1. APRIL ‘07
Evolving Wireless Markets & The Software Challenge, or
Software Is The Answer, But What Is The Question?
Willie Anderson and Rowland Reed
Qualcomm CDMA Technologies
DSP Development Group, Austin
PAGE 1
2. Overview
• Motivation: The Evolving and Huge Wireless Market Opportunities
• Hexagon DSP Introduction
• Hexagon DSP V2 Verification
• The Software Challenge & Its Effect On Overall Market
• Practical Approaches
Structured Specification
Functional Coverage for Software / System Test
Open-Source RTOS?
Software System Simulation
• Conclusion and Questions
PAGE 2
3. Wireless: Key Growth Drivers for Semiconductor Industry
Converging
600+ Million Mobile
Multimedia & Mobility Multimedia Users (2006)
Combining location & mobility with
audio, camera, video & gaming
Migrating Voice to Data 1+ Billion Data
3G Network Deployments around the globe: DO, Users (2007)
WCDMA, TD-SCDMA
Increasing Tele-density ~1.8 Billion Voice
Extending Voice Services into emerging markets Users (2008)
Growth Drivers
Sources (top to bottom): Yankee Group; ARC Group; Yankee Group
PAGE 3
4. Wireless: Key Growth Drivers for Semiconductor Industry
Converging
The Wireless Internet Will Soon HaveMillion Mobile
600+
Multimedia & Mobility
Greater Impact Than& The Wired Internet Users (2006)
Combining location mobility with
Multimedia
audio, camera, video & gaming
• 1B+ Mobile Data User By End Of 2007
With Anytime, Anywhereto Data
Migrating Voice Connectivity 1+ Billion Data
3G Network Deployments around the globe: DO, Users (2007)
• Vast Portions Of The World Will Have
WCDMA, TD-SCDMA
Access To Only Wireless Internet
Increasing Tele-density Mobile ~1.8 Billion Voice
• Laptops Now Embedded With markets
Extending Voice Services into emerging Users (2008)
Wide Area Broadband Modules
• Over 600M Phones Already Have
Multimedia CapabilityDrivers
Growth
Sources (top to bottom): Yankee Group; ARC Group; Yankee Group
PAGE 4
5. “Tomorrow’s Cell Phone Will Entertain, Amaze…
and Even Make Calls” (Fortune Magazine, October 2006)
Live TV Channels
Live TV broadcasts
Health and Fitness
Monitor your heart rate
Tagging
Share your favorite restaurant
with friends
Personal Identification
Gain access to data
Digital Wallet
Mobile commerce
Media Center
Get connected with other
Source: Fortune / Samuel Velasco / 5W Infographics
in-home devices through Wi-Fi
PAGE 5
6. Mobile Phone Shipments Outstrip All Other Devices
Global Device Sales (M's)
1400 1286
1252
1205
1156
1200 1086
986
1000
833
800
Millions
676
600 530
437
400
200
0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Sources
• Mobile phone sales – Strategy Analytics, Oct’06
• Laptop and Desktop data – IDC PC tracker – Q1’07
PAGE 6
7. Moving Beyond Voice
Wireless WAN Evolution From 1G to 4G
UMB
LTE
10,000 DO-rev B
HSPA+
DOrB
HSDPA
1000 HSDPA 7.2
Average Throughput
3.6
DO-rev A
(Kbps)
EVDO
WCDMA
100
EDGE
CDMA 1x
GPRS
10 GSM CDMA
AMPS
1
1980 1985 1990 1995 2000 2005 2010
PAGE 7
8. 3G Data Services Generating Increasing
Revenue for Carriers
250 100%
Premium Revenues 90% • Wireless data
Messaging revenues
200
Data as % of revenues
80% accounted for
17% of service
Data as % of revenues
70%
Revenues ($B)
150 60%
revenues in 2006
50%
100 40%
• Messaging (SMS,
30% MMS, IM, Email)
50 20% is still the key
10% driver for data but
0 0% share of premium
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 data revenues
was at 19%
(Gaming, Audio,
Video, Other
personlization) in
2006
PAGE 8
Source: Strategy Analytics, Dec’06
9. Wireless WAN Evolution
All The Power Of A Laptop…In Your Pocket 1 GHz Scorpion + ARM 9 CPUs
600 MHz DSP
3000+ MIPS*
3000
Paging, Messaging, Voice Mail MSM7200
2000 Personal Information Manager Convergence Dual-Core
ARM9 + ARM11
Up to 740 MIPS*
(400 MHz)
MSM6550
1600 Enhanced Platform ARM9
Up to 250 MIPS*
MIPS
(225 MHz)
MSM6500
1200 Multimedia Platform ARM9
Up to 160 MIPS*
(146 MHz)
800 MSM2 MSM2300 MSM3000
Intel 80186 ARM MCU ARM7TDMI
10 MIPS < 20 MIPS 23 MIPS* (27 MHz)
(~2.5 MHz)
400
0
1996 1998 2000 2002 2004 2006 2008
PAGE 9
10. Wireless WAN Evolution
All The Power Of A Laptop…In Your Pocket 1 GHz Scorpion + ARM 9 CPUs
600 MHz DSP
3000+ MIPS*
Exponential Complexity Growth In
3000 Wireless Embedded Systems…
• Driven By Moore’s Law & Economies Of Scale In
Paging, Messaging, Voice Mail MSM7200
