lp trend08 godwin
TRANSCRIPT
Predictable Success
Low Power Trends and
Methodology
Godwin Maben
© 2008 Synopsys, Inc. (2)
Predictable Success
• Introduction
• Summary of Low Power Techniques
Designer’s Arsenal
• Details on each Technique
Technology Highlights
Advantages/Trade-Offs
Challenges
• Synopsys Low Power Flow Summary
Agenda
© 2008 Synopsys, Inc. (3)
Predictable Success
Reasons for Low Power Demand
It’s all about
battery life
250nm180nm130nm 90nm 65nm 45nm
Leakage Power
Dynamic Power
Sourc
e: In
tel
© 2008 Synopsys, Inc. (4)
Predictable Success
Reasons for Low Power Demand
13%
22%
41%
48%
0%
25%
50%
75%
Consumer Communications Computer/
Peripherals
Other
2004 2006 2007
1% 1%
30%
10%
1%
10%
26%
21%
0%
20%
40%
60%
>250nm 250nm 180nm 130nm 90nm 65nm 45nm 32nm
Last Current Next
Application Driven Technology Driven
Source: SNUG 2007
© 2008 Synopsys, Inc. (5)
Predictable Success
Summary of Low Power Techniques
(Designer’s Arsenal)
© 2008 Synopsys, Inc. (6)
Predictable Success
Factors Governing Power
Minimize Ileak by:
Reducing operating voltage
Fewer leaking transistors
Minimize Iswitch by:
Reducing operating voltage
Less switching cap
Less switching activity
Total Power
Dissipation
Dynamic PowerDissipation
Static PowerDissipation
Ileak
Iswitch
dtfCVIVE cDDleakDD
t
)( 2
0
t
leakDD dtIV0
Total Power
Dissipation
Dynamic Power
Dissipation
Static Power
Dissipation t
cDD dtfCV0
2
Ileak
Iswitch
© 2008 Synopsys, Inc. (7)
Predictable Success
Basic Low Power Techniques
Summary
f
d
b
b
a
c
a
c
c
d
b
16 bit 64 bit
Ripple 90 366
CLA 100 405
Carry
Skip108 437
Carry
Select161 711
Carry
Save218 1323
FF
Q
EN
CLK
D
FF
Q
EN
LT
ICG
LT
ICG
LT
ICGCLK
D
Clock Gating
-
Delay
Le
ak
ag
e C
urr
en
t
Low VTH
Nominal VTH
High VTH
-
Delay
Le
ak
ag
e C
urr
en
t
Low VTH
Nominal VTH
High VTH
Multi-Threshold
© 2008 Synopsys, Inc. (8)
Predictable Success
Advanced Low Power Techniques
Summary
Non minimum size gate lengths VTCMOS
A Z
Stack Effect
Power Gating
A Z
VDDB
VSSB
nSLEEP
Virtual VDD
Virtual VSS
SLEEP
0.8V0.7V
Multi-Voltage (MV)
1.0V
© 2008 Synopsys, Inc. (9)
Predictable Success
Details on each Technique
• Clock Gating
Basic Understanding
Advantages
Concerns in the flow
© 2008 Synopsys, Inc. (10)
Predictable Success
Understanding Clock Gating
CLK
D
ENFF
Q
CLKCLK
D
EN FF
Q
FF
Q
EN
CLK
D
FF
Q
EN
LT
ICG
LTLT
CLK
D
© 2008 Synopsys, Inc. (11)
Predictable Success
• Enable Timing
• Insertion delay increase
Impact on OCV
• Peak Power during ON/OFF of clock
IR drop as well
• Verification Impact
Clock Gating Flow Challenges
© 2008 Synopsys, Inc. (12)
Predictable Success
Managing Enable Timing
© 2008 Synopsys, Inc. (13)
Predictable Success
Clock Gating Impact on Verification
• RTL functional Simulation
None
• Gate level Simulation
FF
Q
EN
CLK
D
FFEN
LT
ICG
LT
ICG
LT
ICG
DReset
“0/1”
“X”
“0/1”
FF
© 2008 Synopsys, Inc. (14)
Predictable Success
Details on each Technique
• Multi-Threshold Leakage Optimization
Basic Understanding
Advantages
Concerns in the flow
© 2008 Synopsys, Inc. (15)
Predictable Success
• Multiple Threshold Cells
Understanding Multi Threshold
16ps
Vdd 0.9v
Vth 0.3v
10ps
Vdd 0.9v
Vth 0.22v
