TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 6063 ~ 6082
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.553
8
6063
Re
cei
v
ed
De
cem
ber 3
1
, 2013; Re
vi
sed
March 27, 20
14; Accepted
April 13, 201
4
Advances on Low Power Designs for SRAM Cell
Labonn
ah F
a
rza
n
a Rah
m
an*, Moha
mmad F. B.
Amir, Mamun Bin Ibne Reaz,
Mohd. Marufuzzaman, Ha
fizah
Hus
a
in
Dep
a
rtment of Electrical, El
ec
tr
onic an
d S
y
st
ems Engi
ne
eri
ng,
F
a
cult
y
of Engi
neer
ing a
nd Bu
ilt Enviro
nment
, Universiti Ke
b
angs
aa
n Mala
ysia,
436
00 Ba
ngi,
Sela
ngor, Mal
a
ysi
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: labo
nn
ah.de
ep@
gmai
l.com
A
b
st
r
a
ct
As the devel
op
me
nt of comp
l
e
x meta
l oxid
e
se
mic
ond
ucto
r (CMOS) tech
nol
ogy, fast low
-
pow
er
static rand
o
m
access
me
mor
y
(SRAM) has
beco
m
e a
n
i
m
p
o
rtant co
mpon
ent of
ma
n
y
very lar
ge s
c
ale
integr
ation
(VL
S
I) chips. L
o
t of app
lic
ations
preferre
d to us
e the
6T
SRA
M
beca
u
se
of i
t
s robustn
ess
an
d
very hi
gh s
pee
d. How
e
ver, th
e le
aka
ge curr
ent has
incr
ea
sing w
i
th th
e in
crease S
R
AM
si
z
e
.
It consu
m
e
s
mor
e
pow
er w
h
ile
in stan
dby
conditi
on. T
h
e
pow
er
dissi
pa
tion has
bec
o
m
e
an i
m
p
o
rta
n
ce cons
id
erat
ion
due
to th
e
incr
ease
int
egrati
o
n, op
erati
n
g
sp
eeds
a
nd t
he e
x
plosiv
e grow
th
of battery
op
erated
a
ppl
ia
n
c
es.
T
he o
b
jectiv
e
of this
pap
er
is
to revi
ew
an
d
disc
uss sev
e
r
a
l
met
hods
to
overco
me the
pow
er d
i
ssip
a
ti
on
prob
le
m
of SR
AM. Low
pow
er SRAM c
an
be
prod
uce
d
w
i
th improve
m
ent i
n
ter
m
of
pow
er d
i
ssip
a
ti
o
n
duri
ng the sta
ndby co
nditi
on
, write
operati
on an
d read o
perat
i
on. Disc
harg
i
ng a
nd c
harg
i
ng of bit l
i
nes
consu
m
es mor
e
pow
er duri
n
g
w
r
ite ‘
0
’
a
nd
‘1
’
co
mpar
ed to read o
per
ation.
One
of the methods to pr
oduc
e
low
pow
er S
R
AM des
ign
is
w
i
th mak
e
mo
dificati
on c
i
rcui
t at a sta
n
d
a
rd
6T
SRAM c
e
l
l
.
T
h
is
mo
difica
tion
circuit w
ill hel
p
to decrease
p
o
w
e
r dissip
a
tio
n
and l
eak
age
current. Sever
a
l metho
d
w
a
s discusse
d in this
pap
er for und
e
r
stand the
met
hod to pro
duc
e
low
pow
er
design of SRAM c
e
ll. Rec
o
mmen
datio
ns for future
researc
h
are a
l
so set out. T
h
is review
gives
some i
dea for
future researc
h
to impr
ove the des
ign of lo
w
pow
er SRAM cell.
Ke
y
w
ords
:
le
a
k
age curr
ent, p
o
w
e
r dissip
a
tio
n
, read w
r
ite operati
on, SRA
M
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
In mode
rn
world, inte
grat
ed
circuit (I
C) is an
adva
n
ce
d ele
c
tri
c
circuit
by p
a
tterned
diffusion
of trace ele
m
ent
s into the thi
n
su
rfa
c
e of
a se
mico
nd
uctor mate
ria
l
. All electro
n
ic
equipm
ent today is used IC and it ha
s revolutioni
ze
d the worl
d o
f
electroni
cs [1-8]. Due to t
h
e
int
r
odu
ct
ion
o
f
compl
e
x
me
t
a
l ox
ide
sem
i
con
d
u
c
t
o
r
(C
MOS
)
t
e
chn
o
l
ogy
,
semi
co
n
duct
o
r d
e
v
i
ce
s
spread th
rou
gh every di
scipline of en
gi
neeri
ng
a
nd CMOS
me
mo
ries be
come core
compo
n
ent
[9-15]. As
CMOS tech
nol
ogy develop
ment, the m
e
mory b
e
co
mes the
mai
n
po
wer i
n
the
System-O
n-Chip (SO
C
) a
nd pa
rt of t
he chip
i
s
t
he mem
o
ry
circuit. The
memory
chi
p
is
descri
bed
a
s
integrated
circuit
pla
c
ed
on
a
Pri
n
te
d Ci
rcuit Bo
ard
(PCB
).
Ran
dom A
ccess
Memory
(RA
M
) is the
common name
for the m
e
mo
ry.RAM is fu
n
damental
sy
stem com
pon
e
n
ts
use
d
in th
e
transfe
r a
nd
stora
ge
of d
a
ta thro
ugh
o
u
t com
puter
system. In v
e
ry large
scale
integratio
n
(VLSI) technol
o
g
y, memo
ries devel
o
ped
in
CM
OS p
r
o
c
ess a
r
e
utilized in
a
numb
e
r
of application
s
as the
sou
r
ce
s of store
data in RFID applications
[16-23].
