Short-circuit dissipation of static CMOS circuitry and its impact on the

advertisement
468
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-19, NO. 4, AUGUST 1984
Short-Circuit Dissipation of Static CMOS
Circuitry and Its Impact on the Design
of Buffer Circuits
HARRY
J. M. VEENDRICK
Abstract —This paper gives a detailed dkcussion of the short-circuit
compouentin the totaf powerdissipationin CMOS circuits, on the basisof
an elementaryCMOS inverter. Designconsiderationsare givenfor CMOS
buffer circuits, based upon the results of the dissipation discussion,to
increasecircuit performance.
LIST OF PARAMETERS
A
a
b
process-determined
ratio between
@ of NMOS
P,
h
~ of PMOS
~ of nMOST
input
ALP
rise or fall time of an input
To
rise or fall time of an output
Tr
rise time of a signal
v dd
supply
~,
input
v out
VT
VT.
v
output voltage
threshold voltage
transistor
channel width
with
capacitance
equal channel
length
INTRODUCTION
of the Nth
inverter
of a
D
of a string
total power
URING
the last
on node N
of inverters
dissipation
periods
when no signal transients
Manuscript received October 19, 1983; rewsed December 28, 1983.
The author is with Philips Research Laboratones, 5600 JA Eindhoven,
The Netherlands.
0018 -9200/84/0800-0468
CMOS
technology
has
technologies
for
is its
low
occur.
However,
static
during
during
signal there will always be a short-cir-
cuit current flowing from supply to ground in static CMOS
circuits. So far only limited analyses and discussions have
appeared in the literature
on this power component
of
static CMOS circuits [1].
In integrated circuits it is always necessary to drive large
circuitry,
etc.), often at
capacitances (bus lines, “off-chip”
high clock frequencies. Such driving circuits (buffers) will
take a relatively large part of the total power consumption
of the chip. It is clear that optimization
requires a different approach as compared
of CMOS
dynamic power dissipation
short-circuit
power dissipation
period-time
of a signal ( = l/~)
years
due to the absence of dc currents
an edge of an input
current
five
become one of the most dominant
VLSI circuits.
The most important
reason for this
power dissipation,
mean value of the short-circuit
current
maximum value of the short-circuit
current
gate length of the nMOST
gate length of the pMOST
gate length minus effective channel length of the
nMOST
gate length minus effective channel length of the
pMOST
P
Pl
P2
T
of the Nth inverter
and
( = I/T)
number
of the Nth inverter
of the Nth (symmetri-
gate oxide capacitance
N
of the pMOST
of a string
total capacitance on node N
input capacitance of the first inverter
short-circuit
signall
voltage
w PN
and pMOST
signall
voltage
G:N
string
load capacitance
PN
fall time of a signall
T,
width
frequency
ALn
and
rise or fall time of a signall
T-
transistor
gate capacitance
parasitic
I
I mean
I max
Ln
Lp
in (11)
time
r
threshold voltage of nMOST
threshold voltage of pMOST
channel width of the nMOST
/3 of a transistor
channel
CL
CN
co
c 0x
defined
Lp and Ln [(A9)]
cal) inverter
c gN
constant
ratio between parasitic
nodal capacitance
load capacitance
gain factor (pA/V2
) of an MOS Itransistor
i
lb
USED
t
tion
logic [2]. These buffer
to obtain
detailed
inverter
minimum
power
discussion on power
will be given first.
circuits
of such circuits
to optimization
need extra atten-
dissipation.
dissipation
Therefore,
a
of a basic CMOS
lAlthough the rise and fall times are commonly defined to be the time
between the 10 and 90 percent level of the signal extremes, in this paper
these parameters are defined as the total duration of a linearized edge.
$01.00
01984 IEEE
469
VEENDRICK: SHORT-CIRCUIT DISSIPATION OF STATIC CMOS CIRCUITRY
-Wjd
the
-G-
~“
transistor
will
be in
saturation
this
during
of time.