WirelessInformation Manager Electronics Segment ARM9 + ARM11
2000 Personal Consumer Convergence Dual-Core
Up to 740 MIPS*
(400 MHz)
•1600
Nearly 10X Growth In CapabilityEnhanced Platform ARM9 In Less
InMSM6550
Devices
Up to 250 MIPS*
Than 4 Years
MIPS
(225 MHz)
MSM6500
1200 Multimedia Platform ARM9
• Mobile Device Storage Lags PC Industry By < 8 Years
Up to 160 MIPS*
(146 MHz)
MSM3000
• 800 Intel 80186Device Performance Lags PC Industry By < 5
MSM2
Mobile MSM2300 23 MIPS* (27 MHz)
ARM MCU ARM7TDMI
10 MIPS < 20 MIPS
Years
400
(~2.5 MHz)
• This Growth Portends The Current And Future Growth
0
Of Complexity Of Mobile Embedded Software
1996 1998 2000 2002 2004 2006 2008
PAGE 10
11. The Mobile Experience Requires Many Capabilities…
• Productivity (MS Office)
• Social Networking (Email, IM, VOIP, VT)
• Multimedia:
Streaming Audio/Video- MPEG4, MP3, AAC+, WMP…
Gaming
Image and Video Capture
• Content Transfer:
Local Networking (LAN/PAN)
Peripherals, Removable Storage
• Other Wireless Capabilities
Location-Based Services
Broadcast / Multicast
• Fundamental Characteristics
Stylish, Ergonomic and Easy-To-Use
Low Power
Low Cost
PAGE 11
12. Which Require A “Laundry List” Of Technologies…
• Multimode, Multi-band WWAN and WiFi (data bandwidth)
• Multimedia Processor:
Audio/Video CODECS- MPEG4, MP3, AAC+, WMP…
High Resolution Graphics Display
Image Sensor and Processor
• WPAN (BT, UWB, etc.) for data and audio support
• Peripherals (USB, Removable Storage)
• Large Internal Memory
• GPS
• Broadcast (DVB-H, MediaFLO)
• Low Power, Size, Cost
PAGE 12
13. Handsets Increase in Capability while Prices
Continue to Decline
Wireless Handset ASPs
$400
Smartphone
Enhanced Phone
$300 Low-End
Ultra-Low-End
Industry Average
$200
$100
$-
2004 2005 2006 2007 2008 2009 2010 2011
Source: ABI Research, August 2006
PAGE 13
15. Hexagon V2 Verification Methodology Pillars
• Code And Functional Coverage
The Metric For Verification Completion
• Constrained-Random Stimulus
To Find The Dark Corners Missed By Functional Coverage
• Unit- And Core-Based Environments
• Checking And Coverage Independent Of Stimulus
Allows Portability From Unit To Core
PAGE 15
16. Random Stimulus Without Functional Coverage
Random stimulus without the visiblity provided by functional coverage
is like shotgun blasts in the dark:
Design space hit by
random stimulus
Design Space
PAGE 16
17. Functional Coverage without Random Stimulus
Using directed tests as the primary mechanism to hit functional
coverage points leaves you exposed to bugs in areas you don’t have
coverage for:
Functional Coverage Points
Design Space
PAGE 17
18. The Magic Combination
Functional Coverage in combination with constrained
random stimulus yields high-quality, repeatable
verification results:
Design space hit by
random stimulus
Functional Coverage Points
Design Space
PAGE 18
19. Hexagon V2 Unit Verification
• Testbench
Synopsys NTB, Ported From VERA
• Stimulus
Vera-based Constrained Random
Limited Directed Testing
• Correctness Checking
Custom Models And Checkers
Assertions
– Blackbox: Developed By Verification Engineer
– Whitebox: Developed By Design Engineer
• Functional Coverage
Blackbox: Developed By Verification Engineer
Whitebox: Developed By Design Engineer
PAGE 19
20. Hexagon V2 Core Verification
• Testbench • Functional Coverage
NTB, Verilog Architectural Functional Coverage
• Stimulus – Automatically Generated From XML
Architecture Description
Random
– Leveraging Both Internal And Vendor
Unit-level Coverage Ports Up
Generators
Directed • Emulation
– Targets Specific Verification Plan Line FPGA-based Emulation
Items
– Targets Hard-to-hit Coverage Points Used To Run Applications Like H.264
Video Decode
• Correctness Checking
Golden Model: Instruction Set
Simulator (ISS)
Checking Ported Up From Units
Additional Core-specific Checking –
Generally Assertions
PAGE 20
21. Hexagon V2 Coverage Development
• Code/Functional Coverage Is Our Fundamental Metric For Verification
Completeness
• Verification Plan Used To Define Functional Coverage
Subject To Extensive Review By Design And Verification Engineers
• ~20k Core-level Architectural Coverage Points
• ~240k Total Core-level Coverage Points
Including Coverage Points Ported From Units
• How Much Functional Coverage To Create?
