SUMMARY

CPU

PD / KEIT PD / KEIT PD

CPU: vs ARM

PC OS , x86 CPU ARM CPU

x86 Atom ,

PC SW

ARM v7 Out-of-order , v8 64 , big.LITTLE ,

CPU Application Processor(AP)

, AP , 2011 82$ 2015 362$ PC CPU

2-3 , IT

, / (Heterogeneous core), big little

, , SW

54 55

PD ISSUE REPORT JULY 2012 VOL 12-6

1.

CPU( )

CPU //UMPC PDA PC , SoC AP(Application Processor)

- CPU, GPU (OS) ,

AP CPU, 2D/3D GPU, ISP(Image Signal Processor) , DSP, IP

* VPU(Visual Processing Unit or Video Processing Unit)

[ 3-1] AP

MPU GPU

DSP

I/D Cache

SIMDShaderShader

ShaderShader

2D3D

Connectivity

Timer

Keypad

I2S/AC97/SPDIF

USB/IrDA/MMC

Haptics

UART/I2C/SPI

SDRAM/Flash

GSM/HSDPA

HQ Audio Decoder

Effector(BBE)

WCDMA

HQ Audio Encoder

Codec modules

MMP

Control processor

Video Encoder

Video Decoder(720p, 1080p)

ISP, JPEG

BB

- CPU OS , SW , ARM Cortex-A

- GPU OpenVG, OpenGL ES Imagination Technologies PowerVR ARM Mali GPU IP

- H.264, MPEG2, DivX, Xvid IP

54 55

ISSUE 3 CPU

CPU

CPU PC ARM

- CPU , PC 2011 5 , ARM 500$

ARM [1, 2]

- Calxeda 32- ARM 64- ARM AppliedMicro

- , 4 Lava International Atom

-

CPU 2006

[ 3-2] Pollack's rule

4 2.8 4

4 4 4

1 1 1

2 2 2

2 1. 4 2

CPUCPU

Freq

uenc

y

Freq

uenc

yArchitectureEnhancement

Tran

sist

or

Tran

sist

or

Pow

er

Pow

er

Perf

orm

ance

Perf

orm

ance

Perf

orm

ance

Wat

t

Perf

orm

ance

Wat

t

Die size(Transistors) 2xPerformance 1.4-1.5x

* : 25 5(2010. 10)

- Pollacks rule

* Pollacks rule: () 2 1.4( ) , ,

56 57

PD ISSUE REPORT JULY 2012 VOL 12-6

ARM CPU

- PC ,

- CPU HD , AV , , , 3D

vs. ARM

CPU PC x86 ARM

- PC CPU 86%, AMD 12% x86 , MS OS CPU

x86 CISC(Complex Instruction Set Computing) ARM RISC (Reduced Instruction Set Computing)

* CISC RISC , RISC

* 30 RISC CISC

, , ARM

- Out-of-Order

* Out-of-Order Xeon In-Order Atom 2.5 [2]

ARM

*

- Calxeda ARM , Xeon 1/10 1.5W ,

56 57

ISSUE 3 CPU

* Calxeda HP 8 288 Calxeda ARM

- , 3-4 3D

PC , OS

- AP ARM , (, , , , TI 600 ) ARM AP

/ x86 ARM

- x86 Medfield CPU, GPU, , I/O 2012 22 3D Tri-gate Medfield 2013

- ARM CPU , IDM(Integrated Device Manufacturer)

(Win-Tel) MS ARM Windows 8 Windows RT(RunTime) CPU

- Windows 8(Windows RT) ARM CPU Windows SW ARM , MS Silverlight, Adobe Flash10 AIR, Mozilla Firefox Windows 8 ARM SW ARM CPU

- 2015 13% , PC(/) ARM 8%

* PC CPU 107$ 20$ , 2015 2.2B$

58 59

PD ISSUE REPORT JULY 2012 VOL 12-6

2. CPU

IT AP

AP , CPU SoC 5

- AP 45.0% 2015 362$

- AP , 2015 AP PC CPU( 342$)

, , 3D PC CPU

, AP 2011 (4.9 ), (5,500 ), TV(700 ) , 2015 (9.6 ), (2.6 ), TV(0.9 )

8 16 MCU(Micro Controller Unit) , AP , PC CPU

- AP , 2015 (0.3 ), (0.3 ), (0.2 ) AP

AP PC CPU , AP

- 2009 13% 2015 48% 9.6 ,

58 59

ISSUE 3 CPU

[ 3-1] AP

2009 2010 2011E 2012E 2013E 2014E 2015E

Device

( )

Total 890 1,125 1,384 1,615 1,848 2,022 2,276YoY 11.9% 26.5% 23.0% 16.7% 14.5% 9.4% 12.5%PC 316 363 381 398 425 451 478

180 307 485 620 749 845 960 4 16 55 109 155 191 256

TV 170 205 217 226 238 242 268 156 162 169 176 185 191 198

64 72 77 83 88 92 95 - - - 3 8 10 20

AP

PC - - - 1% 8% 12% 15% 80% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%

TV - 1% 3% 8% 15% 25% 35% - - - 1% 3% 12% 18%

1% 3% 4% 7% 10% 16% 25% - - - 100% 100% 100% 100%

AP ( )

