Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gi
n
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
5
,
Octo
ber
201
9
, pp.
4035~4
043
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
5
.
pp4035
-
40
43
4035
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
A
c
ompr
essive
s
en
sing
a
l
gorithm
f
or
h
ardware
t
rojan
d
ete
ctio
n
M.
Pri
yathari
shini
,
M.
Nir
mala
D
e
vi
Depa
rt
m
ent
o
f
E
le
c
troni
cs
and
C
om
m
unic
at
ion
E
ngine
er
ing,
Am
rit
a
School
of
En
gine
er
ing,
Coim
bat
ore
,
Am
rit
a
Vishw
a Vid
y
ap
eetha
m
,
I
ndia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ju
l
12
, 2
01
8
Re
vised
A
pr
17
, 2
01
9
Accepte
d
Apr
2
8
, 201
9
Tra
di
ti
ona
lly
m
a
n
y
fab
le
ss
companie
s
outsourc
e
t
he
fab
ri
ca
t
ion
of
IC
design
to
the
foundri
es,
which
m
a
y
not
be
truste
d
a
lwa
ys.
In
orde
r
to
en
sure
truste
d
IC’s
it
is
m
o
re
s
igni
ficant
to
dev
el
op
an
eff
i
cient
te
chni
qu
e
tha
t
det
e
ct
s
th
e
pre
senc
e
of
har
dware
Troj
an
.
Thi
s
m
al
ic
ious
insert
ion
c
ause
s
the
logic
var
iation
in
th
e
net
s
or
leaks
som
e
sensiti
ve
inf
orm
at
ion
from
the
chi
p
,
which
red
uce
s
th
e
relia
bil
ity
of
the
s
y
s
t
em.
T
he
conv
en
ti
onal
te
sting
algorithm
for
gene
ra
ti
ng
t
est
vec
tors
red
u
ce
s
t
h
e
det
e
ction
sensiti
vity
du
e
to
h
igh
proc
ess
var
iati
ons.
In
this
work,
we
pre
sent
a
compress
ive
sensing
appr
oac
h,
wh
ich
ca
n
si
gnif
ic
an
tly
g
ene
ra
te
op
tim
al
te
st
pa
ttern
s
compare
d
to
the
ATPG
vec
tors
.
Thi
s a
p
proa
ch
m
axi
m
izes
the
prob
ability
of
Trojan
ci
rcu
it
a
ct
iv
ation
,
with
a
high
l
evel
of
Trojan
dete
ct
ion
r
at
e
.
Th
e
side
ch
anne
l
anal
y
sis
such
as
power
signat
ur
e
s
are
m
e
asure
d
at
d
iffe
r
ent
ti
m
e
stamps
to
isol
ate
th
e
Tro
ja
n
eff
ects.
Th
e
eff
e
ct
of
proc
ess
noise
is
m
ini
m
iz
ed
b
y
thi
s
po
wer
profile
compari
son
appr
oac
h,
which
pro
vide
s
high
det
e
c
ti
on
sensiti
vi
t
y
f
or
var
y
i
ng
Troj
an
size
and
el
iminates
th
e
r
e
quire
m
ent
of
gol
den
ch
ip.
The
pr
oposed
te
st
gene
ra
ti
on
a
pp
r
oac
h
is
va
li
da
t
ed
on
ISCA
S
benc
hm
a
rk
circ
uit
s,
which
ac
hi
eve
s
Trojan
det
e
ct
ion
cove
r
a
ge
on
an
av
era
g
e
of
88.
6%
r
edu
ct
ion
in
te
st
le
ngth
when co
m
par
ed
to
ran
do
m
pat
te
rn
.
Ke
yw
or
d
s
:
Com
pr
essive s
ensin
g
Hardwa
re
secu
rity
Hardwa
re
t
roja
n
Self r
e
fer
e
nci
ng
Test
g
e
ner
at
io
n
Copyright
©
201
9
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
:
M.
Pr
iy
at
ha
rishin
i,
Dep
a
rtm
ent o
f El
ect
ro
nics
and C
omm
un
ic
ation
En
gin
ee
rin
g,
Am
rita
Sch
ool
of Enginee
rin
g, Coim
bator
e
,
Am
rita
V
ishw
a
V
idya
peetham
, In
dia.
Em
a
il
:
m
_p
riy
a
tharishi
ni@c
b.a
m
rita
.ed
u
1.
INTROD
U
CTION
The
pres
um
pti
on
of
fabrica
te
d
ICs
no
t
be
in
g
sub
j
ect
ed
to
any
trade
off
in
secu
rity
was
pr
ese
nt
f
or
a
long
tim
e.
To
lowe
r
the
fabri
cat
ion
co
sts,
th
ird
par
ty
fa
br
ic
at
ion
in
dustrie
s
are
in
vo
l
ved
in
the
pr
oduction
o
f
So
C
(S
yst
em
on
C
hip).
So
m
e
of
these
f
oundr
ie
s
m
a
y
act
as
an
a
dversa
ry
a
nd
m
igh
t
insert
m
alici
ou
s
ci
rc
uitry
into
the
chi
ps
durin
g
t
he
fa
br
ic
at
ion
pr
ocess
.
These
ad
diti
onal
m
od
ules
a
r
e
eff
ic
ie
ntly
de
sign
e
d
s
uc
h
th
at
they
evad
e
detect
ion
durin
g
post
m
anu
fact
uri
ng
te
sts
and
ar
e
dep
loye
d
in
fiel
d
app
li
cat
ion
.
T
he
m
a
liciou
s
IP
-
C
or
es
a
par
t
fr
om
cor
r
up
ti
ng
the
functi
onal
it
y
of
the
So
C
m
a
y
cause
con
se
quences
in
the
fiel
ds
wh
ic
h
require
e
xtre
m
e
inform
ation
sec
ur
it
y
suc
h
as
de
fen
ce
,
m
edici
ne
an
d
c
omm
un
ic
ation
.
To
re
duc
e
the
vu
l
ner
a
bili
ty
of
the
So
C,
hardw
a
re
secu
rity
and
Tr
ojan
de
te
ct
ion
are
inc
orp
or
at
ed
to
s
afeguar
d
an
d
protect
the se
ns
it
ive
de
sign
s
.
Hardwa
re
secu
rity
involves
the
m
ajo
r
re
sea
rch
areas
s
uch
as
detect
ion
a
nd
dia
gnos
is
of
hardwar
e
Troja
ns
al
ong
with
d
esi
gn
of
secur
e
d
ha
rdwar
e
.
Cl
assifi
cat
ion
of
Hard
war
e
T
roja
ns
and
var
i
ou
s
th
r
eat
s
are
analy
zed
in
[1]
.
The
la
ck
of
avail
abili
ty
of
go
l
den
c
hip
s
a
s
a
ref
ere
nce
ci
rcu
it
is
the
m
ai
n
chall
eng
e
f
or
the
researc
hers
to
detect
the
Troj
an.
T
he
Hard
w
are
Tr
oj
a
n
has
b
een
cl
assifi
e
d
[2
]
into
three
diff
e
re
nt
ty
pes
as
an
internal
tri
gg
e
r
,
stora
ge
a
nd
H
ar
dware
T
roj
an
dri
ve
r.
Wa
ng
et
al
.
dev
el
op
e
d
a
n
el
ab
orat
e
hardw
a
re
Troja
n
ta
xono
m
y
in
[3
]
wh
ic
h
cat
eg
or
iz
e
d
the
Tr
oj
ans
into
physi
cal
,
act
iva
ti
on
a
nd
act
io
n
c
harac
te
risti
cs
.
In
[
4]
the
act
ivati
on
m
ec
han
ism
is
fu
rth
er
cat
eg
or
iz
e
d
into
di
gital
and
analo
g
Tr
oja
ns.
The
im
pact
of
d
igit
al
Tr
oj
a
n
m
ay
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4035
-
4043
4036
aff
ect
t
he
lo
gi
c
val
ues
of
the
sp
eci
fied
i
ntern
al
node
or
it
m
ay
al
so
al
ter
the
sto
re
d
c
onte
nt
i
n
the
m
e
m
or
y
un
it
.