Using
v@Jt
1
NMOS
period
PMOST
the simple
MOS formula,
this leads to
NMOST
-
Fig. 1.
(4)
v.~
Basic CMOS inverter
This current
will reach its maximum
half the supply
“nk=i-k-n
tion
voltage
that the inverter
of the input
!
1! !
I _=2*;
Assuming
Fig. 2.
Current behavior of an inverter without load
input
A static CMOS
OF A BASIC CMOS
inverter
the absence of transients
on the input:
f’I(t)dt
t~
equal
conduct.
However,
during
there will be a time period
pMOST
from
will
supply
without
voltage
~,
conduct,
to ground,
load.
(Vi.)
This
in which
causing
both the nMOST
(1)
as long
is higher than a threshold
and lower than a threshold
If we load the inverter
(Fig.
tance C~,’ then the dissipation
it can be derived
and
short-circuit
Clearly,
as the
input
Equations
of the circuit
Pz = Imem”V.
(2)
I
Since there is a difference
without
in the short-circuit
load
and that
dissipation
of an inverter
with
load, we start our discussions on the basis of an inverter
with zero load capacitance. For simplicity
we assume that
the inverter
fundamentally
is symmetrical
Bn=Bp=B
During
current
(an asymmetrical
different),
the period
which
inverter
O to l~m,
Fig. 2 that
and
(7)
t2=;.
t-v~)+
(8)
the output
1 .~”(vdd–2vT
‘em = fi
vdd
)3”;.
(9)
p2=+”(vdd–2v~)3”;
.
(lo)
As l/T=
f,
(10) shows that this dissipation
also proportional
to the frequency
Vdd and VT are process-determined,
component
is
of switching,
Because
the only design param-
eters that affect P2 are /.? and the input rise and fall times
(~) of the inverter.
For an inverter
with capacitive
load, the ~‘s of the
transistors are determined by requirements
on output rise
and fall times. In this case the short-circuit
dissipation
(3)
depends only on the duration of the input signal edges. As
will be shown further on, these edges should not be too
long, especially in the case of driver circuits that have a
means that
and VTH=– VT,=VT.
from
(6)
is not
(tl – t2;Fig. 2) in which the short-circuit
Z increases
between
its transients
From (2) and (9) the following expression can be derived
for the short-circuit
dissipation of a CMOS inverter without
load:
due to para-
design.
of an inverter
relation
(5), (6), and (7) lead to
PI does not depend on
contributions
sitic output capacitances,
such as junction
capacitances).
Pz, however, strongly depends on
The second component
the inverter
~f = ~) of the
(q=
has the solution
consists of two
dissipation:
from
(5)
a capaci-
(1)
component
times2
and a linear
I mean —
-~~;:”’’”d’(+r-v,)d(~
which
of Fig. 1 with
Pl = C~”V2” f and
design (apart
(~n(t)–vT)2dt.
voltage ( V~n) above
dissipation:
the dynamic
the inverter
from
VT
tl=.—— -r
v dd
components:
dynamic
rise and fall
as
to flow
( lV~Pl) below Vdd.
output
of
T (equal to one period
voltage (Vin) and time (i) during
on the input,
a short-circuit
flows
result
fin(t)=+t,
as shown in Fig. 2 for an inverter
current
Another
power during
when the input
a transient’
a time
=;J:’;
(symmetrical)
1) is at high level ( V~~), only the NMOS transistor conducts, and when the input is at low level, only the pMOST
will
during
INVERTER
does not dissipate
due to the assump-
was symmetrical.
signal) can thus be written
signal
the input
DISSIPATION
value when ~n equals
= Vdd/2),
this assumption
is that the current behavior during the
with respect to the
time period tl – t3 will be symmetrical
time t2.
The mean current
If;
(~.
large ~ value. In the derivation
of (10), we started with an
voltage
(V&,) will be larger than the input voltage ( ~,) minus the
threshold voltage (VT) of the nMOST. As a consequence,
2The definitions of rise and fall times used here are different from those
in common use (see the list of parameters used).
