Lecture 2: Power Consumption in a CMOS Circuit

Download Report

Transcript Lecture 2: Power Consumption in a CMOS Circuit

ELEC 5970-003/6970-003 (Fall 2006)
Low-Power Design of Electronic Circuits
(ELEC 5270/6270)
Power Consumption in a CMOS Circuit
Vishwani D. Agrawal
James J. Danaher Professor
Department of Electrical and Computer Engineering
Auburn University
http://www.eng.auburn.edu/~vagrawal
[email protected]
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
1
Components of Power

Dynamic

Signal transitions
Logic activity
 Glitches



Short-circuit
Static

Leakage
8/22/06 and 8/24/06
Ptotal =
Pdyn + Pstat
Ptran + Psc + Pstat
ELEC5970-003/6970-003 Lecture 2
2
Power of a Transition: Ptran
VDD
Ron
ic(t)
vi (t)
R=large
vo(t)
CL
Ground
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
3
Charging of a Capacitor
R
t=0
v(t)
i(t)
C
V
Charge on capacitor, q(t)
=
C v(t)
Current, i(t)
=
C dv(t)/dt
8/22/06 and 8/24/06
=
dq(t)/dt
ELEC5970-003/6970-003 Lecture 2
4
C dv(t)/dt =
[V – v(t)] /R
dv(t)
V – v(t)
───
=
─────
dt
RC
dv(t)
dt
∫ ───── = ∫ ────
V – v(t)
RC
-t
ln [V – v(t)]
=
── + A
RC
i(t)
=
Initial condition, t = 0, v(t) = 0 → A = ln V
-t
v(t) =
V [1 – exp(───)]
RC
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
5
v(t) =
i(t)
8/22/06 and 8/24/06
=
-t
V [1 – exp( ── )]
RC
dv(t)
C ───
dt
=
V
-t
── exp( ── )
R
RC
ELEC5970-003/6970-003 Lecture 2
6
Total Energy Per Charging
Transition from Power Supply
Etrans =
=
8/22/06 and 8/24/06
∞
∫ V i(t) dt =
0
∞ V2
-t
∫ ── exp( ── ) dt
0 R
RC
CV2
ELEC5970-003/6970-003 Lecture 2
7
Energy Dissipated per Transition in
Resistance
∞2
R ∫ i (t) dt
0
8/22/06 and 8/24/06
=
V2 ∞
-2t
R ──
∫ exp( ── ) dt
2
R 0
RC
=
1
─ CV2
2
ELEC5970-003/6970-003 Lecture 2
8
Energy Stored in Charged Capacitor
∞
∫ v(t) i(t) dt
0
8/22/06 and 8/24/06
∞
-t V
-t
= ∫ V [1-exp( ── )] ─ exp( ── ) dt
0
RC R
RC
1
= ─ CV2
2
ELEC5970-003/6970-003 Lecture 2
9
Transition Power

Gate output rising transition



Gate output falling transition



Energy dissipated in pMOS transistor = CV 2/2
Energy stored in capacitor = CV 2/2
Energy dissipated in nMOS transistor = CV 2/2
Energy dissipated per transition = CV 2/2
Power dissipation:
Ptrans =
Etrans α fck =
α fck CV2/2
α
8/22/06 and 8/24/06
=
activity factor
ELEC5970-003/6970-003 Lecture 2
10
Components of Power

Dynamic

Signal transitions
Logic activity
 Glitches



Short-circuit
Static

Leakage
8/22/06 and 8/24/06
Ptotal =
Pdyn + Pstat
Ptran + Psc + Pstat
ELEC5970-003/6970-003 Lecture 2
11
Short Circuit Current, isc(t)
VDD
VDD - VTp
Vi(t)
Volt
Vo(t)
VTn
0
Iscmaxf
isc(t)
Amp
0
8/22/06 and 8/24/06
tB
tE
ELEC5970-003/6970-003 Lecture 2
1
Time (ns)
12
Peak Short Circuit Current




Increases with the size (or gain, β) of
transistors
Decreases with load capacitance, CL
Largest when CL= 0
Reference: M. A. Ortega and J. Figueras,
“Short Circuit Power Modeling in
Submicron CMOS,” PATMOS ’96, Aug.
1996, pp. 147-166.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
13
Short-Circuit Energy per Transition
tE

