC.
T
he
sixt
h
sect
ion
will
discuss
an
d
analy
se
the
f
us
ibil
it
y
and
th
e
diff
e
ren
t
res
ults
of
our
syst
e
m
.
Finall
y,
we
will
con
cl
ude
and
expose
our f
ut
ur
e
works.
2.
RELATE
D
W
ORK
Op
ti
cal
net
wor
k
on
chi
p
pro
vid
e
a
pr
om
is
i
ng
so
l
utio
n
to
increase
the
r
equ
i
rem
ents
in
ultra
-
hi
gh
band
width
an
d
lo
wer
pow
er
co
nsum
pti
on
i
n
the
MP
So
C.
Acc
ordin
g
the
pro
gr
ess
of
te
ch
no
l
og
y
,
nano
photonic
wav
e
guide
a
nd
op
ti
cal
switc
h
are
prese
nted
i
n
the
c
hip,
w
hi
ch
they
are
t
he
key
com
pone
nts
f
o
r
ONoC.
I
n
the
la
st
decad
e,
m
a
ny
researc
hers
fo
c
us
their
work
s
to
pro
po
se
a
per
f
orm
ed
arch
it
ect
ur
e
of
O
N
oC
i
n
obj
ect
ive
t
o
sat
isfy
the
dif
f
eren
t
re
qu
irem
ents
of
MPS
oC
pe
rfor
m
ance.
H
oweve
r,
the
cr
os
sta
lk
nois
e
is
one
of the m
os
t i
m
portant iss
ues
t
hat r
esea
rc
her
s
f
ace i
n dev
el
opin
g op
ti
cal
ne
twork
on
-
c
hip.
I
n
the
li
te
ratu
re,
m
os
t
of
th
e
resea
rch
e
rs
f
ocus
their
w
o
r
k
on
analy
sis
and
m
od
el
ing
of
c
r
os
sta
l
k
no
ise
s in
duce
by the o
ptica
l dev
ic
es in
ONoC. Furth
e
rm
or
e, m
any stud
ie
s isolat
e their w
ork for m
od
el
ing
and
analy
zi
ng
the
c
ro
sstal
k
no
ise
in
op
ti
cal
dev
ic
es
sep
aratel
y
a
s
r
ou
te
r
or
w
av
egu
i
de.
Yiyua
n
Xie
a
nd
al
.
an
al
yse
and
op
ti
m
iz
e
cornstal
k
in
5x
5
Hitl
ess
Sil
ic
o
n
-
Ba
se
d
O
ptica
l
Rou
te
r
[
7].
Hen
ce
,
they
an
al
yse
d
cro
sstal
k
noise
at
dev
ic
e
le
vel
and
r
ou
te
r
le
ve
l.
Ba
sed
on
t
he
detai
le
d
a
na
ly
sis,
they
pro
po
s
ed
a
ge
ner
a
l
analy
ti
cal
m
o
del
to
stud
y
the
tran
s
m
issi
on
loss,
c
ro
s
sta
l
k
noise
,
op
ti
cal
sign
al
-
to
noise
rati
o
(
OSNR),
a
nd
bit
err
or
rati
o
(B
ER).
Indee
d,
they
us
ed
the
cr
os
si
ng
ang
le
s
of
60
⁰
or
120
⁰
in
ste
ad
of
the
c
onve
nt
ion
al
90
⁰
c
ro
s
s
ing
a
ng
le
to
de
sign
the opti
cal
rou
t
er. Usin
g
t
his
m
et
ho
d OS
NR
is im
pr
ov
e
d b
y about
10 d
B
[7
]
.
Fabri
zi
o
Gam
bin
i
a
nd
al
.
pr
opos
e
d
a
phot
on
ic
m
ulti
-
m
icr
ori
ng
N
oC
,
w
hich
t
he
the
oret
ic
al
m
od
el
base
d
on the tran
sfe
r
m
at
rix
m
et
ho
d,
has
be
en
validat
e
d
th
rou
gh
e
xp
e
rim
ental
r
esults i
n
te
rm
s
o
f
trans
m
issi
on
sp
ect
ra
[8
]
.
T
r
ansm
issi
on
s
at
10
GB/s
ha
ve
been
as
sesse
d
in
te
rm
s
of
BER
for
both
si
ng
le
-
wa
velen
gt
h
an
d
m
ul
ti
-
wav
el
e
ngth
co
nfi
gurati
on
s
.
The
integ
rated
N
oC
co
nsi
sts
of
8
the
r
m
al
l
y
tun
ed
m
ic
ro
ri
ngs
co
up
l
ed
to
a
central
rin
g,
w
her
e
the
BER
m
easur
em
ents
sh
ow
pe
rfor
m
ance
up
to
10
−9
at
10
G
b/s
with
li
m
it
ed
cro
s
sta
lk
and p
e
nalty
(b
el
ow
0.5 dB)
induce
d by an
interf
e
rin
g
tr
ans
m
issi
on
[8].
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
491
2
-
4921
4914
Be
sides,
m
any
works
deal
w
it
h
the
cr
os
sta
lk
in
the
w
ho
le
ONoC.
Ma
inl
y,
it
is
ho
w
th
e
cro
sstal
k
no
ise
s
ef
fect
t
he
perf
or
m
ance
of
the
netw
ork.
Partic
ular
ly
,
these
st
ud
i
es
analy
se
a
nd
m
od
el
the
c
r
os
sta
lk
pro
gr
ess
in
t
he diffe
re
nt p
a
rts
of the
ne
tw
ork, an
d pr
ese
nt th
ei
r
ef
fect t
o
BE
R an
d
S
NR
of
the n
et
wor
k.
Ma
hd
i
Ni
ksda
t
and
al
.
syst
e
m
at
ic
ally
stud
y
and
c
om
par
e
the
w
or
st
ca
se
as
well
as
the
ave
rag
e
cro
sstal
k
noise
an
d
S
NR
in
t
hr
ee
well
kn
own
opti
cal
interco
nnect
arc
hi
te
ct
ur
es,
m
esh
-
base
d,
f
old
e
d
-
tor
us
-
base
d,
a
nd
fat
-
tree
-
ba
sed
O
N
oCs
us
i
ng
WD
M
[9
]
.
T
he
an
al
yt
ic
al
m
od
el
s
for
the
worst
case
an
d
the
a
ver
a
ge
cro
sstal
k
noise
and
S
NR
in
the
dif
fer
e
nt
ar
chite
ct
ur
es
ar
e
pr
ese
nted
.
Furtherm
or
e,
the
propose
d
anal
yt
ic
al
m
od
el
s
are
int
egr
at
e
d
into
a
new
ly
de
ve
lo
pe
d
crosst
al
k
noise
an
d
loss
analy
sis
platform
(CLAP
)
to
analy
ze
the
cro
s
sta
lk
noise
an
d
SN
R
in
W
DM
-
bas
e
d
O
No
Cs
of
a
ny
netw
ork
siz
e
us
in
g
an
ar
bi
trary
op
ti
cal
ro
ute
r.
The
a
naly
ses’
r
esults
dem
on
st
rate
that
the
c
r
os
sta
lk n
oise
is
of
c
riti
cal
concern
t
o
WD
M
-
base
d
O
N
oCs:
in
th
e
worst
case,
the
cro
sstal
k
nois
e
power
exce
e
ds
the
sig
nal
powe
r
in
al
l
three
WD
M
-
base
d
O
No
C
arc
hitec
tures,
even w
he
n
the
nu
m
ber
of pr
oc
esso
r
c
or
es
is
sm
a
ll
, e.
g.