470
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-19, NO. 4, AUGUST 1984
V(jd=!iv
[WI]
pMosT VT.Vfn= -Vlp. 0.8 Vo[t
v,” -it
‘“Y:
~ ‘/% ’/$=170
UA/v2
/
L.
I
~
2
Fig. 3.
T,= TO= 10 n5ec
The inverter used in the example
1(+.-
;,.:
,,,~
/
r
,’
71= TO= 5 nsec
6\
~
240
7, .5
220
[nsec]
/
CL=0
20
~o 1
18i
dynamic
&
L’
/
dlsslpatlon
=0.s..)
(~,
/’
2
i’
2
,
L68yj2
T
‘
Fig. 6.
Inverter dissipation
time -
Fig. 4.
Short-circuit
current
as a function
capacitances.
of different
inverter
‘rO.
transient
capacitances
-f - 5@c]
to ground
it is obvious
of the input
hme
what
happens
when
following
we load
will
signal.
depend
with zero load
versus load
on the rise and fall
Fig. 6 shows this characteristic
values of the input
rise and fall times ( 71). The
the load capacitance
example
the inverter
with
examines
different
to equal input
and out-
characteristics
are
on the figure.
From
these characteristics
operation
input
capacitances.
corresponds
put rise and fall times for the different
indicated
The
inverter
dashed line shows the dynamic dissipation
(~= 10 MHz),
while the solid lines show the actual inverter dissipation
(dynamic plus short-circuit
dissipation).
The points where
Inverter output voltage behawor for different inverter load
capacitances.
load.
a high level
that the dissipation
characteristic
for different
without
from
with a fall time of 5 ns.
capacitance
times
values of the load
is switched
As expressed in (10) for a CMOS
inverter
of the inverter load capaci-
curves for different
when the input
capacitance,
Fig. 5.
2FT]
load
the output
[v]~
as a function
tance.
“d
of the inverter
we can conclude
that if the
is such that the output
signal and
signal have equal rise and fall times, the short-circuit
dissipation
will
be only
a fraction
( <20
percent)
of the
Example
total dissipation.
However, if the inverter is more lightly
loaded, causing output rise and fall times that are relatively
The discussions that follow are based upon the inverter
whose parameter values are shown in Fig. 3. Its operation
short as compared to the input rise and fall times, then the
short-circuit
dissipation
will increase to the same order of
was simulated
are presented
magnitude as the dynamic dissipation.
Therefore, to minimize dissipation, an inverter used as part of a buffer should
current
with a circuit analysis program, Some results
in Fig. 4. The figure shows the short-circuit
behavior,
during
a time interval
tl – td (see Fig. 2),
as a function of the load capacitance CL, for input rise and
fall times of 5 ns. Curve @ shows the behavior of the
inverter
without
load. At any time this current is the
maximum
short-circuit
current that can occur. This means
that all other current
characteristics
for different
load
capacitances must be within this curve. Curve @) shows
the short-circuit
current behavior of the inverter when it is
loaded with a characteristic
capacitance CL of 500 fF. In
this case the rise and fall
times
on the output
equal to the rise and fall times on the input.
node
are
Fig. 5 shows
be designed in such a way that the input
are less than
or about
equal
rise and fall times
to the output
rise and fall
times in order to guarantee a relatively small short-circuit
dissipation.
Fig. 7 shows the linear relationship
[according to (10)]
between the short-circuit
dissipation and the input rise and
fall times (7, ) derived by means of circuit simulations.
In
this case the kverter
of Fig. 3 was loaded with a capacitance of 500 fF. From Fig. 5 it was known that a load
capacitance
output
of 500 fF causes 5 ns rise and fall times of the
signal,
when the inverter
input
rise and fall times
VEENDRrCK: SHORT-ClRCUIT Dissipation
OF sTATIc cMos CIRCUITRY
[NJ
CL= 500 fF
160.
dynamic
lLO.
~
(
‘–-
120.----------------------------
4-——i~
,“t*r”al
s
100
0
n.
