Title: Pr
1Prélude
2Le Boléro de Maurice Ravel, la recette originale
dune addiction.
En 1927, quand la très célèbre danseuse Ida
Rubinstein demanda une pièce de ballet à Maurice
Ravel, celui-ci dut résoudre un formidable
problème Comment créer et retenir lattention
du public pendant 17 minutes sur une musique de
ballet? La solution de Maurice Ravel est devenue
lœuvre classique la plus jouée au monde. Parce
quelle utilise un mécanisme daddiction
implacable.
http//fr.wikipedia.org/wiki/BolC3A9ro_(Ravel)
Le tempo 2 mesures répétées 169 fois.
3La mécanique dun succès
Maurice Ravel propose dabord un thème simple
compris de tous. Avant que le public ne se
lasse, il propose une nouvelle variante, en
ajoutant un instrument. Il prend le soin de
soutenir la musique par un tempo particulièrement
explicite. Par ce tempo, le public comprend vite
quune nouvelle variante va arriver et en devine
linstant. Mieux, il apprend à lattendre. Il
lattend sans se rendre compte dailleurs que le
volume augmente. Le public suit le rythme et se
refuse inconsciemment à envisager une fin, une
fin pourtant inéluctable. Lorchestre
sépoumone, quelques sons discordants et en
quelques notes la musique séteint. 18, cest le
nombre de variantes que le public a appris à
attendre avec impatience.
4Un succès qui ne peut être eternel
Mais la recette de ce succès contient sa fin. Il
nest pas possible de répéter cette mécanique à
linfini. La musique sarrête, non pas par
manque dinspiration mais parce quil nest pas
possible daugmenter encore le volume
5La mécanique dun succès
- Maurice Ravel a utilisé une méthode simple qui se
décompose en - quelques points clés
- Proposer quelque chose de simple, facilement
compréhensible. - Anticiper un besoin de renouveau en proposant
juste à temps une variante plus attractive. - Augmenter le volume pour satisfaire loreille.
- Utiliser une cadence rigoureuse pour créer
lattente et ne pas décevoir.
6La loi de Moore ?
- La microélectronique est liée à la loi de Moore.
- Elle doit son succès à un mécanisme très
similaire au Boléro de Ravel. - Un thème simple, le circuit intégré, compris de
tous. - Un tempo implacable.
- De nouvelles fonctions offrant des variantes
toujours plus attractives. - Avec des performances toujours plus grandes.
- Et cela à un prix similaire.
La loi de Moore serait-elle en fait la loi de
Maurice ?
7Un Soupir
8Power and Energy Management ECOFAC 2010 Yves
Leduc Texas Instruments France
Challenge de la conception low powerdun point
de vue industriel
9High Level Agenda
Consumption
Heat
Waste
10Power I a Concern
Sun Black Box
- Systems Power consumption could be simply too
much. - - Data centers spend twice the money .. 100 MW
- 1 to power the servers
- 1 to cool the server room
-
- Desk-top computers need fan .. 100 W
- Fan is noisy
- Reliability issue
- Mobile phones cannot dissipate more than .. W
11Power I a Concern
- ICs Power consumption could be also too much.
- - Some package needs heat sink or fan
- ICs could run hot and cannot accept new
functionalities - Die stacking is limited by power dissipation
capability - Current density could be too high for the wiring
12Power I a Concern
- Home, office
- - Standby of household appliances
- - Low profile cheap appliances in activity
- waste our money in silence
13Energy I a Concern
- Portable systems It hurts.
- Mobile phones, portable computers, .. must be
charged regularly - Pacemakers need battery to be replaced
- Energy harvesting systems search for scarce
energy - Batteries are heavy, expensive, not reliable
-
14Energy I a Concern
- - Standby of household appliances
-
- - Low profile cheap appliances in activity
- Abuse of precious resources all around the world.
15Energy I a Concern
Last but not least, world should be green, of
course
Doc NASA
16 No Brute Force Solutions
Energy ? Batteries. Little hope for a
significant breakthrough soon. Fuel cells.
They have a small 20 efficiency. 80 heat?
Alas Energy harvesting. An ultra low power
niche Heat Dissipation ? Silicon
carbid. Running hot, the heat dissipation is
more efficient. Far to be in low cost high
volume production. Fluidics to evacuate
heat. Expensive! Limited efficiency!
