CSE140: Components and Design Techniques for Digital...
Transcript of CSE140: Components and Design Techniques for Digital...
CSE140: Components and Design Techniques for Digital Systems
Tajana Simunic Rosing
Sources: TSR, Katz, Boriello, Vahid
1
Announcements and Outline
• Check webct grades, make sure everything is there and is tcorrect
• Pick up graded homework at TA’s or my assistant’s office• Final exam Tuesday June 10th at 3pm same location as• Final exam – Tuesday, June 10th, at 3pm, same location as
the class– Everything covered in lectures, whole book & all handouts– Format:
• Problems similar to HW and previous exams• Multiple choice and/or T/F questions on the assigned reading
– Discussion session will go over the previous year’s final
• Today’s topic: Power and Energy
Sources: TSR, Katz, Boriello, Vahid
2
CSE140: Components and Design Techniques CSE140: Components and Design Techniques for Digital Systems
Power & Energy
Tajana Simunic Rosing
Sources: TSR, Katz, Boriello, Vahid
3
Overview
• Motivation for design constraints of power consumptiong p p• Power metrics• Power consumption analysis in CMOS• How can a logic designer control power?
Sources: TSR, Katz, Boriello, Vahid
Phone + Messenger + PDA
Blackberry 8310 • Quad-band GSM/GPRS/EDGE,
wi-fi, Bluetooth™ 2.0• 2 megapixel camera s with 5x
zoom and built-in flash • MicroSD memory card slot• MicroSD memory card slot• Email, IM, SMS, Media player• Docs: Word, Excel, PDF and , ,
JPEG• 4 hours talk time, 17 days standby
Sources: TSR, Katz, Boriello, Vahid
Phone + Messenger + PDA
• Quad-band GSM™ phone; 802 11 b/g EDGEiPhone
Quad band GSM phone; 802.11 b/g, EDGE & Bluetooth® v2.0+EDR
• View PDF, JPEG, Word and Excel docs• Chat-style SMS text messaging• 2.0 megapixel camera • Screen Resolution: 480 x 320 pixels (163 ppi) • Talk time: Up to 8 hours • Standby time: Up to 250 hours• Standby time: Up to 250 hours • Internet use: Up to 6 hours • Video playback: Up to 7 hours
Sources: TSR, Katz, Boriello, Vahid
Video playback: Up to 7 hours • Audio playback: Up to 24 hours
Important (Wireless) p ( )Technology Trends
“S t l Effi i ”“Spectral Efficiency”:More bits/m3
Rapidly increasingtransistor densityy
Rapidly decliningsystem cost
Sources: TSR, Katz, Boriello, Vahid
In the Physical World: Sensor Devicesy
Sources: TSR, Katz, Boriello, Vahid
Important (Wireless)Technology TrendsTechnology Trends
Sp d Dist n C stRapid Growth: Machine-to-
Machine Devices
Sources: TSR, Katz, Boriello, Vahid
Speed-Distance-CostTradeoffs
Machine Devices (mostly sensors)
Why Worry About Power?Why Worry About Power?• Portable devices:
– Handhelds laptops phones MP3 players cameras all need to run for– Handhelds, laptops, phones, MP3 players, cameras, … all need to run for extended periods on small batteries without recharging
– Devices that need regular recharging or large heavy batteries will lose out to those that don’t.
• Power consumption important even in “tethered” devices – System cost tracks power consumption:
• Power supplies, distribution, heat removal– Power conservation environmental concerns– Power conservation, environmental concerns
• In 10 years, have gone from minimal consideration of power consumption to (designing with power consumption as a primary design constraint!
