This document discusses the concept of diminishing returns in technology and its implications. It notes that while complexity was never an obstacle when design costs were low, limits are now upon us. This leads naturally to discussions about multicore processors and their role as technology moves out of being the primary driver of markets. The intent is to provide context for discussions around efficiency and diminishing returns to better understand multicore and questions about future directions.

1. Abstract ...
 Why does it take millions of transistors to realise the broadcast radio receiver
done in five in the 70's? Why are there millions of lines of software in products that
are not programmable? And why do we throw things away before they break? The
last 40 years saw the exponential growth of silicon capacity and the technology
lead markets that ensued. Complexity was never an obstacle whilst design cost
was 2nd order, but now that limits are upon us it is also apparent that p p buy
, p pp people y
products not technology ... so we strive to deliver elevated expectation despite
Diminishing Returns.
 My intent is to give context to subsequent Multicore discussions in and beyond
this event. I will do this by looking at 'Efficiency' in a context of increasingly
Diminishing Returns This will lead quite naturally to Multicore (CMP); its rationale
Returns.
and its role. But will also raise questions about the way(s) forward ... as
technology moves out of the market driving seat.
1

2. 1v0
Prof. Ian Phillips
Principal Staff Eng’r,
Progress:
ARM Ltd
ian.phillips@arm.com
Visiting Prof. at ...
Despite the Law of
p
Diminishing Returns
OR
Contribution to Industry
Award 2008
Whatever happened to the 6 transistor radio?
Multi-Processor Workshop
5sep11
2

3. Things Haven’t Always
Been Like This ...!
3

4. A Moment of Retrospection
 I learned all I know in the past!
... I just have more of it than most people!
 I’ll take you back 36yrs to 1975 ...
 New: Degree, Home, Job, Wife, Baby, Car, Phone ...
Mk1: H man bab
Mk1 Human baby C1975 Mark’N’: Mud-Hut
M k’N’ M d H t C1975
Aka Dylan-Paul Phillips Aka: 5 Manor Close.
Mk1: Ian Phillips
C1975 ... M
More-or-less th same t d !
l the today!
4

5. 1975: Technology Products
Vauxhall Viva HB SL90
GPO Type 706 Telephone
... Recognisably the same, but very different today.
5

6. 1975: Semiconductor Electronics
 Domestically we had...
 Portable Radio
 Pocket Calculator
 Hi-Fi (Partial)
 Colour TV (Partial)
... That’s about all!
 Professionally we had... TI SR 51 Calculator
c1975
1975
BeoVision 3500 c1975
 Computers & First PCs
 Radio Receivers
 Satellites
 Under-Sea Cables
 Transmitters (Partial)
 TV C
Cameras (P ti l)
(Partial)
 Telephony (Partial) Stuart 5 Transistor Radio
1975
IBM 220PX c1975
6

7. 2010: Electronics Everywhere but Nowhere
7

8. Confusing Technology
with P
Products ...
8

9. The Computer ... Or Is It?
9

10. Computer: A Machine for Computing ...
Computing ...
... A general term for algebraic (mathematical) manipulation of data ...
Numerated Processed Data/
Phenomena y=F(x,t,s)
F( t ) Information
OUT (y)
IN (x)
... State and Time are factors in this.
 It can include phenomena ranging from human thinking to calculations
with a narrower meaning. Wikipedia
 Usually used it to exercise analogies (models) of real-world situations;
Frequently in real-time.
q y
... No mention of Implementation Technology in this!
10

11. Planet Motion Computer – Orrery c1700
Mechanical
Technology
• Inventor: George Graham (1674-1751)
• Single-Task, Continuous Time, Analogue Mechanical Computing (With backlash!)
11