Total 148 328 550 761 996 1,198 1,461YoY 43.5% 121.4% 67.8% 38.3% 30.9% 20.3% 21.9%PC - - - 4 34 54 72

144 307 485 620 749 845 960 4 16 55 109 155 191 256

TV - 2 7 18 36 61 94 - - - 1 6 23 36

1 2 3 6 9 15 24 - - - 3 8 10 20

Device ()

PC - - - 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

TV - 1 1 1 1 1 1 - - - 1 1 1 1

1 1 1 1 2 2 3 - - - 1 1 1 1

AP ( )

Total 148 328 550 761 1,005 1,213 1,509YoY 43.5% 121.4% 67.8% 38.3% 32.1% 20.7% 24.4%PC - - - 4 34 54 72

144 307 485 620 749 845 960 4 16 55 109 155 191 256

TV - 2 7 18 36 61 94 - - - 1 6 23 36

1 2 3 6 18 29 71 - - - 3 8 10 20

ASP() 18 16 15 18 20 22 24YoY -10.0% -11.1% -6.3% 20.0% 11.1% 10.0% 9.1%

AP ( ) 27 52 82 137 201 267 362CPU ( ) 323 355 384 407 399 369 342

YoY(%, AP) 29.2% 96.8% 57.3% 66.0% 46.8% 32.8% 35.7%YoY(%, CPU) -1.8% 9.8% 8.2% 6.1% -2.2% -7.3% -7.3%

* : Gartner,

60 61

PD ISSUE REPORT JULY 2012 VOL 12-6

- PC PC CPU AP

20W x86 ARM CPU AP 0.5W ,

- 46.9% 2015 2.6

- , , OS AP

7 , , AP AP 1 5 AP

- , AP OS 1.7

AP 2015 18%(3.6 ) AP

PC

- PC AP Windows 8(Windows RT) ARM CPU AP

PC CPU 3.0GHz, , 64 , ARM AP 1.5 GHz, , 32 5 ,

Windows 8(Windows RT) ARM

ARMv7 Cortex-A15 2GHz ,

- ARM AP CPU SoC ,

CPU 608$, AP 499$ 18% , SoC 20$

60 61

ISSUE 3 CPU

, 40% 600$ ,

- ARM AP

CPU AP

PC , AP HD AP PC CPU

- AP CPU ( )

* AP A5(122mm2) A4(53mm2) 2

3. CPU

CPU

Intel Architecture(IA) 1978 PC CISC(Complex Instruction Set Computing) 8086 32 IA-32 64 x86 -64 [3]

* IA (ISA: Instruction Set Architecture) x86 x86 ISA AMD, Via ownership IA

* IA-32 x86 32 1986 80386 2000 Pentium4 [4]

* IA-64 1994 HP VLIW(Very Long Instruction Word) EPIC(Explicit Parallel Instruction Computing) 64 Itanium . IA-64 x86

* x86 64 IA-64 x86-64 AMD AMD64 Athlon 64, Opteron . AMD64 IA-32e, EM64T 2006 Intel 64 Intel 64 2004 Xeon [5]

- CPU Atom x86

62 63

PD ISSUE REPORT JULY 2012 VOL 12-6

- Atom Core2Duo I/O Controller Hub(ICH) Memory Controller Hub(MCH) System Controller Hub(SCH)

Atom IA-32 x86-64 CPU 2008 45 CMOS , UMPC, 32 High-K Metal Gate (HKMG) [6]

- , x86 PC

Atom 2-issue x86 CISC RISC In-order Bonnell

- Bonnell ALU load/store , P6 Netburst

- CPU P5 i486 , hyper - threading

[ 3-3]

Atom Processor

CPU

SystemMemory

Memory, GFX, High SpeedI/O, Legacy and LowerSpeed I/O Controller

Peripherals

BIOSHigh SpeedI/O

FSB

Components : IOH I/O Hub

ICH I/O Controller Hub

SCH System Controller Hub

MCH Memory Controller Hub

Interfaces : FSB Front Side Bus

QPI Quick Path Interconnect

DMI Direct Media Interface

Core 2 Duo Processor Core i7 Processor

High Speed I/O

Controller

Legacy andLower

Speed I/OController

SystemMemory Peripherals

BIOSHigh SpeedI/O

CPU

IOH

DMIIntel

OPI

ICH

SystemMemory

Memory,GFX & HighSpeed I/OController

Legacy andLower

Speed I/OController

Peripherals

BIOSHigh SpeedI/O

FSB DMICPU

1 45 UMPC/MID Silverthorne(Atom Z5xx) Classmate PC Diamondville(Atom N2xx)

62 63

ISSUE 3 CPU

- Atom , Sony Atom

2 45 Pineview HD GMA X3100 GMA 3100 GPU 7W 40% 6 10

- Pine Trail-M Atom Pineview-M Tiger Point Platform Controller Hub , 3 2

- 2010 1 / N4xx DDR2 GPU , hyper- threading , Menlow Atom 1/2

2011 5 Whitney Point Moorestown Oak Trail 2 MID Lincroft

32 3 Cedarview 2011 11 , NM10 Southbridge Cedar Trail

- 3 Blu-ray 2.0 DirectX 9 GPU , 1080p , HDMI DisplayPort , TDP

3 MID UMPC Medfield 3D HD 2012 5

- 240M pixels per second ISP(Image Signal Processor), low frequency mode, high frequency mode, max frequency mode