Escf =∫

Escf = tf (VDD - |VTp| - VTn) Iscmaxf /2

Escr = tr (VDD - |VTp| - VTn) Iscmaxr /2

Escf = 0, when VDD = |VTp| + VTn
8/22/06 and 8/24/06
tB
VDD isc(t)dt = (tE – tB) IscmaxfVDD /2
ELEC5970-003/6970-003 Lecture 2
14
Short-Circuit Energy



Increases with rise and fall times of input
Decreases for larger output load
capacitance
Decreases and eventually becomes zero
when VDD is scaled down but the
threshold voltages are not scaled down
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
15
Short-Circuit Power Calculation



Assume equal rise and fall times
Model input-output capacitive coupling
(Miller capacitance)
Use a spice model for transistors

T. Sakurai and A. Newton, “Alpha-power Law
MOSFET model and Its Application to a CMOS
Inverter,” IEEE J. Solid State Circuits, vol. 25,
April 1990, pp. 584-594.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
16
Short Circuit Power
Psc =
8/22/06 and 8/24/06
α fck Esc
ELEC5970-003/6970-003 Lecture 2
17
Psc vs. C
0.7μ CMOS
45%
Psc/Ptotal
3ns
0%
35
8/22/06 and 8/24/06
Decreasing Input
rise time
0.5ns
C (fF)
ELEC5970-003/6970-003 Lecture 2
75
18
Psc, Rise Time and Capacitance
VDD
Ron
ic(t)+isc(t)
vo(t)
vi (t)
tf
CL
R=large
Ground
8/22/06 and 8/24/06
tr
vo(t)
───
R↑
ELEC5970-003/6970-003 Lecture 2
19
isc, Rise Time and Capacitance
Isc(t) =
8/22/06 and 8/24/06
-t
VDD[1- exp(─────)]
vo(t)
R↓tf (t)C
──── = ──────────────
R↑tf (t)
R↑tf (t)
ELEC5970-003/6970-003 Lecture 2
20
iscmax, Rise Time and Capacitance
i
Small C
vo(t)
Large C
vo(t)
1
────
R↑tf (t)
iscmax
t
tf
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
21
Psc, Rise Times, Capacitance



For given input rise and fall times short
circuit power decreases as output
capacitance increases.
Short circuit power increases with increase
of input rise and fall times.
Short circuit power is reduced if output
rise and fall times are smaller than the
input rise and fall times.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
22
Technology Scaling


Scaling down 0.7 micron by factors 2 and
4 leads to 0.35 and 0.17 micron
technologies
Constant electric field assumed
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
23
Constant Electric Field Scaling


B. Davari, R. H. Dennard and G. G.
Shahidi, “CMOS Scaling for High
Performance and Low Power—The Next
Ten Years,” Proc. IEEE, April 1995, pp.
595-606.
Other forms of scaling are referred to as
constant-voltage and quasi-constantvoltage.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
24
Bulk nMOSFET
Polysilicon
Gate
Drain
W
Source
n+
n+
L
p-type body (bulk)
SiO2
Thickness = tox
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
25
Technology Scaling



A scaling factor (S ) reduces device dimensions as
1/S.
Successive generations of technology have used a
scaling S = √2, doubling the number of transistors
per unit area. This produced 0.25μ, 0.18μ, 0.13μ,
90nm and 65nm technologies, continuing on to
45nm and 30nm.
A 5% gate shrink (S = 1.05) is commonly applied
to boost speed as the process matures.
N. H. E. Weste and D. Harris, CMOS VLSI Design, Third Edition, Boston:
Pearson Addison-Wesley, 2005, Section 4.9.1.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
26
Constant Electric Field Scaling
Device Parameter
Scaling
Length, L
1/S
Width, W
1/S
Gate oxide thickness, tox
1/S
Supply voltage, VDD
1/S
Threshold voltages, Vtn, Vtp
1/S
Substrate doping, NA
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
S
27
Constant Electric Field Scaling (Cont.)
Device Characteristic
Scaling
W / (L tox)
β
Current, Ids
β (VDD – Vt )
S
2
1/S
Resistance, R
VDD/ Ids
1
Gate capacitance, C
W L / tox
1/S
Gate delay, τ
RC
1/S
Clock frequency, f
1/ τ
S
Dynamic power per gate, P
CV 2 f
Chip area, A
1/S
2
1/S
2
Power density
P/A
1
Current density
Ids /A
S
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
28
Technology Scaling Results
L=0.17μ, C=10fF
Psc/Ptotal
70%
60%
L=0.35μ, C=20fF
37%
16%
12%
4%
1%
0.4
8/22/06 and 8/24/06
L=0.7μ, C=40fF
Input tr or tf (ns)
ELEC5970-003/6970-003 Lecture 2
1.6
29
Effects of Scaling Down