Dependability? Power Consumption ? Lower
operating voltages. Still no solution to lower
voltages in the mV range without speed
degradation and good enough noise margin.
17Moores Law
We have to offer more applications to preserve
ourbusiness. As our ICs are at the limit of
the power dissipationor of the energy
consumption, we MUST reduce themto get a chance
to add a new application. There is no
alternative.
18A Realistic Solution Energy Aware Design
Power Dissipation AND/OR Energy Consumptioncould
be too much. In both cases, LIMITING the
energy consumed by the system is the
solution. It is all about EFFICIENCY.
19Tracking Efficiency Everywhere
Specifications, Constraints, Expectations, Nice
to have
20Room for Improvement Everywhere
Efficiency is a design care all along the supply
chainfrom the power plant to the expected
output. We will review a few area where
designers are at work and where innovation is
needed.
21Power Distribution
- There are a few ways to distribute
- electrical power.
- We tap power from
- Electrical outlet
- Primary / secondary battery
- Inductive coupling
- and a few others
22Electrical Outlet
AC-DC Adapter transformer or high voltage
electronics.
Wall wart power adapter
A small transformer, such as a plug-in wall
wart " power adapter commonly used for low-power
consumer electronics devices, may be as low as
20 efficient, with considerable energy loss even
when not supplying any power to the device.
Though individual losses may be only a few
watts, it has been estimated that the cumulative
loss from such transformers in US alone exceeded
32 TWh in 2002. Standby losses must be tracked
as a serial killer.
Reference and recommended reading http//www.effi
cientpowersupplies.org/pages/NRDC_power_supply_rep
ort.pdf
23 Interlude
1 W all around the year ? 1
instance 1 million instances during 1
hour 0.001 kWh 1.0 MWh during 1
day 0.024 kWh 24.0 MWh during 1
year 8.8 kWh 8.8 GWh 32 TWh in US? for 1
person 110 kWh 12 W per person all around the
year ltltltlt LOSSES
cumulative consumption home, office, ..
24The Importance of the Scale Effect
12 W all around the year ? with 1 kW at 0.1
i.e a bill of 110 kWh each year per person
1 instance 300 million instances during
1 year 11.0 3.3 G ltltltlt LOSSES
It does not hurt too much ?
This is a good driver
25Standby and Idle Mode Power Consumption
Standby is activated when an equipment is turned
off but the user wants a quick start. In analog,
there is no particular difficulty at the
exception. For example, of the potential risk of
a popping noise in earphones or loudspeakers. In
digital, there is a cost associated to save
machine states in appropriate registers the
states. In software, disk drive or non volatile
memories are used to make a snapshot of the
system. This potential source of subtle bugs
cannot be underestimated. Systems may be in
standby mode all around the year with a power
consumption ideally at zero. Idle mode. The
system must insure a quick wake-up. This mode is
much more complex to manage in mobile equipment
where significant energy savings are expected. A
part of the system, as small as possible, remains
active and is responsible for the quick wake-up.
State machines managing the idle mode are often
very complex, desing of the parts remaining
active are challenging.
26Standby Power Consumption
Regulation authorities are driving the power
reduction
There is no severe technical obstacle to power
reduction but there is a COST BARRIER to
smash. All products should be designed to
minimize the standby power consumption. Regulation
authorities have triggered the momentum. Now
that this is started,the effort is in the
development of low cost techniques. Forgetting
the standby power consumption is a BUSINESS
KILLER. Customers are becoming sensitive to this
specification, although they are not willing to
pay more. Being the best on this aspect is
becoming smoothly a strong BUSINESS
DIFFERENTIATION.
27Idle Power Consumption
Mobile applications are the most active drivers
All techniques to reduce losses in a portable
device are rapidly implemented. There is a
significant RD effort world-wide on this
topic. Leakages in advanced IC technologies
impact significantly idle power consumption. This
is a formidable challenge that industry is
fighting actively on a daily basis.
28Battery Charger and Battery Management
It is all about battery electrochemistry,
metrology and energy management
Efficiency charger itself is not the main
concern as it is plugged to an external
source. Precision and Metrology how to estimate
precisely the remaining energy in an aging
battery idle mode consumption is a severe
requirement (battery gauge..) Safety how many
are forgetting their charger on an electrical
outlet? Integration how to mix power electronics
with digital and analog applications
29A Typical Example the USB Battery Charger
There are a lot of reasons to prefer
ONE single good charger to a set of cheap
solutions. The battery may
be also charged from any
USB master. For a design point of view, the
USB link is a source of headaches to maintain a
good conversion efficiency. But the USB is also
an ideal data link to exchange precious
information back and forth!