[ 3-4] ATOM

64 65

PD ISSUE REPORT JULY 2012 VOL 12-6

[ 3-5] Medfield (Z2460)

CPU

2D/3D Graphics

Video Decode1080p30

I/O

MIPI-HSIUARTULPI

I/O

LPDDR2eMMC

MIPI-DSIHDMI 1.3aMIPI-CSI

UART

Video ImageEnhance

L2 $512KB

LP-DDR2Ctrl

Video Encode1080p30

Display Ctrl3 Pipes

Image Signal ProcessorProgrammable

IOSF - OCP Bridge

Security Programmable ExecutionEnvironment & Cryto Engine

[ 3-2] ATOM

Platform

- Stealey A100 90nm - TDP: 3W : 600, 800MHz

1

Dimondville*(2008. 03)

N2xx 45nm - TDP: -2.5W : 1.6GHz FSB: -533MHz

2xx 45nm - TDP: 4W 64 - bit

Silverthrone**(2008. 03) Z5xx 45nm menlow

TDP: 0.65 and 2.4W( 0.1W) : 0.8 to 2.0GHz FSB : 533MHz

Dual Dimondville*(2008. 09) 3xx 45nm, dual -

TDP: 8W : 1.6 to 1.66GHz(64-bit) FSB: -667MHz

2

Pineview*(2009. 12)

N4xx 45nm

Pine Trail

TDP: 6.5W : 1.83GHz(64-bit) GPU: 200MHz

D4xx 45nm TDP: 10W : 1.66GHz(64-bit) GPU: 400MHz

N5xx 45nm, dual TDP: 8.5W 64 -bit GPU: 200MHz

D5xx 45nm, dual TDP: 13W : 1.66GHz(64-bit) GPU: 400 MHz

Lincroft**(2010. 05) Z6xx 45nm Moorestown&Oak Trail

TDP: 1.3-3W GPU: 400MHz

64 65

ISSUE 3 CPU

Platform

3

Cedarview*(2011. 11)

D2500 32nm, dual

Cedar Trail

TDP: 10W 64-bit GPU: 400MHz

D2700 32nm, dual TDP: 10W 64-bit GPU: 600MHz

N2600 32nm, dual TDP: 3.5W : 1.6GHz(64-bit) GPU: 400MHz

N2800 32nm, dual TDP: 6.5W : 1.86GHz(64-bit) GPU: 600MHz

Medfield**(2012. 05) Z24xx 32nm -

: 1.6GHz GPU: 400MHz

* : //Classmate PC, ** : MID, UMPC,

* TDP: Thermal Design Power

4. ARM

ARM

RISC ARM feature Thumb ARMv4 ARM7-TDMI(ARM7-Thumb+Debug+Multiplier+ICE)

2011 ARM ARMv7 Cortex ,

- 2011 10 ARM Cortex-A7 big.LITTEL energy

- Application A-Profile(ARMv7-A), Real-time R-Profile (ARMv7-R), Mid-end Low-end Micro-controller M-Profile(ARMv7-M)

- A-Profile Cortex-A8 Cortex-A9 SMP(Symmetric Multi-Processor) , , ARM926 Cortex-A5 ,

- ARMv7 ecosystem , SW backward compatible

66 67

PD ISSUE REPORT JULY 2012 VOL 12-6

-

ARMv7, ARMv8 , ARM PC CPU

- Cortex-A9 ARM PC , ARM

, ARM 32 ARMv7 Cortex-A15 LPAE(Large Physical Address Extensions) (Virtualization)

- 32 4KB 40 LPAE , CPU GPU 4GB

- , feature SW

ARMv8 [7]

ARMv8 Cortex-A9 Cortex-A15 32 ARMv7 64 (virtual addressing)

- ARMv7 Real-time(R-Profile) Micro-controller(M-Profile) ARMv8 Cortex A-Profile(ARMv7-A)

- A-Profile Cortex-A15 LPAE , LPAE 40 , ARMv8-A 48 , Cortex-A15 SW

ARMv8 64 A64 , SW

- Backward compatibility , 64 SW 64

66 67

ISSUE 3 CPU

A64 32 , (operand)

- JIT(Just-in-Time) , 31 64

- , LDM/STM(load/store multiple) ,

- SW , SIMD 64

ARMv8 64 (Exception) OS TrustZone Hypervisor HYP ARMv7-A Cortex-A15

- VM hypervisor 32 64

AArch64 MMU(memory management unit) 48 Cortex-A15 , 48 4

- AArch64 4KB 64KB 4 2 , 64 4GB 256TB

- Cortex-A15 , ARMv8 MMU (VA: Virtual Address), (IPA: Intermediate Physical Address), (PA: Physical Address)

A64 30 , X0-X30 64 ,

- , X30 PLR(Procedure Link Register) 32 64 32 64 OS Hypervisor 32

ecosystem tool SW, ARMv8 OS A-Profile

- 32 64 Linux OS , Linux 64 , ecosystem 64 OS 32

68 69

PD ISSUE REPORT JULY 2012 VOL 12-6

[ 3-6] ARM

ARMv8-A

A32+T32 ISAs

AArch32

A64 ISA

AArch64

CRYPTOARMv8 - A-profile only (at this time) - 64-bit architecture support

CRYPTO

including:- Scalar FP (SP and DP)- Adv SIMD (SP Float)

including:- Scalar FP (SP and DP)- Adv SIMD (SP+DP Float)