© 2008 Synopsys, Inc. (16)
Predictable Success
Multi-Threshold Flow Challenges
• Looking at a different view of the problem, optimizations push towards zero-slack
• High-Vt cells tend to be weaker and can be more susceptible to variability
+ve slack-ve slack
Power critical
range
Slack
© 2008 Synopsys, Inc. (17)
Predictable Success
Multi-Threshold Flow Challenges
• Implant Spacing
Violations
• Min Width Violations
• Chip Finishing
requires Proper filler
Cell Insertion
© 2008 Synopsys, Inc. (18)
Predictable Success
Details on each Technique
• Power Gating
Basic Understanding
Advantages
Concerns in the flow
© 2008 Synopsys, Inc. (19)
Predictable Success
Understanding Power Gating
VDDG
Gating
Control
Switch Fabric
Switch Fabric
Sw
itch
Fab
ric
Power Gated
Block
Always On
Block
VSS
ISO
© 2008 Synopsys, Inc. (20)
Predictable Success
Architectural Tradeoffs
• Amount of leakage power savings that is
possible
• Entry and exit time penalties incurred
• Energy dissipated entering and leaving such
leakage saving modes
© 2008 Synopsys, Inc. (21)
Predictable Success
Activity Profile for Power Gating
WAKESLEEPSLEEP WAKE SLEEP
Dynamic Power
Activity “X”
Leakage Power
Activity “X”
Leakage Power
Power Gated
Dynamic Power
Activity “Y”
Leakage Power
Activity “Y”
Dynamic Power
Activity “Z”
Leakage Power
Activity “Z”Leakage Power
Power Gated
TIME
Po
wer
© 2008 Synopsys, Inc. (22)
Predictable Success
Realistic Profile for Power Gating
WAKESLEEPSLEEP WAKE SLEEP
Dynamic Power
Activity “X”
Leakage Power
Activity “X”
Leakage Power
Power Gated
Dynamic Power
Activity “Y”
Leakage Power
Activity “Y”
Dynamic Power
Activity “Z”
Leakage Power
Activity “Z”Leakage Power
Power Gated
Po
wer
TIME
© 2008 Synopsys, Inc. (23)
Predictable Success
Architectural Tradeoffs
• Gate VDD or
VSS?
• Retention
Mechanism
© 2008 Synopsys, Inc. (24)
Predictable Success
• Functional Verification
Power Gating Flow ChallengesClock wiggling while
power-gated
Island Power-gated
(Z)
Save
Restore
RETENTION Rail
Registers RESTORED
Rail ramp-up
Rail with Standby
value (0.6 volts)Rail ramp-down
© 2008 Synopsys, Inc. (25)
Predictable Success
Power Gating Sequence/Coverage
Voltage States: 8/8
Logical States: 2/3
Is the wakeup/shutdown
sequence correct ?
© 2008 Synopsys, Inc. (26)
Predictable Success
VDD
on/off VDD_BACKUP
Shut
Down
RR
CP
D
SI
SE
LDRS
Q
•Logic Synthesis
Power Gating Flow Challenges
GATE
VSS
OFF VDD
Active
Logic
Active
Logic
X ISOEN
save
restore
Insert Isolation Cell Insert Retention RegisterHookup Retention Register
© 2008 Synopsys, Inc. (27)
Predictable Success
Power Planning the Switches
MAXIR
(mV)
Area
(%) Resistance
# PWR
SWITCHE
S
X_INC
R
Y_INC
R LIB_CEL_NAME
37.2 0.4 650.0 328.0 26.3 3.6 SXXZZYX8L
42.7 0.3 1300.0 328.0 26.3 3.6 SXXZZYX4L
43.6 0.2 650.0 164.0 26.3 7.2 SXXZZYX8L
53.2 0.2 2600.0 328.0 26.3 3.6 SXXZZYX2L
55.3 0.2 1300.0 164.0 26.3 7.2 SXXZZYX4L
72.5 0.2 650.0 164.0 52.6 3.6 SXXZZYX8L
53.2
39.9
© 2008 Synopsys, Inc. (28)
Predictable Success
NSLEEP
VDD
VVDD
t0 t1 t2
Power-up Normal - Power-on
NSLEEP
VDD
VVDDPower-up Normal - Power-on
IR-Drop Management
IN
VSS
NSLEEP
VDD
VVDD
NOT_IN
Limit the voltage