T
he
perf
or
m
ance,
po
w
er
an
d
no
ise
m
arg
ins
are
th
e
par
am
et
er
wh
ic
h
gets
af
fec
te
d
du
e
to
the
analo
g
payl
oad
T
roja
n.
I
n
-
orde
r
to
secur
e
the
vulne
rab
le
syst
e
m
,
the
resear
cher
s
m
ai
nly
fo
c
us
es
on
act
ivati
on
m
echan
ism
of
Tr
oj
a
n
ci
rc
ui
t.
The
act
ivat
ion
m
echan
is
m
includes
tr
ansiti
on
p
roba
bili
ty
based
Troja
n
trigg
e
rin
g
[
5],
in
wh
ic
h
the
node
sel
ect
io
n
te
ch
nique
f
or
the
T
r
oj
a
n
insertio
n
is
propose
d
f
or
de
te
ct
ion
process
.
T
o
e
nsure
the
prese
nce
of
Tr
ojan
m
od
ule
us
i
ng
conve
ntion
al
te
sti
ng
a
ppr
oac
h
tri
gg
e
rs
t
he
i
nter
nal
nodes
a
nd
pro
pag
at
io
n
of
lo
gic
va
riat
ion
s
in
the
nets
due
to
Tr
ojan
ef
fec
ts
to
the
payl
oa
d
m
us
t
be
e
nhance
d.
The
pr
ob
le
m
of
detect
in
g
the
Troja
n
m
od
ule
us
in
g
co
nventi
on
al
te
st
vect
or
appro
a
ch
is
a
chall
eng
i
ng
ta
sk
f
or
extracti
ng
Tr
ojan
trig
ger
i
ng
vec
tors
by
ta
ki
ng
fa
ult
m
aski
ng
log
ic
int
o
co
ns
ide
rati
on.
I
n
add
it
io
n,
the
inter
nal
nodes
an
d
Tr
oj
a
n
t
rig
ger
s
are
de
pende
nt
fact
or
s
,
w
hich
i
nd
ic
at
es
i
ns
te
ad
of
conve
ntion
al
te
sti
ng
,
a
com
pr
essive
sensing
a
ppr
oa
ch
for
s
pa
rse
te
st
patte
r
n
gen
e
rati
on
f
or
trig
ge
rin
g
the
T
roja
ns
a
re
form
u
la
t
ed
i
n
a f
easi
ble m
ann
er
.
In
the
pro
pos
ed
wor
k,
a
n
op
ti
m
al
te
st
patte
rn
gen
e
rati
on
te
ch
nique
us
in
g
c
om
pr
es
sive
se
ns
in
g
appr
oach
has
been
at
te
m
pte
d
to
ai
d
i
n
th
e
detect
io
n
of
ha
rdwar
e
Tr
oja
n.
T
he
pro
pose
d
c
om
pr
essive
sensi
ng
appr
oach
is
pe
rfor
m
ed
on
th
e
i
np
ut
patte
rns
to
prov
i
de
optim
al
te
s
t
patte
rn
s
that
can
disti
nguish
a
Troja
n
infected
c
hip
f
ro
m
a
go
lden
c
hip
.
T
his
appr
oa
ch
is
per
f
or
m
ed
to
extract
th
e
Trojan
trig
ge
rin
g
te
st
vectors
the
t
are
ra
re
a
nd
is
al
so
s
par
se
.
T
hu
s
by
at
te
m
pt
ing
t
he
propos
ed
c
om
pr
e
ssiv
e
sensi
ng
ap
proach,
the
sp
a
rs
it
y
of
the
te
st
vectors
is
achieved
a
nd
ca
n
eff
e
ct
iv
el
y
deter
m
ine
the
rev
eal
in
g
te
st
patte
rn
s
f
rom
the
la
rg
e
pool
of
te
st
vector
sp
a
ce.
This
a
ppr
oa
ch
gen
e
rates
com
pact
te
st
s
et
,
w
hich
m
ini
m
iz
es
the
tim
e
com
plexity
fo
r
te
sti
ng
,
wh
il
e
m
axi
m
i
zi
ng
t
he
probabil
it
y
of
Tr
oj
an
detect
ion
c
ov
e
ra
ge.
T
he
transiti
on
pr
obabili
ty
al
go
rithm
is
form
ulate
d
to
extract
the
sig
nificant
inter
na
l
nodes
in
the
net
-
li
st
for
ins
erti
on
of
T
r
oj
a
n
m
od
ule.
T
he
m
ai
n
insig
ht
of
m
eas
ur
i
ng
t
he
powe
r
pr
of
il
e
in our
w
ork
is
to
cl
as
sify
the
IC
is T
roja
n
f
ree
if
t
he
powe
r
sig
nat
ur
e
is
const
ant
and
T
roja
n
infected
i
f
it
has
so
m
e
a
no
m
al
y
a
t
diff
eren
t
tim
e
instances.
Sim
ulatio
n
re
su
lt
s
sho
ws
that
the
pro
po
se
d
c
om
pr
essive
se
ns
in
g
ap
proac
h
is
eff
ect
ivel
y
us
ed
f
or
dete
ct
ing
both
c
om
bin
at
ion
al
as
well
as
seq
uen
ti
al
Tro
ja
ns
.
The
rest
of
the
pap
e
r
is
orga
ni
zed
as
fo
ll
ows
:
Sect
ion
2
de
s
cribes
the
sta
te
-
of
-
t
he
-
a
rt
te
chn
iq
ues
f
or
hard
war
e
T
roj
an
detect
io
n
sc
hem
es
and
cha
ll
eng
es
ba
sed
on
c
om
pr
essiv
e
sensing
te
ch
niques.
T
he
P
r
o
pose
d
com
pr
essive
s
ensin
g
based
te
st
set
app
r
oa
ch
f
or
detect
ion
T
roja
n
m
odule
is
prese
nted
in
Sect
io
n
3.
The
Sim
ulati
on
res
ults
f
or
var
i
ou
s
IS
C
A
S
be
nch
m
ark
ci
rcu
it
with
de
ta
il
ed
ob
se
rvat
ion
s
is
desc
r
ibed
in
Sect
ion
4. At t
he
e
nd, S
ect
i
on 5
c
oncl
udes t
he
p
a
per.
2.
BACKG
ROU
ND
Hardwa
re
Tr
ojan
s
a
re
em
bed
de
d
i
nto
the
ori
gin
al
finit
e
sta
te
m
achi
ne
by
the
Tr
oj
a
n
desi
gn
eng
i
neer
s
by
c
om
po
sin
g
a
hi
gh
le
vel
desi
gn
descr
i
ptio
n
of
the
Tr
oj
a
n.
Th
us
a
Tr
ojan
FSM
is
in
j
ect
ed
by
m
erg
in
g
their
sta
te
s
with
the
or
igin
al
desig
n
a
nd
is
ind
ist
inguisha
ble
fro
m
the
fu
nctionalit
y
of
t
he
golde
n
c
hip
.
This
m
et
ho
d
of
Tr
ojan
insertio
n
will
hi
de
the
pr
ese
nc
e
of
Tr
oj
a
n
a
nd
is
ha
r
d
to
det
ect
by
the
co
nv
e
ntio
na
l
authe
ntica
tio
n
te
c
hn
i
ques. Th
e
c
ounterm
easur
e
s
agai
ns
t
these
T
roja
ns
ar
e
cl
ass
ifie
d
int
o
tw
o
ty
pes
su
c
h
a
s
de
te
ct
ion
a
nd d
e
sign f
or trust
.
2.1.
Det
ec
tion
The
ha
rdwar
e
Troja
n
detect
ion
is
devel
ope
d
i
n
the
hardware
sec
ur
it
y
c
om
m
un
it
y
and
are
broa
dly
cat
egorised
i
nt
o
destr
uctive
a
nd
no
n
-
dest
ruct
ive
[
4]
as
s
how
n
i
n
the
F
ig
ur
e
1.
T
he
des
tru
ct
ive
a
ppr
oa
ch
is
analy
sed
to
ve
rify
the
chip
de
sign
by
an
opti
cal
m
et
ho
d.