Key featureARMv7-A

compatibility

Thumb-2

VFPv2

TrustZone

Jazelle

SIMD

ARMv5

VFPv3/v4

NEONAdv. SIMD

ARMv5 ARMv5ARMv6 ARMv-7A/R

big.LITTLE

CPU (big) CPU (LITTLE) Cortex-A15 Cortex-A7

[ 3-7] big.LITTLE Cortex-A7 Cortex-A15

Performance and Energy-Efficiency

LITT

LEbi

g

Most energy-efficient processor from ARM Cortex-A7

Cortex-A15

- SImple, In-order, 8 stage pipeline- Performance better than today's mainstream, high-volume smartphones

- Complex, out-of-order, multi-issue pipeline- Up to 5x the performance of today's mainstream, high-volume smartphones

Highest performance in mobile power envelope

- Cortex-A CPU 1-4 CPU L2 , SCU(Snoop Control Unit) , big.LITTLE 2 Cortex-A15 2 Cortex-A7

68 69

ISSUE 3 CPU

- Cortex-A7 Cortex-A15 , Cortex-A15 LPAE

[ 3-8] big.LITTLE

GIC-400

CCI-400 (Cache Coherent Interconnect)

Interrupts

Memory Controller Ports System Ports

Interrupts

Cortex-A15Core

Cortex-A7Core

Cortex-A15Core

Cortex-A7Core IO

CoherentMasterL2 L2

big.LITTLE coherency Cortex-A15 Cortex-A7 L1 L2 ,

- coherency ARM CCI-400 GIC(Generic Interrupt Controller)-400

OS Task Migration Model

- Task Migration OS DVFS(Dynamic Voltage and Frequency Scaling) ( ) , Cortex-A15 Cortex-A7

* : Cortex-A15 Cortex-A15 Cortex-A7

- Inbound Outbound L2 Inbound , Outbound Outbound

70 71

PD ISSUE REPORT JULY 2012 VOL 12-6

[ 3-9] Cortex-A15, Cortex-A7 DVFS Curves

Power

Overdrive condition

Highest Cortex-A15 Operation Point

Lowest Cortex-A15 Operating Point

Highest Cortex-A7 Operating Point

Lowest Cortex-A7 Operating Point

Performance

Cortex-A15

Cortex-A7

CPU MP Cortex-A15 Cortex-A7 , OS Dispatcher big.LITTLE OS

big.LITTLE , LG, TI, , , ST,

[ 3-10] ARM CPU-GPU

Scalable Mobile Processor Evolution25

20

15

10

5

0

2009 2010 2011 2012 2013 2014 2015 2016

Superphone

Entry LevelCortex-A8Mali-200

Cortex-A8Mali-300

Cortex-A8Mali-400 MP

Dual Cortex-A9Quad Mali-400 MP

Cortex-A5Mali-300

Quad Cortex-A7Quad Mali-400 MP

Dual Cortex-A7Dual Mali-400 MP

Cortex-A5Mali-400 MP

Quad Cortex-A9Quad Mali-T604

Dual Cortex-A15Dual Cortex-A7Quad Mali-T658

Dual Cortex-A16Dual Cortex-A7Eight Mali-T658

Cortex-A15Cortex-A7Dual Mali-T604

Rela

tive

Perf

orm

ance

Mid Range

70 71

ISSUE 3 CPU

[ 3-3] ARM

Cache(I/D), MMU Typical MIPS

ARM11

ARMv6 ARM1136J(F)-S variable, MMU 740@532665 MHz(i.MX31 SoC), 400528 MHz

ARMv6T2 ARM1156T2(F)-S variable, MPU -

ARMv6Z ARM1176JZ(F)-S variable, MMU+TrustZone965 DMIPS@772 MHz,

up to 2 600 DMIPS with four processors

ARMv6K ARM11 MPCore variable, MMU -

Cortex-A ARMv7-A

Cortex-A5(MPCore)

4-64KB/4-64KB L1, MMU+TrustZone 1.57 DMIPS / MHz per core

Cortex-A7 MPCore

32KB/32KB L1, 0-4MB L2, L1&L2 have Parity&ECC,

MMU+TrustZone-

Cortex-A816-32KB/16-32KB L1,

0-1MB L2 opt ECC, MMU+TrustZone

up to 2,000(2.0 DMIPS/MHz in speed from 600 MHz to greater than 1 GHz)

Cortex-A9 MPCore

16-64KB/16-64KB L1, 0-8MB L2 opt Parity,

MMU+TrustZone

2.5 DMIPS/MHz per core, 10,000 DMIPS@2 GHz on Performance

Optimized TSMC 40G(dual core)

Cortex-A15 MPCore

32KB/32KB L1, 0-4MB L2, L1&L2 have Parity&ECC,

MMU+TrustZone

At least 3.5 DMIPS/MHz per core (Up to 4.01 DMIPS/MHz depending on

implementation)

Cortex-R ARMv7-R

Cortex-R40-64KB/0-64KB L1, 0-2 of 0-8MB TCM,

opt MPU with 8/12 regions-

Cortex-R5 (MPCore)

0-64KB/0-64KB L1, 0-2 of 0-8MB TCM,

opt MPU with 12/16 regions-

Cortex-R7 (MPCore)