The first ge
n
e
ration
of
RAM not
chip
s a
t
all,
but rathe
r
ferrite ri
ng
s
use
d
to
store
data by
cha
ngin
g
thei
r pola
r
ity to re
pre
s
ent a
0 o
r
a
1 as data. RAM
was not
packa
ged as
memo
ry
chi
p
s
until the co
ming of the inte
grated
circuit.
Memory
ch
ip
s a
c
t as a m
a
trix of switche
s
that sto
r
e th
e
state of one
bit as voltage
. These state
s
are a
rep
r
e
s
entatio
n of the data curre
n
tly being sto
r
ed
in RAM. These me
mory
chips tran
sfer to
and from the different
devices conne
cted t
o
a
comp
uter
system. Memory
chip
s provid
e a fast acce
ss a
r
ea am
on
g the hard d
r
i
v
es and CP
U o
f
a
compute
r
whe
r
e data can
be sto
r
e
d
du
ring
pro
c
e
ssi
ng
with
out in
curring
the pe
rform
ance
penaltie
s
of sl
owe
r
me
cha
n
i
cal ha
rd d
r
ives [24-25].
The two
majo
r of RAM a
r
e
dynamic
RA
M (D
RAM)
a
nd stati
c
RA
M (SRAM
)
. The SRAM
and DRAM u
s
e differe
nt tech
nolo
g
ies t
o
hold the d
a
ta. The co
mmon type is use is DRAM.
Ho
wever, if t
he spee
d i
s
i
m
porta
nce, SRAM is
faste
r
than
DRAM.
SRAM do
es
not ne
ce
ssary to
be refresh
ed
but DRAM m
u
st to be refresh
ed thou
sa
nds of times
per second.
SRAM can gi
ve
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 606
3 –
6082
6064
acce
ss tim
e
s as lo
w a
s
1
0ns,
DRAM
only su
ppo
rt
acce
ss tim
e
s of about
60
ns. SRAM i
s
not
comm
on use like DRAM b
e
ca
use SRAM is more ex
pen
sive, despite it faster. Both type of
RAM
are volatile. Volatile mean
whe
n
the po
wer tu
rne
d
off the data cont
ent will lose.
In this stu
d
y, the static
RA
M (SRAM
)
is
fo
cu
sed. Adv
ance in
CMO
S
technol
ogy
make it
potential to d
e
sig
n
SRAM
for de
crea
se
power
con
s
u
m
ption,
incre
a
se sp
eed p
e
rform
a
n
c
e
a
n
d
high integ
r
ati
on den
sity. To rea
c
h th
e
s
e obj
ective
s, the charact
e
risti
c
si
ze d
e
vice ha
s be
en
desi
gn to very small dime
nsio
n and fe
ature
s
. The tech
nolo
g
y scaling re
sult
s in a significa
nt
increa
se lea
k
age current o
f
CMOS device
s
[26].
As the inte
gration
den
sity of transi
s
tors in
crea
se, in today’
s
SOC
de
si
gn an
d
pro
c
e
s
sors, l
eakage
po
we
r ha
s be
co
m
e
an imp
o
rta
n
ce
co
ncern.
Con
s
id
era
b
l
e
attention h
a
s
been p
a
id to
the desi
gn of
high pe
rform
ance and
lo
w powe
r
SRA
M
s, as they
are imp
o
rta
n
c
e
comp
one
nts
in both
hig
h
pe
rform
a
n
c
e p
r
o
c
e
sso
rs and
ha
ndh
eld d
e
vice
s.
Different d
e
s
ign
reme
die
s
ca
n be imple
m
enting; the
dynamic
p
o
we
r consu
m
ption can
redu
ce d
r
a
s
t
i
cally
decrea
s
e
by
po
wer supp
ly voltage. Howeve
r, with
an
agg
re
ssi
ve scaling
in
tech
nolo
g
y a
s
predi
cted
by
the tech
nolo
g
y roa
d
map,
sub
s
t
antial
probl
em
s ha
ve alre
ady b
een e
n
count
ered
whe
n
the co
n
v
entional 6T
SRAM config
uration i
s
use
d
at an ultra-l
o
w po
we
r su
pply [26]. This
cell indi
cates
weak stability at
very small feature si
zes.
This
pap
er
m
o
re fo
cu
se
s
o
n
po
we
r di
ssi
pation. Th
e p
o
we
r di
ssipati
on ha
s
bee
n
a majo
r
con
s
id
eratio
n
cau
s
ed to th
e ope
rating
speed
s an
d in
cre
a
sed inte
g
r
ation, a
s
wel
l
as cau
s
ed t
o
the explosive
gro
w
th of ba
ttery operatin
g applia
nc
es.
The majo
r d
r
ives of fast lo
w po
we
r de
si
gn
are
supply and processes
scaling
still. This research
studied diss
ipation in power,
whi
c
h can be
use
d
in
co
nju
n
ction
for
sca
ling to a
c
hi
eve fast lo
w
po
wer op
eratio
n
s
. The
po
we
r
dissipatio
n was
happ
en
duri
n
g write, read
and
mode
op
eration,
so
in
this
pap
er showed th
e
several
metho
d
s
whi
c
h u
s
e by rese
arch
er to redu
ce p
o
wer diss
ip
ation
during
write,
read o
r
mod
e
operation a
n
d
finally some recom
m
en
dati
on for future
work.
2. Backg
rou
nd
2.1.
