Design of Integrated Circuits with Low-Power Consumption
Lecture By
Bengt Oelmann
Contents
• Motivation – Three examples
– high performance ICs with limited energy resources
– medium-to-low performance ICs with basically no energy resources – Complex systems integrated on a single chip
• Projects in Low-Power at Electronics Design Division
– Event-driven designs using asynchronous circuits – Dynamic power management for controller design – Construction of VLCs and their CODECs
– Circuit design for photon-counting pixel-array detectors – Digital circuits for low-noise/low-power
The First Example
• The mobile terminal
– Evolution of MIPS requirements in cellular
GSM GPRS 3G
Communication Application DSP MIPS
10000
1000
100
10
1
Power consumption is proportional to MIPS
The Energy Gap
1 10 100 1000 10000 100000 1000000 10000000
Battery Capacity (i.e. Eveready’s Law) Algorithmic
Complexity 3G
Signal Processor Performance (~Moore’s Law)
More Energy-Efficient Alternatives to Processors
Dedicated HW
Flexibility 1
10 100 1000
0.1
Reconfigurable processor Pleiades: 10-80 MIPS/mW
MIPS/mW
2V DSP: 3 MIPS/mW
ASIP / DSP
SA110
0.4 MIPS/mW
Embedded processors
Reducing the Energy Gap
• Signal processing in digital mobile communication
– Implemented as a combination of Software, programmable DSPs, and dedicated hardware (ASIC) for the most demanding algorithms
– To meet the power requirements in 3G systems, no more than 10% of the algorithms can be implemented in software.
– Most of the algorithms must be implemented in dedicated hardware in ASICs
– Only ASICs is not really the solution:
• Very costly
• Inflexible: 20-30 standards should be covered, should be scalable with date-rates
Reducing the Energy Gap
• Flexibility comes at an enormous cost (power)
• Specific hardware comes at an enormous cost (design effort)
• To achieve energy reduction and flexibility
– Research in efficient programmable processor architectures targeting low power consumption
– Research in efficient architectures for dedicated hardware that can be incorporated in programmable structures
– Careful analysis of what kind of flexibility is actually needed
The Second Example
• Distributed Sensor Networks
A Wireless Sensor Node
Solar Cell
Solar Cell
Mirror Controller
Accelerometer
0.25µm CMOS – double poly, 5 metals
Photodiode
Receiver
Temperature Sensor
ADC
Charge Pump
Source: Smart Dust Project
A Wireless Sensor Node
Source: Smart Dust Project
360µm
30 0 µ m
Photodiode
Pad to CCR
VddPad
GND Pad/
LFSR
Power-on Reset
Charge Pump
Wireless Sensor Node – Design Constraints
• Low data-rates << 10kBit/s
• Self-configuring , maintenance-free, and robust
• Cost
– Very low-cost device – should be deployed everywhere
• Size
– Should not intrude on the environment it is deployed
• Power/Energy
– Operate in self-contained fashion for the life-time of the product – Energy must be ”picked-up” from the surroundings
– Electronics must consume extremely low levels of static and dynamic power
Dynamic Power Consumption
• Reduce dynamic power consumption
– Obtain high MIPS/mW
– Dedicated processor architecture
• Low-level optimization
• Dynamic power management
– Make the sensor node reactive – wake it up on external events – Data/Instruction dependent activation of circuits
• Speed is not an issue
– Speed can be traded for low power
Static Power Consumption
• A sensor node is idle most of the time
– Computation and communication < 1% of the time
• Also static power consumption must be reduced
– Power consumption when it is idle
• Reduce leakage currents in transistors
Reduction of leakage
• Lower limit for the processor – leakage
Processor characteristics:
-Processor core: 15 000 transistors -102412bit Memory: 74 000 transistors -0.25m CMOS, vt=0.55V
• Standby power reduction through substrate bias control
Core SRAM
6 nW 79 nW
0.35nW 6.4nW
|Vbs|=0.0V |Vbs|=0.5V
12-20 times reduction
Power reduction for sensor nodes
• To achieve power reductions in digital circuits
– Known techniques to reduce dynamic power must be applied for low-level optimization on gate-, circuit-, and physical-levels
– Application specific processor and memory architectures must be designed – Static power dissipation dominates – Leakage currents must be controlled
The Third Example
• Integration of complex systems on single chip
– System-on-Chip (SoC)
• Traditionally, one chip per function
– Memory
– Microprocessor, DSP, video- and audio processors – A/D D/A converters
– RF – Etc.
• SoC is the technology integrating all functions on
one chip
CMOS IC technology enables SoC
400480088080 8085 8086
286386
486 PentiumP6
0.001 0.01 0.1 1 10 100 1000
1970 1980 1990 2000 2010
Transistors (MT)
2X growth in 1.96 years!