Thu
s
t
he
chi
p
is
exam
ined
la
ye
r
by
la
ye
r
in
this
m
et
ho
d
a
nd
ea
ch
c
hip
is
t
o
be
te
ste
d
in
div
i
du
al
ly
.
Hen
ce
this
ap
proac
h
i
s
ap
plica
ble
only
fo
r
the
IC’s
wh
ic
h
is
fabrica
te
d
unde
r
un
t
ru
ste
d
fou
ndry.
T
he
requirem
ent
of
sp
eci
al
iz
ed
e
quipm
ent
and
the
c
os
t
are
the
m
ajo
r
lim
it
at
ion
s
of
this
analy
sis.
The
non
-
de
str
uctive
a
ppr
oac
h
is
a
naly
sed
by
c
onside
ring
the
cha
ract
erist
ic
beh
a
viou
r
of
the
chi
p
an
d
the
prese
nce
of
Tr
oj
a
n
is
id
entifi
ed
by
m
app
i
ng
with
the
ex
am
ined
prof
il
e
.
This
m
e
tho
d
of
hard
war
e
T
rojan
is
f
ur
t
her
c
la
ssifie
d
into
three
ty
pes
s
uc
h
as
desig
n
ti
m
e
app
r
oach,
r
un
ti
m
e
appr
oach an
d
t
est
tim
e app
r
oa
ch.
Figure
1
.
Cl
assifi
cat
ion
of H
a
rdwar
e
Tro
j
an
detect
ion [
4]
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
A com
pr
essiv
e
sensing
algorit
hm for h
ardwa
re troja
n detec
ti
on
(M.
Priy
at
ha
ris
hin
i)
4037
2.1.1.
Design a
ppr
oach
A
pre
-
sil
ic
on
m
et
ho
d
of
ha
r
dw
a
re
T
roja
n
detect
ion
is
e
m
plo
ye
d
in
thi
s
ap
proac
h,
whic
h
detect
s
th
e
threats
ass
ocia
te
d
with
intel
le
ct
ual
pro
per
t
y
(I
P
).
A
hard
war
e
Tr
ojan
m
od
ule
is
inse
rted
to
the
de
sign
of
go
l
den
I
P
without
the
desig
ner’s
knowle
dge
an
d
these
Troja
ns
be
ha
vi
our
will
be
hid
de
n
durin
g
norm
al
ver
ific
at
io
n
te
s
t.
The
desig
n
ti
m
e
app
r
oac
h
is
furthe
r
di
vid
e
d
into
f
or
m
al
ver
ific
at
ion
a
nd
desig
n
f
or
secu
rity
.
The
process
of
pro
per
ty
c
hec
king
is
a
f
or
m
al
ver
ific
at
io
n,
in
wh
ic
h
the
I
P
pro
pe
rtie
s
of
the
des
ig
n
un
der
te
st
are
analy
sed
w
it
h
the
or
igi
nal
ref
ere
nce
I
P.
In
pap
e
r
[
6]
propose
d
a
m
e
tho
d
for
diag
nos
is
wh
ic
h
locat
es
the
m
al
ic
iou
s
m
odule
in
the
thir
d
pa
rty
IP
.
T
he
isolat
ion
of
infected
t
hird
par
ty
IP
from
the
authe
ntic
on
e
is
exp
e
rim
ental
l
y
pro
ved
i
n
t
his
m
et
ho
d
a
nd
the
locat
io
n
of
Troja
n
m
od
ule
is
al
so
ef
fecti
vely
identifie
d
on
l
y
wh
e
n
t
he
T
roj
an
a
re
trig
ge
re
d.
Hard
war
e
T
roja
ns
inte
rn
al
an
d
e
xter
nal
act
ivati
on
delivers
the
c
ha
ra
ct
er
of
Troja
n
m
od
ule
and
it
s
var
i
ous
ty
pes
of
act
ivati
on
m
echan
is
m
are
li
st
ed
in
[
7].
Th
e
tri
gg
e
rin
g
in
put
of
t
he
Troja
n
ci
rc
uit i
s in
dep
e
ndent
f
ro
m
the origi
na
l ci
rcu
it
and it
is re
pr
ese
nted
as alway
s
on T
roja
n.
2.1.2.
Ru
n
time
a
ppr
oa
c
h
The
r
un
tim
e
a
ppr
oach
es
a
re
us
e
d
to
detect
to
detect
the
T
roja
n
by
inse
rting
a
se
nsor
m
odule
to
th
e
or
i
gin
al
c
hip
a
nd
the
act
ivit
y
of
the
IC’s
a
r
e
co
ntinuo
us
ly
m
on
it
or
ed
.
T
he
var
ia
ti
on
i
n
the
c
har
act
erist
ic
s
of
the ch
i
p
un
der
te
st fr
om
the e
xp
ect
e
d
re
fe
re
nce c
har
act
eris
ti
cs w
il
l i
nd
ic
at
e the prese
nce
o
f
m
alici
ou
s
m
od
ule.
A
hy
br
i
d
m
et
h
od
of
r
un
tim
e
is
p
r
opos
e
d
[
8],
w
hich
c
ombines
a
desi
gn
tim
e
co
m
po
ne
nt
with
t
he
r
un
ti
m
e
m
on
it
or
ing
s
uch
as
blu
e
chip.
An
on
-
li
ne
m
on
it
or
i
ng
a
ppr
oac
h
of
ha
rdwa
re
Tr
oj
a
n
de
te
ct
ion
is
propose
d
in
[9
]
,
wh
ic
h
m
ai
nly
fo
c
us
es
on
or
i
gin
al
f
unct
ion
al
it
y
of
the
ci
rcu
it
an
d
detect
s
a
ny
va
riat
io
n
from
the
exp
ec
te
d
log
ic
value
s
at
the
su
sp
ect
ed
inter
nal
node
s.
A
Tw
o
rail
check
e
r
m
od
ul
e
are
dev
el
op
e
d
an
d
inserted
in
t
he sy
stem
to
m
on
it
or
t
he
lo
gic m
al
functi
on in
it
s v
ic
init
y.
2.1.3.
Te
st
time
a
pp
roa
c
h
The
pr
ese
nce
of
m
al
ic
iou
s
m
od
ule
is
detect
ed
in
te
st
tim
e
app
r
oach
durin
g
po
st
m
anufactu
ri
ng
process
.
The
t
est
tim
e
app
r
oa
ches
are
cl
ass
ifie
d
into
tw
o
ty
pes
su
c
h
as
f
un
ct
io
nal
te
sti
ng
a
ppr
oach
a
nd
si
de
channel
analy
s
is.
The
pr
e
sen
ce
of
Tr
ojan
c
hanges
the
f
unct
ion
al
it
y
of
the
desig
n
is
detect
ed
us
i
ng
log
ic
ver
ific
at
io
n
known
as
functi
onal
te
sti
ng
a
ppro
ac
h
[
10]
.
Th
e
m
ai
n
dr
a
wb
a
ck
of
this
m
et
ho
d
is
t
hat
the
Troj
a
n
inserted
will
not
al
te
r
the
f
un
ct
ion
al
it
y
of
th
e
ci
rcu
it
unti
l
it
is
trigg
ere
d.
Com
pr
essive
s
ensin
g
te
ch
nique
is
an
e
m
erg
in
g
m
et
ho
dolo
gy
wh
ic
h i
s r
el
at
ed wit
h t
he fu
nctio
nal
te
sti
ng
a
pproac
h [11].
2.2.
Design f
or t
ru
st
The
desig
n
for
secu
rity
te
chni
qu
es
is
a
ppli
cable
f
or
desi
gning
t
he
str
uctu
re
in
-
orde
r
t
o
enh
a
nce
t
he
le
vel
of
sec
uri
ty
in
hard
war
e.
The
desi
gn
f
or
secu
rity
is
cl
assifi
ed
su
c
h
a
s
l
og
ic
e
ncr
y
ptio
n,
IC
cam
ouflagin
g,
sp
li
t
m
anu
fact
ur
i
ng
a
nd
T
roj
an
act
ivati
on.
In
pap
e
r
[
12
]
,
pro
po
se
d
a
des
ign
m
et
ho
dolo
gy
in
w
hich
th
e
IP
’
s
are
protect
ed
by
ob
f
us
cat
in
g
the
ori
gin
al
ne
t
-
li
st.