drop across the switch
TimeT
T- T+
Volts
Turn On All Switches at Once
Sequentially Turn On Switches
© 2008 Synopsys, Inc. (29)
Predictable Success
Details on each Technique
• Multi-Voltage
Basic Understanding
Advantages
Concerns in the flow
© 2008 Synopsys, Inc. (30)
Predictable Success
• Static Voltage
• Multi-Level Voltage Scaling
• Dynamic Voltage and Frequency Scaling
• Adaptive Voltage Scaling
Multi-Voltage Design Style
Mode
Control
Monitor Monitor
Monitor
Voltage Island Voltage Island
A B
C
Voltage
Regulators
Voltage Island
Voltage Island Voltage Island
A
1.2 V, 350 MHz
Voltage Island
B
1.0 V, 250 MHz
C
1.5 V, 500 MHz
Mode
Control
Voltage Island Voltage Island
A B
C
Voltage
Regulators
Voltage Island
Pro
gra
mm
able
© 2008 Synopsys, Inc. (31)
Predictable Success
Understanding Multi-Voltage/Scaling
• Avoid completing tasks early – energy is wasted
• Use only enough energy required to complete task in time
Time
VoltageReduce
Voltage
Reduce
Voltage
Reduce
Voltage
Task 1 Task 2 Task 3Idle
Energy
Energy
Saved
Energy
Run Task Slow
as Possible
Run Task in
Available Time
© 2008 Synopsys, Inc. (32)
Predictable Success
• Voltage/Frequency are not changed at runtime
Change of V/F is defined either through s/w or h/w
Voltage/Frequency changed in a specific direction based on the “amount of change”
Wait till Voltage/Frequency are locked
Designs at run time will operate on locked Frequency/Voltage
• Identify the various operating voltages under which blocks will be operating at any point of time
• Do we have libraries characterized at these V points?
Voltage Scaling Facts
Control
Unit(CU)
requests VCU to
ramp up Voltage
VCU Interrupts
CU when Target
Voltage Reached
CU programs clock
generator to maximum
Frequency
CU requests
VCU to ramp
down voltage
and reprograms
clock generator
to lower
Frequency
0V
1.08V
0.7
6V
© 2008 Synopsys, Inc. (33)
Predictable Success
Voltage Scaling Flow Challenges
When Vsrc is between VIH
And VIL propagate “X”
Vsrc1.0v
0.6v
0.0v
HiX ZSIM
Output
• Accurately represent effects of dynamic change
in voltage values during simulation
Logic0
X
Logic1
Vdst
Vsrc
1.0V
0.0V
0.6VVIH
VIL
© 2008 Synopsys, Inc. (34)
Predictable Success
Voltage Scaling Flow Challenges• Synthesis
Multi-Mode Optimization
Optimal Voltage for Timing/Area/Power Optimization
Level Shifter Optimization based on Mode
• Power Planning
Optimal Power Grid
• Choose the appropriate Power Numbers
© 2008 Synopsys, Inc. (35)
Predictable Success
Voltage Scaling Flow Challenges
• Placement, CTS and Route
Minimize Skew
MCMM
Placement of Level Shifters
Temperature Inversion
0
1000
2000
3000
3.5 3.8 3.9 4 4.2 4.4 4.5
# Fl
op
s
Large Skew Across Domains
0
1000
2000
3000
3.5 3.9 4.1 4.4 4.7
# Fl
op
s
Reduced Skew Across Domains
VDDMax
VDDMin
VDDMax
Mode1
Mode2
Mode3
© 2008 Synopsys, Inc. (36)
Predictable Success
Voltage Scaling Timing Challenges
• Functional or Performance Failures
DFF
D Q
CLK
DFF
D Q
CLK
Combinatorial
DFF
D Q
CLK
DFF
D Q
CLK
VDD=1.0 VDD=1.0 VDD=1.0 VDD=1.0
VDD=1.0
VDD=1.0VDD=1.0
D
CLK
Set-up Time Met
© 2008 Synopsys, Inc. (37)
Predictable Success
Voltage Scaling Timing Challenges
