Embedded Computing: Differences and Performance in General-Purpose Systems, Study notes of Electrical and Electronics Engineering

Download Embedded Computing: Differences and Performance in General-Purpose Systems and more Study notes Electrical and Electronics Engineering in PDF only on Docsity!

ECE 563

Advanced Computer Architecture

Fall 2007

Lecture 13: Embedded Processors

Important stuff … ‰^ Final exam on 12/19 8:00—11:00PM in classroom(yet to confirm) ‰^ Project and report due on the same day after theexam!!!

What is an Embedded Computer?^ ‰^ A computer not used to run general-purpose programs, butinstead used as a component of a larger system. Usually,user does not change the computer program

(except for

manufacturer upgrades). ‰ Example applications:^ ƒ^ Toasters^ ƒ^ Cellphone^ ƒ^ Digital camera (some have several processors)^ ƒ^ Games machines^ ƒ^ Set-top boxes (DVD players, personal video recorders, ...)^ ƒ^ Televisions^ ƒ^ Dishwashers^ ƒ^ Car (some have dozens of processors)^ ƒ^ Internet router (some have hundreds to thousands ofprocessors)^ ƒ^ Cellphone basestation^ ƒ^ .... many more

Early Embedded Computing Examples

-^ Intel 4004, 1971 -^ developed for Busicom 141-PFprinting calculator –^ Intel engineers decided thatbuilding a programmable computerwould be simpler and more flexiblethan hard-wired digital logic -^ MIT Whirlwind, 1946-51^ –^ Initially developed for real-time flightsimulator^ –^ IBM later manufactured versions for SAGEair defense network, last used in 1983

What is different about embedded computers? ‰^ Embedded processors usually optimized toperform one fixed task with software from systemmanufacturer ‰^ General-purpose processors designed to runflexible, extensible software systems with codefrom third-party suppliers^ z^ applications not known at design time ‰^ Note, many products contain both embedded andgeneral-purpose processors^ z^ e.g., smartphone has embedded processors for radiobaseband signal processing, and general-purposeprocessors to run third-party software applications

Lesser emphasis on software portability in embedded applications^ ‰^ Embedded systems^ z^

can usually recompile/rewrite source code for different ISA,and/or use assembler code for new application-specificinstructions z processor pipeline microarchitecture and memory capacity andhierarchy known to programmer/compiler z mix of tasks known to writer of each task, usually static: usescustom run-time system z each task usually “trusts” others, can run in same addressspace

‰^ General-purpose systems^ z^

must have standard binary interface for third-party software z compiler doesn’t know about this particular microarchitectureor memory capacity or hierarchy (compiled for general model) z unknown mix of tasks, tasks dynamically added and deletedfrom mix: uses general-purpose operating system z tasks written by various third-parties, mutually distrustful,need separate address spaces or protection domains

What is Performance? ‰ Latency (or response time, or executiontime)^ z^ time to complete one task ‰ Bandwidth (or throughput)^ z^ tasks completed per unit time

Performance Measurement

Average rate:

A^ >^ B

>^ C

Worst-case rate:

A^ <^ B

<^ C

Processing Rate (Inputs/Second) Inputs

A B C

Which is best for desktop performance? _______ Which is best for hard real-time task?

_________

Average rates

Worstcaserates

Power Measurement^ ‰^ Energy measured in Joules^ ‰^ Power is rate of energy consumption measured inWatts (Joules/second)^ ‰^ Instantaneous power is Volts * Amps^ ‰^ Battery Capacity Measured in Joules

ƒ^ 720 Joules/gram for Lithium-Ion batteries (best energyper weight ratio) ƒ^ 1 instruction on Intel XScale processor takes ~1nJ

⇒^ ~1 billion executed instructions weigh ~1mg

V

I

Fall 2007

Power versus Energy^ ƒ^ System

A^ has higher peak power, but lower total energy

ƒ^ System

B^ has lower peak power, but higher total energy

Power

Time

Peak A^ Peak B

Integratepowercurve to getenergy

Power Dissipation in CMOS (low static power) Primary Components: ‰^ Capacitor Charging (~85% of active power)^ ƒ^ Energy is 1/2 CV

2 per transition ‰^ Short-Circuit Current (~10% of active power)^ ƒ^ When both p and n transistors turn on during signal transition ‰^ Subthreshold Leakage (dominates when inactive)^ ƒ^ Transistors don’t turn off completely, getting worse with technology scaling^ ƒ^ For Intel Pentium-4/Prescott, around 60% of power is leakage^ ƒ^ Optimal setting for lowest total power is when leakage around 30-40% ‰^ Gate Leakage (becoming significant)^ ƒ^ Current leaks through gate of transistor ‰^ Diode Leakage (negligible)^ ƒ^ Parasitic source and drain diodes leak to substrate

CL

Diode Leakage Current Subthreshold

Leakage Current

Short-CircuitCurrent Capacitor Charging Current

Gate LeakageCurrent

Reducing Switching Power

Power

∝^ activity * 1/2 CV

2 * frequency

ƒ^ Reduce activity ƒ^ Reduce switched capacitance C ƒ^ Reduce supply voltage V ƒ^ Reduce frequency

Reducing Switched Capacitance^ Reduce switched capacitance C^ z^

Different logic styles (logic, pass transistor, dynamic) z Careful transistor sizing z Tighter layout z Segmented structures

A^

B^

C

Bus

Shared bus driven by Aor B when sending

values to C Insert switch to isolatebus segment when Bsending to C

A^

B^

C

Reducing Frequency^ ‰^ Doesn’t save energy, just reduces rate at which itis consumed^ z

Some saving in battery life from reduction inrate of discharge