In
this
te
chn
i
qu
e
,
the
nor
m
al
m
od
e
of
operati
on
is
pos
sible
on
ly
when
the
valid
key
is
pro
vid
e
d
to
th
e
key
gates
w
her
e
the
delay
an
d
a
rea
a
re
the
m
ai
n
cons
trai
ns
.
S.
Dup
uis
,
et
a
l.
,
[
13,
14]
pr
opose
d
an
enc
ry
ption
al
gorith
m
wh
ic
h
m
ini
m
ise
s
the
num
ber
of
rar
e
occ
urren
c
e
values
of
the
inter
nal
node.
I
n
this
log
ic
en
cryp
ti
on
te
chni
qu
e,
the
probabil
it
y
of
al
l
t
he
inter
nal
nodes
i
s
com
pu
te
d
in
-
orde
r
to
i
de
ntify
the
low
co
nt
ro
ll
abili
ty
nodes.
The
pro
pose
d
m
et
ho
d
ov
e
rc
om
es
the
ab
ov
e
sp
eci
fied
draw
backs
an
d
al
s
o
pr
otect
s
the
IC’
s
from
il
l
egal
over
pro
duct
ion.
A
se
ve
ral
key
te
c
hn
ol
og
ie
s
associat
ed wit
h netw
ork
sec
uri
ty
are
pr
op
os
e
d
in pape
r
[
15
]
, which
sat
isfie
s the co
m
pu
ta
ti
on
al
sec
ur
it
y aspect
in
el
ect
ronic
voti
ng,
data
m
i
ning
a
nd
ot
her
fiel
ds
.
T
he
s
pl
it
m
anu
factu
ring
[
16]
te
chn
i
qu
e
is
a
pp
li
ca
ble
in
la
yout
le
vel
in
wh
ic
h
the
o
rigi
nal
desig
n
is
s
plit
into
two
la
ye
rs
an
d
it
is
fab
ricat
ed
unde
r
dif
fer
e
nt
f
oundries.
The
a
dv
e
rsa
ry
can
no
t
acce
s
s
the
com
plete
detai
ls
of
t
he
sp
li
t
desig
n
a
nd
hen
ce
t
he
in
serti
ng
a
H
a
r
dware
Troja
n
is
not p
os
sible i
n
this
techn
i
qu
e
.
3.
PROP
OSE
D MET
HO
DOL
OGY
The
m
ai
n
go
a
l
of
the
propo
sed
m
et
ho
dolo
gy
is
to
ge
nerat
e
the
te
st
patte
rn
s
that
can
excit
e
the
Troja
n
trig
ger
i
ng
nodes
m
or
e
eff
ect
ively
.
The
Prob
a
bili
ty
distrib
ution
of
each
net
in
the
ci
rcu
it
un
de
r
te
st
is
cal
culat
ed
in
order
to
deter
m
ine
the
extra
Troj
an
gate
in
the
IC.
The
H
ar
dware
Troja
n
detect
ion
us
i
ng
pro
po
se
d
m
et
ho
dolo
gy
is s
ho
wn in
F
ig
ur
e
2.
3.1.
Redu
ce
d Tes
t set Ge
nera
tio
n u
sin
g CS
Algorithm
1
de
scribes
t
he
m
ajor
ste
ps
in
th
e
pro
posed
ge
ner
at
io
n
of
opt
i
m
al
te
st
vector
s
f
or
Tr
oja
n
detect
ion.
The
gate
le
vel
net
-
li
st
is
con
si
de
red
as
t
he
in
put
file
f
or
the
conve
ntion
al
ATPG
to
ol.
T
he
te
st
vecto
rs
obta
in
ed
from
the
ATP
G
are
furth
er
com
pr
esse
d
us
in
g
pro
pos
ed
com
pr
essiv
e
sensing
al
go
rithm
.
The
input
sig
na
l
to
the
co
m
pr
essive
se
ns
in
g
al
gorithm
mu
st
be
sp
a
rs
e
in
any
do
m
ai
n.
The
sp
arsit
y
of
the
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4035
-
4043
4038
sign
al
is
achieved
by
incorp
orat
ing
basis
m
a
trix
(
Ψ
)
to
the
input,
w
hich
c
onve
rts
the
sign
al
in
on
e
do
m
ai
n
to
ano
t
her
do
m
ain
w
her
e
it
is
sp
ars
e
.
Th
us,
the
N
nu
m
ber
of
te
st
patte
r
ns
gen
e
rated
f
r
om
ATP
G
is
fu
rt
her
com
pr
essed
in
to
M
nu
m
ber
of
te
st
patte
r
ns
(M<<
N)
af
te
r
com
pr
essiv
e
sensing
(CS
).
T
he
res
ult
of
th
e
pro
po
se
d
CS
pa
tt
ern
ge
ne
rati
on
m
et
ho
d
is
a
optim
al
te
st
set
that
identif
ie
s
the
prese
nc
e
of
Tr
ojan
an
d
al
s
o
i
m
pr
oves the
t
est
covera
ge fo
r
T
roja
ns
c
om
par
ed
w
it
h o
rigi
nal test
p
at
te
r
ns.
Figure
2
.
Pro
pose
d
h
ar
dw
a
re
Troja
n detec
ti
on
flo
w
Algorith
m
1.
G
ener
at
e
re
du
ce
d
te
st set
us
in
g com
pr
essive
s
ensin
g for T
r
ojan Dete
ct
ion
Inpu
t:
Ci
rc
uit
Net
-
li
st, li
st o
f ran
dom
p
at
te
rns (
T
),
c
onsta
nt(
c),
li
st o
f n
od
e
(R).
Out
p
ut:
Re
duced test
patte
rns (
Tc
).
1.
Re
ad
the
circ
uit net
-
li
st.
2.
for
al
l ra
ndom
p
at
te
rn
s
in
T
do
3.
C
om
pu
te
the s
parsi
ty
p
ar
a
m
et
er k
4.
Com
pu
te
M wi
th c as
constant
v
al
ue
s
5.
≥
∗
∗
log
(
/
)
6.
Com
pu
te
Mi
nim
al
v
al
ue of
M
<N.
7.
Gen
e
rate
Φ
m
a
trix
€
M
x N.
8.
for
al
l bit
s i
n T d
o
9.
C
om
pu
te
re
du
ce
d set
Tc
10.
Tc=
Φ
m
at
rix * T
m
at
rix
11.
en
d f
or
12.
end f
or
3.2.
N
ode i
den
tific
at
i
on
fo
r
H
ar
dware Tr
ojan i
nsertio
n
Transi
ti
on
pro
bab
il
it
y
value
play
s
a
ver
y
si
gn
i
ficant
analy
sis
in
insertin
g
the
Tr
oj
a
n
m
odule
to
the
ci
rcu
it
.
T
he
i
ntern
al
nodes
wi
th
lo
w
tra
ns
it
io
n
probabil
it
y
(TP)
a
re
m
or
e
s
us
ce
ptible
f
or
trig
ger
in
g
t
he
T
roj
a
n
and
payl
oa
d.
It
is
al
so
hi
gh
ly
li
kely
that
an
adv
e
rsa
ry
m
igh
t
insert
a
Hard
war
e
T
r
oj
a
n
(
HT)
i
n
these
l
ow
TP
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
A com
pr
essiv
e
sensing
algorit
hm for h
ardwa
re troja
n detec
ti
on
(M.
Priy
at
ha
ris
hin
i)
4039
nodes
beca
us
e
of
t
heir
lo
w
switc
hing
ac
ti
viti
es,
wh
ic
h
trigg
e
rs
the
Troja
n
m
od
ul
e
in
rar
e
i
ns
t
ances.
The
al
gorithm
2
pro
vid
es
the
cal
cula
ti
on
f
or
transiti
on
pro
ba
bili
ty
.