80
ix
~1
60 ~
v
LO
*
~
short-c!rcwt
4+6
8
T,=TO=5
nsec
1012
14
16
1’820i22L
262830
$2@Se~
Fig. 8.
—
Inverter dissipation as a function of the input nse and fall times.
~, = TO= 5 ns, is indicated
7 and the corresponding
short-circuit
only about 10 percent of the dynamic
This
result
of designing
way that the input
a string
and output
on
dissipation
is
dissipation.
capacitance
and therefore
dissipation,
it can also be applied
of inverters
mum,
is for the maximum
T, = TO. It has empirically
designs
the short-circuit
calculated
cgN+
Cparwl
Inverter string acting as a buffer circuit.
in such a
not
lead to an optimum
design.
and silicon
Design
optimization
minimum
dissipation
approach,
as will be shown in the following.
It has been derived
area requires
in the Appendix
of
the inverter
dissipation
design”
determined
by the following
for
a different
that a “minimum
string
is completely
three equations:
rise and fall times of each
inverter are equal to obtain minimum
dissipation can very
well be applied to the design of static CMOS buffers.
Although (10) was derived for an inverter with zero load
with
#
71
are equal to 5 ns, This point,
Fig.
bo.dlny
the nodal capoc!tances miude the
POrasrtlc trodd capacltOneef.(CpOq )
CM=
2
Fig. 7,
L8u 11., or
str, n~)
%cL+cpdrN
20.
o
+buff., @wwt.r
log ,. yt.
dissipation
short-circuit
to inverters
been found
is half
where
1
vdd–vTn”
designed
that
for
such
‘=
of the maxi-
(
2VT
2(vdd–vTfl)-v
vdd-iT
,,
‘ln
v
[
1}
(11)
with (10).
is constant
for a given technology.
DESIGN CONSIDERATIONS
In integrated
capacitances,
circuits
it is always necessary to drive large
like bus lines or “off-chip”
circuitry.
More-
1%.1
—=*.
b.
(l+b). cox.LnN
. (LnN(l+3a2)-ALnN-3aALPN}
over, this must often occur at high speed, which will take a
relatively
large part
chip.
Particularly
large
number
of the total
power
dissipation
in the case of bus lines, which
of inputs
of the different
control
subcircuits
d
and
on a
b,v N —
CN
—.
()
chip, it is necessary to have short signal rise and fall times
[(10)] to minimize dissipation.
‘
Suppose
we want
line or “off-chip”
to drive
fiN-1
such a bus interconnection
a signal coming from an where ~,?J represents
internal node A. Let us assume that the logic gate, having
6N- l/~N
is the Weriw
node A as output, is capable of charging a capacitance CO rise and fall times and
in a time t to 95 percent of the supply voltage. From the string. For a practical
foregoing results we know that if we use an inverter string
ions are made for the
as a buffer
circuitry
with
vTn=–vT,
rise and fall times on the bus line (or bonding
(Fig.
problem
8) loaded
~
of
the
last
inverter
stage,
factor at equal input and output
N is the number of inverters of the
application
the following
parameters
of (11), (12), and (“13):
assumpt-
now
with
is how
to design
a capacitance
an inverter
CL, with
~ ns rise and
between the P‘s of the successive inverters (tapering
was needed to guarantee a‘ minimum
propagation
vdd=5v
string
fall times on each node, driven from an internal logic gate
capable of charging and discharging the input capacitance
CO in the same time. In [3] it was derived that a factor of e
factor)
delay
time for such an inverter string. However, it is well understood that in terms of dissipation and silicon area this will
=lv
V= 0.05* Vdd [in (11)]
pad) to be driven.
The
the
(13)
co
between node A and the bus line, the rise
circuit
and fall times on each node of the string should be equal to
the required
1(12)
of the
&
=42 pA/V2
A =1
I
/3,0 =14 PA/V2
COX= 700 pF/m2
Ln = 2.5 pm
Required
ALn = 0.5 pm
rise and fall times:
the constants
ALP =1 pm.