Conv
entiona
l 6t SRAM
Cell
The
co
nventional
6T S
R
A
M
is quite
sim
ilar to
t
he
stat
ic S
R
lat
c
h. It is th
e
com
b
i
nation
of
two inve
rter
and t
w
o t
r
an
sisto
r
s for
co
ntrol th
e
circuit call
Word
Line.
The
standard S
R
A
M
requi
re
s six tran
sisto
r
s pe
r bit, two transi
s
tors (NM
O
S and PMOS) for ea
ch
inverter and
two
transi
s
to
rs
NMOS for access. The Wo
rd Line (WL)
wa
s repl
aci
n
g
the clock to acce
ss the cell by
enabl
ed the li
ne. It is controlled by two
pass tra
n
si
st
or M
5
and M
6
,
sha
r
ed b
e
tween the read
and
write o
p
e
r
atio
n. The SRA
M
can
achiev
e high m
e
mo
ry
den
sity by size it as
small as
po
ssi
ble.
Figure 1 sh
o
w
s the
circuit
conve
n
tional
6T SRAM cell
[27-28].
Figure 1. Con
v
entional 6T
SRAM Cell [2
5]
A conventio
n
a
l 6T SRAM
cell ha
s three co
ndition
operation
s
. It is stan
dby condition
(ci
r
cuit in idle), write o
p
e
ration
(up
d
a
ting the da
ta) and rea
d
con
d
ition
(data ha
s b
een
requ
este
d).
T
he "write
sta
b
ility" and "reada
bilit
y" sh
ould
have
du
ring
The
SRAM ope
rated
in
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Advan
c
e
s
on
Low Po
we
r Desig
n
s for S
R
AM Cell (La
b
onna
h Farza
na Ra
hm
an)
6065
write an
d rea
d
conditio
n
. Table 1 sh
ows the ON/OFF transi
s
to
r duri
ng write an
d read ope
ratio
n
.
The thre
e sta
t
es ope
ration
as follo
ws:
In the
stand
b
y
mode, th
e t
r
an
sisto
r
s M
5
and
M
6
a
r
e
turn
‘OFF’
an
d di
sconn
ect
the cell
from the bit lines [29]. As l
ong a
s
the two cross-co
upl
ed inverte
r
s
(M
1,
M
2,
M
3
and M
4
) co
nne
cted
to the supply,
it will contin
u
e
to reinforce
each other.
Assu
me valu
e at Q node is 1 and Q bar node is‘0
’. T
he rea
d
pro
c
ess is beg
un by the bit
lines
are
both
initially floating hig
h
, then
the word li
nes is ‘H
IGH’, trans
is
tors
M
1
and
M
2
ar
e
tur
n
‘ON’. Value ‘
1
’stored in Q
note is pa
ss to the bit
lines by leaving BL at its pre
-
charg
e
value
and
BL bar was d
i
scharge via tran
sisto
r
s M
2
and M
6
to a g
r
oun
d [30]. At BL, the transisto
rs M
1
and
M
3
pull th
e bi
t line to
‘1’.
On the
othe
r
hand,
wh
en t
he valu
e at
Q no
de i
s
‘0’
and
Q
ba
r n
o
d
e
is‘1’, the BL
pull to ‘1’
while BL
bar to ‘0’.
Between BL
bar and
BL will
have a
minor difference of
delta. After that, these lines reach a
sense
amp
lifier, whi
c
h
will sense
amplifier
whi
c
h line has
highe
r voltag
e. Thu
s
will
inform the
r
e
wa
s ‘0’
o
r
‘1
’ stored. Th
e
faster th
e
speed
of re
ad
operation is b
e
rea
c
h if the highe
r the
se
nsitivity of sense am
plifier
[29].
If value ‘1’
ne
eds to b
e
writ
ing, first th
e v
a
lue ‘
1
’ shoul
d be
ap
plying
to the
bit line
s
. Th
e
trans
is
tor M
6
is turn ‘O
N’ a
nd di
scha
rge
the cha
r
ge
o
n
Q b
a
r
nod
e
via M
2
and
M
6
. Q bar is ‘
0
’;
the tran
sisto
r
M3 is turn ‘ON’ an
d tra
n
s
isto
r M
4
is
t
u
rn ‘OFF’. In this
s
i
tuation, the c
h
arge was
store
d
on the
Q node.
Conve
r
sely, value ‘0’ ne
ed
s to write, the
value ‘0’ sho
u
ld be ap
ply to the bit line. Whe
n
Wo
rd Li
ne
(WL
)
is ‘HIG
H’, tran
si
stors M
1
a
nd M
4
is tu
rn ‘
O
N’
and
ch
arg
e
d
so
re
at th
e BL
discha
rge to
grou
nd via tra
n
si
stors M
1
a
nd M
4
. The n
ode Q i
s
‘0’ so the tran
sist
or M
5
is turn ‘ON’
and tran
si
stor M
6
is turn ‘O
FF’. The ch
arge stores at
B bar line
s
.
Table 1. Tran
sisto
r
Tu
rn O
n
/Off dur
ing
Rea
d
and
Wri
te Operation [29]
M1 M2
M3
M4 M5 M6
Read 1
ON
ON
ON
OF
F
OF
F
ON
Read
0
OF
F
OF
F
OF
F
ON
ON
ON
Wr
it
e
1
O
FF O
N
O
N
O
FF
O
FF O
N
W
r
i
t
e 0
ON
OF
F
OF
F
ON
ON
OF
F
The
ca
use of
the
po
wer di
ssi
pation
du
e
to th
ree
fact
ors;
the
dyna
mic
and
stati
c
p
o
wer
dissipatio
n an
d subth
r
e
s
h
o
l
d
leakage
ch
annel. While
t
he dynami
c
p
o
we
r di
ssi
pation is
cla
ssifie
d
into two cat
egori
e
s, on
e
is short ci
rcuit pow
er a
nd se
con
d
is power con
s
umptio
n during
swit
chin
g.