12. Babbage's Difference Engine 1837
Mechanical
(Re)construction Technology
c 000
c2000
 Th diff
The difference engine consists of a number of columns, numbered f
i i t f b f l b d from 1 t N E h column is able to store one decimal number. The only operation
to N. Each l i bl t t d i l b Th l ti
the engine can do is add the value of a column n + 1 to column n to produce the new value of n. Column N can only store a constant, column 1 displays
(and possibly prints) the value of the calculation on the current iteration.
Computer for Calculating Tables: A Basic ALU Engine
12

13. Enigma ~1940
Mechanical
Technology
Data Encryption/Decryption Computer
13

14. Colossus Computer 1944
Valve/Mechanical
Technology
Code-Breaking
Code Breaking Computer: A Data Processor
14

15. Digital Computer – Baby 1947 (Reconstruction)
Valve/Software
Technology
General Purpose, Quantised Time and Data, (Digital) Electronic Computing
15

16. Analogue Computer – AKAT c1960
Transistor
Technology
General Purpose, Continuous Time, Approximate (Analogue) Electronic Computing
16

17. Evolution of Radio
Tele Verta
Tele-Verta Radio
4 Valves
1 Rectifier Valve
BTH c1945
Crystal Set
1 Diode Evoke DAB Radio
c1925 100 M Transistors
2-3 Embedded Processors
Bush Radio c2005
7 Transistors
1 Diode
c1960
Ian’s ‘Span’
17

18. Radio as Computation ...
Integrated Circuit
Transistor
Valve
Technology
Vi
Vrf=Vi*100
Vrf Vro='Bandpass'(Vif*1000)
Vif
Vro
Vif=Vrf*Vlo
Vlo
Vlo=Cos(t*1^6)
Single-Task, Continuous Time, Approximate (Analogue) Electronic Computing
18

19. Products Make Business
 21c Businesses have to be
 Operations and Competition is Global and so are Investors
 Nationality has little meaning
 Business needs
 End-Customers buy Products, not Technology
 Technologies enable Product Options
 Business-Models make Money
 New Products are
 Design is a Cost/Risk to be Minimised
 New Technology increases Cost/Risk ...
... But does not always increase Value
 HW, SW, Mechanics, Optics, etc are (just) means to an end!
... New Technology ≠ Market Success (Any More)
19

20. High Performance Computing (HPC)
 The exponential progression of
Moores Law has enabled the
fantastic computation power we
now take for granted ...
 AF T
Few Tens of h dli
f headline-grabbing
bbi
 A Few Hundred Million visible
... BUT...
 Tens of Billions of invisible
ubiquitous
... Gives impression that General Purpose
Digital Computation is what it is all about.
20

21. Products are Solutions;
Products
Not The Other-Way
Other-
Round ...
21

22. Embedded Computing Is ...
 Entertainment
 Remote Control
 Security
 Televisions
 ID C d
Cards
 Memory
 Logistics
 Transport
 Banking
g
 Manufacturing
 Energy
 Communications
 Medical
... etc, etc, etc.
t t t
22

23. High Performance (Embedded) Computing
Obvious: Less-Obvious:
 Business Model  Infrastructure ...
 Manufacture and Distribution
 Aesthetics
A th ti
 Road network, Fuel supply, Tyres
 Performance  Service network & Training
 Brand & Image  Sales and Marketing, etc
 Finance schemes
Fi h
 Technologies
 Dealership  Internal Combustion Engine ...
 Warranty, etc  Bearings, Casting, Metal forming,
Paint, Aerodynamics Glass Rubber,
Paint Aerodynamics, Glass, Rubber
Suspension ...
 Manufacturing, Reliability, Quality ...
 Electronic Systems
23