30Inductive Coupling
Efficiency is an obvious design care. New and
very exciting developments to come!
31The Root Causes of the Power Consumption
- The application !
- Search for the right algorithms in the right
architecture! - The active power.
- ? CV2f
-
- ? represents the activity of the IC.
Capacitance C is correlated to the complexity.
Higher speed (f) is requiring higher voltage
(V2). - The static power.
- Leakages
- Leakages are larger in advanced process which are
used for the - largest applications.
32Operating Voltage
- Operating voltage is an important factor in the
power consumption I - Limit overdesign !
- Clean supplies power AND ground
- Use efficient voltage regulators
-
33Overdesign
The root causes of overdesign are related to a
poor analysis of themarket or of the customer
requirements, on a loosely defined product or a
loose design. ? The system is targeting too
many applications. We have to verify that the
most demanding targets are realistic. ?
A system is often overspecified to compensate
for the weakness of the analysis. ?
Designers are tempted to build too larger
design margins Overdesign has a MAJOR impact to
the size and the power consumption of IC modules.
Design margins are necessary but should be
justified by solid statistical analysis.
SO EASY TO UNDERESTIMATE
34Limit Power Supply Voltage
FMAX depends on the gate delay which depends
directly on the current in the CMOS gates which
depends on the supply voltage. Choose
thesmallest operating voltage suited to reach
specified performances. CMOS gate
delay may be approximated in many ways. This is
the Taur-Nin model
Y. Taur and T. H. Ning, Fundamentals of
Modern VLSI Devices. Cambridge University Press,
1998.
35Power Supply Noise Impact on ICs Performances
Power supply noise limits the IC maximum
operating frequency. It is important to stress
that,despite a common belief, externalbypass
capacitors are not effectiveon all the supply
noise bandwidth. This is a deadly
MISTAKE. Package is blocking high
frequencies.Supplies must be decoupled
internallyusing integrated bypass
capacitors. Another belief is that a larger
bypasscapacitance provides automatically a
better decoupling. The powerdistribution network
is a complex RLC structure and is prone to
resonances.
36Clean Supplies
Bad supplies cause IR drop, therefore losses but
also produces ringing. Both have to be
compensated by larger supply voltages, an
additional power consumption. The art or science
of decoupling the supplies was the distinctive
characteristic of the good electronic
engineer. In the microelectronics era, it was
stated that this is the reserved domain of the
PCB design. It WAS indeed true in the TTL era.
This is not true anymore.
Each power supply distribution strip is
bypassed with a 100 nF capacitor at the top and
bottom. This should provide sufficient bypassing,
but in case of suspected problems, you may want
to place a 100 nF bypass capacitor at each TTL
package.
37(No Transcript)
38Impedance of a Power Supply Network
Power Supply ImpedanceNormalized Magnitude
NB Noise bandwidth depends on the slopes
of signals BWnoise ? 0.35 / slope
No Bypass Caps
1 Bypass Cap on PCB
10
An example of rough hierarchical bypassing
Rtarget
1
TTL era
0.1
100KHz
1MHz
10MHz
100MHz
1GHz
10KHz
Frequency
MHz
tens of MHz
GHz
39Bypassing techniques
Bypassing techniques are more sophisticated that
what we have presented. This is a science.
Bypassing strategy, modeling and simulations are
possible. But as EDA tools are lagging, brute
force simulations are overflowed by the
complexity of the designs. This is therefore also
an art. Good decoupling not only guarantees a
correct functionality but increases the speed
that a circuit is able to reach without
increasing the supply voltage! This is a
distinctive area for differentiation and
therefore a confidential topic.
40Voltage Converter and Regulation Power
management is a system in the system. With
supply switches, Voltage converters and
regulators act as the actuators of the systems
power management. DC-DC converters and linear
voltage regulators are the typical
solutions. Currently, at the exception of low
power ICs, the vastmajority of the voltage
regulation modules are notperfectly integrated
as a few external devices still resist to a low
cost integration. We have to distinguish the
voltage converter which has the main function to
convert the voltage with a minimum of losses and
the voltage regulator, which has to regulate the
output voltage. Both are expected to minimize
losses as far as possible.