Sources: TSR, Katz, Boriello, Vahid
Power and Energy Basics
• Power supply provides energy for charging and discharging wires and transistor gates The energy supplied is stored & then dissipated astransistor gates. The energy supplied is stored & then dissipated as heat.
dtdwP /≡ Power: Rate of work being done over timeRate of energy being used
W tt J l / dtEP Δ• If a differential amount of charge dq is given a differential increase in
energy dw, the potential of the charge is increased by
Watts = Joules/secondstEP Δ=
• Given that current:• Power is work done over time:
dqdwV /=dtdqI /=
dqdwPower is work done over time:
• Energy is:
IVPdtdq
dqdwdtdw ×==×=/
∫=t
Pdtw
Sources: TSR, Katz, Boriello, Vahid
∞−
Basics• Warning! In everyday language, the term “power” is used
incorrectly in place of “energy”incorrectly in place of energy• Power is not energy• Power is not something you can run out ofg y• Power can not be lost or used up• It is not a thing, it is merely a rateg y• It can not be put into a battery any more than velocity can
be put in the gas tank of a car
Sources: TSR, Katz, Boriello, Vahid
Heats 1 gram of water This is how electric tea pots work ...
0.24 Calories per Second0.24 degree C
1A1 Joule of Heat
Energy per Second
+1V -
1 Ohm ResistorResistor
20 W rating: Maximum power the package is able to
Sources: TSR, Katz, Boriello, Vahid
p gtransfer to the air. Exceed rating and resistor burns.
Cooling an iPod nano ...Like a resistor, iPod relies on passive transfer of heat from case to the air
Why? Users don’t want fans in their pocket ... p
To stay “cool to the touch” via passive cooling, power budget of 5 Wp g
If iPod nano used 5W all the time, its battery would last 15 minutes
Sources: TSR, Katz, Boriello, Vahid
...
Powering an iPod nanoBattery has 1 2 W hour rating:Battery has 1.2 W-hour rating:Can supply 1.2 W of power for 1 hour
1.2 W / 5 W = 15 minutes
M W h i bi b tt More W-hours require bigger battery and thus bigger “form factor” --it wouldn’t be “nano” anymore!
Real specs for iPod nano ‘05 : 14 hours for music, 4 hours for slide shows4 hours for slide shows
85 mW for music
Sources: TSR, Katz, Boriello, Vahid
300 mW for slides
0.55 ounces
12 hour 12 hour battery life
1 GB1 GB
Sources: TSR, Katz, Boriello, Vahid
20 hour battery life for audio, 6.5 hours for movies (80GB version)
24 hour battery life for audio
5 h b tt 5 hour battery life for photos
12 hour battery life
Sources: TSR, Katz, Boriello, VahidCS 150 - Spring 2007 – Lec #28 –P 17
y
Notebooks ... now most of the PC marketApple MacBook -- Weighs 5.2 lbs
8.9 inpp M W g .
1 in
12.8 inPerformance: Must be “close enough” to desktop performance ... many people no longer own a desktop
Size and Weight: Ideal: paper notebook
Heat: No longer “laptops” -- top may get “warm”, bottom “hot”.
Sources: TSR, Katz, Boriello, Vahid
g p p p y gQuiet fans OK
Battery: Set by size and weight limits ...Battery rating: 55 Battery rating: 55 W-hour
GH l At 2.3 GHz, Intel Core Duo CPU consumes 31 W running a heavy load running a heavy load - under 2 hours battery life! And, just for CPU!just for CPU!
46x energy than iPod nano. iPod lets you listen to music for 14 hours!
Almost full 1 inch depth. Width and height set by
At 1 GHz, CPU consumes 13 W tts “En s ” pti n
Sources: TSR, Katz, Boriello, Vahid
Width and height set by available space, weight.
13 Watts. Energy saver” option uses this mode ...
Battery Technologyy gy• Battery technology has developed slowly• Li-Ion and NiMh still the dominate technologiesLi Ion and NiMh still the dominate technologies• Batteries still contribute significantly to the weight of
mobile devices
Handspring PDA - 10%Nokia 61xx -
33%
Sources: TSR, Katz, Boriello, Vahid
Toshiba Portege3110 laptop - 20%
55 W-hour battery stores the energy of
1/2 a stick of dynamite1/2 a stick of dynamite.
Sources: TSR, Katz, Boriello, VahidCS 150 - Spring 2007 – Lec #28 –P 21
If battery short-circuits, catastrophe is possible ...
CPU Only Part of Power Budget
Notebook running a full workload.