24. The Evolution of Customer(kind) Universe – 13.6Byr
Earth – 4.5Byr
 (Us!) – appeared 35,000 yr ago
 ‘Developed’ from Homo-Sapien (Wise Human) 100,000 yr ago
 Primary Objective: Survive Nature (1,000 generations)
 - appeared ~2,000 yr ago
 Pythagoras Socrates Plato Aristotle Archimedes ...
Pythagoras, Socrates, Plato, Aristotle, Archimedes,
Pythagoras
 Objective: Understand Nature (100 gen.)
 - appeared ~1,000 yrs ago
 Galileo, Descartes, (1000 ad)
 Electricity - William Gilbert (1600ad) Galileo
 Objective: Manipulate Nature (50 gen.)
 - just 260 yrs ago
 Industrial Revolution (1750: 8 gen’n)
 Year 0: Science Meets Exploitation
 Objective: Exploit Nature (10 gen.) Brunell
... Remember: Real (Cro-Magnon) Customers Don't Buy Technology!
24

25. The iConic Must-Have Product ...
25

26. ... Cool Design(California) & Manufacture(China) !
26

27. ... Most Design Never Noticed or Valued
27

28. The Threshold of Magic 1: Clarke: Any sufficiently
advanced technology is
indistinguishable from magic.
 Everybody has a threshold, beyond which Functionality is
Indistinguishable From Magic1!
 Chemical Systems
Ch i l S t
 Biological Systems
 Economic Systems
y
 Electronic Systems
 The Incandescent Light:
is the
for most non-scientific,
but well educated people!
... Its not a crime, to Not Understand Technology!
... The crime is not realising people don’t when you
are the one who suffers as a result!
28

29. All Technologies Are
Important ...
p
29

30. Exciting Technology ... At the Module
Inside the Case
iPhone 4's vibrator motor. rear-facing 5 MP camera with
720p video at 30 FPS, tap to
focus feature, and LED flash.
,
30 Source ... http://www.ifixit.com

31. Exciting Technology ... At the Module
Inside the Case
The Control Board.
31 Source ... http://www.ifixit.com

32. Exciting Technology ... Inside the Module
Inside The Control Board (a-side)
 Visible Design-Team Members...
 A4 PProcessor, specified b A l d i
ifi d by Apple, designed and manufactured b S
d d f t d by Samsung ...
 The central unit that provides the iPhone 4 with its GP computing power.
 Inc. ARM A8 600 MHz CPU (also other ARM CPUs and IP?)
 ST-Micro (3 axis gyroscope) - (ARM Partner)
 Broadcom (Wi-Fi, Bluetooth, and GPS) - (ARM Partner)
 Skyworks (GSM)
 Triquint (GSM PA)
 Infineon (GSM Transceiver) - (ARM Partner)
GPS
Bluetooth,
EDR &FM
32 Source ... http://www.ifixit.com

33. Exciting Technology ... Inside the Module
Inside The Control Board (b-side)
 Visible Design-Team Members ...
 Samsung (flash memory) - (ARM Partner)
 Cirrus Logic (audio codec) - (ARM Partner)
g ( ) ( )
 AKM (Magnetic Sensor)
 Texas Instruments (Touch Screen Controller and mobile DDR) - (ARM Partner)
 Invisible Design-Team Members ...
g
 OS & Drivers, GSM Security; Graphics, Video and Sound ...
 Manufacturing, Assembly, Test, Certification ...
33 Source ... http://www.ifixit.com

34. Exciting Technology ... Inside ‘The Chip’
Memory
‘Package’
2 Memory Dies
Glue Processor SOC Di
P Die
4-Layer Platform
Package
Package’
The A4 SIP Package (Cross section)
(Cross-section)
 The processor is the centre rectangle. The silver circles beneath it are solder balls.
 Two rectangles above are RAM die, offset to make room for the wirebonds.
 Putting the RAM close to the p
g processor reduces latency, making RAM faster and cuts p
y, g power.
 Unknown Mfr (Memory)
 Samsung/ARM (Processor)
 Unknown (SIP Technology)
34 Source ... http://www.ifixit.com