,
64 [1
0].
Yiyuan
Xie
an
d
al
.
analy
sed
and
m
od
el
le
d
t
he
cr
os
sta
lk
noise
,
s
ig
nal
-
to
-
no
ise
rati
o
(SNR),
a
nd
bit
error
rate
(BE
R)
of
o
ptica
l
r
ou
te
rs
an
d
O
N
oCs
[
11
]
-
[12]. Th
e
analy
ti
cal
m
od
el
s
fo
r
cr
osst
al
k
no
ise
,
m
ini
m
u
m
SN
R,
a
nd
m
axi
m
u
m
BER
in
m
esh
base
d
O
No
Cs
are
de
ve
lop
e
d
wh
ic
h
a
n
a
uto
m
at
ed
cr
os
sta
lk
a
naly
se
r
f
or
op
ti
cal
routers
is
dev
el
ope
d.
They,
fi
nd
that
cro
sstal
k
no
is
e
sign
ific
a
ntly
lim
it
s
the
scal
a
bili
ty
of
ONoC
s.
For
exam
ple,
due
t
o
c
ro
s
sta
lk
no
i
se,
the
m
axi
m
u
m
BER
is
10
-
3
on
the
8×
8
m
esh
base
d
ONo
C
us
in
g
a
n
opt
i
m
iz
ed
cro
ss
ba
r
-
base
d
opti
cal
router.
To
ac
hiev
e
th
e
BER
of
10
-
9
for
reli
able
tr
ansm
issi
on
s,
t
he
m
axi
m
u
m
ONoC
siz
e is 6×
6.
As
we
pr
e
sent
ed
ab
ove
the
r
esearch
w
orks
in
opti
cal
netw
ork
on
c
hip
ha
s
lim
it
ed
to
an
al
yse
and
to
m
od
el
cro
sstal
k
noise
in
both
op
ti
cal
co
m
po
nen
ts
an
d
the
wh
ole
ne
twork
.
Howe
ve
r,
Ed
oard
o
F
us
el
la
,
Alessan
dro
Ci
la
rdo
pr
e
sent
di
ff
e
ren
t
st
ud
y
i
n
ONoC
wh
ic
h
they
pro
pose
d
a
ne
w
m
app
i
ng
syst
em
fo
r
ONoC
wh
e
re
it
c
on
si
der
the
direct
e
ff
ect
of
the
c
rossta
lk
noise
t
o
the
arc
hitec
tur
e
of
N
oC
[13].
They
pro
pose
a
cl
ass
of
al
gorithm
s
that
aut
om
at
i
cal
ly
m
ap
the
ap
plica
ti
on
t
asks
onto
a
ge
ner
ic
m
esh
-
ba
sed
phot
on
ic
N
oC
arch
it
ect
ure
s
uc
h
that
t
he
wor
st
-
case
cr
os
sta
l
k
is
m
ini
m
iz
ed
.
F
ur
th
erm
or
e,
the
res
ults
s
how
that
t
he
c
ro
s
sta
lk
no
ise
ca
n
be
si
gn
i
ficantl
y
reduced
by
ad
op
ti
ng
t
he
awa
re
-
c
ro
sstal
k
m
app
ing
syst
em
,
ther
eby
al
lowi
ng
higher
netw
ork
scal
a
bili
ty
,
and
ca
n
ex
hib
it
enc
oura
ging
im
pr
ov
em
ents
over
app
li
cat
ion
-
obli
vious
arc
hitec
tures
[13]
-
[
14]
.
3.
BASI
C
OPTI
CA
L
DE
VICES A
ND N
ET
WORK
MO
D
EL
Op
ti
cal
com
mu
nicat
io
n
syst
e
m
of
fer
s
a
high
ba
ndwidt
h
an
d
high
Qu
al
it
y
of
Se
rv
i
ce
(
QoS)
.
Howe
ver,
op
ti
cal
co
m
po
ne
nt
s
in
ONoC
intr
oduce
cr
os
sta
l
k
noise
.
I
n
this
sect
ion
,
we
de
scribe
the
pro
gress
of
cro
sstal
k
in
the
opti
cal
d
e
vice
m
or
eov
er
in
t
he
who
le
netw
orks
[15
]
.
3.1.
Opt
ic
al de
vice
s
mo
del
Wav
e
guides
a
nd
m
ic
ro
-
re
sonat
or
(MRs)
ar
e
the
two
ba
sic
op
ti
cal
el
e
m
ents
an
d
exte
nsi
vely
us
ed
t
o
const
ru
ct
basic
opti
cal
switc
hi
ng
el
em
ents
and
opti
cal
r
ou
t
ers.
I
n
par
ti
cul
ar,
we
ha
ve
t
wo
ty
pes
of
opti
cal
switc
hes:
the
pa
rall
el
switc
hi
ng
el
em
ent
(P
SE)
a
nd
t
he
c
ros
sing
sw
it
chi
ng ele
m
ent
(CSE).
The
se
el
em
ents
are
1 x 2
op
ti
cal
s
witc
hing
base
d t
o
MR
s a
nd
w
aveguide
cr
os
s
ing
[
16
]
-
[17].
Figure
1
(
a
)
a
nd
(
b
)
prese
nt
the
str
uctu
re
of
PS
E
a
nd
C
S
E
res
pecti
vely
in
the
t
wo
sta
te
s
OFF
a
nd
ON.
Wh
en
the
sta
te
of
MR
n
is
ON
,
we
sel
e
ct
to
switc
h
the
wav
el
en
gth
t
o
form
the
first
wav
eg
uid
e
to
the
seco
nd
this
ca
ll
ed
no
de
in
s
ta
te
ON.
How
ever,
wh
e
n
th
e
sta
te
MR
n
i
s
O
FF
we
ha
ve
not
a
ny
s
witc
hing
op
e
rati
on a
nd in
this
case, t
he node is
in
sta
te
O
F
F [1
7].
M
R
1
M
R
n
M
R
1
M
R
1
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
λ
n
O
F
F
S
t
a
t
e
M
R
1
M
R
w
M
R
1
M
R
n
M
R
w
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
λ
n
O
N
S
t
a
t
e
(
a
)
M
R
1
M
R
n
M
R
1
M
R
1
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
O
F
F
S
t
a
t
e
M
R
1
M
R
w
λ
n
M
R
1
M
R
n
M
R
w
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
λ
n
O
N
S
t
a
t
e
(
b
)
M
R
1
M
R
n
M
R
1
M
R
1
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
λ
n
O
F
F
S
t
a
t
e
M
R
1
M
R
w
M
R
1
M
R
n
M
R
w
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
λ
n
O
N
S
t
a
t
e
(
a
)
M
R
1
M
R
n
M
R
1
M
R
1
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
O
F
F
S
t
a
t
e
M
R
1
M
R
w
λ
n
M
R
1
M
R
n
M
R
w
.
.
.
.
.
.
D
r
o
p
A
d
d
I
n
p
u
t
T
h
r
o
u
g
h
λ
n
λ
n
O
N
S
t
a
t
e
(
b
)
Fig
ure
1
.