“other”If our CPU took no power at all to run, that would only double battery life!CPU
otherGPU
only double battery life!CPULCD Backlight
LCD
Sources: TSR, Katz, Boriello, Vahid
Servers: Total Cost of Ownership (TCO)Machine rooms are Machine rooms are expensive … removing heat dictates how many servers to put many servers to put in a machine room.
Electric bill adds up! Powering the servers + powering the air conditioners is a big part of TCO
Reliability: running computers hot makes h f l f
Sources: TSR, Katz, Boriello, Vahid
them fail more often
How Do We Measure and Compare Power Consumption?Consumption?
• One popular metric for microprocessors is: MIPS/watt– MIPS, millions of instructions per second
• Typical modern value?– Watt, standard unit of power consumption
• Typical value for modern processor?– MIPS/watt reflects tradeoff between performance and power– Increasing performance requires increasing power– Problem with “MIPS/watt”
• MIPS/watt values are typically not independent of MIPS– Techniques exist to achieve very high MIPS/watt values, but at very
low absolute MIPS (used in watches)low absolute MIPS (used in watches)• Metric only relevant for comparing processors with a similar performance
– One solution, MIPS2/watt. Puts more weight on performance
Sources: TSR, Katz, Boriello, Vahid
Metrics
• How does MIPS/watt relate to energy?• Average power consumption = energy / time• Average power consumption = energy / time
– MIPS/watt = instructions/sec / joules/sec = instructions/joule
– Equivalent metric (reciprocal) is energy per operation (E/op)
• E/op is more general - applies to more that processorsE/op is more general applies to more that processors– also, usually more relevant, as batteries life is limited by total energy
draw.– This metric gives us a measure to use to compare two alternative
i l t ti f ti l f tiimplementations of a particular function.
Sources: TSR, Katz, Boriello, Vahid
Power in CMOS
pullupnetwork
Vdd
VddSwitching Energy:energy used to switch a node
C
network
pulldownnetwork
10
i(t)
v(t) t0 t1
v(t)energy used to switch a node
network
GND
t0 t1
Energy dissipated in pullup:
111 ttt
∫∫∫222 2121
)()()()()(
1 1
1
0
1
0
1
0
t t
t
t dd
t
t dd
t
tsw
VVVddV
dtdtdvcvVdttivVdttPE =⋅−=⋅−==
∫ ∫
∫∫∫
Energy supplied Energy dissipatedEnergy stored
21210 0
ddt t dddddd cVcVcVdvvcdvcV =−=⋅−= ∫ ∫
Sources: TSR, Katz, Boriello, Vahid
An equal amount of energy is dissipated on pulldown
Switching Powerg• Gate power consumption:
– Assume a gate output is switching its output at a rate of: f⋅αg p g p
1/f
fαactivity factor clock rate(probability of switching on any particular clock period)
swavg ErateswitchingtEP ⋅=Δ=
any particular clock period)
Pavg221 ddavg cVfP ⋅⋅=αTherefore:
clock f
ddavg f
221 ddVcfnP ⋅⋅⋅= αChip/circuit power consumption:
Sources: TSR, Katz, Boriello, Vahid
21 ddavgavgavg VcfnP αnumber of nodes (or gates)
Other Sources of Energy Consumptiongy p
• “Short Circuit” Current: • Junction diode leakage:
Vout
I
T i t d i iVin
I
VoutVin
I
Transistor drain regions“leak” charge to substrate.
Vin
I
DiodeCharacteristic10-20% of total chip power
V
~1nWatt/gate
Sources: TSR, Katz, Boriello, Vahid
1nWatt/gatefew mWatts/chip
Other Sources of Energy Consumptiongy p• Consumption caused by “DC leakage current” (Ids leakage):
Vout=VddVin=0
Ids
Ioff VgsT i t /d d t gVthTransistor s/d conductance
never turns off all the way
Low voltage processes much worse
• This source of power consumption is becoming increasing significant as process technology scales down
• For 90nm chips around 10-20% of total power consumption Estimates put it at up to 50% for 65nm
Sources: TSR, Katz, Boriello, Vahid
Estimates put it at up to 50% for 65nm
Controlling Energy Consumption: What Control Do You Have as a Designer?Control Do You Have as a Designer?