1-16% short-circuit power at 0.7 micron
4-37% at 0.35 micron
12-60% at 0.17 micron
Gate delay and rise/fall times decrease with
scaling and that prevents short-circuit power
from increasing.
Reference: S. R. Vemuru and N. Steinberg,
“Short Circuit Power Dissipation Estimation for
CMOS Logic Gates,” IEEE Trans. on Circuits and
Systems I, vol. 41, Nov. 1994, pp. 762-765.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
30
Summary: Short-Circuit Power




Short-circuit power is consumed by each
transition (increases with input transition time).
Reduction requires that gate output transition
should not be faster than the input transition
(faster gates can consume more short-circuit
power).
Increasing the output load capacitance reduces
short-circuit power.
Scaling down of supply voltage with respect to
threshold voltages reduces short-circuit power;
completely eliminated when VDD ≤ |Vtp|+Vtn .
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
31
Components of Power

Dynamic

Signal transitions
Logic activity
 Glitches



Static

8/22/06 and 8/24/06
Short-circuit
Leakage
ELEC5970-003/6970-003 Lecture 2
32
Leakage Power
IG
Ground
Gate
VDD
R
Source
Drain
n+
Bulk Si (p)
Isub
IPT
IGIDL
n+
ID
nMOS Transistor
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
33
Leakage Current Components





Subthreshold conduction, Isub
Reverse bias pn junction conduction, ID
Gate induced drain leakage, IGIDL due to
tunneling at the gate-drain overlap
Drain source punchthrough, IPT due to
short channel and high drain-source
voltage
Gate tunneling, IG through thin oxide;
may become significant with scaling
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
34
Subthreshold Current
Isub = μ0 Cox (W/L) Vt2 exp{(VGS –VTH ) / nVt }
μ0: carrier surface mobility
Cox: gate oxide capacitance per unit area
L: channel length
W: gate width
Vt = kT/q: thermal voltage
n: a technology parameter
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
35
IDS for Short Channel Device
Isub= μ0 Cox(W/L)Vt2 exp{(VGS –VTH + ηVDS)/nVt}
VDS = drain to source voltage
η: a proportionality factor
W. Nebel and J. Mermet (Editors), Low Power Design in Deep Submicron
Electronics, Springer, 1997, Section 4.1 by J. Figueras, pp. 81-104
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
36
Increased Subthreshold Leakage
Log (Drain current)
Scaled device
Ic
Isub
0 VTH’ VTH
8/22/06 and 8/24/06
Gate voltage
ELEC5970-003/6970-003 Lecture 2
37
Summary: Leakage Power




Leakage power as a fraction of the total power
increases as clock frequency drops. Turning
supply off in unused parts can save power.
For a gate it is a small fraction of the total
power; it can be significant for very large
circuits.
Scaling down features requires lowering the
threshold voltage, which increases leakage
power; roughly doubles with each shrinking.
Multiple-threshold devices are used to reduce
leakage power.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
38
A Design Example


A battery-operated 65nm digital CMOS device is found
to consume equal amounts (P ) of dynamic power and
leakage power while the short-circuit power is
negligible. The energy consumed by a computing task,
that takes T seconds, is 2PT.
Compare two power reduction strategies for extending
the battery life:
A.
B.
Clock frequency is reduced to half, keeping all other
parameters constant.
Supply voltage is reduced to half. This slows the gates down
and forces the clock frequency to be lowered to half of its
original (full voltage) value. Assume that leakage current is
held unchanged by modifying the design of transistors.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
39
A. Clock Frequency Reduction


Reducing the clock frequency will reduce
dynamic power to P / 2, keep the static
power the same as P, and double the
execution time of the task.
Energy consumption for the task will be,
Energy = (P / 2 + P ) 2T = 3PT
which is greater than the original 2PT.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
40
B. Supply Voltage Reduction


When the supply voltage and clock frequency
are reduced to half their values, dynamic power
is reduced to P / 8 and static power to P / 2.
The time of task is doubled and the total energy
consumption is,
Energy = (P / 8 + P / 2) 2T = 5PT / 4 =1.25PT
The voltage reduction strategy reduces energy
consumption while a simple frequency reduction
consumes more energy.
8/22/06 and 8/24/06
ELEC5970-003/6970-003 Lecture 2
41