VLSI design and its technology choice. It becomes increasingly more important
..... Prentice-Hall,. 1975, ch, 2. S. M. Kang,. “A design of CMOS polycells for LSI.
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-21, NO. 5, OCTOBER 1986
Regular
Correspondence
Accurate
Simulation of Power Dissipation in VLSI Circuits
SUNG Abstract VLSI
—Power
design
important
MO KANG,
dissipation
and
its
increases.
Therefore
highfy
desirable
while circuits
SPICE. circuit.
Tfte
power
drawn from
for CMOS
Power tion.
techniques
significantly
power
“power
disturbing
the
power
technology
chip
useful will
circuit
circuits
introduce
Fig. average large source.
value
1 shows power capacitor
an
in
meeting cost,
power
circuit
other
where
and
circuit,
dissipa-
T
yet accuwithout
is inserted
in
series
with
monitoring
the
In
supply
impedance.
enough
When
not
to introduce
the series impedance
voltage
across the voltage
current
as
any
this voltage
is small,
significant
0.25 mA 0.00 mA
XIO-8
the case
corresponding
the calculated
CMOS
to the machine
cycle of the
value of lDD depends
strongly
on CX
I. For a large value of C’, the corresponding
T needs to be increased
period
proportionally
not even be detectable.
since otherwise
The results shown in Table I
that this scheme is not usable. methods
power
for
estimating
For example,
dissipation
the power the frequently
consumption
are
used formula
for
is [3]
a
cTv&f
P=
v~~
series
the average
where
CT is the total
change
in a machine
The application a moderate estimate tion
I ~D =
0.027 x10-8 O.m
based on formulas.
the steady-state
VX can be used to calculate
1.OF
in Table
V, (T) would
It appears to be a viable means so long as the capacitance is large
T
mA
0.58
10 F
is the period
as shown
METHODS for
O.35OX1O-8
and k’. (0) is set to zero,
In reality,
simulators
a simple
circuits.
IDD =
is
indicate
scheme
integrated
C, V.(T)
“;11
OIF
behavior.
often-mentioned
dissipation
I wrm THE METER IN FIG. 1 in Fig, 4 and T= 600 nS,)
at com-
as a subcircuit
IN CONVENTIONAL
is shown
c,
Others DRAWBACKS
OBTAINED
circuit
designers
device
with
CURRLNTS
measured
pay
circuit
force in the
VLSI
while area,
(The
of the major
circuits
to simulate
that can be inserted
the origirtaf
One
a technology,
are analyzed
[2]. This paper
meter”
TABLE AVERAGk
by the
designers
in CMOS
dissipation
as speed,
circuits
limited
to become a major dissipation
it is quite
while
II.
Load
to the
and apply
dissipation.
technology
such
related
[1], circuit
of technology
the power
Therefore
such as SPICE rate
results are shown
is closely
dissipation
than that in other
minimize objectives
accurately
c
reads the average
has been often
the power
is that lower
reliability. tion
circuits
speed. Even after choosing to
+7-+ Cif%it
+ T)
= Vin(t
RC
design, and also the scale of integra-
to the choice
to reduce
VLSI
Vin(t)
the power
and a parallel
C-xvr(7-)/T.
integrated
reasons for the CMOS
need
for simulating
source
Device
Input
such as
INTRODUCTION
circuit
maximum
attention
design
is
Vx(o)=o
Fig. 1. An often-proposed scheme for measuring the average current drawn from the voltage source VDD. The idea is to calculate the average power by
dissipation,in
acceptable
parable
method
dissipation
T-. –J
L
more in
Vxl
lx*
in
of devices
simulators
source. Simulation
Since the scale of integration
current
of power
across the capacitor
the supply voltage
of technology,
special
increasingly
with circuit
current
factors
circuits.