• Largest contributing component to CMOS power consumption is switching power:
221 VfP α• Factors influencing power consumption:
n: total number of nodes in circuit
21 ddavgavgavg VcfnP ⋅⋅⋅= α
– n: total number of nodes in circuitα: activity factor (probability of each node switching)
– f: clock frequency (does this effect energy consumption?)– Vdd: power supply voltageVdd: power supply voltage
• What control do you have over each factor? • How does each effect the total Energy?
Sources: TSR, Katz, Boriello, Vahid
Our design projects do not optimize for power consumption
Scaling Switching Energy per GateMoore’s LawMoore s Lawat work …
Due to reduced V and C (l n th nd C (length and width of Cs decrease, but plate distance plate distance gets smaller)
Recent slope Recent slope reduced because V is scaled less
Sources: TSR, Katz, Boriello, VahidFrom: “Facing the Hot Chips Challenge Again”, Bill Holt, Intel, presented at Hot Chips 17, 2005.
aggressively
Device Engineers Trade Speed and Power
We can reduce CV2 (Pactive) b l i Vby lowering Vdd
We can increase speed by raising Vdd andlowering Vt
We can reduce leakage (Pstandby) by raising Vt
Sources: TSR, Katz, Boriello, Vahid
From: Silicon Device Scaling to the Sub-10-nm RegimeMeikei Ieong,1* Bruce Doris,2 Jakub Kedzierski,1 Ken Rim,1 Min Yang1
Customize processes for product types ...
Sources: TSR, Katz, Boriello, VahidFrom: “Facing the Hot Chips Challenge Again”, Bill Holt, Intel, presented at Hot Chips 17, 2005.
Intel: Comparing 2 CPU Generations ...Find enough tricks, and you can afford to can afford to raise Vdd a little so that you can raise the clock speed!
Cl ck sp d
Sources: TSR, Katz, Boriello, Vahid
Clock speed unchanged ... Lower Vdd, lower C,
but more leakage
Design tricks: architecture & circuits
Switching Energy: Fundamental PhysicsE l i t siti dissi t s
Vdd
V
Every logic transition dissipates energy
C
Vdd
12
C E1
212
C E0
22 V
dd1-
>0=
2 Vdd
0-
>1=
Strong result: Independent of technology
How can we limit
switching
(1) Slow down clock (fewer transitions). But we like speed ...(2) Reduce Vdd. But lowering Vdd lowers the clock speed ...
Sources: TSR, Katz, Boriello, Vahid
switching energy? (3) Fewer circuits. But more transistors can do more work.
(4) Reduce C per node. One reason why we scale processes.
Second Factor: Leakage Currents
Even when a logic gate isn’t switching, it burns power …
Isub: Even when this nFetis off, it passes an Ioffleakage current.
0V = We can engineer any Ioffwe like, but a lower Ioff also results in a lower Ion and thus results in a lower Ion, and thus the lower the clock speed.
Intel’s current processor designs, l k it hi
Igate: Ideal capacitors have
leakage vs switching power
A lot of work was done to get a ratio
Sources: TSR, Katz, Boriello, Vahid
Igate Ideal capacitors have zero DC current. But modern transistor gates are a few atoms thick, and are not ideal.
done to get a ratio this good ... 50/50 is common.
Bill Holt, Intel, Hot Chips 17.
Engineering “On” Current at 25 nm ...
Vd I
We can increase Ion by raising Vdd and/or lowering Vt.
I ds
VV
g
I ds
Vs 1.2 mA = I
on
0 25 ≈ V0.25 ≈ Vt
Ioff
= 0 ???
Sources: TSR, Katz, Boriello, Vahid
0.7 = Vdd
Plot on a “Log” Scale to See “Off” Current
Vd
We can decrease Ioff by raising Vt - but that lowers Ion
IdsV
Vg
I ds 1.2 mA = I
on
s0.25 ≈
VVt
Ioff
≈ 10 nA
Sources: TSR, Katz, Boriello, Vahid
0.7 = Vdd