Seems Like More Would Be Better
But, You Can Bury Yourself In Data If You’re Not Careful
PAGE 21
22. Hexagon V2 Coverage Analysis
• Initial Focus More On Code Coverage
Can Identify Missing Functional Coverage Points
• Transition Focus To More Functional Coverage
• Review Process Is Meeting/Time Intensive
Iteration Cycle-time Determines How Quickly You Can Converge On Coverage
Goals
– Review Data
– Identify Changes Required To Improve Coverage
– Run Cycles Against Updated Design -- Collect Data
– Repeat
• Synopsys .html Format Is One Interesting Slice Of Coverage Data, But Has
Its Limits
A Real Database To Slice And Dice Is Much More Useful
• Coverage Convergence
Newly Implemented Coverage Requires Debugging
Coverage Definition Will See Refinement And Clean Up
Stimulus Will Need To Be Modified/Enhanced To Hit Uncovered Areas
PAGE 22
28. Better RTOS And Specification Capture Are Needed
• Ideally, There Would Be An Open Source RTOS Which Is The “Moral
Equivalent” To Linux, But This Has Yet To Happen
“Real-Time” Variants Of Linux Exist
Many Open Source RTOSes Exist
None Have Become As Ubiquitous As Linux
• “Real Time” Requirements Are Vastly Different For Different Systems
Automotive, Wireless Modems and Multimedia All Have Significantly
Different Real-Time Requirements
• RTOSes Are Like Standards: There Are So Many To Choose From
But Lack Of A Popular Open-Source RTOS Has Had Significant Impact
• Structured Specification Capture For Embedded Software Modules
Could Have Dramatic Impact On Quality Of Delivered Product
Parseable Specification Can Enable Functional Coverage In SW Testing
Functional Coverage May Be Able To Improve Software Development As
Much As It Has Hardware Development
PAGE 28
29. Functional Coverage May Be Key Technology for Software
• Functional Coverage Is Already Widely Used In ASIC Verification
• Functional Coverage Finds Defects In Hardware Which Would Likely
Never Be Seen In Field Testing Or Even Normal System Operation
• Functional Coverage Is A “Results” Approach To Verification1
Describes System Design From User’s Viewpoint
Monitors (Using “Coverage Checkers”) Stimulus Scenarios, Errors, Corner
Cases, State Transitions, Protocols, etc.
Tracks Combinations (“Crosses”) of Stimulus Points and Groups
• Why Does Functional Coverage Work?
It Focuses Engineer’s Attention On Areas Of Lowest Coverage
It Can Tell You When Your Design Is “Good Enough”
It Decouples Stimulus From Measurement, And Allows High Quality Stimulus
• Tools: Focus, Meteor (IBM), PurifyPlus (Rational), xSuds (Telcordia)
• IBM Has Website On Functional Coverage Methods for Software:
http://www.haifa.il.ibm.com/projects/verification/coverage_advisor/index.html
1- Some definitions come from Design Verification with e by Samir Palnitkar
PAGE 29
32. What Does Functional Coverage Ideally Look Like?
Tests
Tests
Tests
Scenario
Monitor
Generator
Transactors Monitor
Drivers Monitor
Functional
Coverage
Analysis
System Under Test
PAGE 32
33. What Does Functional Coverage Ideally Look Like?
Tests
Tests
Tests
Scenario
H
Monitor
Generator
NC
BE
Transactors Monitor
ST
TE
Drivers Monitor
Functional
Coverage
Analysis
System Under Test
PAGE 33
34. What Does Functional Coverage Ideally Look Like?
Tests
Tests
Tests
Scenario
Monitor
RS
Generator
KE
Transactors Monitor
EC
Drivers CH Monitor
Functional
Coverage
Analysis
System Under Test
PAGE 34
35. What Does Functional Coverage Ideally Look Like?
Tests
Tests
Tests
Scenario
Monitor
Generator
Transactors Monitor
ER
Drivers Monitor
YZ
Functional
AL
Coverage
AN
Analysis
System Under Test
PAGE 35
36. If Software Takes Too Long To Develop, Start Earlier
Engineering Resources
Typical Embedded System Development
$€¥£…
Software
Hardware
Customer
Integration
Time
PAGE 36
37. If Software Takes Too Long To Develop, Start Earlier
Engineering Resources
Better Embedded System Development
$€¥£…
Software
Hardware
Customer
Integration
Time
PAGE 37
38. If Software Takes Too Long To Develop, Start Earlier
• This Better Approach Requires
Motivation To Improve The Overall Embedded
System Development Process
A Software Simulation System Capability
• A Software Simulation System Provides Simulators For All
Programmable Cores (CPU, DSP) And Hardware In The System
• Simulation Of The Full System On A Software Simulator Offers Several
Significant Advantages In Addition To Better “Time To Money”
Allows Improved Hardware and Software Architecture Tradeoff Analysis
Affords Embedded Software Developers Better Visibility Into Behavior
Affords Customers A Very Early Start Into Integration
• What Is Missing?
An Open Source Standard (Like GCC, Linux, etc.) Software Simulation
Platform To Encourage Reuse And Create A Simulator “Ecosystem”
• “Cycle Approximate” Is Good Enough – The “80/20” Rule Applies
PAGE 38
39. Summary
• Opportunities In Wireless Embedded Products Are Excellent!
A Plethora Of Wireless Communications Protocols For Personal, Local and
Wide Area Networks Is Creating Huge New Product Demand
New Capabilities Are Migrating To Complex Handheld Systems, Driven By
Consumer Desire For Wireless Internet Everywhere And Digital Multimedia
The Challenge: Engineering Must Deliver Complex Products That Consume
Very Low Power And Provide Gigabytes Of Software and Billions Of
Transistors For Very Low Cost And Very Low Defect Densities
• Embedded Software Development Will Continue To Present Challenges
A Ubiquitous Open-Source Simulation Platform and RTOS Is Desirable
Better, More Structured Methods For Specification Capture Can Facilitate
Better Quality In Embedded Software, Particularly When These Specs Can
Drive A Functional Coverage-Based Test System
Software System Simulation Can Facilitate Better Overall Lifecycle Costs
1. H. Sackman, W.J. Ericson, and E.E. Grant, “Exploratory Experimental Studies Comparing Online and Offline Programming Performance,” CACM,
Vol. 11, No. 1, Jan. 1968, pp. 3-11.
PAGE 39
40. Summary
• Opportunities In Wireless Embedded Products Are Excellent!
A Plethora Of Wireless Communications Protocols For Personal, Local and
Wide Area Networks Is Creating Huge New Product Demand
Great designs come from great designers. Software construction is a
New Capabilities Are Migrating To Complex empower and liberate the By
creative process. Sound methodology can Handheld Systems, Driven
Consumer Desire For Wireless Internet Everywhere And Digital Multimedia
creative mind; it cannot inflame or inspire the drudge.
The Challenge: Engineering Must Deliver Complex Products That Consume
Very Low Power And not minor – they are rather like the differences
The differences are Provide Gigabytes Of Software and Billions Of
Transistors For Very Low Cost And Very Low Defect Densities very best
between Salieri and Mozart. Study after study shows that the
• Embedded Software structures that are faster, smaller, simpler, cleaner,
designers produce Development Will Continue To Present Challenges
and produced with less effort1. The differences between the great and
A Ubiquitous Open-Source Simulation Platform and RTOS Is Desirable
the average programmer approach an order of magnitude.
Better, More Structured Methods For Specification Capture Can Facilitate
Better Quality In Embedded Software, Particularly When“No Silver Bullet”
- Fred Brooks, These Specs Can
Drive A Functional Coverage-Based Test System
Software System Simulation Can Facilitate Better Overall Lifecycle Costs
1. H. Sackman, W.J. Ericson, and E.E. Grant, “Exploratory Experimental Studies Comparing Online and Offline Programming Performance,” CACM,
Vol. 11, No. 1, Jan. 1968, pp. 3-11.
PAGE 40