Ba
sic
opti
cal
d
e
vices in
ONoC:
(a) PSE,
(b) C
SE
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d Mo
nitorin
g In
t
ra
/I
nter Cr
os
st
alk in O
ptical
Ne
tw
or
k on C
hi
p
(
Ah
med Jedi
di)
4915
3.2.
Net
w
or
k m
odel
Figure
2
pr
ese
nts
an
ove
rv
ie
w
of
a
n
opti
cal
No
C
c
omm
un
ic
at
ion
betw
e
en
tw
o
proces
so
rs
C0
an
d
C1.
Wh
e
n,
th
e
process
or
C
0
deci
des
to
e
sta
blish
co
nne
ct
ion
with
t
he
cor
e
C1
a
n
opti
cal
sign
al
will
be
gen
e
rated
with
a
sp
eci
fic
wa
ve
le
ng
th
λ.
Ele
ct
rical
-
Op
ti
cal
(E
-
O)
inter
face
on
the
pa
rt
of
the
co
re
C
0
sta
rts
to
conve
rt
the
el
ect
rical
sign
al
in
op
ti
cal
signa
l.
Af
te
r,
t
he
op
ti
cal
sign
al
s
passes
thr
ough
the
opti
cal
on
-
c
hi
p
netw
ork
fl
ow
i
ng
s
pecific
nodes.
Fi
nally
,
the
op
ti
cal
sig
na
l
detect
ed
with
photodect
ors
i
n
O
ptica
l
-
al
ect
rical
(O
-
E)
inte
rf
ac
e.
Thus,
opti
cal
nodes
are
th
e
key
el
e
m
ent
in
op
ti
cal
N
oC
and
m
any
of
them
are
pr
opose
d
i
n
diff
e
re
nt ty
pes of net
work [
18
]
-
[19].
O
p
t
i
c
a
l
N
e
t
w
o
r
k
o
n
c
h
i
p
P
r
o
c
e
s
s
o
r
C
0
P
r
o
c
e
s
s
o
r
C
1
E
-
O
c
o
n
v
e
r
t
e
r
O
-
E
c
o
n
v
e
r
t
e
r
Fig
ure
2
.
O
verview
of
op
ti
cal
co
m
m
un
ic
at
ion
via O
n
oC
To
e
valuate
our
syst
em
we
con
si
der
t
he
network
m
od
el
as
show
n
in
Fig
ur
e
3.
We
as
sum
e
that
the
desig
n
of
t
he
ne
twork
c
om
posed
by
sta
ges
,
wh
ic
h
the
num
ber
of
sta
ge
s
is
direct
relat
ed
with
t
he
num
ber
of
process
or
co
re
s
N.
Mo
reove
r,
the
nu
m
ber
of
sta
ges
P
gi
ve
n
by
the
eq
uatio
n
1,
w
her
e
N
i
s
the
total
nu
m
ber
of
process
or co
re
s.
1
2
l
o
g
l
o
g
*
2
N
P
(
1)
Accor
ding the
nu
m
ber
of stag
es P
i
n
O
N
oC t
he
total
nu
m
be
r of
node
s R is:
1
2
*
2
1
2
1
p
i
i
N
R
(
2)
The
i
nn
e
r
a
rch
i
te
ct
ur
e o
f
node
router
com
po
s
ed
by
pa
rtic
ula
r
num
ber
β of MR
accor
ding the
num
ber
of
wa
velen
gth
s
ʎ use
d
i
n ONo
C.
C
0
C
1
C
2
C
w
.
.
.
C
w
+
1
C
n
.
.
.
C
n
-
1
C
w
+
2
C
3
C
w
-
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
s
t
a
g
e
1
s
t
a
g
e
2
s
t
a
g
e
p
.
.
.
M
R
Figure
3
.
Str
uc
ture of
Opti
cal
Netw
ork o
n
C
hip
4.
CROSST
AL
K NOISE I
N O
NOC
Op
ti
cal
crosst
al
k
is
pr
ese
nted
in
ONoC
com
pone
nts
an
d
de
gr
a
des
the
qu
a
li
ty
of
sig
nals,
increasi
ng
their
BER
(Bit
Error
Ra
te
)
pe
rfor
m
ance
as
they
travel
th
r
ough
the
netw
or
k
[
20]
.
I
n
ad
di
ti
on
,
c
ro
s
sta
lk
no
ise
increases
sig
na
l
-
to
-
no
ise
rati
o
(SNR)
an
d
a
ffec
ts
the
qu
a
li
ty
of
se
rv
ic
e
(Qo
S)
of
O
N
oC.
I
n
fact,
bo
t
h
for
m
s
of
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
491
2
-
4921
4916
op
ti
cal
crosst
al
k
can
arise
in
ONOC
r
ou
te
rs
:
inter
-
cr
os
sta
lk
an
d
intra
-
cr
osst
al
k
[5
]
.
W
e
con
si
der
bo
t
h
intra
-
channel a
nd int
er
-
c
hannel i
n o
ne
c
ro
s
sta
lk fo
rm
an
d we call
it
cro
s
sta
lk
no
i
se.
The
intri
ns
ic
char
act
erist
ic
of
photonic
de
vices
al
lows
the
cro
sstal
k
i
nduc
ing
to
the
opti
cal
sign
al
accor
ding
the
wav
el
e
ng
t
hs
c
ro
ssi
ng
i
n
the
op
ti
cal
de
vices
.
W
e
ass
um
e,
that
the
rate
of
the
cr
os
sta
lk
no
is
e
∆
λ
is
i
nduce
d
by
the
node
i
with
wa
velen
gth
λj
an
d
the
rate
of
the
cr
os
sta
l
k
noise
∆
λ
is
in
duced
by
t
he
wav
e
guide l
wi
th w
a
velen
gt
h λj
[21]
-
[22]
.
We
co
ns
ide
r
the
w
or
st
case,
wh
e
n
each
nodes
an
d
wa
veguides
of
the
ne
twork
im
m
ediat
el
y
ind
uce
cro
sstal
k
nois
e
(inter/i
nt
ra
cro
sstal
k)
c
orrespo
nd
i
ng
wi
th
a
s
pecific
wa
velen
gth
λi.
The
proce
ss
of
com
m
un
ic
at
ion
betwee
n
co
r
es
beg
i
n
by
th
e
init
ia
ti
on
fr
om
the
pr
ocess
or
c
or
e
M
(CM)
to
connect
with
the
process
or
c
or
e
N
(CN
).
At
this
tim
e,
ON
oC
def
ines
the
route
betwee
n
CM
and
CN
a
nd
sel
ect
s
a
spe
ci
fic
wav
el
e
ng
t
h
for
this r
e
quest
.
This
r
ou
te
is
c
om
po
sed
by
a
sp
eci
fic
num
ber
of
node
s
an
d
wav
e
guides
.
T
her
e
fore,
t
he
longest
r
oute
is
com
po
sed
by
P
no
des
a
nd
P+1
wa
ve
gu
i
de
s.
Howe
ver,
the
shortest
r
oute
is
con
sti
tuted
by
1
no
de
an
d
two
wav
e
guides
.
A
s
m
entioned
be
fore,
th
e
op
ti
cal
com
po
nen
t
s
induce
direct
ly
the
cro
sstal
k
noise
to
the
op
ti
cal
sign
al
pas
sed
t
hro
ugh
them
.