0-64KB / 0-64KB L1, 0-128KB of TCM,

opt MPU with 16 regions-

Cortex-M

ARMv6-M

Cortex-M0 No cache, No TCM, No MPU 0.9 DMIPS/MHz

Cortex-M1 No cache, 0-1024KB I-TCM, 0-1024KB D-TCM, No MPU136 DMIPS@170 MHz,

(0.8 DMIPS/MHz FPGA-dependent)

ARMv7-M Cortex-M3 No cache, No TCM, opt MPU with 8 regions 1.25 DMIPS/MHz

ARMv7-ME Cortex-M4 No cache, No TCM, opt MPU with 8 regions 1.25 DMIPS/MHz

72 73

PD ISSUE REPORT JULY 2012 VOL 12-6

ARM AP

Windows PC OS , OS

- OS 2010 38%, 23%, RIM 17%, iOS 16%, Windows Mobile 4% , 2015 43.8%, iOS 16.9%, Windows 20.3%, RIM OS 13.4% OS

ARM x86 IP ARM AP

[ 3-4] x86 vs. ARM

x86 Atom ARM Cortex-A9 AP

/ 35$ 15$

70mm2 45nm 55mm2 45nm

/ 950 8 1,150 8

ARM

ARM IP

- , ARM

* , MS, ,

ARM AP

ARM AP , AP

- PC CPU , 2010 52% 1, TI 23%, 13% 2, 3

- 2007 AP AP 1 , 2010 , AP Baseband AP

72 73

ISSUE 3 CPU

* AP , AP

- AP TI AP AP , TI AP

AP ARM , AP

- AP , 1.2-1.5GHz ,

5.

CPU

AP CPU Cortex-A , Atom //, MIPS DTV

[ 3-5]

()

ARM(Cortex-A8/A9/A15), MIPS(MIPS32), IBM(PowerPC), (Atom), ITC(Godson, ),

Ceva(Ceva-X), Tensilica(Xtensa), ADChips(EISC)

AP(A4, A5), (SnapDragon), (S5PC111, Exynos), Freescale(i.MX53), AMD(Au1250), (Tegra2/3/4), (MV8657), ADChips(EAGLE)

( )

(iPhone, iPad), (Galaxy S/S2), HTC(Desire), LG(Q), RIM(Blackberry), (N95), (Moto QRTY), MS(KIN)

[ 3-6] ARM CPU

ARM11 ARM Cortex-A7 ARM Cortex-A8 ARM Cortex-A9 Scorpion Krait

Decode single-issue partial dual-issue 2-wide 2-wide 2-wide 3-wide

Pipeline Depth 8 stages 8 stages 13 stages 8 stages 10 stages 11 stages

Out of Order Execution N N N Y Partial Y

Pipelined FPU Y Y N Y Y Y

74 75

PD ISSUE REPORT JULY 2012 VOL 12-6

ARM11 ARM Cortex-A7 ARM Cortex-A8 ARM Cortex-A9 Scorpion Krait

NEON N/A Y(64-bit wide)Y

(64-bit wide)Optional MPE (64-bit wide)

Y(128-bit wide)

Y(128-bit wide)

Process Technology 90nm 40nm/28nm 65nm/45nm 40nm 40nm 28nm

Typical Clock Speeds 412MHz 1.5GHz 600MHz/1GHz 1.2GHz 1GHz 1.5GHz

iPhone 3GS AP 2008 P.A.Semi , 2009 Intrinsity

- Intrinsity Dynamic logic EDA SW Fast14 1GHz Cortex-A8 , Intrinsity S5PC111 , A4 iPad iPhone4, TV CPU

- 2011 2 iPad2 ARM Cortex-A9 CPU A4 PowerVR 545 GPU A5

- A AP iPhone iPad

iPad 3(The new iPad) ARM Cortex-A9 A6 , Cortex-A9

[ 3-11] Apple AP(A4, A5, A5X) die size

A4(53.3mm2)7.3mmx7.3mm

ChipworksApple A4 Polysilicon

ChipworksApple A4 Polysilicon Apple A5x Polysilicon Die Photo

source: chipworksDDR SDRMInterface

ProcessorData Path

2 GPIO

GPIOGPIOUSB

Arm Core Arm Core

Audio

VideoDAC

WiFi

I/OI/O

ProcessorData Path

1

ProcessorData Path

2

ProcessorData Path

1

DigitalLogic

Blocks

GPUCore

GPUCore

GPUCore

GPUCore

GPUSpine

(preliminary)

GPUSpine

(preliminary)

ARMCortex9

Core

ARMCortex9

Core

A5(122.2mm2)10.09mmx12.15 mm

A5X(165mm2)12.90mmx12.79mm

* : www.chipworks.com

74 75

ISSUE 3 CPU

- 2011 2 iPad2 iPhone 4S AP A5(S5L8940) 45 1GHz ARM Cortex-A9 CPU A4 PowerVR543-MP2 GPU die 122.2mm2