41Linear Voltage Regulators
The voltage regulator is a kind of bidirectional
filter. Low frequency current noise for instance
is kicked back to the input. Therefore, linear
voltage regulators must be included in the
modeling of the power supply network. Models
should include their non linear behavior!
42Voltage Converters
- Two major categories
- Step-down and step-up DC-DC converters using LC
tanks - Switched Capacitors DC-DC converters (e.g.
voltage doubler,) - The energy density in LC elements is usually
- not compatible with integration of LC tanks.
- Integrated solutions are now proposed in somelow
power mobile applications. - Switched capacitors converters are limited to
smaller power.
43DC-DC Converters using LC Tanks
One example of Step-up converter (aka Boost
Converter) Tip the hydraulic ram and boost
converter are based on a similar idea.
Hydraulic Ram
In hydraulic analogy, there is no inductor. The
water has a significant mass which could store
kinetic energy. In electricity the mass of
the electron is not significant. A dedicated
inductor stores thekinetic energy of
the electrical current.
44Hydraulic Ram
DC-DC Boost Converter
45DC-DC Converters
An example of a step-down step-up converter
aka Buck-Boost or SEPIC
As DC-DC Converters are based on an energy
exchange between the potential energy stored in
the capacitor and the kinetic energy stored in
the inductor, there is no dissipation but the
dissipation due to parasitic resistances, control
circuitry and switching scheme inaccuracies. Effi
ciency is quite good excepted in low current
regime. DC-DC converters are among the important
modules which are profiting from a healthy RD
work of labs and industry.
46The Power Distribution Network aka PDN
From external supply (battery, ) to the internal
modules, all the chain must be carefully
designed, modeled and simulated. It is
impossible to simulate the system at the
transistor level. The PDN is including highly non
linear modules and resists to accuratelinear
approximation. It is very difficult to realize a
good decoupling to limit noise at an acceptable
level. For the power consumption estimation, the
modeling and simulation are quitesimple. The
major issue to get the best compromise between a
simple butinefficient architecture and a costly
more efficient solution. System partition, IC
processes, Power Management, PCB complexity
and cost are the major factors to consider. A
good PDN is the 1st condition to keep power
consumption under control!
47The PDN Consumers
From specifications, IC processes and current
know-how, the voltage of each modules is chosen
to the best values. Analog modules are powered
by a dedicated network. Supply noise is the major
concern. They are traditionally considered as the
victims althoughsome class D amplifiers for
example could be considered as aggressors.Analo
g modules are traditionally powered on/off by
control signals. Fast digital I/O (like DDR3)
are powered by another dedicated network. They
are truly aggressors for all other modules.
Decoupling is needed! Digital modules are
powered by a set of dedicated networks. Each
networkmay be switched on or off to limit power
consumption despite high transistorleakages.
Voltages may be chosen separately. For SRAM
modules in retentionmode, the voltage may be
lowered to keep the information stored in the
array while limiting the power consumption. Inter
nal Busses and Clocks. Heavily loaded, they are
significant contributorsto the power consumption
of a digital IC.
48The Power Down Issue
- IC modules may be tricky as the power consumption
could dependon the value of external voltages or
on external loads. - There are a few mistakes to avoid.
- Internal floating nodes.
- I/O nodes. Loosely defined system may connect 2
ICs in a wrong way.
?
Hi Z
Hi Z
OK
49A Power Up Issue
The IC, of course, has a global reset. Hardware
or Software? Lets imagine that a system
engineer decides that a software reset is
enough. When supplies are activated, the IC is
waiting for a software reset. So far so good...
The IC is powered on, but its internal bus is
idle. The software resetis supposed to be
transmitted through the bus to position a control
bitsomewhere. This control bit will eventually
reset the bus? Perhaps not
Supply is OVERLOADED !!!
50Soft Floating Node the Hidden Killer of Analog
ICs
51Power Management in Large ICs
- In large ICs like microprocessors, DSPs or SoCs,
the power managementrepresents a significant
design effort. Power management is often shared - by dedicated hardware, embedded software and
application. - There are at least 4 operating modes
- active mode
- idle mode
- standby mode
- off mode
- There are several techniques used together to
limit the power consumption - in each mode. Statistics modeling and simulation
are worth to be used.