35. The Phone: Hetrogeneous Computation ...
• About 20 Chips in a Smart-
Phone
• Processing:
• Audio, Video, RF,
Touch, Temperature,
Orientation, G-Force,
Orientation G Force
Magnetism, Power
• Core Functions:
• GSM GPS WiFi
GSM, GPS, WiFi,
3/4G Net, BlueTooth
• Application Functions:
• Applets Games Mail,
Applets, Games, Mail
Diary, Address-book,
etc.
... Multi Processing before we
Multi-Processing
open the ‘App’n Processor’!
... Partitioning: The difference between a good and bad Product!
35

36. Commodity HMP In Qual. Today...
Pocket ‘Super-Computer’ ...
 10 Programmable Processors
 4 x A9 Processors (2x2): ~10,000 MIP
P (2 2) 10 000 Block-Diagram for a typical 40nm Mobile
 4 x MALI 400 Fragment Proc: ~1Gp/s Computing & Smart-Phone Platform Chip
 1 x MALI 400 Vertex Processor
 1 x MALI Video CoDec
 Plus Dedicated Processors
 Smart MMUs
 Smart Interrupt Controllers
 Smart DMA Engines
 Smart QoS and Power Mgt
 Smart Cache & Memory Repair
Plus ...
 Customer Additions/Peripherals!
... ~15 Proc ~2GHz 1-2W
Proc., ~2GHz,
... Strong Application Focus
36

37. 30nm Target Architectures
About 50MTr
About 50KTr
... D li i ~5x speed (A hit t
Delivering 5 d (Architecture + P
Process + Cl k)
Clock)
37

38. There’s D
Th ’ Design;
and there’s
Technical Design ...
38

39. Design A Part-Formalised Process ...
 Partition and Refine until every Thread has identified an Established
(Reuse) path to Physical Implementation
 Then C
Th Construct and V if ...
t t d Verify
Concept Phone Actual Phone
Known-Links from
K Li k f
AAA cell
TFT-LCD
Model-to-Reality (Reuse)
 High
Electronics
GaAs Front End
H
Std Radio Chip
Low  Hierarchy of Reuse
Baseband
Signal Processing
ARM CPU
MALI MPU
H
HW Support
FP Engine
Gates/MC.Code
39

40. Functional
F2
F1 F5 Analysis
A l i
F4
F3
(F2) ( )
(F5)
Thread
(F1) (F3)
HW1 HW2 HW3 HW4 RTOS/Drivers
Hardware I t f
H d Interface
Bus(es) Processor(s)
Execution Pl tf
E ti Platform
40

41. The Real-Time Execution of Models
 It is about creating a Functional-Model and an Execution-Platform
for it, to meet Functional and Non-Functional needs.
 D i A hi
Design: hierarchical mathematical process of M d l R fi
hi l th ti l f Model Refinement and
t d
Verification; based on (Heuristic) Architectural decisions.
 Implementation: Process of ‘bringing up’ and Validating the Functional-
Model on the Physical Execution-Platform.
 A good Solution is one that ...
1. Meets a valued human need
2. Is Manufacturable to support a Competitive Price/Biz Model
3. Works at least as well as your Competitors
4. Scores well on Aesthetic (Non-Functional) criteria
 A bad Solution is one where the Technology Shows!
41

42. And so to
Multi-P
Multi-Processors ...
42

43. The Argument for (C)MP
 Potentially Much Better Power Efficiency than Large/Fast Uni-Processor
 Power is a Major Problem ... On Die and In System.
 But 10x-100x improvement required!
 Potential to deliver Higher Performance than Uni-Processor
 Can Amdahl’s Law be broken?
Amdahl s
 For GP applications difficult to see improvement after 3/4 processors
 Potential to handle Redundancy Schemes
 Many Processor ‘Tiles’ with NoC Connectivity
 Potentially a small % malfunction can be ‘re-routed’
 Potential to offer a Scalable Standard Implementation
Scalable,
 Reduces Chip Design/Masks/Qual’n/Production Cost by ~90%
 Needs a new (tbd!) GP Software Methodology
 Must work with Legacy (90% of designs are inherited).
... Can CMP actually deliver any (let alone all) of this?
43