I.
choice
simulation
an accurate
voltage
becomes
as the number
are simnlated
use of a dependent
steady-state
It
dissipation
accurate
This paper presents with
choice.
the pbwer
VLSI
I#+i
SENIOR MEMBER, IEEE
is one of the most important
technology
to rednce
dissipation
889
Cx~(T) T
of this formula
of state-changing
f
mainly
Therefore,
to the accurate integrated
to disturb
Not
with
$01.00
01986
regions
along with
of power
to
identificacapacibackgate
the above formula
monitoring
NONINVASIVE
the magnitude supply
IEEE
not only
cannot
dissipation
in
circuits.
in the simulation,
the power
of even
logic gates, but also their effective in practice,
supply
circuits
because it is difficult
of CT requires
biasing
III.
by the logic
cycle frequency,
to real integrated
and drain
effects.
affected
is the machine
tances in the gate regions
large-scale
0018 -9200/86/1000-0889
capacitance
size is not practical
CT. The estimation
be applied
Manuscript received October 22, 1985. This work was supported in part by the Semiconductor Research Corporation (SRC) under Grant RSCH 84-06. 049-5. The author is with the Coordinated Science Laboratory and the Department of Electrical and Computer Engineering, University of Illinois at UrbanaChampaign, Urbana, IL 61801. IEEE Log Number 8609887.
effective cycle and
POWER METERS
of current
the subcircuit
should
drawn
from
the power
to be inserted
not have any loading
effect.
in series Such a
[EEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-21, NO. 5, OCTOBER 1986
890
idd(t)~
‘D
[;
:F!z3?jvy ------------
{
9
+“ Device
Device
Input
Vin (t) = Vin (t + T)
Fig,
A
2,
subcircuit
R C circuit
allel
Input
c Load T
a current-controlled
measuring
the
average
current power
source
and
consumption
by
a parreading
Fig. 3. A sub circuit with a voltage-controlled current source and a parallel RC circuit for measuring the average power consumption by reading VV(T),
subcircuit
A.
c Losd
*T
*
k’,(T).
dent
or Circuit
Vin(t) = Vi”(t + T)
Cir%uit
with
for
(”)’”
can be realized
by using
a current-controlled
source or a voltage-controlled
Power
Meter
with
a Current-
dependent Controlled
depen-
Current
RC
parallel
Some earlier versions
circuit.
current-controlled
source [4]. Source
2 shows a subcircuit that is composed of a currentcontrolled current source and a parallel l?C circuit. The shunt to provide a dc path. resistor R, is required in the simulation
of SPICE
can be used instead.
The series resistance
that
in i~d is negligible.
the disturbance
be modified
did not allow
sources. In such cases, the subcircuit rx should
in Fig. 3
be small
For this circuit,
so
(2) can
as
Fig.
The insertion
of O-V independent
the intrinsic of
the
in the circuit.
idd
dependent
loading.
With
thevoltage
source,
a proper
choice
average
power
as follows.
consumption
source does not chahge
due to theunidirectionality
the subcircuit
across thecapacitor
derived
voltage Also,
(7)
does not
of the current
when any R&
gain parameter~,
C,, inthesubcircuit P,vg.r The proper
impose will
value
>> T,
att=T
read the (8)
of /3 can be
First we note that which
suggests that
(1)
From
thesubcircuit
in Fig. 2
(2)
or
(9) where
i,, =iJ,,
and V,(0)=O.
Solving
for ~Y in(2)
yields IV.
(3) To compare where
a = I/I?
then
at
(4)
t= T -FL
e-”(’–’)idd).
of
) d.
(5)
[5]
inverters.
F’avg when
v..;
=
wiring
VDDCJ.
(6)
Power
Meter
with
a Voltage-Controlled
Current
VDDCV f/rK,
or
Source
Fig. 3 shows a subcircuit for monitoring the average power consumption with a voltage-controlled current source and a
composed total
we will use C,,f
fan-out done
the
gates with
the
drawn
and logic
drawn power
or
CPf/rx
to
effective only
of
capacitance wiring
are tied subcircuit
the
logic
same
in Fig.
states
type.
involved
nodes. 3 at
in
It
in individual
capacitances
in
is nontrivial
to
switching.
is specified The
and
gate is the sttm of l-pF
gate-to-substrate the
capacitance
to output
NOR gates,
V~~ source is the
the
CL of each
the totaf gates
NAND,
from
to switch
capacitance
capacitance
simulation,
of multiple-input current
required
loading
fan-outing
calculate cuit
B.
VDDCY f
The
The
three /3=
the results
modification should not cause any confusion of ~ or gm, is simply scaled down by VD~; Le.,
sum of currents gates.
~y (T)=
with
the average current. The power dissipation is simply VDD~ (T) tinder this arrangement. Fig. 4 shows a CMOS polycell
circuit
suggests that
approaches
monitor
.)