The
sho
r
t
circuit p
o
wers i
s
the
inp
u
t si
gnal
ha
s a li
mited sl
ope
and
ca
n
cau
s
e
dire
ct
path current f
l
owin
g via th
e gate for a
sho
r
t time
du
ring the
switching o
peratio
n.For thi
s
sh
ort
time frame, t
here
is a
sho
r
t ci
rcuit path
betwe
en V
DD
and
gro
und
and th
e
circuit usin
g a
la
rge
numbe
r of po
wers [30
-
31].
Figure 2. Short Circuit Power Di
ssi
pation
[31]
The e
quatio
n
the en
ergy
con
s
um
ption
per
switchi
n
g
can
be
de
rive from refer to the
Figure 2.
∗
∗
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 606
3 –
6082
6066
∗
(
2
)
Whe
r
e,
V
dd
= voltage sup
p
ly,
I
peak
= Peak current,
t
cs
= time period po
wer
con
s
umptio
n
Short circuit power can b
e
redu
ce wit
h
matc
hin
g
the fall and ri
se times. Bu
t in real
worl
d,
the
ti
mes are not
match
ed,
since optimizi
ng
for pro
p
agation dela
y
can re
sult
in
unmatched ti
mes. The
r
efo
r
e, in the digi
tal circuit,
the circuit po
wer is a bigg
er
source of po
wer
con
s
um
ption [30],
[32-33].
Powe
r con
s
u
m
ption d
u
rin
g
switchi
ng i
s
a
cha
r
g
e
i
s
flow from
V
DD
to the V
OUT
of the
inverter, afte
r an
d curren
t input tran
saction,
with
this V
OUT
was pull to V
DD.
The lum
p
e
d
cap
a
cita
nce CL ca
uses
from
g
a
te cap
a
citan
c
e
s
an
d from
pa
ra
si
tic wi
re
capa
citan
c
e
of the
gate
cap
a
cita
nce of the logic g
a
tes drive
n
b
y
inverter, wh
ich is
sho
w
in
Figure 3.
Figure 3. Dyn
a
mic Po
wer
Dissip
ation [3
1]
Whe
n
the
op
posite
switch
over of th
e in
put, tr
an
sisto
r
NM
OS is turn ‘O
N’ a
nd transi
s
to
r
PMOS is turn ‘OFF’.
The charge on
CL i
s
di
scharge to ground.This
situati
on i
s
calling
the
dynamic
po
wer di
ssi
pation
.
In CMOS circuit, it is
the large
s
t source of en
e
r
gy dissip
atio
n.
Con
c
lu
de tha
t, one rise an
d following fa
ll transiti
on of
the output consume en
ergy is [31], [3
4-
36]:
∗
∗
(
3
)
Whe
r
e,
C
LOAD
= the load ca
pa
citan
c
e an
d
V
DD
= the power
sup
p
ly.
If f is the clo
ck freque
ncy
and the ave
r
age n
u
mb
er of ‘HIGH’ to ‘LOW’ o
r
‘L
OW’ to
‘HIGH’ of the
node is de
n
o
ted by
α
the
n
the power
con
s
um
ption
due to ca
pa
citive switchi
n
g
is
given by:
(
4
)
Whe
r
e,
α
=
ac
tivity fa
c
t
or,
V
DD
=voltage swi
ng of the output nod
e,
C
LOAD
= effective capa
citan
c
e of the outpu
t load and
F = switchi
ng
freque
ncy.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Advan
c
e
s
on
Low Po
we
r Desig
n
s for S
R
AM Cell (La
b
onna
h Farza
na Ra
hm
an)
6067
The static co
mpone
nt
of
p
o
we
r con
s
um
pti
on ha
s b
e
en ign
o
re
d in
static
CMOS
. But, a
numbe
r of l
e
aka
ge m
e
ch
anism
s
begi
n
to gain
si
gn
ifican
ce. Mo
st of these
m
e
ch
ani
sms a
r
e
indire
ctly or d
i
rectly ca
use to the
small d
e
vice ge
omet
ries [31], [34-35].
(a) I
hot
= gate
curre
n
t
(b) IPT = cha
nnel pu
nch throug
h lea
k
ag
e
(c
) I
sub
= the subthre
s
h
o
ld l
eakage
curre
n
t
(d) I
gate
= the gate oxide tu
nnelin
g
(e) IGIDL = g
a
te indu
ced d
r
ain lea
k
a
g
e
(f) I
rev
= reverse bia
s
pn ju
nction le
akag
e
Sub-threshol
d lea
k
a
ge via
a M
O
S devi
c
e
ch
ann
el called
th
e se
cond so
urce o
f
leakage
curre
n
t. De
sp
ite a tran
sisto
r
is tu
rn ‘OF
F
’, but
still have lea
k
age
cu
rre
nt, via the cha
nnel at th
e
microsco
pic l
e
vel. Thi
s
current i
s
call
ed,
as
the
sub-t
here
s
h
o
ld l
e
a
k
ag
e
curre
n
t
due it
hap
pe
ns
durin
g the gat
e voltage is b
e
low V
th.
Equ
a
tion of leaka
ge cu
rrent is:
1
)
(
5
)
From th
e e
q
u
a
tion we
can
see
directly t
he
su
b
-
thre
sh
old lea
k
a
ge
current i
s
redu
ce if V
th
increa
se a
n
d
vice versa.