~ = 5 ns. Practical
a and b are a = 3/2.5
values for
and b = 0.1.
472
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-19, NO, 4, AUGUST 1984
With
(12) this leads to
q$-qjiq+’.
B.
—=11.5.
&-1
This tapering
dependent;
factor,
nearly
as it is often called, is strongly
all parameters
the process. Different
to different
CMOS
tapering
of the driver
is as follows
according
by
Fig. 9.
Co= 100 fF and CL (Fig.
5 ns rise and fall times (r)
(inverter
(with
string),
8)
COMPARISON OF THE PERFORMANCES OF Two
on
then the design
I
c.f
( humber
(l+
7.
5
I
b). C.=llpF
stage (N)
is determined
by
power
I
P.. =PPN=fi~=
4.4*~O-3
A/V’,
creased
and
will
by 1 ns as compared
LP– ALP
Wp
~=/3PD =315’
The (N – i) inverters
‘0
_ 630pm
LP
~–
(-)
are now determined
3pm
In this paper
static
A-l
For this example,
as shown
therefore,
in Fig.
(This,
needed for this example
of
course,
has to be
string should be
9 to guarantee
a very
small
short-circuit
dissipation and a minimum area consumption,
The parasitic
nodal capacitances
are also depicted
in
Fig. 9.
By means of circuit
simulations
circuit
performance
found
the mean power dissipa-
a simple
when
formula
circuits.
A detailed
with
that if each inverter
way that the input
short-circuit
dynamic
of
a quick
dissipation
dissipation
of
of this short-
capacitances.
of the
It was
of a string is designed in such a
and output
( <20
rise and fall times are equal,
will
be much
percent).
This
less than
result
the
has been
applied to a practical
design of a CMOS driving circuit
(buffer),
which is commonly
built up of a string of inverters. An expression has also been derived for a tapering
factor between two successive inverters of such a string to
minimize
parasitic
power
dissipation.
power, and area) than is possible
propagation
delay.
e is a much
discharge
parasitic
for
dissipation
discussion
different
tapering factors: a factor of 11.5, which is derived in this
paper (process-determined)
from optimization
of power
dissipation
and area, and a factor e, which is derived [3]
from optimization
of the propagation
delay.
In this example the most important
improvement
due to
choosing a tapening factor equal to 11.5 instead of a factor
and a reduced
is derived
is given based upon the behavior
loaded
tion has been calculated at a clock frequency of ~ = l/T =
10 MHz. Table I shows the results of a comparison of two
area ( < 1/4)
(power,
by optimiza~ion
short-circuit
cluded that optimization
leads to a better overall
smaller
of the
like buffers,
AND CONCLUSIONS
the maximum
dissipation
inverter
the
the inverter
of
CMOS
circuit
fF and C~ = 11 pF we find from
will be equal to N =1.83.
rounded off to N = 2.)
optimization
circuits,
equal
delay.
SUMMARY
by the tapering
calculation
of inverters
been in-
“
B.
—=11.5.
(13) that the number
overall
con-
the power
factor
driving
and area) then can be achieved
the propagation
factor
the given CO =100
power
minus
to a tapering
we can state that
of CMOS
lead to a better
delay,
.&
parasitic
is the total power consumption
In summarizing,
‘Oepower dissipation
‘“(%l.=%:
Wp
This
I
dissipation which was actually meant (C~V2~).
In our case, the propagation
delay has only
or (see Appendix)
(&~n)N=%=105
1
( = 1/4).
consumption
sumption
designed
I
to (11):
Thus, the last inverter
With
e
I
invertersl
t’
()
factor
{actortl.5
the above assumptions):
CN=CL+CPXN=
INVERTER STRINGS
WITH DIFFERENT TAPERING FACTORS
to Fig. 8:
according
The designed inverter string for a practicaf example.
lead
TABLE I
situation,
equals 10 pF and we want
procedure
in (12) are determined
processes may therefore
factors.