The
prob
a
bili
ty
cal
culat
ion
for
basic
log
ic
gates [
5] is e
xpresse
d
as
(
1
)
=
1
(
1
)
∗
2
(
1
)
(1)
(
1)
(
0
)
=
1
(
0
)
+
2
(
0
)
–
[
1
(
0
)
∗
2
(
0
)
]
(
2)
The
Hardw
a
re
Troja
ns
s
uch
as
com
bin
at
ion
al
an
d
3
bit
counter
se
que
ntial
m
od
ules
are
desi
gn
e
d,
wh
ic
h
get
act
i
vated
at
r
are
conditi
ons
an
d
cause
c
hange
s
in
the
p
er
f
orm
ance
of
the
chosen
be
nchm
ark
ci
rcu
it
s.
Node
s
with
lo
w
T
r
ansiti
on
Pro
ba
bili
ty
(TP)
a
nd
hi
gh
co
nnec
ti
vity
are
sel
ect
ed
an
d
the
de
sign
e
d
Hardwa
re
Tr
ojan (HT
)
m
odules are
i
ns
erte
d at
these
nodes for
validat
io
n process
.
Algorithm
2
. C
al
culat
ing
T
ra
nsi
ti
on
prob
a
b
il
it
y value
for
ea
ch net
Inpu
t
: C
ircuit
net
-
li
st, m
od
ul
e li
st of
t
he
ci
r
cuit.
Out
p
ut:
T
ra
nsi
ti
on
pr
ob
a
bili
ty
(
TP
) values.
Step1
:
Scan
th
e m
od
ule li
st.
xlsx fil
e.
Step2
:
Extr
act
the prim
ary inp
ut
(PI)
, out
pu
t
(
P
O) an
d
t
he g
at
es o
f
the ci
rc
uit.
Ste
p3
:
I
niti
al
is
e the
pro
bab
il
it
y of p
rim
ary inp
uts
n
et
s a
s
0.5
.
Step4
:
Determ
i
ne
the
n
et
i
nde
x.
Step5
:
If (
net i
nd
e
x
=
n
et
i
ndex of
PI) t
hen
Step6
:
St
ore the T
P
values
in
a
n
a
rr
ay
Step7
:
else
Step8
:
for
e
ach
net in t
he n
et
-
li
st do
Step9
:
Find the ty
pe
of
gate an
d
i
ndex
values
of in
put
Step10:
C
al
culat
e the
prob
a
bili
ty
at the outp
ut n
et
acc
ordin
g
t
o
the
gat
e ty
pe
Step11:
C
al
culat
e the tra
ns
it
ion p
roba
bili
ty
v
al
ues
f
or
each
net
Step12:
St
or
e
n
et
s t
ran
sit
ion
pro
ba
bili
ty in
asce
ndin
g o
rd
e
r.
Step13:
e
nd
for
Step14: e
nd if
3.3.
Log
ic
t
estin
g
Lo
gic
te
sti
ng
i
s
perf
or
m
ed
by
com
par
ing
t
he
log
ic
values
at
the
no
des
in
t
he
gold
en
ci
rc
uit
with
that
of
t
he
Tr
ojan
I
Cc
ircuit
for
al
l
possible
te
st
pa
tt
ern
s
obta
ine
d
f
ro
m
propos
ed
al
gorithm
.
The
Ta
ble
1
prov
i
de
s
the
com
par
iso
n
of
the
l
og
ic
values
for
T
roj
an
in
fected
a
nd
golde
n
ci
rc
ui
t.
A
c
om
bin
at
i
on
al
ty
pe
of
Tr
oj
a
n
i
s
inserted
in
the
low
tra
ns
it
io
n
pr
ob
a
bili
ty
node
a
nd
the
log
ic
var
ia
ti
on
s
in
the
nets
are
obse
rv
e
d
wh
ic
h
ind
ic
at
e
s the
pr
esence
of Tr
oj
an
m
od
ule.
Table
1
.
C
om
par
iso
n of o
utpu
t functi
on in
L
og
ic
Test
in
g
Inp
u
t
Pattern
C1
7
Gold
en
C1
7
T
rojan
cir
cu
it
Internal No
d
es
Pri
m
a
r
y
ou
tp
u
t
Internal No
d
es
Pri
m
a
r
y
ou
tp
u
t
N6
N7
N8
N9
N1
0
N1
1
N6
N7
N8
N9
N1
0
N1
1
1
1
1
1
0
0
1
1
1
1
0
0
1
7
1
0
1
1
0
0
1
1
1
1
0
1
14
1
0
1
1
0
0
1
1
1
1
1
1
In
-
or
der
t
o
va
li
date
the
ef
fici
ency
of
dete
ct
ion
of
Hard
war
e
Tr
oj
a
n
usi
ng
c
om
pr
ess
ive
sen
sin
g,
m
et
rics
su
ch
as
Tru
e
P
os
it
iv
e
Ra
te
(TPR)
and
P
roba
bili
ty
of
detect
ion
(P
D
)
[5
]
are
app
li
ed
to
the
ci
rcu
it
unde
r
te
st.
In
c
ase
of
bin
a
ry
c
la
ssific
at
ion
,
t
he
T
ru
e
posit
ive
(TR)
valu
e
identifie
s
t
he
num
ber
of
T
roj
an
nets
as
m
alici
ou
s
ne
ts
it
sel
f
an
d
t
he
False
ne
gat
ive
(
FN)
value
s
sho
ws
the
num
ber
of
T
roj
an
nets
ide
ntif
ie
d
as
norm
al
n
et
s b
y m
ist
ake.
3.4.
Power
a
n
aly
si
s
The
com
pr
esse
d
te
st
set
gen
er
at
ed
from
the
pr
op
os
ed
al
gorithm
are
fo
rce
d
to
the
de
sig
n
un
der
te
st
at
diff
e
re
nt
tim
e
windows.
A
S
ynopsys
pri
m
e
tim
e
too
l
is
us
ed
to
determ
i
ne
the
le
a
kag
e
powe
r
f
or
e
ve
ry
te
st
vecto
r.
T
he
ha
r
dw
a
re
Tr
ojan
c
ho
s
en
for
this
analy
sis
is
a
three
bit
co
un
te
r
and
it
s
im
pact
will
be
no
ti
ce
d
on
ly
durin
g
certai
n
cl
ock
cy
cl
e.
The
obse
rv
e
d
powe
r
pro
file
will
var
y
for
the
Tr
ojan
IC
at
so
m
e
tim
e
stam
p
wh
e
reas
it
is
un
al
te
re
d
for
t
he
T
roja
n
f
ree
IC.
T
he
var
i
at
ion
of
t
he
powe
r
p
r
of
il
e
i
s
m
a
inly
du
e
to
the
app
li
cat
io
n
of
certai
n
te
st
set
s
wh
ic
h
act
ivate
s
the
Tr
oj
a
n
m
od
ule.
This
m
et
ho
d
of
side
channel
ap
pro
ach
f
or
detect
ing Tr
oja
n
m
od
ule
does
no
t
re
qu
ire
s a
ny
g
ol
den n
et
li
s
t as re
fer
e
nce c
ircuit
.
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4035
-
4043
4040
4.
RESU
LT
S
A
ND
DI
SCUS
S
ION
The
pro
po
se
d
m
et
ho
dolo
gy
is
validat
ed
us
i
ng
IS
C
AS
85
com
bin
at
ion
al
benchm
ark
ci
rcu
it
s
wit
h
com
bin
at
ion
al
and
se
qu
e
ntial
Tr
oj
a
n
m
odul
es,
wh
ic
h
al
te
r
s
the
functi
ona
li
ty
of
the
ci
rc
uit.
Ta
ble
2
li
sts
the
com
par
ison
of
te
st
vect
or
re
du
ct
io
n
f
or
I
S
CAS
be
nch
m
ark
ci
rc
uits
f
or
c
onve
ntion
al
AT
PG
patte
r
ns
a
nd
pro
po
se
d
com
pr
essi
ve
sen
sin
g
(CS
)
ap
proa
ch.
It
is
obser
ve
d
that
the
pro
po
s
ed
te
st
set
us
in
g
CS
achie
ves
an
aver
a
ge
te
st
l
eng
t
h
re
duct
io
n
of
88.
6%,
wh
il
e
m
axi
m
i
zi
ng
t
he
T
roja
n
trig
ge
rin
g
r
at
e.