52Active Mode
For the thermal management, this mode is of
course the most demanding. Active mode drains
the maximum current from the supplies. Many
techniques are used to keep the power consumption
to a minimum. These techniques will be
described later
53Idle Mode
- Parts of the system are stopped but their
information is retained in a fast - access memory (SRAM, Flash, ..). System is
reactivated from time totime to refresh
synchronization with external links. The system
is - supposed to wake up extremely quickly.
- This mode could be the major contributor to the
energy consumptionof the system for 3 reasons - This system is mainly operating in idle mode
(e.g. phones) - Real time clock and counters are running.
- - Process leakage.
- A good design of the system architecture is key
to keep the powerconsumption to a reasonable
value.
54Standby Mode
The system (or part of a system) is stopped but
its informationis retained in a fast access
memory (SRAM, Flash, ..). The system needs some
time to start-up to resynchronize toexternal
links like RF, ADSL All combinatorial logic is
switched off or at least the clocks arestopped.
SRAM are in a retention mode with a supply
maintainedat a minimal retention voltage. There
is no activity. The system is waiting for a
software or a hardware on/off command. Leakage
control is the main issue. It will be
described later
55Off Mode
Although the system is still physically connected
to theexternal supply, it is not supposed to
drain any significant current.Previous state of
the system may be stored in some non
volatilememory (disk drive, flash
memory..). There is no activity. The system is
waiting for a hardware on/offcommand. Floating
nodes are the traditional killers of the Off Mode
esp. inmixed signal modules.
56Reducing Active Power
P ? CV2 f leakages
- The product ? f corresponds to the activity of
the system. - Architecture and algorithms must be tuned.
- Supply voltage V is obviously important. Limit
overdesign at least! - C is the capacitance of active wires, 2
solutions - Process scaling More Moore
- 3D-IC More than Moore
- In the recent CMOS processes, leakages cannot be
ignored anymore!
57High Importance of Algorithms and Architecture
There are more expectations to reduce active
power by selecting at high level the best
algorithms or to use the best hardware
solutions. HARDWARE or SOFTWARE? Application
specific hardware is always much More efficient.
But it is obviously restricted tolow level
designs by the lack of reconfigurability. FPGA
offers some of the expected reconfigurability but
performances and costare deceiving. It
exacerbates THE problem of 2D circuits the
weakness of theconnection system. Software
solutions are very poor performers in speed and
power efficiency.Their reconfigurability makes
them nevertheless a successful choice.
58Reducing Active Power by Parallelization
The major contributor is the system architecture.
This is where efforts should be directed
first. A few hints system architecture,
algorithm, data move Parallel architecture
should be preferred, although there is not yet
really satisfactoryparallelization scheme of
serial algorithms
Example what is the more efficient bus?
or
59Prefer Parallel Solutions, a Simple Example
What is the more efficient bus?
Active power Consumption is proportional to
transitions C V2 C of 1 wire !
Active power lower Static power higher
60Parallel Solutions ? Not so Simple !
- Some applications are obviously parallel. For
cost pressure, they havebeen painfully
serialized making real-time software pathetically
complex. - The power reduction targets are now pushing to
parallel processing. - Data transfer, memory, are making parallel
processing very complex. - Cost is another major barrier to
parallelization. - Many algorithms are serial by nature.
Parallelization is more than challenging.
Breakthroughs are still regularly announced - Multicore processors are replacing single core
processor.
61Reducing Active Power Process Scaling
rchannel (e/eox) l / w cgate
(eox/e) w l
w w / ? l l / ? e e
This is not as good for the connections
62Reducing Active Power 3D-IC More than Moore
2D - IC
Shorter wires Lower capacitances
3D - IC
Die 3
gt1000s Through Silicon Vias / mm2
It is all about systempartitioning,
architecture,business model, cost
One Chip
Die 2
Die 1
63Reducing Active Power Data Transfer
LIMIT the transfer of the data ! Significant
improvements could be obtained during the design
of the high level architecture! SoCs may take
profit of differentiated structures. This is a
hot subject although a significant breakthrough
is still to be found! One example among MANY
proposals Time Triggered Architecture aka
TTA
http//en.wikipedia.org/wiki/Transport_triggered_a
rchitecture
64Data Transfer, Minimize Voltage
- It is important to choose efficient solutions.