If, in a practical
the output
process
Finally,
it is con-
in terms of power dissipation
performance
(in terms of speed,
by minimization
of the
APPENDIX
It can easily be derived
a capacitive
[1] what
transistor
is needed to
load C~ from the supply voltage
Vdd
473
VEENDtUCK: SHORT-CIRCUIT DISSIPATION OF STATIC CMOS CIRCUITRY
to a voltage
From
V in the time ~~:
2VT
% = $ “ ~dd2vT
2(vdd–vTn)–v’
“
“ { vdd–;Tn+ln
v
[
)
w)
(21) and (22) we derive
C~_l = (1 + b). (WP&~
This, combined
cN_,
with
(A1O) yields
*l+a.
(3aL~~
2vTn
1
Vdd–v=
1)
2(vdd–vTn)–v
n “ {, vdd–vT
n ‘ln
Equations
-3ALP~)
.(l+b).cox.
LnN – AL~~
(-
=A
v
[
[A13)
(J’LN”LJ
=
where
+ W~#~&X.
(A14)
)
(A4), (AS), and (A14) result in
(A2)
is a constant
for a given technology.
6%1
Thus,
= + “(l+
b)”cox”
Lw”:”fl;
‘N
“ ( LnN - AL~N +3a2. LnN –3a.ALPN).
(A3)
&=~.A
Finally,
and for the (N – l)th
bN-I
(A4)
BN-1=~.A.
assuming
the inverter
pn=pp=p;
and,
from
(A.15)
(18) and (25) we derive
inverter:
c~-~
Again,
‘N
and
because of the difference
~r=~f=T
in mobility
(A5)
of holes and
electrons:
WP– AWP
Wn– Awn
Lp – ALP ‘3*
Ln– ALn”
Wn
=3*
Ln – AL.
r“~no * ‘1 + b)”cOx”L”N
With equations (Al)
pletely determined.
Th~
number
‘N
and (A16)
of inverters
PN
the inverter
(N),
ratio in (26), is now determined
-3a.AL
which
}
“
(A16)
string is com-
depends
on the
by
(A.17)
(A6)
where Co and CN represent the input capacitance and the
output load capacitance of the inverter string, respectively
(Fig. 8).
As Wn >> AWn and WP>> A WP, (A6) reduces to
Wp
A
. (LnN(I+3a2)-AL
to be symmetrical:
V,n=-v=,
LP – ALP
~N
=
(A7)
“
ACKNOWLEDGMENT
With
The
Wn=(Ln–
ALn)~
b.
(A8)
❑
and given a lines
relation
between
LP=a.
Ln and LP
[1]
[2]
we find
a“Ln–
‘P” LP = ‘n ’Ln”3a”
[3]
ALP
Ln – AL.
“
wishes
to acknowledge
the helpful
made by A. T. van Zanten
suggesancl C.
Rf3FERENCES
(A9)
Ln
author
tions and contributions
Hartgring.
(A1O)
M. I. Elmasry, “ Digitaf MOS integrated circuits: A tutoriaf;’ Digital
MOS Integrated Circuils.
New York: IEEE Press pp. 4-27,
A. Kapuma, ” CMOS circuit optimization,” Solid-State Electron., vol.
26, no. 1, pp: 47-58, 1983.
C. Mead and L. Conway, Introduction
to VLSI
Systems.
New
York: pp. 12-15.
Harry J. M. Veendrick was born in Humme”lo en
Keppel, The Netherlands, on March 8, 1950; He
received the Ing. degree from the Zwolle l?olytechnicaf College, Zwolle, The Netherlands, in
of
1971, and graduated from the Department
Electronic
Engineering
at the Eindhoven University of Technology,
Eindhoven, The Netherlands,
in 1977.
.
c
PmN–l
=b” CgN and
CpuN=b. C~.
(A12)
In 1977, he joined Philips Research Laboratories, Eindhoven, The Netherlwds, where he is
working in a digital integrated-circuit
design
group on MOS circuit design.
Download