Therefore
asymmetric
con
f
iguration de
als with
du
al threshold
volt
age
in each SRA
M
cell.
Re
sea
r
che
r
s have done
rese
arch to
over
com
e
this problem
and propo
se
several
method
s
of solution. T
h
is
pape
r
will
un
cover a
nu
m
ber of
ways t
o
solve th
e p
r
oble
m
of
po
wer
con
s
um
ption.
2.2.
Single Bitline 6t SRAM Cell
Majumd
ar an
d Basu a
r
e
usin
g the theory with d
e
c
re
ase bit lin
e cap
a
cita
nce of the
SRAM cell wi
ll lower dynamic power
di
ssi
pation
with
out deg
ra
din
g
the p
e
rfo
r
mance to d
e
s
ign
their pro
p
o
s
e
d
SRAM cell. So these de
si
gns SRAM
o
n
ly using a si
ngle bit line for write and read
operation. M
ean that only
singl
e
bit pro
c
e
ss
wa
s inv
o
lved for di
scharg
e
an
d ch
arge
du
ring t
heir
operation.
The l
e
a
k
age
power and
a
c
tive power co
ntrol
th
e p
o
wer
co
nsumed
du
ring
the
o
peratio
n
pha
se. A lea
k
ag
e po
we
r i
s
the p
o
wer
con
s
um
ed
when
cha
r
ge
s
leak via tran
sistor th
at is t
u
rn
‘OFF’. Whil
e an active p
o
w
er i
s
the p
o
w
er,
con
s
um
ed wh
en b
o
th
pull-d
o
wn an
d pull-up net
work
are a
c
tive, creating a direct current flow
from V
DD
to ground [37].
Figure 4. Low Power Singl
e Bit Line 6T SRAM cell [3
7]
For eve
r
y wri
t
e operation
of the conve
n
t
ional,
before
the pre
c
ha
rg
e starte
d, bot
h of the
bit line is pla
c
ed
com
p
lem
entary data i
s
. Depen
d on
the data valu
e only one of
the bit lines
will
cha
r
ge. The
circuit is assumed that the capa
cito
r is discha
rg
e after the writ
e is compl
e
te. So
durin
g a write
operation th
e power di
ssi
pation is
dou
ble. Wh
en re
ad ope
ratio
n
, both the bit lines
once ag
ain a
r
e
cha
r
ge
d a
nd then
one
is di
scharged
duri
ng readi
ng a ‘0’,
whil
e the othe
r
i
s
discha
rge
d
after the ope
rat
i
on is complet
e
.
For th
e p
r
o
p
o
se
d 6
T
SRAM cell,
only
a
singl
e bit
l
i
ne i
s
eith
er
cha
r
ge
d if it i
s
a
‘1’
or
doe
s not get
cha
r
ge
d at al
l assumin
g
th
at t
he data was al
rea
d
y prese
n
t before
the pre
c
h
a
rg
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 606
3 –
6082
6068
circuit ha
s be
en bega
n. Assume
dthe bit line is disc
h
a
rged after the write ‘1’ ope
ration. In a read
‘1’ condition,
the bit line
s
a
r
e p
r
e
c
h
a
rg
e
d
. Mor
eover,
assume
d the
bit lines a
r
e d
i
scharged
after
the read
pro
c
e
ss i
s
over.
Durin
g
a re
ad ‘0’ cycl
e,
the bit lines
cap
a
cita
nce is disch
a
rg
ed
and
pre
c
ha
rg
ed via the cell [37]
.
Figure 5. Fou
r
Tra
n
si
stor
(Single Ende
d
)
SRAM Cell [
37]
The
circuit p
u
rpo
s
e
d
6T
SRAM cell d
e
sig
ned
by
way remove
M
6
(PMOS t
r
ansi
s
tor
control
WL
for BL
b
a
r) a
nd M
1
(NMO
S tran
sisto
r
f
o
r
co
nne
ctio
n Q
no
de to
grou
nd
) a
nd
wa
s
repla
c
in
g by two NM
OS transi
s
tor i
s
calling R
ead
Drive
r
Tra
n
si
stor (MRA
) a
nd Re
ad Dri
v
er
Tran
si
stor (M
RD). Both of these tra
n
si
st
ors
cr
eate co
nne
ction bet
wee
n
BL line
s
to grou
nd. The
advantag
e of removing M
6
from circuit is it can be redu
ce
d by a factor of two of power
con
s
um
ption
from chargin
g
. As only on
e bit line
is
charg
e
du
ring
a rea
d
ope
ra
tion not of two
and the line i
s
ch
arge du
ri
ng write ope
ration abo
ut
o
f
time (assu
m
e equ
al pro
bability of wri
t
ing
‘1’ and ‘0’) i
n
stead of eve
r
y time when
a write o
pera
t
ion is re
quired. The cell area
s redu
ce
by
one-bit line
and o
ne tra
n
s
isto
r. The
a
d
vantage
wh
en M
1
tra
n
si
stor bein
g
taken a
w
ay is t
he
further redu
ct
ion
po
wer co
nsum
ption.
If Q content
‘1’ and Q b
a
r
‘0’, both of
me
mory no
des
will lock each
other
at their
respe
c
tive vo
ltage. Assum
e
Q i
s
‘0’
and
Q ba
r ‘1’,
Q
is floatin
g. See the
Figu
re
4, the le
aka
ge
curre
n
t viatransi
s
tor M
2
sho
u
ld be lo
wer
comp
are of transi
s
tor M
5
to make
sure Q still stay at ‘0’.