Fu
rt
her
m
or
e,
we
de
fine
the
total
cr
os
sta
lk
no
ise
a
dd
e
d
to
the
opti
cal
si
gn
al
ʎ
induced
b
y t
he diffe
re
nt w
a
ve
gu
i
des
is:
ʎ
=
∑
∑
∆
λ
∄
=
1
=
1
w
he
re k
is the n
um
ber
of w
a
ve
gu
i
de
w
hich
c
on
sti
tute t
he
r
oute
b
et
wee
n
CM
an
d
CN and
ʎ
j
is the w
avel
eng
t
h
us
e
d
in
ONoC
deprive
of
ʎ
i.
Additi
on
al
ly
,
we
prese
nt
the
total
cro
sstal
k
no
ise
a
ppen
de
d
to
the
opti
cal
sign
al
ʎ
in
du
ce
d by th
e d
if
fer
e
nt ro
ut
er
nodes
is:
ʎ
=
∑
∑
∆
λ
∄
=
1
=
1
w
he
re
L
is
t
he
nu
m
ber
o
f
r
oute
r
nodes
w
hich
they
are
a p
a
rt
of
the
r
oute
be
tween
CM
an
d
CN.
T
hen,
the
total
cro
sstal
k no
ise
pow
e
r
i
n
the
CN for t
he
opti
cal
sign
al
ʎ
is
presente
d by:
ʎ
=
ʎ
+
ʎ
ʎ
=
∑
∑
∆
λ
∄
=
1
=
1
+
∑
∑
∆
λ
∄
=
1
=
1
Finall
y, the S
N
R of the
opti
cal
carr
ie
d o
n
t
he
w
a
velen
gth
ʎ
is
:
λ
=
1
0
log
(
λ
λ
)
w
he
re,
λ
is t
he
opti
cal
p
owe
r
si
gn
al
a
nd
λ
is t
he
cr
os
sta
lk
noise
pow
e
r.
To
point
out
the
da
ng
e
rous
aspect
of
the
c
ro
sstal
k
on
the
net
work,
we
evaluate
t
he
S
NR
a
nd
the
cro
sstal
k
pow
er
in
O
N
oC.
B
esi
des,
we
use
CLAP
to
ol
s
for
the
sim
ulati
on
w
hich
de
velo
ped
by
[
9
]
.
W
e
expose
th
e
SNR
and
the
pow
er
of
the
cr
os
s
ta
lk
no
ise
acc
ordi
ng
th
e
siz
e
of
the
netw
ork
and
t
he
num
ber
of
wav
el
e
ng
t
h us
ed
in
ON
oC. P
recisel
y, we
c
om
par
e b
et
wee
n t
he
lo
ngest
ro
ute and t
he
s
ho
rtest
one.
The
F
ig
ur
e
4
(a)
an
d
(
b)
pr
esent
the
S
NR
an
d
C
ro
s
sta
lk
powe
r
noise
accor
ding
t
he
siz
e
of
the
netw
ork
f
or
t
he
nu
m
ber
of
ʎ
equ
al
8
a
nd
16
res
pecti
vely
.
No
ti
ce
that
the
cro
s
sta
lk
powe
r
noise
sli
gh
tl
y
increase a
cco
r
ding the
size
of n
et
w
ork
, but
t
he
S
NR
decr
ea
ses.
Fu
rt
her
m
or
e,
i
t
is
cl
ear
that
the
cro
sstal
k
power
no
ise
is
gr
eat
er
f
or
the
longest
r
oute
than
the
sh
ort
est
o
ne
, is d
ue
to the add
it
ive n
at
ur
e of cro
sstal
k
noise
w
hich
it
induc
es each ti
m
e
w
hen
they has
c
r
os
s in
the r
oute
. Cons
equ
e
ntly
,
the cro
sstal
k
powe
r
no
ise
acc
um
ul
at
es thr
ough th
e o
ptica
l ro
ute
betwee
n
sourc
e cor
e
and
destinat
io
n
on
e.
We
noti
ce
that
the
SNR
beco
m
e
so
hig
h
w
he
n
the
num
ber
of
wav
e
le
ng
th
em
plo
ye
d
i
n
the n
et
wor
k
ar
e increase
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d Mo
nitorin
g In
t
ra
/I
nter Cr
os
st
alk in O
ptical
Ne
tw
or
k on C
hi
p
(
Ah
med Jedi
di)
4917
(a)
num
ber
of
ʎ=8
(b) nu
m
ber
of
ʎ=16
Figure
4
.
Cr
os
s
ta
lk noise
pow
er a
nd S
NR ac
cordin
g
t
he netwo
rk size
We
e
valuate
th
e
cr
os
sta
lk
po
wer
noise
a
nd
the
S
NR
acc
ordin
g
t
he
wav
el
eng
t
h
num
ber
e
m
plo
ye
d
i
n
the
O
N
oC
as
dep
ic
t
i
n
the
F
igure
5.
Partic
ularly
,
we
fixe
d
th
e
netw
ork
siz
e
by
16x16
cor
es
a
nd
we
swee
p
wav
el
e
ng
t
h
nu
m
ber
fr
om
2
to
32.
O
n
acc
ount
of
the
inc
rea
se
of
t
he
wa
vel
eng
t
h
num
ber
,
the
cr
os
sta
lk
powe
r
no
ise
ta
ke
s
a
n
e
xponentia
l
pro
gr
es
s.
H
oweve
r,
the
S
NR
co
ns
i
der
a
bly
dec
reases
w
hen
the
nu
m
ber
of
wav
el
e
ng
t
h
in
crease.
Def
i
nitel
y,
the
pres
en
ce
of
hi
gh
nu
m
ber
of
op
ti
ca
l
sign
al
s
in
th
e
sam
e
wav
eg
uid
e
or
op
ti
cal
c
om
po
nen
t
ca
us
e
s
a
di
rect eff
ect
t
o
c
ro
sstal
k
in
duce
d
in
the c
orres
pondin
g wav
el
eng
t
h.
Finall
y,
we
c
oncl
ude
that
t
he
cr
os
sta
lk
no
ise
becam
e
crit
ic
al
wh
en
we
increase
the
wav
el
e
ng
t
h
n
um
ber
or the
netw
ork
siz
e.
Accor
dingly
, det
ect
an
d m
on
it
or
c
rossta
lk in
ONo
C
becam
e ind
is
pe
ns
able
.
Figure
5
.
Cr
os
s
ta
lk noise a
nd
SN
R acc
ordin
g t
he nu
m
ber
of
wav
el
e
ng
t
hs
for 16x
16 cores
4.
CROSST
AL
K NOISE
DE
TE
CTION A
ND M
ONI
T
O
R
S
Y
STE
M
A
s
m
entioned
befor
e
,
crosst
a
lk
no
ise
is
seriou
s
obsta
cl
es
t
o
de
velo
p
opti
cal
network
on
chip
w
hic
h
the
reli
abili
ty
and
pe
rfor
m
ance
of
t
he
MPS
oC
will
be
cu
rb.
As
a
Re
s
ult,
fi
nd
a
syst
em
that
has
a
capa
bi
li
ty
t
o
detect
and
m
on
it
or
the
c
r
os
st
al
k
noise
in
O
No
C
is
esse
ntial
and
vital
.