44. Amdahl’s Law is Alive and Well
32 100% parallel
28 Speedup on parallel processors is limited by the
sequential portion of the program
24
peedup
20
Sequential portion need not be large
to significantly constrain speedup!
Maximum sp
16
12 95% parallel
M
8
90% parallel
4
75% parallel
50% parallel
0
0 4 8 12 16 20 24 28 32
Number of cores
44

45. Many (C)MP Technologies ... even in the UK
 Transputer (Inmos 1978)
 Highly Parallel Apps (Graphics)
 Lots of history; no success ...
y
 Pixelfusion –
 1,536 processors/chip
 Clearspeed –
p
 192 full 64 bit arch/chip
 Picochip –
 Aimed at 3G Pico-Cells
 Spinnaker –
 18 processors (Scal. to 1^6)
 18,000 neurons (Scal. To 10^9)
 XMOS –
 New version of Transputer
 Occam
 HandleC – HW synthesys (See also SystemC)
 OpenCL – Smartphone's / Tablets (GPGPU)
... Success(?) requires a End-Product and Market of appropriate Scale.
45

46. OpenCL Enables Heterogeneity
 Plug-in architecture
 Mali-T604 Vithar GPU
 ARM/NEON
 Custom Device
 Video Decoder
 Discovery of
computational units
 Scheduling of work
... It does not automatically
solve “which computation
where”
46

47. Architecture: A Viable Mix of Technology
 YES: Power is a Major Concern ...
 Power-Efficiency the way of recovering it.
 Away from.. ..towards.. ..wherever possible.
 But so is Productivity, NRE Cost, TTM, Quality ...
 Reuse: As much as possible:
Mech, Elect, SW, Acoustic, RF, Stacks, OS, Displays, Keyboards, etc.
 Teams: Use people who know how to do the work (duh!)
 Use External Expertise: It is seldom a differentiating factor in your Product.
 Producible: Make something that can be economically made (duh!)
 Performance: Competitive; don’t push the bounds of possibility.
don t
 New Technology: As little as possible.
 And so are Aesthetics ...
 Colour, Style, Package, Availability, Quality, Business Model, etc ...
... Remember: The Product is the way to deliver
a Compelling End-Customer Experience.
47

48. Conclusions
 Multi-Processing makes sense in lots of Products today...
 It will seldom be entire solutions (ie: Small markets)
 It will seldom be homogeneous
 If the Work-Load and Programming Models are good.
 Physical Concurrency makes sense in lots of Products
Products...
 Mechanical, CPU/GPU, Optical, RF, MEM, SAW, etc
 Simplifies Productivity, Design, Qualification, Quality and Reuse
 Few Products (None) have the luxury of a Clean-Sheet Design...
 Legacy is unavoidable
 CPU is not the answer for everything...
everything
 ‘Software’ is amongst the least energy efficient technology
 DSP, Video HW and GPU can be (much) better
 But Analogue and Mechanical are best
... Products can be Enabled or Disabled MP Technology
48

49. The END
Th EN ...
Th k for L
Thanks f Listening
49

50. Reading & References
 The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail (Disruptive Tech.)
 by Clayton M. Christensen: HBS Press, 1997
 Open Innovation: The New Imperative for Creating and Profiting from Technology (Research in 21C)
 by Henry Willi
b H William Ch b
Chesbrough : HBS Press, 2003
h P
 The World Is Flat (Globalisation)
 by Thomas L. Friedman: Penguin, 2005
 Staying Power (Business)
y g ( )
 by Michael Cusumano: Oxford, 2010
 A Short History of Nearly Everything (A different view on what we know)
 by Bill Bryson: Black Swan, 2003
y y
 The Voyages of the Beagle (Scientific Observation)
 By Charles Darwin,1860
 An Essay on the Principles of Population (Natural Competition)
 By Thomas M l h 1789
B Th Malthus,1789
50