If value ‘1’ need
s to write
,
the word li
ne is
cha
r
g
e
to V
DD
,
sinc
e NMOS
t
r
a
n
si
st
or i
s
stron
g
e
r
driv
er than PMO
S
transisto
r, no matter is
incu
rred whil
e
writing a ‘0’ into the cell. The
lack of
the
p
u
ll do
wn
NM
OS tran
si
stor for
memo
ry
Q allo
ws
writ
ing a
‘1’ i
n
to
the
cell
ea
si
ly.
Writing a ‘1’ i
s
don
e by prech
a
rgi
ng bit
line ‘1’ to V
DD
. The Purpo
s
ed SRAM a
nd Conve
n
tio
nal
6T SRAM
ha
ve a
sam
e
way to do
o
peration
write
‘0’
.
The bit li
ne
BL is di
scharge a
nd th
en
wo
rd
line WL i
s
ch
arge
d to V
DD
.
Alternately, if value at Q node i
s
‘0’ b
e
fore
read th
e value at Q
node
s, the bit line is
cha
r
ge to V
DD
. The MRD t
r
an
sisto
r
wa
s turn ‘O
N’. T
he value
at Q ba
r is ‘
1
’ so tran
sisto
r
MRA
wa
s turn o
n
and
will d
r
ain
the cha
r
ge
on the
bit lin
e thro
ugh
tra
n
si
stor
MRD
to gro
und. T
hat
means the bi
t line has just read a ‘0’.Assume va
lue at Q node i
s
‘
1
’ and Q bar
will be ‘0’. T
he
cha
r
ge in bit l
i
ne ca
nnot pa
ss th
roug
h fro
m
MRD to g
r
ound b
e
cau
s
e of transi
s
tor MRA is ‘OFF
’.
These te
ch
ni
que
s
can
de
cre
a
se
of dy
namic po
we
r co
nsumption
to only
40%
to 60%
(be
s
t ca
se / worst case) compa
r
e of a conve
n
tional
6T SRAM cel
l
. Table 2 sh
ows the re
sul
t
of
simulatio
n
do
ne by Majum
dar an
d Basu
.
Table 2. Power Di
ssi
pation
betwee
n
Pro
posed
SRAM
and Co
nvent
ional 6T SRA
M
Cell [37]
Conventional
6T
SRAM
cell
Proposed SRAM
cell
WRITE ‘0’
162µW
0µW
WRITE ’1’
162µW
81µW
READ ‘0’
243µW
162µW
READ ‘1’
243µW
81µW
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Advan
c
e
s
on
Low Po
we
r Desig
n
s for S
R
AM Cell (La
b
onna
h Farza
na Ra
hm
an)
6069
2.3.
SRAM
Cell For Portable
Dev
i
ces
This de
sign
wa
s fo
cu
s to
redu
ce
po
we
r dissip
ation d
u
ring
write op
eration.
Up
ad
hyay
et
al. wa
s
su
gg
est to i
n
cl
ude
two
more trai
l tran
sisto
r
s
(M
7
and M
8
)be
t
ween
pull
do
wn
network a
nd
grou
nd level for prope
r dischargi
ng an
d chargi
ng a bit lines a
s
sho
w
n in Figure 6 [25].
Figure 6. SRAM Cell of Upadhyay et al
. [25]
The su
b-th
re
shol
d, which is flowing in the ci
rc
uit during trans
i
s
t
or, is
in c
u
t off
region
can
re
du
ced
with u
s
e
the
two p
u
ll d
o
wn tran
si
stors.
Threshold
voltage i
n
fluen
ce th
e le
aka
g
e
curre
n
t in
ci
rcuit. When
sub-th
re
shol
d
curre
n
t is
in
crea
se, th
e thresh
old volta
g
e
is de
crea
ses.
Becau
s
e of d
r
ain ind
u
ced barrier lo
we
ri
ng (DIBL
)
in the MOSFET the drain volt
age incre
a
se
s if
the thre
shold
voltage de
cre
a
se
s. The eq
uat
ion can be
derived a
s
b
e
low [25, 38]:
′
(
6
)
Whe
r
e,
=
DIBL c
o
effic
i
ent,
= thre
shol
d voltage an
d
= drai
n voltag
e.
The in
crea
se
the threshol
d voltage
by
usin
g
two
trai
l tran
sisto
r
s h
a
s
red
u
ced t
he d
r
ain
voltage with
decrea
s
e
d
su
b-threshold
current. De
cre
a
se
d sub-th
resh
old
can
b
e
red
u
ci
ng p
o
w
er
dissipatio
n. The main fa
ct
or to e
n
sure
prop
er
fu
ncti
oning
of SRAM is the
si
ze of tran
sist
ors.
Acco
rdi
ng to
Thumb
Rule,
the Wi
dth ratio of tra
n
si
sto
r
s
M
5
an
d M
3
, M
3
and
M
1
, M
6
and
M
4
an
d
M
4
and M
2
, is equal to 1.5.
1
.
5
and
1
.
5
(
7
)
This
size con
f
iguration
pro
v
ides the p
r
o
per
driving v
o
ltage to trans
is
tors
for turn ‘ON’
and ‘O
FF’
co
ndition. Th
e
M
7
and M
8
tra
n
si
stors
were
function
to redu
ce p
o
wer dissipatio
n
with
cut off the circuit duri
ng write ‘1’ to ‘0’ and ‘0’ to ‘1’.
Assu
me want
to write valu
e ‘1’. Have 2
situat
ions, from ‘1’ to ‘1’ or ‘0’ to ‘1’. Node B
must
write to
‘0’, to get n
ode B ‘0’, BL
sho
u
ld
settin
g
to ‘0’an
d
a
s
serting
WL.