Ma
inly
,
this
syst
e
m
m
us
t
resp
ect
the
fo
ll
owin
g req
ui
rem
ents
:
Eff
ic
ie
ncy
Scal
abili
ty
Faci
li
ty
to
i
m
ple
m
ent
To
detect
a
nd
m
on
it
or
crosst
al
k
in
O
N
oC,
we
pro
pose
th
e
Cros
sta
lk
De
te
ct
ion
an
d
Mon
it
or
Syst
em
(CDMS
).
T
he
m
ai
n
idea
of
C
DMS
is
to
m
on
it
or
c
onti
nuously
the
dif
fer
e
nt
i
m
pairm
ents
(intra/i
nter
c
r
osst
al
k)
in
the
w
ho
le
ne
twork
.
T
o
r
ea
ch
the
se
obj
ec
ti
ves
CDMS
i
s
com
po
se
d
by
sever
al
C
ro
s
sta
lk
Detect
io
n
Bl
oc
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
491
2
-
4921
4918
(CDB)
distri
bute
d
i
n
the
ne
twork
acco
rdi
ng
a
sp
e
ci
al
local
iz
at
ion
syst
e
m
as
sh
own
in
t
he
F
i
gure
6
.
Fu
rt
her
m
or
e,
CDB
is
place
d
betwee
n
tw
o
op
ti
cal
no
de
s
in
the
sam
e
le
vel
w
hich
C
DB
s
plit
s
the
var
i
ou
s
op
ti
cal
s
ig
nals
passe
d
th
r
ough
these
op
ti
cal
rou
te
rs.
The
proc
ess of
functi
on
of
C
DB is
gi
ven b
y t
he
al
go
rithm
1.
First,
we
s
plit
the
op
ti
cal
sign
al
s
from
t
he
input/
outp
ut
of
the
appr
opriat
e
opti
cal
nodes
.
To
real
iz
e
thi
s
op
e
rati
on
we
us
e,
a
photode
te
ct
or
m
od
ule
t
o
co
nvert
t
he
opti
cal
sign
al
s
in
el
ect
rical
sign
al
s.
Sec
ond,
t
he
diff
e
re
nt
input/
ou
t
pu
t
sig
nals
passe
d
to
CDB
wh
ic
h
the
pr
oc
ess
of
detect
ion
a
nd
local
iz
at
ion
of
the
c
r
os
sta
lk
no
ise
la
unc
he
d.
T
hird,
a
syst
e
m
of
cro
sst
al
k
cl
assifi
cat
i
on
beg
i
n
to
m
on
it
or
an
d
to
cl
assify
the
diff
e
ren
t
detect
ed
i
m
pairm
ents
according
the
val
ues
of
the
crosst
al
k
no
ise
s
.
More
ov
e
r,
the
cro
sstal
k
noi
ses
are
cl
assifi
ed on 3
ty
pes
as s
how
n i
n
the
Ta
ble
1.
C
0
C
1
C
2
C
4
C
5
C
3
s
t
a
g
e
1
s
t
a
g
e
2
s
t
a
g
e
4
C
4
C
5
s
t
a
g
e
3
s
t
a
g
e
5
C
D
B
C
D
B
C
D
B
C
D
B
C
D
B
C
D
B
Figure
6. P
rop
os
e
d
Cr
os
sta
lk
Detect
ion
an
d
Locali
zat
ion
S
yst
e
m
in
FON
oC
Table
1.
Im
per
t
m
ent ala
r
m
s ty
pes
I
m
p
er
m
en
t
Alar
m
Descripti
o
n
No
Cros
stalk
No
detectio
n
of
cr
o
ss
talk
(
h
ealth
y
si
g
n
al)
Acceptab
le Cros
st
alk
ʎ
≤
3
.
2
&&
ʎ
∈
ʎ
(
ℎ
ℎ
ℎ
)
Dan
g
en
rou
s Cro
ss
t
alk
ʎ
>
3
.
2
∀
ʎ
(
)
OR
ʎ
≤
3
.
2
&&
ʎ
∈
ʎ
(
ℎ
ℎ
ℎ
)
The
C
DMS
m
on
it
or
t
he
cr
osst
al
k
noise
by
evaluate
t
he
st
at
us
a
nd
featu
r
es
of
the
sig
na
l,
w
hich
t
he
cl
assifi
cat
ion
of
the
cr
os
sta
lk
no
ise
is
esse
ntial
.
Indeed,
w
he
n
the
cr
os
sta
l
k
noise
po
wer
is
gr
eat
than
3.2
db
the
cr
os
sta
lk
noise
is
cl
assifi
ed
as
da
nger
ous
and
we
m
us
t
gen
e
rate
an
al
a
rm
wh
at
ever
th
e
local
iz
at
ion
of
t
he
op
ti
cal
com
ponen
t
ho
w
res
pons
i
ble
for
it
.
Be
sides,
this
sit
uation
nam
ed
Dange
rous
cr
osst
al
k
le
vel.
H
ow
e
ve
r,
wh
e
n
t
he
c
ro
s
s
ta
lk noise
pow
er is less t
han
3.2 db,
we hav
e two si
tuati
on
:
1.
If
the
opti
cal
com
po
ne
nt
how
has
the
respo
nsi
bili
ty
of
this
i
m
pair
m
ent
is
locat
ed
in
the
uppe
r
-
half
r
ou
t
e,
and
we
cl
assi
f
y
this
crosst
al
k
no
ise
as
acc
eptable.
Be
si
de
s,
the
acce
pta
ble
cr
os
sta
lk
l
evel
al
low
to
t
he
op
ti
cal
sig
nal
r
each the
dest
in
at
ion
without
a
la
rm
b
ut they
m
us
t be ex
am
i
ned.
2.
If
we
detect
the
cr
os
sta
lk
noise
at
the
lo
wer
-
hal
f
r
ou
te
and
al
arm
is
gen
e
rated
a
nd
we
cl
assify
this
cro
sstal
k
as
da
ng
e
r
ou
s
.
Be
ca
us
e,
th
e
pro
ba
bi
li
t
y
to
ind
uce
m
or
e
cro
sstal
k
is
hig
h
a
nd
ex
trem
e
ly
aff
ect
the
op
ti
cal
sig
na
l.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d Mo
nitorin
g In
t
ra
/I
nter Cr
os
st
alk in O
ptical
Ne
tw
or
k on C
hi
p
(
Ah
med Jedi
di)
4919
Algorithm
1
. C
ro
sstal
k no
ise
Mon
it
ori
ng
FOR
ʎi=
1
to
32
IF
(C
ro
sstal
k d
et
ect
ion
=
1)
T
hen
IF (
ʎ
>
3
.
2
) T
hen
Im
per
m
ent A
la
rm
= D
an
gerous Crosst
al
k
El
se IF
(
ʎ
≤
3
.
2
&&
ʎ
∈
ʎ
) T
he
n
Im
per
m
ent
Ala
rm
= D
an
gerous Crosst
al
k
El
se IF
(
ʎ
≤
3
.