This
CIRCUITS
This
the value
instead V,(T)
the results of proposed
consumption. since
~T
TO CMOS
shown in Table I, we chose to monitor the average current from the power supply voltage instkad of the average
,,C,,. If we choose Rp and Cy such that R ,,C,, >> T
APPLICATION
SPICE
and
In
cir-
the three
simulation
was
T = 95° C, VDD= 4.5 V with
R,, = 10120, C, = 1 pF, and T= 600 nS. This choice satisfies the R ,,CP>> T. Three different values of rx, 10– 5, 10 – 6,
inequality
IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-21, NO, 5, OCTOBER 1986
891
v ~~ = 4.5
volts
idd(t) + r- I -a I Current I lL Meter ! - l-- A
V,n(t CL
t
Fig. 4.
ACMOSpolycell
circuit used tosimtdate from
TABLE AVERAGE
the average
I2x1o–4
OBTAINED IN FIG.
WITH
L
IDD = v;T)
10-7 Q 10-6 Q
O 124 mA 0.124 mA 0.124 mA
Q
drawn
I
I
I
I
I
I 1.0
I 2.0
I 3.0
I 4.0
I 5.0
3
(The measured cmw,t M shown in Fig. 4 and T= 6(XJ nS.)
10-5
currerrt
source.
11
CURRENT THF METER
the voltage
?0 –
8-
6-
4–
! 2–
1
1 VIN
3,6
2.6
1,6
0.6
o
–2 –
-4 0.0
Time
Fig, 6, SPICE output waveforms for idd and ~,. Here the steady-state value VV( T ) represents the average current drawn from the voltage source over the period 7’.
VLSI
5.
Input
voltage
waveform
used
in
SPICE
simulation
of
circuit
circuit
X1O–7
Time (s)
Fig.
choosing
Fig,
tion
10– 7 0,
average
were
current
AS
remains
which
indicates
shows
the input
corresponding
tried.
that
in
constant
Table
11,
throughout
the proposed
voltage
outputs
shown
approach
waveform
the
the
measured
three
is reliable.
cases, Fig.
used in the simulation.
5
The
are shown in Fig. 6 for both instantaneous
source
renders
accurate
end
source,
of the
results
circuit
so long
the
method
the
a current-controlled
average
by reading
period
while
By properly
gn, in the case with power
through
a
consump-
the voltage
T. This method
as demonstrated
The proposed
paper
any interference such as SPICE.
conductance
can be measured
at the
electric
transfer
current
of circuits
this paper.
causing
a simulator
gain /3 in the case with
or the
capacitor
this
without with
the transfer
current
4 circuit,
model
is simulated
voltage-controlled
and
6.0 X1O–7
(s)
across
the
is simple
yet
an example
in
can be used in any electronic
or
as the inequality
constraint
mentioned
in
is satisfied.
zdd ( ~) ~d Y(t). Note that as expected ~,(t) reaches its steadystate value at t = T, which is the average current drawn from the
VDD source
over
the period
T= 600 nS.
REFE32f3NcEs [1]
V. In this paper, average
power
we presented consumption
current-controlled junction power
with
a simple
measurement.
RC circuit This
circuit
in integrated
or voltage-controlled a parallel
[2]
REMARKS
CONCLUDING
circuits,
dependent
yielded
subcircuit
for measuring
a simple
The
the
use of
source in consubcircuit
can be inserted
into
for any
W. Keyes, “Physical limits in semiconductor electronics,” Electronics; The Continuing Evolution. Washington, DC: AAAS, 1977, pp. 197-202. L. W, Nagel and D, 0, Pederson, “Simulation program with integrated circuit emphasis (SPICE),” in Proc. 16th Jfidwe$t Symp. Circuit Theory, R.
Apr.
[3]
D:
1973. A.
Hodges
tegrated
[4]
L. 0.
Chua
Englewood
[5]
S. M. Circuits
and
Circuits. and Cliffs,
Kang, and
“A
H.
G.
New P.-M. NJ: design
Systems,
Jackson,
York: Lin,
Computer-Aided
Prentice-Hall, of CMOS vol.
A rtalysis
McGraw-Hill, 1975, polycells
and
Design
1983,
ch. 3.
A na[ysis
of Digital
of Electronic
In-
Circuits.
ch, 2. for
LSI
circuits,”
CAS-28, no. 8, pp. 838-843,
Aug.
IEEE 1981.
Tram.