The
r
un
t
i
m
e
fo
r
execu
ti
ng t
hi
s
al
gorithm
is al
so
c
om
pu
te
d.
Table
2
.
T
est
pat
te
rn
r
e
duct
io
n usin
g
CS
alg
or
it
hm
Ben
ch
m
ark
circuits
No
.
o
f
pri
m
ar
y
in
p
u
ts
Po
ss
ib
le
Test vecto
rs
Test vecto
rs us
in
g
ATPG
(N
)
Test vecto
rs
u
sin
g
CS
(M)
Co
m
p
r
ess
io
n
ratio (CR)
Proces
sin
g
Ti
m
e
(
s)
C1
7
5
2
5
7
x
5
3
x 5
9
9
.4
0
.01
8
2
C4
3
2
36
2
36
6
3
x 3
6
2
3
x 3
6
9
2
.33
0
.58
7
C4
9
9
41
2
41
5
6
x 4
1
1
9
x
4
1
9
3
.66
0
.79
C8
8
0
60
2
60
1
4
8
x 6
0
5
5
x 6
0
8
1
.66
1
.00
3
C1
3
5
5
41
2
41
1
0
0
x 4
1
3
7
x 4
1
8
7
.66
0
.17
8
C3
5
4
0
50
2
50
2
6
5
x 5
0
9
7
x 5
0
7
0
.6
0
.16
2
C6
2
8
8
32
2
32
3
4
x 3
2
1
3
x
32
9
5
.6
1
.56
The
Tra
ns
it
ion
Pr
obabili
ty
(TP)
value
s
are
evaluated
for
di
ff
e
ren
t
be
nc
h
m
ark
ci
rcu
it
s
and
lo
w
T
P
values
al
ong
with
the
high
connecti
vit
y
nodes
a
re
al
so
li
ste
d
in
the
T
able
3.
The
Lo
w
TP
value
s
are
consi
der
e
d
as
po
te
ntial
node
s
and
t
hese
no
des
are
c
hose
n
by
the
at
ta
cke
rs
at
the
desi
gn
sit
e
to
intr
oduce
the
Troja
n
gate
f
or
m
a
li
ci
ou
s
intenti
on.
F
or
the
si
m
ulati
on
of
powe
r
prof
il
e
s
om
e
of
the
locat
ion
s
a
re
c
onsid
ered
for
T
roja
n
in
se
rtion an
d
t
he re
su
lt
s ar
e
v
al
ida
te
d
f
or d
et
ect
io
n process
.
Ta
ble
3
.
T
ar
get nod
e
s ide
ntifi
cat
ion
us
in
g Tr
ansiti
on Pro
ba
bili
ty
v
al
ues
The
C
om
par
ison
of
the
dete
ct
ion
pro
bab
il
it
y
m
et
rics
su
ch
as
Tr
ue
P
os
i
ti
ve
Ra
te
(TR
P)
a
nd
trig
ge
r
cov
e
ra
ge
for
t
he
c
onve
ntio
na
l
te
st
gen
e
rat
ion
AT
PG
an
d
the
propose
d
c
om
pr
essive
sensi
ng
a
ppr
oach
is
analy
zed
in
Ta
ble
4.
The
T
r
ojan
is
i
ns
erte
d
at
diff
e
ren
t
nodes
wit
h
lo
w
TP
an
d
it
is
obser
ve
d
that
th
e
te
st
vecto
r
ob
ta
ine
d
by
propose
d
C
S
a
ppr
oach
is
capa
ble
of
determ
ining
t
he
Tr
ojan
m
od
ule
m
or
e
ef
fec
ti
vely
durin
g
lo
gic
te
sti
ng
.
It
is
al
so
obser
ve
d
tha
t
red
uce
d
te
st
patte
rn
pro
vide
s
high
tro
j
an
trigg
eri
ng
co
ve
rag
e
com
par
ed
t
o
ATPG
patte
rns,
w
hich
pro
ve
s
the
CS
a
pp
ro
ac
h
has
rev
e
al
ed
the
o
pti
m
al
te
st
vectors
for
t
he
trigg
e
rin
g
t
he Tro
j
a
n
m
od
ule
and im
pr
oves
the d
et
ect
io
n r
at
e.
Table
5
li
sts
t
he
Prob
a
bili
ty
of
Detect
ion
(P
D
)
rate
for
the
IS
C
AS
be
nc
hm
ark
ci
rcu
it
.
Key
nodes
with
lo
w
TP
va
lues
are
sel
ect
ed
f
or
T
roja
n
insertio
n
a
nd
t
he
resu
lt
s
a
re
va
li
dated.
It
is
ob
serv
e
d
that
f
or
m
os
t
of
the
patte
r
n
the
detect
ion
pr
oba
bili
ty
is
hig
h,
w
hich
ind
ic
at
es
the
eff
ect
ive
te
st
set
gen
erate
d
by
the
pro
po
se
d
al
gor
it
h
m
, w
hile m
axim
iz
ing
the
T
roja
n detec
ti
on r
at
e.
The
Fi
gure
3
s
hows
the
Pow
er
sig
natu
re
f
or
va
rio
us
I
SCAS
ben
c
hm
ark
ci
rcu
it
s.
A
se
qu
e
ntial
3
-
bit
counter
is
desi
gn
e
d
in
or
der
t
o
obser
ve
the
var
ia
ti
on
of
po
wer
prof
il
e
dur
ing
s
pecific
cl
ock
pu
lse
to
va
li
date
the
Troja
n
pr
e
sence.
T
he
propose
d
CS
te
st
set
is
app
li
ed
to
ci
rcu
it
unde
r
te
st
and
the
power
pro
file
is
m
easur
ed
for
di
ff
ere
nt
tim
e
st
a
m
ps
.
It
is
obs
erv
e
d
that
the
value
of
the
m
easur
e
d
power
ov
e
rlaps
for
T
roja
n
fr
ee ci
rc
uit and fo
r
s
om
e set o
f
te
st patt
ern
s it
sh
ows so
m
e
var
ia
ti
on in
po
wer
fo
r
Tr
oj
a
n i
nf
ect
ed
ci
rc
uits. T
he
pro
po
se
d
sp
a
rs
e
te
st
p
at
te
rn
s
trigg
e
rs
seq
ue
nt
ia
l
Tro
j
a
n
an
d
is
ref
le
ct
ed
by
non
-
overlap
pi
ng
of
po
wer
prof
il
e
wh
ic
h
m
axi
m
i
zes the
pro
bab
i
li
ty
o
f
Tr
ojan
det
ect
ion
.
Ben
ch
m
ark
circuits
Mini
m
u
m
TP
v
alu
es
No
d
es with
low T
P valu
e
Hig
h
Co
n
n
ectiv
ity
no
d
es
C1
7
0
.18
7
5
N6
,N7
N6
,N7
C4
3
2
0
.08
3
6
6
9
1
N2
5
9
,N26
2
,N263,N2
6
6
,N26
9
,N272,
N2
7
8
,N28
1
,N284,N2
8
7
,N28
8
,N289
,
N2
9
0
,N29
1
,N292,N2
9
3
,N29
4
,N299,
N3
0
0
,N30
1
,N302,N3
0
3
,N30
4
,N305,
N3
0
6
,N30
7
N2
9
5
,
N2
9
9
,
N3
0
0
,
N3
0
1
,
N3
0
2
,
N3
0
2
,
N3
0
8
,
N3
1
8
.
C4
9
9
0
.05
8
5
9
3
8
N3
8
2
,N38
3
,N384,N3
8
5
N3
7
8
,N37
9
,N380,N3
8
1
N2
8
9
,
N3
2
5
,
N3
1
2
,
N3
7
2
,
N3
7
8
,
N3
7
9
.
C8
8
0
0
.00
0
0
6
1
0
3
1
4
N5
0
7
,
N5
0
8
,
N5
0
9
,
N5
1
0
,
N5
1
1
,
N5
1
2
,
N5
1
3
,
N5
1
4
.