Figure 7
sho
w
s
write ‘1’ ope
ration. Situation 1: from ‘1’ to ‘1’ writ
e to cell. This is not
possible be
cause both no
de
B and BL a
r
e at ze
ro p
o
t
ential. Situation 2: from
‘
0
’ to ‘1’ write to cell. In
Upa
dhyay et al.
prop
osed cell, it is simple
to flip the cell state
from ‘0’ to ‘1’ by b
e
fore as
serting WL, turn
ON
trans
is
tor M
7
with s
e
t WS is
‘1’.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 606
3 –
6082
6070
Figure 7. Path for Write
“1” [25]
Ho
wever, if someon
e wa
nts to write val
ue 0,
then th
ere have t
w
o
situation
s
, from ‘0’ to
‘0’ or ‘
1
’ to ‘0’.
Nod
e
B mu
st
write to
‘1’, to get B n
ode
‘1’ BL shoul
d
set to V
DD
an
d asse
rting
WL.
Figure 8
sh
o
w
s write ‘0’
operation. Situation 1:
fro
m
‘0’ to ‘0’
write to
cell. S
i
nce
nod
e B
is
initially high, t
herefore this
write pa
ttern is
not poss
ible. Situation
2:
from ‘
1
’ to ‘0’ write to cell.
In
Upa
dhyay et
al. propo
sed
cell, it i
s
simp
le to flip
the
cell state from
‘1’ to ‘0 by
set WS to ‘
0
’,
so
that cha
r
g
ed
can
not di
sch
a
rge
thro
ugh
M
2
to gro
und
[25]. The fun
c
tion of
WS
signal i
s
to ma
ke
s
sure the
right
value of sig
n
a
l and the
ri
ght ope
ra
tion
before
asse
rting WL, tra
n
sition fro
m
1 to 0
and 0 to 1 ca
n be ea
sily allowe
d.
Figure 8. Path for write
“0” [25]
The po
wer di
ssi
pation du
ri
ng write ‘0’ a
nd ‘1’
operation is more, causes on
e of the two
bit lines of
co
nventional 6
T
SRAM cell
should b
e
di
scharg
ed to lo
w rega
rdle
ss written value. In
prop
osed SRAM, by control a sign
al WS, transisto
r
M
7
or M
8
can
be turn ‘OF
F
’. Proper
cont
rol of
sign
al WS ca
n avoid any si
ngle bit line from being
di
scha
rge
d
du
rin
g
write
‘0’ or write ‘1’ mode.
The re
sult of comp
ari
s
o
n
betwee
n
pro
p
o
s
ed SRAM a
nd co
nventio
nal 6T SRAM
cell on
different freq
uen
cy du
ring
write
ope
ratio
n
are
sho
w
n
at Table
3. During
the
swit
chin
g a
c
tivity the
prop
osed SRAM cell, dissi
pate lower dy
namic
po
wer.
Comp
are to
conve
n
tional
6T SRAM cel
l
,
the power di
ssipatio
n for propo
sed SRA
M
cell re
du
ce
d by 12% to 38% [25].
Table 3. Power Di
ssi
pation
betwee
n
Co
nventi
onal 6T
SRAM and Propo
se
d SRA
M
Cell [25]
Freque
ncies
Conventional 6T
SRAM cell
(uW
)
Proposed SRAM
celll
(uW
)
Percentage decr
ease (%
)
500MHz 3.95
2.45
38.00
1GHz
6.01
4.97
19.8
2GHz
9.83
8.6
12.8
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Advan
c
e
s
on
Low Po
we
r Desig
n
s for S
R
AM Cell (La
b
onna
h Farza
na Ra
hm
an)
6071
2.4.
7T SRAM Ce
ll
Akashe and Jain proposed the circuit to
reach improvements in
performance,
stability
and
p
o
wer d
i
ssi
pation co
mpared with
a
conventio
nal 6T S
R
A
M
cell. T
h
e
7T SRAM
ce
ll is
prop
osed for
innovative prech
a
rgi
ng an
d bit li
ne balanci
ng sche
me durin
g write operatio
n and
maximum sta
ndby po
wer
saving in a SRAM array. Fig
u
re 9 sho
w
s
prop
osed 7T
SRAM cell [2
6].
The mai
n
so
urce of
stand
by power fo
r
a SRAM
cell
is lea
k
a
ge
cu
rre
nt whi
c
h th
e majo
r
comp
one
nts i
s
sub
-
thresho
l
d lea
k
a
ge, th
e gate
di
re
ct
tunnelin
g le
a
k
ag
e a
nd th
e
reve
rse bi
ased
band
to ba
nd
tunneli
ng ju
n
c
tion l
e
a
k
age
. The
subth
r
e
s
hol
d lea
k
a
g
e
, whi
c
h
is de
fined a
s
a we
ak
inversi
on
con
ductio
n
curre
n
t of the CM
OS tran
si
sto
r
whe
n
Vgs < Vth, represe
n
ts a
signifi
cant
leakage
curre
n
t compo
nent
in the off-state [26];
1
)
e
(
8
)
Figure 9. Pro
posed 7T SRAM cell of Akash
e
and
Jai
n
[26]
From
Equatio
n (8), the
V
ds
of the MOS
F
ET influe
nce the
su
bthre
s
hol
d
curre
n
t. Whe
n
V
ds
incre
a
se, the subth
r
e
s
hol
d cu
rrent
also in
crea
se. So Anie Jain and Shy
a
m Aka
s
he
wa
s
addin
g
M
5
(a
dditional tra
n
s
isto
r) to redu
ce the sub thresh
old curren
t [26].