- 2012 3 A5 2 S5L8942 TV iPad2 2 , 32 S5L8940 41% 69.6mm2 die

- 2012 3 iPad3 1 GHz ARM Cortex-A9 CPU PowerVR543-MP4 GPU S5L8945 , 45 GPU 210%

1990 DEC MPU , 2000 iPhone AP AP

- 2007 iPhone 2G, 3GS AP , iPhone4 iPad AP AP

[ 3-12] Exynos

Display / Camera CortexA9 Dual core Memory I/F

High speed I/F

System Peripheral

Graphic / Video / ISP

Wireless BB

Modem I/F

2-ch MIPI CSI input

TV out : Composite

HDMI v1.3a output

Single WXGADual WSVGA

LPDDR2 / DDR2 / DDR3

USB Host 2.0

Dynamic addressing

SCU and ACP

CPU 01.0GHz

32KB / 32KBNEON

CPU 11.0GHz

32KB / 32KBNEON

1MB L2 Cache

JPEG HW codec

3D / OpenVG / 2D HW

1080p 30fps codec

GPS

eMMC4.4 DDR 8-bit

SATA 1/2

Memory Interleaving

32x DMA

PCI Express (SCP)

4PLLs

Timers / 4x PWM

SRAM / ROM/NOR

HSIC

LP co-processor

OneNAND / SLC

USB Dev 2.0

CPU cache coherence

HSIC, DPSRAM

SecureRAM

CryptoEngine

SecureROM

PPMU

External Peripheral

HS-SPI

14x8 keyboard

UART

MIPI Slimbus

SD2.0 / MMC4.3

I2C

I2C/PCMAC97/S/PDIF

Multi-Layer AXI/AHB Bus

76 77

PD ISSUE REPORT JULY 2012 VOL 12-6

Imagination Technologies 3D GPU , Intrinsity 2009 4 1GHz ARM Cortex-A8 Hummingbird (S5PC110 S5PC111) [8]

- S5PC111 MMP MFC(Multi-Format Video Codec)

2010 S5PC111 45 Cortex-A9 Mali-400 MP GPU Exynos 4210

- AP Exynos , Imagination Technologies PowerVR GPU ARM Mali GPU

- 2012 Exynos 4210 32 HKMG LP 4212, Cortex-A15 Mali-T400 Mali-T604 GPU 5250 , Cortex-A9 Mali-T604 GPU 4412 S3

- Mali-T604 GPU 4 Shader Tri-pipe OpenVG 1.1, OpenGL 1.1, 2.0 DirectX11 Mali-400 5

PC

- ARM CPU GeForce Tegra MID , MID

- 2003 , , Mesia Q, 2005 ULi Electronics, 2006 2D/3D Hybrid Graphics, 2007 , PortalPlayer CPU

60% AP , 3D

- Tegra, Tesla, PC GeForce

- ARM CPU IP IP, ARM11 Tegra600 ARM Cortex-A8 , ARM Cortex-A9 AP Tegra2

- , Shader 3D IP GPU Shader , ARM7 Tegra 2

76 77

ISSUE 3 CPU

[ 3-13] Tegra2

Cortex-A9Processpor

HD VideoDecode

Processor

HD VideoEncode

Processor

AudioProcessor

2D/3DGraphics

Processor

ARM7

Cortex-A9Processpor

ImageProcesspor

Cortex-A9Dual-Core

Image Processor - Still/Video Camera driver

Audio decoder(Tech, by acquiring Portalplayer 2007)

Video Encoder/Decoder - Encoding of 1080p H.264 - 1080p HP H.264 decode (NVIDIA says no one else can do 1080p decode power-efficiently like them)

Chip management - Dataflow, power management and other similar tasks

GPU- OpenGL ES 2.0 support- 32b 333MHz LPDDR2

Tegra 3 Cortex-A9 Cortex-A9 Companion , MPE(media processing engine) [9]

- Tegra 3 TSMC 40 LPG , Cortex-A9 G(generic) , Companion LP

* TSMC 40 40 LP

[ 3-14] 40nm LP(CPU B) G(CPU A) vs

CPU B

CPU A

POW

ER

PERFORMANCE

CPU on Fast Process-has higher leakage power in active standby

CPU on low power process-has lower leakage power but

consumes more power at hihger perfromance ranges

POW

ER

PERFORMANCE

Companion Core = OFFMain Cores = ON

Max Quad CorePerformance

Companion Core = ONMain Cores = OFF

- , Variable SMP OS , DVFS CPU Hot-Plug companion

78 79

PD ISSUE REPORT JULY 2012 VOL 12-6

vSMP companion activate coherency OS , hysteresis thrashing [9]

* Thrashing CPU

- Companion L2 nano sec. ,

- , vSMP OS CPU , voltage regulator ,

[ 3-15]

Single(Companion) core , , ,

Single core performance, 2D , ,

Dual core performance , ,

Quad core performance , ,

Core 1Companion

Companion

Companion

Companion

Core 2

Core 2

Core 2

Core 2

Core 1

Core 1

Core 1

Core 1 Core 3

Core 3

Core 3

Core 3 Core 4

Core 4

Core 4

Core 4

Core 1

Core 1

Core 1

Tegra 3 vertex shader 4 3D pixel shader 4 8 GPU

- ARM App. NEON Tegra 2 Tegra 3

78 79

ISSUE 3 CPU

[ 3-16] vSMP

100%

80%

60%

40%

20%LPO* MP3 Playback HD video playback Gaming

Tegra 2 Project Kal-EI

28%14%

lower

61%lower

34%lower

* LP0:

CPU AP

- 2011 5 USB 3G/4G Icera , CPU Tgra+Icera

- OS OS 3.0 Tegra 2 Tegra AP , Windows RT Tegra 3 kal-El+ Wayne OS

[ 3-17] Tegra Roadmap

T2

2011

Smar

tpho

necl

amsh

ells

Tabl

ets

Supe

rpho

ne

2012 2013

Kal-EI+

Tegra+Icera

T2

Tegra Roadmap

Kal-EI

Tegra+Icera

Grey

Wayne

80 81

PD ISSUE REPORT JULY 2012 VOL 12-6

Baseband AP Baseband one-chip MSM , SnapDragon

SnapDragon , , , Windows OS AP, 1GHz Baseband AP

- SnapDragon MSM8660 1.5GHz ,

ARM ARMv7(Cortex-A8 ISA) Scorpion ARM

- SnapDragon CPU ARMv7 1.5GHz Cortex-A8 , AMD Adreno GPU GPU

- 2011 Krait Adreno225 GPU AP Krait 28 , 1.7GHz Cortex-A9 Krait CPU Adreno320 GPU APQ8064 LG

- 65 S1 , 45 S2 S3 , Krait CPU SnapDragon S4

[ 3-18] SnapDragon S4

Memory (MCP)Memory (MCP)

SDRAMSDRAM

NAND FlashNAND Flash

Multi-core CPUBluetooth

2D/3D GraphicsWiFi

Multimedia Codec

2G/3G/4G Cellular

RF TransceiverRF Frontend

(Antenna, LNA, PAs)

GPS Connectivity

AP (Application Processor)

DSP

Mobile Broadcasting

Display/imaging/Memory Support

DigitalBasdband

Modem(2G/3G/4G Celluar

+ WiFi/BT/FM/GPS)

A/DD/A

(Analog + Digital)

, AP

Snapdragon:(+AP )

80 81

ISSUE 3 CPU

[ 3-19] SnapDragon S4

LTE WorldModem

AdrenoGPU

KraitCPU

VeNum

L1 Cache

L2 Cache

Snapdragon System Fabric

Dual Channel Memory

Snapdragon Adaptive Power Technologies

L1 Cache

VeNum

KraitCPU

HexgonDSP

HexgonDSP

MultimediaProcessor

HexgonDSP

GPS/WiFi/BT/FM Audio/Video HWAccalerators

ModemSubsystem

MulticoreSubsystem

MultimediaSubsystem

SnapDragon S4 MSM8960 LTE Krait CPU SoC TSMC 28 LP S3

- Krait CPU Scorpion 1.6 , S4 S1() 8 [10]

[ 3-20] thermal vs. performance

Thermal Limit

POWER

PERFORMANCE

40G ARM hits thermal limit,starts throttling and results in

lower performance

Krait in 28LP sustains higherperformance with better

thermal performance

40nm G

28nm LP

Krait L2 (aSMP: asynchronous Symmetrical Multi-Processor)

82 83

PD ISSUE REPORT JULY 2012 VOL 12-6

- CPU ( ) , 25-40%

- Krait double-precision VFP(Vector Floating Point) SIMD(Single Instruction Multiple Data) ,

[ 3-7] SnapDragon S4 AP

S4 Play S4 Plus S4 Pro S4 Prime

CPU 1.7 GHz Dual ARM Cortex A5 CPU 1.7 GHz

Dual Krait CPU 1.7 GHz

Dual or Quad Krait CPU 1.7 GHz

Quad Krait CPU

GPU Adreno 230GPU Adreno 305GPU Adreno 320GPU Adreno 320GPU

Video FWVGA 1080p HD 1080p HD 1080p HD

Modem 3G/4G /

LTE3G/4G /

LTE 3G/4G / LTE 3G/4G

/ LTE

Camera 8 MP 20MP, 3D

20MP, 3D

20MP, 3D

GPS gpsOne Gen 7 gpsOne Gen8A gpsOne Gen8A gpsOne Gen8A

USB USB 2.0 USB 2.0

OTG(480Mbps)USB 2.0

OTG(480Mbps)USB 2.0

OTG(480Mbps)

Bluetooth BT3.x BT4.0 BT4.0 BT4.0

Wifi

802.11n(2.4/5GHz)

802.11n(2.4/5GHz)

802.11n(2.4/5GHz)

802.11n(2.4/5GHz)

45nm 28nm 28nm 28nm

Individual Chips

MSM8625MSM8225

APQ8060AMSM8960

MSM8660AMSM8260AAPQ8030MSM8930MSM8630MSM8230MSM8627MSM8227

APQ8064MSM8960T MPQ 8064

App. , TV,

Vertex, pixel geometry shader Unified Shader AMD GPU Adreno 200 GPU ,

82 83

ISSUE 3 CPU

- SnapDragon S4 Adreno 220 50% Adreno 225 GPU , Windows RT Adreno 320 GPU LTE S4 CES2012

TI(Texas Instruments)

TI 2011 Cortex-A9 OMAP4 , Cortex-A15 OMAP5 ARM AP

- TI , SoC , Tegra [11]

[ 3-21] OMAP5430 SoC

OMAP5 OMAP5430 OMAP5432 28 LP Cortex-A15 Symmetric Multi-processing (SMP)