Power Density
4004 8008
8080
8085 8086
286 386
486 PentiumP6 1
10 100 1000 10000
1970 1980 1990 2000 2010
Power Density (W/cm2)
Hot Plate
Rocket
Nozzle
Nuclear
Reactor
Power Problems
• Power delivery
–Large voltage drops in the power supply distribution caused by large currents
• Electromigration
–High current densities may cause reliability problems
• High thermal dissipation in a single chip requires
–Advanced packaging technologies
–Expensive and bulky cooling devices like fans and heat-sinks
Traditional IC Design
• One clock domain
clock
• One power supply voltage
• Global busses for communication
• One type of transistors
• One pMOS and one nMOS
• Homogenous technologies for homogenous functions
SoC Requirements
• SoC characteristics
– Function units with very different requirements on the IC technology are integrated on the same chip
– Fully synchronous designs result in clock distribution networks that consumes approx. 50% if the total power
– Global busses will be slow and power hungry
• SoC is heterogeneous
– Requires a technology that matches the computational model
there is a need for a ”heterogeneous” technology – Novel architectures
• Communication-based design
• Reconfigurable hardware
CMOS Technology
• Application-specific MOS devices
High Voltage Low Leakage
High Speed
Ultra High
Speed EEPro
m
MRa
m
RF
Global communication and synchronization
• Advantages
– Global Clock eliminated – Local clock
– Less stringent skew bounds – Lower Peak current
• Disadvantages
– Local clock generation – Synchronization problems – No established design
methodology
Data Control
Asynchronous communication
Different speed/power supply voltage Synchronous islands
SB
1SB
2SB
3SoC summary
• Integration of heterogeneous functions on a single chip requires new solutions
• Novel architectures
– Power awareness in processor architectures and configurable hardware
• Novel circuit solutions
– Power supply voltage as a design parameter
• Novel design methods
– Communication-based design
Summary on power
• Power and/or energy have become dominant drivers
– Limiting factor for performance and reliability in wall-plugged applications
– Enabler for wide-spread use of distributed computing and data access
• Large energy reduction require a joint optimization
process between application and implementation
Projects in Low-power
• Techniques for building event-driven designs using asynchronous circuits
• Dynamic power management for controller design
• Construction of Variable-Length Codes and their CODECs for high speed and low power
• Low-Power Circuit Design for photon-counting pixel array detectors
• Digital circuits for low-noise/low-power in mixed
analog/digital designs
Techniques for building event- driven designs using
asynchronous circuits
• General model for fully asynchronous circuits
clk
Asynchronous model:
Synchronous model:
• Project includes
– Development of re-usable basic structures
– System-level model for performance evaluation
Asynchronous IC
• DSP for a spread-spectrum digital radio receiver
Design objective: verify the usefulness of proposed structures for “structured” design.
The design provide data for system-level modeling of complex asynchronous circuits.
Idea: Data-flow approach supporting local interconnections and with data rates adapting
to the actual speed needed.
Speed: 49 MHz
Power: 600 mW at 5V
Complexity: 100.000 transistors IC technology: 0.8 m CMOS
[Oelmann, Tenhunen]
Dynamic power management for controller design
S0
S1
S2 S3
S4
0.6
0.2
0.005 0.005
0.01 0.04
0.04
0.1
FSM specification
S
X Y
FSM implementation
clk
LIFS – the CAD-tool
S4 S0T
S2
S1
S2T S3T
FSM #1 FSM #2
S4
S3
S2
S0
S1
Partitioner State probability
computation
Local FSM transformation
Input switching probabilities
Technology information User
constraints State-transition graph
CAD-tool for power optimization
In the 8051 C, the controller consumes
50% of the total power
The Future of LIFS
• LIFS today
– Optimization procedures implemented in a tool written in JAVA – Optimization results: average of 45% power reduction over a set
of standard benchmarks
– Area penalty is a bit too high
• LIFS 2
– New approach that introduces a mixed synchronous/asynchronous state-memory – New optimization procedures
• Partitioning
• State-clustering
• State-encoding
Construction of Variable-Length Codes and their CODECs
• Algorithm-level optimization
– High performance – low-power CODECs for VLCs – Focus on VLCs developed for image and video data
• Golomb-Rice
• Universal VLC
power
w/o suffix 1b suffix 2b suffix VLC decoder
GR-VLC decoder
Low-Power Circuit Design for photon- counting pixel array detectors
• Pixel-arrays
– Consists of a large amount of pixels
• Photon-counting
– Each pixel contains several hundreds of transistors – Large power consumption – Mixed signal problems
• Asynchronous circuits
– Used for detecting events – Reduces power consumption
four times
Discriminator
Counter Output logic Pre-amplifier
Pulse shaper
Detector Comparators
Analog
Digital
Digital circuits for low-noise/low- power in mixed analog/digital
designs
• A project starting up
• Objectives:
– Develop design techniques for digital circuits that induce low levels of noise in a mixed signal environment
– Develop on-chip measurement of digital noise
• The project involves:
– Characterization of sub-threshold and current-mode techniques – Architectural speed optimization
– Design automation aspects
Conclusions
• Three different scenarios have been used to
illustrate the importance of low-power IC design
– Mobile wireless terminal (flexibility – power trade-off)
– ”power-less” sensor nodes (the ultimate low-power application) – System-on-chip (Power defines performance, system cost, and
reliability)
• Projects at the Electronics Design Division, MH
– Generic research
• Asynchronous design methods
• Low-power FSM design
– Application-specific
• VLC CODECS