1
.
(
9
)
From Eq
uatio
n (9),
we
kno
w
the sub thresh
ol
d lea
k
a
ge cu
rrent isprop
ortio
n
al tran
sist
o
r
size. Anie Jai
n
and Shya
m
Akashe
sug
g
e
st with
scali
ng do
wn th
e
pull do
wn
NM
OS can
re
du
ce
the leaka
ge current.The BL
voltage also
affects t
he le
aka
ge cu
rrent
. The leaka
g
e
current ca
n b
e
redu
ce
d by decrea
s
e the voltage.
Both bit line
sho
u
ld b
e
re
store to V
DD
fol
l
owin
g a
writ
e op
eratio
n i
n
the
conve
n
tional 6
T
SRAM to ma
ke
su
re a
su
ccessful
read
operation. T
he write am
pl
if
ier ci
rcuitry
of ensures t
hat
the sel
e
cte
d
bitline is b
a
ck to a hi
gh v
a
lue
by ge
ne
rating a
neg
ative pulse to pre
c
h
a
rge
the
sele
cted
bitlin
e hig
h
afte
r d
r
iving the
bitli
ne ‘L
OW’
to
write
‘0’ i
n
to
a SRAM
cell. The
r
efo
r
e, b
o
th
bit lines (BL a
nd BL bar) wil
l
be resto
r
e
d
to
high state a
fter write op
eration [26, 38].
Duri
ng the
write
ope
rati
on, firstly cu
t off the feedba
ck
co
nne
ction by tu
rn ‘OFF’
trans
is
tors
N
5
. BL bar carries co
mplem
ent of the input data. Tra
n
si
stor N
3
is
kept ‘OFF’ a
nd
trans
is
tor N
4
is tu
rn ‘
O
N’.
The BL
b
a
r tran
sfers th
e
compl
e
me
nt
of input
data
to inve
rter 2
to
develop
Q, cell data, which drive
s
inve
rter l and
dev
elop
s QB. Assume
wa
nt to write ‘0’. BL
bar
is kept ‘HIG
H’ with negli
g
ible write powe
r
co
ns
u
m
ption. Assu
me want to write 1. BL bar i
s
discha
rge
d
to ‘0’ with comp
arabl
e po
wer
con
s
um
ption
to a conventi
onal write.
The
rea
d
o
p
e
r
ation f
o
r
pu
rposed S
R
AM
is
sa
me li
ke
conve
n
tional
6T SRAM
cel
l
and
N
5
is turn ‘
O
N’.
Assu
me valu
e at Q nod
e i
s
‘0’. The
rea
d
path con
s
ists of N
2
an
d N
4
, and exa
c
tly
behave
s
li
ke
the co
nventio
nal cell. Assu
me value at
Q nod
e is ‘
1
’. The read
pat
h
co
nsi
s
ts
of N
1
,
N
5
and N
3
tha
t
repre
s
e
n
ts a
critical
rea
d
path [26].
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 606
3 –
6082
6072
Anie
Jain
an
d Shyam
a
k
ash
e
a
r
e
u
s
i
ng
CADE
NCE for
done
t
he
simulatio
n
and
the
result sho
w
at Table
4
a
nd 5.
The
re
sult
sho
w
th
e
pro
p
o
s
ed
7
T
SRAM
a 4
5
% redu
ctio
n in
power con
s
u
m
ption du
ring
write ope
rati
on com
p
a
r
e
with a co
nven
tional 6T SRAM.
Table 4. Power Co
nsumpti
on between
Conve
n
tional
6T and Pro
p
o
s
ed 7T S
R
A
M
cell [26]
Parameter
Power
Co
nsumpt
ion
6T
2.097mW
7T
1.147mW
Table 5. Lea
kage Current b
e
twee
n Co
nventional
6T a
nd Prop
osed
7T SRAM Cel
l
[26]
Parameter
Leakage
Cur
r
ent
6T
2.311mA
7T
1.148mA
2.5.
Cach
e De
sig
n
Aly and Bayoumi pro
p
o
s
e
d
7T SRAM
cell for redu
ce power
con
s
umptio
n du
ri
ng write
operation. Th
e ch
arg
e
and
discha
rge
of a bit
line influen
ce po
we
r con
s
um
ption
.
The purpo
se
SRAM only u
s
e BL b
a
r fo
r perfo
rms
write operation
and it dep
en
d on cutting
off the feedb
ack
con
n
e
c
tion b
e
twee
n the 2 inverters, inverter 1 an
d
in
verter 2, befo
r
e a write o
p
e
r
ation. Figu
re
10
sho
w
s pro
p
o
s
ed 7T S
R
A
M
cell [39].
Figure 10. Aly and Bayoumi’s 7T SRA
M
Cell
Aly and Bayoumi’s p
r
op
osed ci
rcuit is
same a
s
ci
rcui
t propo
se
d b
y
Akash
e
an
d Jai
n
’s.
Only have
so
me differen
c
e
at
WL
sig
nal
betwe
en
th
ese ci
rcuits.
For ci
rcuit propo
sed
by Aly a
n
d
Bayoumi, the
WL
sign
al is isolate
d
, whi
c
h
controls t
he tran
si
stors N
3
and
N
4
. Ho
wev
e
r
ci
rc
uit
prop
osed by Akashe an
d Jain’s a
r
e u
s
in
g only
singl
e sign
al to cont
rol both tra
n
si
stors.
Figure 11. 7T
SRAM –write
operatio
n of Aly and Bayoumi pro
p
o
s
ed
Evaluation Warning : The document was created with Spire.PDF for Python.