Overdesign should be avoided! - Do not underestimate capacitive coupling between
wires! -
- High-Level Interconnect Delay and Power
Estimation A. Courtay, O. Sentieys, J. Laurent,
and N.Julien - Journal of Low Power Electronics Vol. 4, 2133,
2008
65Data Transfer, Be Smart
- To reduce power, adaptive techniques may also be
applied, e.g.
CV2f
Always the right supply. Never higher! But this
is costly
66Reducing Active Power the Clocks
- Clocks are the fastest, the busiest and the most
loaded signals of a system. - Gated clocks are now well supported by EDA tools
and do not representa design challenge anymore
clocks are distributed only to modules in
activity. - Clock rates are chosen statically or dynamically
to respond to the desiredthroughput of the IC
modules. Clock domains and voltage domains
partitionsystem and IC. - In some applications, asynchronous logic is
preferred. Clock is then used only - as time keeper.
- Real time clock is a low speed and precise clock
used for time keeping. Veryoften, the power
consumption of a real time clock is reduced to a
bare minimum. - Real time clock may be used in idle mode as
system clock. Jitter is a killer.
67Reducing Active Power DVFS
- Dynamic Voltage and Frequency Scaling, aka DVFS
is the current solution to - put the active power consumption of an IC under
control! - There are a few reasons to modulate the speed of
an application. - Under hardware control
- Limit power consumption by running at the
smallest supply voltage - Keep the operating temperature of an IC under
control -
- Under software control
- Limit power consumption by running an
application at the right frequency
68Adaptive Dynamic Voltage and Frequency Scaling
Under the control of the hardware, the voltage
may be tuned finely, with a relatively fine
granularity, to maintain the operation with a
minimal supply. This is a dynamic calibration of
the performances instead to the
traditionalcalibration of the supply voltage to
the specified value. Design is done for
typicalprocess parameters with Adaptive DVFS
insuring the functionality at the
process corners. Design is more aggressive, sizes
are smaller, parasitic capacitances are smaller,
great factor diminishing the power
consumption. On a mobile terminal for instance,
the power consumption is a concern per se.All
tasks do not need to run at the same speed. It is
therefore interesting tolimit the operating
frequency to the minimum and reduce the supply
voltageaccordingly. The granularity of Adaptive
DVFS may be as small as a module. On a portable
computer, the processors may be heavily loaded
andtemperature could reach critical limits.
Slowing down the core and diminishingthe supply
is an efficient way to bring back the IC to a
reasonable temperature. DVFS and Adaptive DVFS
are adding more complexity as modules may
runasynchronously. Under the control of the
application software they are a potential source
of bugs.
69Reducing Static Power Voltage Supply?
To reduce the static power consumption, it is
tempting to reduce the supply voltage
Vsupply reduction ?
Reduce ILeakage !!
70Reducing Static Power Reduce leakage!
- Off mode the ultimate solution is the power
switch. - In Standby or Idle mode where data retention is
mandatory, 3 solutions - Use high threshold voltage devices where they are
not impacting critical paths. - Use low supply voltage in retention mode (in
SRAM, ) - Use dynamic threshold voltage control by back
gate biasing (body effect)
Vbackgate
well
71Leakages in Nanometric Technologies
- Large Variability!
- High LEAKAGES
- Highly UNPREDICTABLE ?
Device Simulation of Random Dopant Effects
inUltra-small MOSFETs Based on Advanced Physical
ModelsToriyama, S. Hagishima, D. Matsuzawa,
K. Sano, N. Simulation of Semiconductor
Processes and Devices, International Conference
on Physical Models, Sept. 2006
72New Transistors ?
1 2 3 4
LETI
5
Hot activity on new concepts! FD SOI FET 1,
Double gate FET, FinFET 2, Junctionless
Nanowire Transistor 3, Graphene Transistor 4,
Carbon Nanotube 5 With a major target
Improving the ratio ( Ion/Ioff ) None of them
are reducing leakages to zero
Freescale
Tyndal
MIT
73Lets Face It. Leakage is a Reality.
74Conclusion, No Simple and Definitive Solution
We can safely state that there is no single
solution to power consumption reduction. A
continuous effort is required. Power
consumption is the other problem of the XXI
century.
Way to success is narrow..
75Thank you for your attention !