2
&&
ʎ
∈
ʎ
) T
he
n
Im
per
m
ent A
la
rm
= A
ccepta
bl
e Cro
sstal
k
END IF
El
se IF
(C
ro
s
s
ta
lk d
et
ect
io
n = 0
)
Then
Im
per
m
ent A
la
rm
= N
o
C
ro
ss
ta
lk
END IF
END FO
R
To
reali
ze
thes
e
op
e
rati
ons
in
real
-
tim
e
fu
nc
ti
on
we
desig
n,
si
m
ulate
and
i
m
ple
m
ent
the
CDMS
in
RTL.
I
ndeed
,
the
in
ner
a
rc
hitec
ture
of
t
he
CDB
sho
wn
in
the
Fig
ur
e
7.
F
ur
the
rm
or
e,
CDB
c
om
po
ses
by
sever
al
MR
i
to
sp
li
t
the
dif
fere
nt
opti
cal
sig
nal
ʎi
w
hich
th
e
num
ber
of
M
Ri
equ
al
t
o
the
num
ber
of
ʎ.
Next
,
we
us
e
d
a
phot
od
et
ect
or
to
conve
rt
op
ti
cal
sign
al
to
el
ect
rical
on
e
and
th
e
nu
m
ber
of
th
e
photodete
ct
ors
ar
e
the
num
ber
s
of
ʎ.
For
the
i
nput
sig
nals
pa
ssed
to
delay
process
t
o
synch
r
on
iz
e
with
the
ou
t
pu
t
s
ign
al
s
.
Finall
y,
a
syste
m
to
detect
t
he
crosst
al
k
noise
pr
ese
nte
d
with
the
com
plexity
and
the
cost
a
directl
y
relat
ed
with the
num
ber
of
wa
velen
gt
h
use
d.
CDMS
is
a
distribu
te
d
syst
em
us
e
the
different
inf
orm
ati
on
s
c
ollec
te
d
f
ro
m
the
CDB
dev
ic
es
.
W
e
centrali
ze
thes
e
inf
or
m
at
ion
s
in
the
C
DMS
,
w
hich
we
proceed
the
m
onit
or
in
g
of
t
he
cro
sstal
k
noise
in
the
whole
netw
ork.
Mor
eo
ver,
the
CDMS
loc
al
iz
e
and
cl
assify
the
diff
er
ent
cro
sstal
k
no
ise
s
in
ob
j
e
ct
ive
to
gen
e
rate t
he
appr
opriat
e ala
rm
s as
dep
ic
t i
n t
he Alg
or
it
hm
1.
MR
1
MR
n
MR
w
OP
w
OP
n
OP
1
D
o
w
n
s
t
r
e
a
m
(
s
p
l
i
t
s
i
g
n
a
l
)
MR
w
MR
n
MR
1
OP
1
OP
n
OP
w
U
p
s
t
r
e
a
m
(
s
p
l
i
t
s
i
g
n
a
l
)
D
e
l
ay P
r
oc
e
s
s
C
r
os
s
t
al
k
D
e
t
e
c
t
i
on
D
e
l
ay P
r
oc
e
s
s
C
r
os
s
t
al
k
D
e
t
e
c
t
i
on
D
e
l
ay P
r
oc
e
s
s
C
r
os
s
t
al
k
D
e
t
e
c
t
i
on
Q
Q
S
E
T
C
L
R
D
Q
Q
S
E
T
C
L
R
D
-
=
=
C
r
p
T
h
1
C
r
p
T
h
2
A
la
r
m
1
0
A
la
r
m
2
1
A
la
r
m
.
.
.
.
.
.
.
.
.
Figure
7. Cr
osst
al
k
Detect
io
n B
lock
5.
RESU
LT
S
AND A
N
ALYSIS
To
disc
us
s
the
feasibil
it
y,
the
reli
abili
ty
,
t
he
scal
abili
ty
and
t
he
cost
of
our
syst
em
,
we
sim
ulate
,
synthesiz
e
an
d
i
m
ple
m
ent
C
DMS
in
FP
G
A.
F
ur
t
her
m
ore,
we
us
e
d
the
STA
RTER
K
it
of
Xili
nx
w
it
h
the
diff
e
re
nt
sim
u
la
ti
on
an
d
syn
theses
to
ols
(
Pr
oject
Na
viga
tor
of
Xili
nx
an
d
Mo
delSi
m
).
Partic
ularly
,
we
sel
ect
ed
SP
AR
TAN
-
3E for
our
work
[23].
We
stu
dy
the
cost,
the
c
om
plexity
and
the
scal
abili
ty
of
CDMS
acco
rd
i
ng
t
he
siz
e
of
the
netw
ork
and
the
num
ber
of
t
he
wav
el
e
ng
t
hs
us
e
d
in
ONoC.
Be
si
de
s,
these
pa
ram
et
ers
e
val
uate
by
the
occ
upat
ion
a
rea
of
t
he
CDMS
in
the
c
hip
as
sh
ow
n
in
t
he
Figure
8.
We
no
ti
ce
that
the
area
occ
upat
ion
i
n
the
c
hip
is
the
nu
m
ber
of
L
U
Ts
us
e
d
on
c
hip
.
Mo
re
ov
e
r,
t
he
total
num
ber
of
LUT
s
ex
pone
ntial
ly
increases
acco
rd
i
ng
the
nu
m
ber
of
pro
cesso
r
cor
es
s
cal
es.
Indeed,
this
evo
l
ution
exp
la
ine
d
by
the
increase
of
the
nu
m
ber
of
CDB
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
491
2
-
4921
4920
accor
ding
the
siz
e
of
the
network.
I
n
add
it
ion
,
we
rem
ar
k
that
the
execu
ti
on
ti
m
es
i
ncr
ease
s
sm
oo
thly
as
functi
on
as
t
he
num
ber
of
co
r
es
w
hich
t
hey
r
eflect
the
hi
gh
scal
ab
il
it
y
of
CDMS.
In
pa
rtic
ular,
t
he
a
ver
a
ge
of
the
exacti
ons
t
i
m
e
is
aro
und
23
m
ic
ro
sec
onds
.
As
res
ul
t,
CDMS
reac
h
a
real
ti
m
e
functi
on
with
a
high
scal
abili
ty
.
In
de
ed,
w
he
n
the
netw
ork
has
2048
pr
ocess
or
cor
es
C
DMS
ne
ed
le
ss
than
4000
LUT
s
an
d
this
value
i
s
0.0
1%
of
the
siz
e
of
the
chi
p
a
nd
th
e
exec
ution
ti
m
e
is
le
ss
than
25
m
ic
ro
seco
nd
s
.
T
he
pe
nalty
fr
om
the
photodete
c
tor
operati
on
sli
gh
tl
y
eff
e
ct
the
real
ti
m
e
f
un
ct
io
n
of
t
he
C
DMS
bec
ause
al
l
de
vices
ar
e
i
m
ple
m
ented
in
the
sam
e
chip
a
nd
t
he
c
on
c
eption
of
the
C
DMS
boos
t
the
real
tim
e
execu
ti
on
by
a
high
RTL
desig
n.
Figure
8
.
Com
plexity
and sca
la
bili
ty
o
f
CD
LS in
ch
i
p
To
bette
r
unde
rstan
ding
the
cost
an
d
the
com
plexity
of
the
CDMS,
we
ex
plore
th
e
cost
an
d
the
scal
abili
ty
of
CDB
accor
ding
the
num
ber
of
the
wa
veleng
t
h
ʎi
use
d
i
n
O
No
C
.