N7
5
2
,N75
3
,N754,
N7
5
5
,N75
6
,N760,N7
6
1
,N77
0
C1
3
5
5
0
.00
7
2
7
2
2
9
N9
9
6
,N10
0
1
,N1
0
0
6
,
N1
0
1
1
,
N1
0
1
6
,N10
2
1
,N1026
,
N1
0
3
1
N7
9
4
,
N7
9
7
,
N8
0
0
,
N8
0
3
,N80
6
,N81
2
,
N8
1
5
,N99
6
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
A com
pr
essiv
e
sensing
algorit
hm for h
ardwa
re troja
n detec
ti
on
(M.
Priy
at
ha
ris
hin
i)
4041
Table
4
.
C
om
par
iso
n of
Tr
ue Posit
ive Rat
e
m
et
rics
and
t
rigg
e
r
c
overa
ge
Ben
ch
m
a
rk
circuits
Tr
o
jan
Ins
erted
No
d
es
P
rop
o
sed
CS Alg
o
rith
m
pattern
s
ATPG
Patterns
Tr
ig
g
er
Co
v
erage
Tr
u
e
Po
sitiv
e
False
Neg
ativ
e
TRP
Tr
ig
g
er
Co
v
erage
Tr
u
e
Po
sitiv
e
False
Neg
ativ
e
TRP
C1
7
7
6
6
.66
2
1
0
.66
7
2
8
.57
2
5
0
.28
5
9
3
3
.33
1
2
0
.33
3
1
4
.28
1
6
0
.14
2
7
,
9
100
2
1
0
.66
7
4
2
.85
2
5
0
.28
5
C4
3
2
300
9
5
.65
3
20
0
.13
1
4
9
.2
4
59
0
.06
3
4
301
9
5
.65
2
21
0
.08
6
9
4
4
.44
2
61
0
.03
1
7
3
0
0
,
3
0
1
9
5
.65
3
20
0
.13
1
5
3
.96
4
59
0
.06
3
4
C4
9
9
378
9
4
.73
6
13
0
.31
5
7
8
3
.92
8
48
0
.14
3
379
9
4
.73
6
13
0
.31
5
7
8
2
.14
8
48
0
.14
3
3
7
8
,
3
7
9
100
6
13
0
.31
5
7
9
1
.07
8
48
0
.14
3
C8
8
0
507
9
4
.54
51
4
0
.92
7
3
9
.18
57
91
0
.38
5
511
9
4
.54
47
8
0
.85
5
3
9
.18
53
95
0
.35
8
5
0
7
,
5
1
1
9
4
.54
52
3
0
.94
5
3
9
.18
58
90
0
.39
1
C1
3
5
5
996
9
1
.89
34
3
0
.91
9
66
66
34
0
.66
1016
8
9
.18
33
4
0
.89
2
65
65
35
0
.65
9
9
6
,1016
9
4
.59
35
2
0
.94
6
73
73
27
0
.73
Table
5
.
Prob
a
bili
ty
o
f
Detect
ion
Met
rics
f
or Tr
oj
a
n detec
ti
on
Ben
ch
m
ark
circuits
Ap
p
lied
T
est
p
atterns
us
in
g
CS
No
.
o
f
T
rojan
s
trigg
ered
Ou
tp
u
t chan
g
es
Prob
ab
ility
o
f
detectio
n
C1
7
1
/3
1
y
es
3
3
.33
%
2
/3
2
no
6
6
.66
%
C4
9
9
1
1
/1
9
3
y
es
100%
7
/1
9
2
y
es
6
6
.66
%
1
/1
9
0
no
0%
C8
8
0
4
6
/5
5
3
y
es
100%
4
/5
5
2
y
es
6
6
.66
%
2
/5
5
1
y
es
3
3
.33
%
3
/5
5
0
no
0%
C3
5
4
0
2
7
/4
3
3
y
es
100%
1
6
/4
3
2
y
es
6
6
.66
%
C6
2
8
8
8
/1
3
3
y
es
100%
4
/1
3
2
y
es
6
6
.66
%
(a)
(b)
(c)
(d
)
Figure
3
.
Me
a
s
ur
em
ent o
f
po
wer p
r
o
file
at d
iffer
e
nt ti
m
e w
indows
(a)
C
432 Tr
oja
n
in
serted
circ
uit, (b
)
C4
32 ci
rcu
it
,
(c) C
499
Troja
n
in
serte
d ci
rcu
it
,
(d)
C
499 ci
rc
uit
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4035
-
4043
4042
(e)
(f)
Figure
3
.
Me
as
ur
em
ent o
f
po
wer p
r
ofi
le
at d
iffer
e
nt ti
m
e w
indows
(e)
C
1355 T
r
ojan
in
se
rted
circ
uit, (f)
C1
355 ci
rcu
it
5.
CONCL
US
I
O
N
In
t
his
w
ork
a
com
pr
essive
sensing
a
ppr
oa
ch
ba
sed
on
T
roja
n
detec
ti
on
is
pr
ese
nted
,
wh
e
re
the
con
ce
pt
of
spa
rsifyi
ng
al
l
po
ssible
in
pu
t
ve
ct
or
s
is
us
e
d
t
o
gen
e
rate
the
rev
eal
in
g
te
st
vecto
r
wh
ic
h
t
rig
ger
s
the
T
roja
n
m
od
ule.
T
he
sim
ulati
on
is
done
on
I
SCAS
’
85
ben
c
h
m
ark
ci
r
cuit,
wh
ic
h
s
hows
the
p
r
opose
d
CS
base
d
te
st set
g
ener
at
io
n
ap
proach
wh
ic
h
achieves a
bou
t 8
8.6% r
e
du
ct
io
n i
n
te
st
le
nght o
ve
r
co
nventio
nal test
set
.
The
detec
ti
on
of
hard
w
are
Tr
oj
a
n
is
ens
ur
e
d
by
va
li
dating
the
pr
ob
a
bili
ty
of
detect
ion
and
t
rig
ger
cov
e
ra
ge
m
et
ri
cs.
T
he
pro
po
s
ed
detect
ion
a
ppr
oach
im
pr
oves
t
he
te
st
qual
it
y
by
m
ini
m
iz
ing
th
e
te
st
l
eng
t
h
and
m
axi
m
iz
in
g
the
trigge
rin
g
rate.
He
side
channel
par
a
m
et
er
are
analy
sed
for
the
CS
based
te
st
pa
tt
ern
s
at
diff
e
re
nt tim
e w
in
dows, w
hic
h
a
vo
i
ds
t
he
re
qu
i
rem
ent o
f g
old
e
n
c
hip
f
or
detect
ion p
ro
ce
ss.
REFERE
NCE
S
[1]
Chakra
bort
y
,
e
t
al
.
,
“
Hardware
T
roja
n:
Threat
s
a
nd
emergi
ng
sol
uti
ons
,”
In
High
Leve
l
Design
Va
li
dati
on
and
Test
Workshop,
2009.
HL
DVT 20
09.
I
EE
E
Inte
rnat
ion
al
,
pp
.
166
-
171
,
2009
.
[2]
Y.
Alkaba
n
i
and
F
.
Kous
hanf
ar
,
“
D
esigne
r’s
har
dware
Tro
ja
n
h
orse
,”
In
Har
dware
-
Or
ie
nte
d
S
e
curit
y
and
Tr
ust,
2008.
HO
ST 2008.
IE
EE Int
ernat
ional
Workshop
On
,
pp.
82
-
83
,
2
008.
[3]
X.
W
ang,
et
al
.
,
“
Dete
ct
ing
m
alici
ous
i
ncl
usion
s
in
sec
ure
har
dware
:
Challenges
and
soluti
ons
,”
In
Har
dwa
re
-
Or
ie
nte
d
S
ec
urit
y
and
Tr
ust,
200
8.
HO
ST 2008.
I
EE
E
Inte
rnat
ion
al
Workshop on
,
pp.
15
-
19
,
2008
.
[4]
M
.
Te
hra
n
ipoor,
“
A
Surve
y
of
Hardware
Troja
n
Ta
xonom
y
an
d
Dete
ction,
”
IEE
E
Design,
Te
st
of
Co
mputers,
2010
.
[5]
J
.
Popat
and
U
.