• Functional or Performance Failures
Set-up Time
Failure
DFF
D Q
CLK
DFF
D Q
CLK
Combinatorial
DFF
D Q
CLK
DFF
D Q
CLK
VDD=1.2 VDD0.95 VDD=0.99 VDD=0.91
VDD=1.0
VDD=0.99VDD=0.98
D
CLK
© 2008 Synopsys, Inc. (38)
Predictable Success
Details on each Technique
• Back Biasing
Basic Understanding
Advantages
Concerns in the flow
© 2008 Synopsys, Inc. (39)
Predictable Success
Understanding Biasing
Short Stop
Body Bias (VTCMOS)
Reduced Leakage
Long Stop
MTCMOS
Zero Leakage
Shut-Down Wake-Up
© 2008 Synopsys, Inc. (40)
Predictable Success
Reducing Leakage using Well Bias
• When VPW < Source
Vth increases
ID decreases
Leakage decreases
FILLBIAS
VDD
VSS
VPW
VNW
Connect N-Well to
VDD
Connect P-Well to
VSS – 0.3V
Source DrainGate
ID
p-well
n+ n+
VPW
Source
Drain
Gate VPW ID
© 2008 Synopsys, Inc. (41)
Predictable Success
Well Biasing Flow Challenges
• Impact on Area due to extra Pick-Up cells
• Multi-Corner Analysis required for Sign-Off
VBB = 1V, VBB = 2V…etc
• Sleep to Transition Power Overhead
• LVS complexity
• Well RC extraction
© 2008 Synopsys, Inc. (42)
Predictable Success
Power Savings Data
Technique Used Technology % Leakage
Savings
% Total Power
Savings
Clock gating 90nm NA 57%
Multi-Vt 90nm 31% NA
Power Gating 130nm 25X 13x
Power Gating 65nm 2x NA
DVFS 45nm 98% during
Standby
NA
Predictable Success
Thoughts on Efficient Leakage Power Management
© 2008 Synopsys, Inc. (44)
Predictable Success
Source:IBM
Efficient Leakage Power
Management
VD
D
VS
S
VDD
VSS VIRTUAL_VSS
VVSS
VDD
Gate Voltage
Gate Voltage
Source:IBM
© 2008 Synopsys, Inc. (45)
Predictable Success
Voltage Generator
Multiple Power Mode Implementation
VD
D
VS
S
VDD
VSS VIRTUAL_VSS
VVSS
VDD
Gate Voltage
S0 S1 S2
S0 S1 S2
1 0 0 Snore
0 1 0 Dream
0 0 1 Sleep
0 1 1 Active
Source:I
BM
© 2008 Synopsys, Inc. (46)
Predictable Success
Activity Profile of Multiple Sleep Modes
Sleep
Snore
Shut-DownWake-Up
DreamDream
Active
© 2008 Synopsys, Inc. (47)
Predictable Success
Recycle Energy - Hybrid Car
Mo
tor
Ge
ne
rato
r
Battery
© 2008 Synopsys, Inc. (48)
Predictable Success
Recycle Energy - Hybrid Car
Mo
tor
Ge
ne
rato
r
Battery
© 2008 Synopsys, Inc. (49)
Predictable Success
Recycle Energy - Hybrid Car
Mo
tor
Ge
ne
rato
r
Battery
© 2008 Synopsys, Inc. (50)
Predictable Success
Recycle Energy - Hybrid Car
Mo
tor
Ge
ne
rato
r
Battery
© 2008 Synopsys, Inc. (51)
Predictable Success
Recycle Leakage
Full Chip
Standard
Cell Area
VDD
GND
© 2008 Synopsys, Inc. (52)
Predictable Success
Recycle Leakage
Full Chip
Standard
Cell Area
VDD
TEMP_GND
Charge
Collector
GND
© 2008 Synopsys, Inc. (53)
Predictable Success
Recycle Leakage
Full Chip
Standard
Cell Area
VDD
TEMP_GND
Charge
Collector
GND
VDD_SPARE
© 2008 Synopsys, Inc. (54)
Predictable Success
RTL
Synthesis
Simulation
Vector
Analysis
TB
Vector
Good?
No
Power
Analysis
Power
Good
Yes
No
Place
and
Route
Sign-Off
Analysis
VCD
Yes Parasitic
Extraction
Re-architect
vectors
Re-architect
Design
Where Best to Manage Power and How?
Architectural Changes
© 2008 Synopsys, Inc. (55)
Predictable Success
Synopsys Low Power Flow Summary
© 2008 Synopsys, Inc. (56)
Predictable Success
Predictable Success