The
Figure
9
pr
e
se
nts
the
com
plexity
of
the
CDB
de
fin
e
as
the
rate
of
the
L
uTs
nu
m
ber
over
the
w
avelen
gth
num
ber
ʎ
use
d
in
ONoC
.
Si
m
il
arly
,
we
pr
ese
nt
the
fl
ow
of
data
proc
ess
as
perf
or
m
ance
a
nd
scal
a
bili
ty
of
CDB
.
W
e
real
iz
e
t
hat
the
com
plexity
of
CDB
increase
sign
ific
a
ntly
wh
en
t
he
num
ber
of
ʎ
is
between
2
a
nd
16
the
n
this
va
riat
io
n
sli
gh
tl
y
sta
bilizes
. I
ndee
d,
thi
s p
rog
ress
of the co
m
plexity
is du
e to the
re
us
e of the h
ar
dware b
l
ocs,
wh
ic
h
the
com
plexity
rate
is
64%
f
or
ʎ
e
qu
al
16
th
an
it
is
62%
for
ʎ
eq
ual
32.
Othe
rw
ise
,
we
noti
ce
that
the
perform
ance
and
scal
a
bili
ty
of
C
DB
are
m
os
tl
y
con
sta
nt
as
functi
on
the
nu
m
ber
of
wa
velen
gth
s
.
As
a
resu
lt
,
the
CDB
offe
r
s
a
hi
gh
scal
a
bili
ty
and
perf
or
m
ance
with
an
a
ppr
opriat
e
com
plex
it
y
wh
en
the
num
ber
of
wav
el
e
ng
t
hs
e
xceeds
16.
Figure
9
.
Com
plexity
and
Da
ta
f
lo
w of
C
D
B
6.
CONCL
US
I
O
N
T
he
te
ch
no
l
ogy
gr
owth
of
tr
ansisto
r
integr
at
ion
has
no
w
reache
d
it
s
lim
it
s.
Moreo
ve
r,
this
high
transisto
r
i
nteg
rati
on
rate
has
pu
s
he
d
t
he
se
m
ic
on
duct
or
in
du
st
ry
to
sh
i
ft
from
the
sing
le
-
co
re
to
a
m
ulti
-
cor
e
in
one
c
hip
(
MPSoC)
.
On
e
of
the
seri
ous
pro
blem
s
of
the
MPS
oC
i
s
the
c
omm
un
ic
at
ion
s
betwe
en
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Detect
ion
an
d Mo
nitorin
g In
t
ra
/I
nter Cr
os
st
alk in O
ptical
Ne
tw
or
k on C
hi
p
(
Ah
med Jedi
di)
4921
diff
e
re
nt
pr
oce
sso
r
’s
co
res.
I
n
this
c
on
te
xt,
Netw
ork
on
C
hip
is
a
prom
ote
so
luti
on
t
o
s
olv
e
t
his
pro
ble
m
bu
t
is
lim
i
te
d
by
th
e
increase
d
nu
m
ber
of
c
ores
i
m
ple
m
ented
on
c
hip
.
O
ptica
l
Netw
ork
on
C
hip
is
a
pr
om
ot
ion
al
so
luti
on
that
s
olv
es
the
pro
ble
m
of
high
rat
e
of
data
exc
ha
ng
e
betwee
n
cor
es
with
le
ss
energy
consu
m
pt
ion
.
Howe
ver,
in
opti
cal
com
m
un
ic
at
ion
ONoC
is
af
fected
by
cr
os
sta
lk
nois
e,
w
hich
is
a
m
ajo
r
pr
ob
le
m
that
hinders
t
he
achievem
ent
and
m
ai
ntenan
ce
of
high
pe
rform
ance.
In
fac
t,
cro
sstal
k
no
ise
deterior
at
e
s
the
qu
al
it
y
of
sig
na
ls
and
he
nce
degra
des
syst
em
’s
per
form
ance.
I
n
this
paper
we
pro
pose
d
a
new
syst
e
m
to
detect
and
m
on
it
or
c
ro
s
sta
lk
no
ise
in
ONoC,
w
hich
t
he
con
tri
bu
ti
on
of
this
work
i
f
to
offe
r
the
first
com
plete
ly sys
tem
to
detect
a
nd
m
on
it
or ala
rm
s ind
uced
by
cro
sstal
k
i
n
ONoC. P
a
rtic
ul
arly
, w
e d
esc
ri
bed
t
he
distrib
uted
a
rc
hitec
ture
a
nd
the
f
unct
ion
of CDMS,
al
s
o,
w
e
f
oc
us
ed
on the
ha
rdwa
re
de
sign
o
f
CDB. Finall
y,
we
im
ple
m
ente
d
a
nd
sim
ulated
our
syst
em
to
evaluate t
her
e
perform
ance. The
resu
lt
s
ha
ve
d
em
on
strat
ed
that,
our
syst
em
of
f
ers
a
hi
gh
scal
abili
ty
with
lo
w
rate
of
occ
upat
ion
in
area
of
the
c
hip
as
well
as
a
real
-
tim
e
functi
on
with
23 m
ic
ro
seco
nds as e
xecu
ti
on t
i
m
e.
REFERE
NCE
S
[1]
E.
Bone
tt
o,
et
a
l.
,
“
Opti
ca
l
te
ch
nologi
es
c
an
improve
the
ene
rg
y
ef
ficien
c
y
o
f
net
works
,
”
Proc
.
35th
Eur.
Con
f
.
Opt.
Comm
un
.
,
pp.
1
–
4
,
2009
.
[2]
M.
J.
Cia
nch
et
t
i,
et
al.
,
“
Phastla
n
e:
A
rap
id
tr
ansi
t
opti
c
al
rout
ing
net
work,”
Proc.
36th
Annu.
ISC
A
,
pp.
441
–
450
,
2009
.
[3]
L
.
Zhou
,
et
a
l.
,
“
A
Joint
-
Coding
Scheme
with
Cross
ta
lk
Avoidance
in
Net
work
on
Chip
,
”
TEL
KOMNIKA
(
Tele
communic
ati
on
Computing
El
e
ct
ro
nic
s and
Control)
, v
ol
/i
ss
ue:
11
(
1
)
,
2013
.
[4]
M
.
F
.
Chat
m
en
,
et
al
.
,
“
New
Design
of
Netw
ork
on
Chip
Ba
sed
on
Virtual
Route
rs
,
”
Indon
esian
Journal
o
f
El
e
ct
rica
l
Eng
in
ee
ring a
nd
Computer
Sc
ie
nc
e
(
IJE
ECS)
,
v
ol
/
issue:
2
(
1
)
,
pp.
115
-
1
31
,
2016
.
[5]
H
.
A.
M.
Harb,
et
al.
,
“
A
Study
of
the
Num
ber
of
W
ave
le
ngths
I
m
pac
t
in
the
Opt
ic
a
l
Bu
rst
Sw
it
chi
ng
Core
Node
,
”
Proce
ed
ing
of
th
e
El
e
ct
rica
l
Eng
ine
ering
Co
mput
er
Sci
ence
and
I
nformatic
s
,
Yog
y
ak
arta,
Indone
s
ia
,
pp.
664
-
667
,
2017
.
[6]
A
.
Jedidi
,
et
al
.