Mehta
,
“
Tr
ansit
i
on
proba
bilis
ti
c
appr
oac
h
for
detec
t
ion
and
d
ia
gn
osis
har
dware
Tr
oja
n
in
combina
t
i
onal
ci
rcu
it
s,
”
I
E
EE
,
2016
.
[6]
M.
Banga
and
M
.
S.
Hs
ia
o
,
“
T
rusted
RTL:
Tro
ja
n
de
te
c
ti
on
m
et
hodolog
y
in
p
re
-
sili
con
design
s
,”
In
Ha
rdwar
e
-
Or
ie
nte
d
S
ec
urit
y
and
Tr
ust (
HO
ST)
,
2010
IEE
E
Inte
rnational
Sy
mpos
ium
on
,
pp.
56
-
59
,
2010.
[7]
Karuna
gar
an
D.
K
.
and
N
.
Moha
nkum
ar,
“
Mali
cious
Hardware
T
roja
n
Det
ec
t
ion
b
y
Gat
e
l
eve
l
m
ini
m
iz
at
ion
90n
m
Te
chno
log
y
,
”
5
th
Inte
rnationa
l
Confe
renc
e
on
Computi
ng
Comm
unic
ati
on
and
Net
work
Te
chno
logy
(ICCCNT)
,
2014.
[8]
M.
Hicks,
et
al.,
“
Overc
om
ing
an
Untrust
ed
Com
puti
ng
Base:
De
te
c
ti
ng
an
d
Removing
Malicious
Hardwa
re
Autom
at
ic
a
lly
,
”
In
IEEE
S
ymposium on
Se
curity
and
Priv
a
cy
,
pp
.
159
-
172.
2010.
[9]
R
.
S
.
Chakr
ab
ort
y
,
e
t
a
l.,
“
A
Flexi
ble
Onl
ine
Che
cki
ng
Te
chn
ique
to
Enc
han
ce
Hard
ware
Tro
ja
n
H
orse
Dete
c
ta
bi
li
t
y
b
y
Rel
ia
b
il
i
t
y
Ana
l
y
sis
,
”
IE
EE Tr
ansacti
ons on Emerging
Topics
in Com
puti
ng,
201
7.
[10]
Chakra
bort
y
,
et
al
.
,
“
MERO:
A
stat
isti
ca
l
appr
o
ac
h
for
har
dwar
e
Troj
an
de
te
c
tion
,”
i
n
C
ryptog
raphic
Har
dwar
e
and
Embe
dded
S
yste
ms
-
CHES
20
09
,
Springer
Ber
li
n
He
ide
lb
erg
,
p
p.
396
-
410
,
200
9.
[11]
S.
Foucar
t
and
H.
Rauhut,
“
An
Invit
a
ti
on
to
C
om
pre
ss
ive
Sensing,
”
Applied
an
d
Numerical
Har
monic
Analysis
,
pp.
1
-
39
,
2013
.
[12]
Chakra
bort
y
,
e
t
al
.
,
“
HAR
POO
N:
an
obfusca
t
io
n
-
base
d
SoC
de
sign
m
et
ho
dology
for
h
ard
ware
prote
c
ti
on
,”
IE
E
E
Tr
ansacti
ons
on
Computer
-
Ai
de
d
Design
of
Inte
grated
Circui
ts
and
Syste
ms
,
vol/
issue:
28
(
10
),
pp.
1493
-
1502
,
2009
.
[13]
S.
Dupuis,
et
a
l.,
“
A
novel
har
dware
log
ic
en
cr
y
pt
ion
te
chn
i
que
fo
r
thwar
ti
ng
i
lleg
al
over
p
rodu
ct
io
n
and
Hardware
Troj
ans
,
”
2014
I
EE
E
20th Inte
rn
ati
onal
On
-
Lin
e T
esti
ng
Symposi
um (
IOLTS
)
,
pp.
49
-
54
,
2014
.
[14]
J
.
Sun,
et
al.,
“
An
Im
prove
d
Publ
ic
Ke
y
En
cr
y
ption
Algorit
hm
B
ase
d
on
Ch
eb
y
s
hev
Pol
y
nom
ia
ls
,”
TEL
KOMNIK
A
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
,
vol
/i
ss
ue:
11(2)
,
pp
.
864
-
870
,
2013
.
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
A com
pr
essiv
e
sensing
algorit
hm for h
ardwa
re troja
n detec
ti
on
(M.
Priy
at
ha
ris
hin
i)
4043
[15]
X
.
Guo,
et
al
.
,
“
Key
T
ec
hnol
ogie
s
and
Applic
a
ti
ons
o
f
Secur
e
Multi
par
t
y
Com
puta
ti
on
,”
TEL
KOMNIKA
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
,
vol
/i
ss
ue:
11(7)
,
pp
.
3774
-
3779
,
201
3
.
[16]
R.
W
.
Jarvis
and
M
.
G.
McInt
y
r
e
,
“
Split
m
anuf
acturi
ng
m
et
hod
for
a
dvanced
sem
ic
onduc
tor
ci
r
cu
it
s
,”
U.S
.
Patent
7,
195,
931
,
Mar
2007
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
M.
Pri
y
atharis
hin
i
is
an
assistant
profe
ss
or
in
the
d
epa
rtment
o
f
El
ectroni
cs
an
d
Comm
u
nic
at
io
n
Engi
ne
eri
ng
at
Am
rit
a
Vishw
a
Vid
y
ap
eetha
m
,
Coim
bat
ore
.
Sh
e
recei
ved
th
e
M.T
ec
h
.
degr
ee
in
Embedde
d
s
y
ste
m
design
fro
m
Am
rit
a
Univer
sit
y
,
Coim
bat
ore
.
She
is
working
towar
d
the
PhD
degr
ee
in
the
E
CE
Depa
r
tment,
Am
rit
a
Vishw
a
Vid
y
apee
tha
m
,
Coim
bat
ore
,
In
dia
.
Her
res
ea
rc
h
int
er
ests
include
har
dware
Trojan
detec
t
ion in
in
tegrat
ed
circui
ts, a
nd
truste
d
har
dw
are
d
esign.
M.
Ni
rm
ala
De
vi
is
a
prof
essor
in
the
d
epa
r
tment
of
Elec
troni
cs
and
Com
m
unic
at
ion
Eng
ine
er
in
g
at
Am
rit
a
Vishw
a
Vid
y
apeet
ha
m
,
Coim
bat
or
e
,
India
.
Her
r
ese
a
rch
intere
st
incl
udes
VLSI
Desi
gn
and
Te
sting
,
Com
puta
ti
onal
Int
e
ll
ige
n
ce,
Hardware
Secur
i
t
y
and
Trust,
Evol
vable
Hardware
and
RF
CMOS
Sy
st
em
Design
.
She
has
publi
shed
ar
ound
55
pape
rs
in
the
Int
ern
atio
nal
Journals
and
Confer
ences
in
her
field
of
exp
ert
ise
.
She
has
serve
d
as
the
re
vie
wer
for
r
efe
r
ee
d
in
te
rna
ti
on
a
l
conf
ere
n
ce
s
and
int
ern
ationa
l
journa
ls
which
inc
lude
th
e
follow
ing;
Springer
Journal
of
the
Instit
uti
on
of
Engi
ne
ers
(Indi
a):
Ser
ie
s
B,
Inde
rscie
n
ce
In
t.
Journa
l
of
Inform
at
ion
an
d
Com
m
unic
at
ion
Te
chno
log
y
.
Sh
e
is
th
e
r
ecipie
n
t
of
th
e
fol
lowin
g
awa
rds
-
Marqu
is
W
ho’s
W
ho
in
the
W
orld
-
201
1
and
2000
Outstandi
ng
In
t
e
l
le
c
tua
ls
of
the
21st
Cent
ur
y
-
20
11
-
Inte
rna
ti
ona
l
Biogra
phi
ca
l
Ce
nte
r,
Cambridge
,
Eng
la
nd
.
She
has
recei
v
ed
th
e
fina
n
ci
a
l
gr
ant
for
the
rese
arch
proposal
from
D
efe
nc
e
R
ese
ar
ch &
Deve
lopment O
rga
niz
a
ti
on,
D
el
hi
,
Indi
a.
Evaluation Warning : The document was created with Spire.PDF for Python.