,
“
Dete
ct
ion
and
loc
al
i
zation
of
cro
ss
ta
lk
in
an
al
l
-
opt
ical
net
w
ork
,
”
Journal
of
opti
cs
,
vol
/i
ss
u
e:
13
(
1
)
,
2011
.
[7]
Y
.
Xie,
et
al
.
,
“
Cross
ta
lk
Noise
Anal
y
sis
and
Optimiza
t
ion
in
5x
5
Hitl
ess
Sili
con
-
Based
Optical
Route
r
for
Opti
c
al
Networks
-
on
-
Chip
(ON
oC)
,
”
Jou
rnal
o
f
Ligh
twave
Techno
logy
,
v
ol/
issue:
30
(
1
)
,
2
012
.
[8]
F
.
Gam
bini
,
e
t
a
l.
,
“
BER
eva
l
uation
of
a
low
-
cro
ss
ta
lk
sili
con
integra
t
ed
m
ult
i
-
m
i
cro
ring
ne
twork
-
on
-
chi
p
,
”
opt
ic
s
ex
press
journal
,
2015
.
[9]
M.
Nikdast
,
et
a
l
.
,
“
S
y
stematic
a
naly
s
is
of
cro
ss
t
al
k
noise
in
fo
ld
ed
-
torus
base
d
o
pti
c
al
ne
tworks
-
on
-
chi
p,
”
IE
E
E
Tr
ans.
Comput.
-
Ai
ded
Design
In
t
egr.
Circu
it
s S
yst.
,
vol
/
issue:
33
(
3
)
,
pp
.
437
–
450
,
2014.
[10]
M
.
Nikdast,
et
a
l.
,
“
Cross
ta
lk
Noise
in
W
DM
-
base
d
Optical
Ne
tworks
-
on
-
Chip:
a
Form
al
Stud
y
an
d
Com
par
ison
,
”
IEE
E
Tr
ansacti
o
ns on
Ve
ry
Lar
ge
Sca
le Int
egrat
i
on
Syste
ms
,
vol
/
i
ss
ue:
23
(
11
)
,
pp.
2552
-
2565,
201
5
.
[11]
Y.
Xie
,
et
a
l
.
,
“
For
m
al
wors
t
-
ca
se
anal
y
s
is
of
cro
ss
ta
lk
noise
in
m
esh
-
base
d
opti
c
al
ne
tworks
-
on
-
chi
p,
”
IEEE
Tr
ans.
Ve
ry
Lar
ge
Sca
le Int
egr
.
(
VLSI
)
Syst
.
,
vol
/i
ss
ue:
21
(
10
)
,
p
p.
1823
–
1836
,
2
013.
[12]
Y.
Xie
,
et al
.
,
“
C
ross
ta
lk
noise an
d
bit
err
or
ra
te
a
naly
s
is f
or
opt
ical
ne
twork
-
on
-
c
hip,
”
Proc. 47t
h
ACM/
IE
EE
DAC
,
pp.
657
–
660
,
20
10
.
[13]
E
.
Fus
el
l
a
and
A
.
Cil
ard
o,
“
PhoN
oCMap:
an
Appl
ic
a
ti
on
Mapping
Tool
for
Photon
ic
Networks
-
on
-
Chip
,
”
D
esign
,
Aut
omation
&
T
est
in
Europe
Co
nfe
renc
e
&
Ex
hibi
ti
on
(
DATE)
,
2
016
.
[14]
E
.
F
usella
and
A
.
C
ilardo
,
“
Cr
oss
ta
lk
-
Aw
are
Autom
at
ed
Map
ping
for
Opti
cal
Networks
-
on
-
Chip
,
”
ACM
Tr
an
s.
Embe
d.
Comput
.
Syst
.
,
vo
l/
issue:
16
(
1
)
,
2016
.
[15]
J.
Chan,
et
a
l.
,
“
Ph
y
sical
-
l
a
y
er
m
odel
ing
and
s
y
s
t
em
-
le
vel
design
of
chi
p
-
sc
al
e
ph
otoni
c
int
er
conn
ec
t
ion
ne
tworks,”
IEE
E
Tr
ans.
Comput.
-
Ai
d
ed
Des
ign
Int
egr.
C
ircu
it
s Sy
st
.
,
vol
/
issue:
30
(
10
)
,
pp
.
15
07
–
1520,
2011
.
[16]
W
.
Ding,
et
al
.
,
“
Com
pac
t
and
low
cro
ss
ta
lk
wave
guide
cro
ss
ing
using
impedanc
e
m
at
ch
ed
m
etam
at
erial
,
”
App
l
.
Phy
s.
Le
tt
.
,
vol
/
i
ss
ue:
96
(
11
)
,
pp.
111114
-
1
–
1111
14
-
3,
2010
.
[17]
Q.
Li
,
et
al
.
,
“
Design
and
dem
onstrat
ion
of
co
m
pac
t,
wide
ba
ndwidth
coupl
e
d
-
resona
tor
filte
rs
on
a
s
il
ic
on
-
on
-
insula
tor
platfor
m
,
”
Opt.
E
xp
.
,
v
ol
/i
ss
ue:
17
(
4
)
,
p
p.
2247
–
2254
,
2
009.
[18]
D
.
Nikolova,
e
t
al.
,
“
Scal
ing
sili
con
photonic
sw
it
ch
fab
r
ic
s
f
or
data
c
ent
er
i
nte
rco
nn
ec
t
ion
net
works
,
”
Opt
.
Ex
press
,
vol
/i
ss
ue:
23(2)
,
pp
.
115
9
–
1175
,
2015
.
[19]
M.
Petracc
a
,
e
t
al
.
,
“
Design
Ex
plora
ti
on
of
Opt
ic
a
l
Int
erc
onne
ction
Net
works
fo
r
Chip
Multi
pro
ce
ss
ors,
”
HOTI
,
2008.
[20]
L.
H.
K.
Duong
,
et
al
.
,
“
A
ca
se
stud
y
of
signal
-
to
-
noise
rat
io
in
ring
base
d
opti
c
al
net
works
-
on
-
chi
p,
”
IEEE
Des.
Test
Comput.
,
v
ol
/i
ss
ue:
31
(
5
)
,
p
p.
55
–
65
,
2014
.
[21]
A.
Parini
,
e
t
al.
,
“
BER
eva
lua
ti
o
n
of
a
passive
S
OI
W
D
M
route
r,
”
IEEE
Phot
on
.
Technol
.
Lett
.
,
vol/
issue:
25(23)
,
pp.
2285
–
2288
,
2013
.
[22]
H.
Gu,
et
al.
,
“
A
Low
-
power
Fat
Tre
e
-
b
ase
d
Optic
a
l
Network
-
on
-
Chip
for
Multi
proc
essor
S
y
stem
-
on
-
Chip
,
”
DATE
,
2009
.
[23]
F.
Xia,
et
al
.
,
“
Ultra
-
compac
t
h
igh
orde
r
ring
re
sonator
fil
t
ers
using
subm
ic
ron
sili
con
photon
ic
wire
s
for
on
-
chi
p
opti
c
al
int
er
conn
ec
ts,
”
Opt
ic
al
E
x
press
,
vol/
issue:
15
(
19
)
,
pp
.
1193
4
-
11941
,
2007
.
Evaluation Warning : The document was created with Spire.PDF for Python.