- TSMC 40 1.3 GHz Cortex-A9 Tegra 3 2GHz Cortex-A15 DMIPS OMAP5 Cortex-A15

- OMAP5430 Cortex-A15 ARM Cortex-M4 , big.LITTLE Cortex-M4 16 Thumb/Thumb-2 32 (1-cycle 32 , SIMD )

84 85

PD ISSUE REPORT JULY 2012 VOL 12-6

OpenGL ES, OpenGL, OpenVG DirectX PowerVR SGX544-MPx GPU

- AP GPU 100-400 MHz 1-16 GPU , iPad 3 Cortex-A9 A5X SGX543-MP4 GPU , TI OMAP5430 SGX544-MP2 GPU

- 2012 6 MWC 2012 TI OMAP5430 iPad 3 1080p GLBenchmark 2.5 iPad3 34 FPS(on-screen) 43 FPS(off-screen), OMAP5430 38 FPS 45 FPS , SGX543 4 SGX544 2

[ 3-8] OMAP5 (5430, 5432)

OMAP5430 OMAP5432

Area-sensitive(, ) Cost-sensitive(, )

28

ARM Cortex-A15 Clock Speed (two) Up to 2 GHz

2D &3D Graphics ,

Video performance(2D) 1080p60 multi-standard

Video Performance(3D) 1080p30 multi-standard

Imaging Performance Up to 24 MP(MIPI CSI-3+ 3x MIPI CSI-2+ CPI interfaces) Up to 20MP(3x CSI-2+ CPI interfaces)

Memory Support 2xLPDDR2 2xDDR3/DDR3L

Peripheral Support UART(6x), HSIC(3x), SPI(4x), MIPI UniPortSM-M, MIPI LLI, HSI (2x)UART(5x), HSIC(2x), SPI(3x)MIPI

UniPortSM-M, MIPI LLI, HSI

14mmx14mm PoP 980 balls 0.4mm

pitch(240-ball, 0.5mm PoP)17mmx17mm BGA 754 balls 0.5mm

pitch (w/depop)

IP

AP CPU GPU IP , CPU 2011 Cortex-A9

- GPU GeForce GPU, Adreno GPU, PowerVR , ARM Mali

84 85

ISSUE 3 CPU

[ 3-9] AP IP

AP Embedded CPU GPU Fab.

A4(Apple) Cortex-A8(single core) PowerVR SGX 535 Samsung 45nm

A5(Apple) Cortex-A9(single core) PowerVR SGX 543MP2 Samsung 45nm

S5PC110(Samsung) Cortex-A8(single core) PowerVR SGX540 Samsung 45nm

Exynos(Samsung) Cortex-A9(dual core) ARM Mali Samsung 45nm

Tegra2() Cortex-A9(dual core) GeForce ULP (4 core) TSMC 40nm

OMAP5(TI) Cortex-A15(dual core) PowerVR SGX544-MP2 TSMC 28nm

MSM8655() Scorpion(ARMv7, single) Adreno 205 TSMC 45nm

MSM8960() Cortex-A9(quad core) Adreno GPU TSMC 28nm

[ 3-10] GPU

Adreno 225 PowerVR SGX 540

PowerVR SGX 543

PowerVR SGX 543MP2

Mali-400 MP4

GeForce ULP

Kal-El GeForce

SIMD Name - USSE USSE2 USSE2 Core Core Core

# of SIMDs 8 4 4 8 4+1 8 12

MADs per SIMD 4 2 4 4 4/2 1 1

Total MADs 32 8 16 32 18 8 12

GFLOPS @200MHz 12.8 GFLOPS 3.2 GFLOPS 6.4 GFLOPS 12.8 GFLOPS 7.2 GFLOPS 3.2 GFLOPS 4.8 GFLOPS

GFLOPS @300MHz 19.2 GFLOPS 4.8 GFLOPS 9.6 GFLOPS 19.2 GFLOPS 10.8 GFLOPS 4.8 GFLOPS 7.2 GFLOPS

2012 Windows RT CPU AP Exynos Snapragon S4, TI OMAP 5 Tegra 3

- Exynos , TSMC 28 TSMC 28 Snapragon S4, OMAP 5 , TSMC 40 Tegra 3 [12]

- Atom Z2460 22

86 87

PD ISSUE REPORT JULY 2012 VOL 12-6

[ 3-11] CPU

SnapDragon S4 OMAP6430 Exynos 5250 Tegra 3 Atom Z2460

28nm LP 28nm LP 32nm HKMG LP 40nm LPG 32nm HKMG LP

CPU Krait Cortex-A15Cortex-M4 Cortex-A15 Cortex-A9 x86 Atom

CPU (+1) (w/HT)

CPU 1.2GHz 2.0GHz 2.0GHz 1.4GHz 1.6GHz

L2 2MB 1MB - 1MB 512KB

LPDDR2 1066 LPDDR2 1066 LPDDR2 1600? LPDDR2 1066 LPDDR2 800

- 8GB - 2GB 1GB

8.5GB/s 8.5GB/s 12.8GB/s 4.26GB/s 6.4GB/s

GPU QualcommAdreno 305PowerVR

SGX544MPxARM

Mali-T604NVIDIA

ULP GeForcePowerVRSGX540

HD 1080p30 1080p60 1080p60 1080p30 1080p30

20MP 24MP - 32MP 24MP

2560x1440 2560x2048 2560x1600 2048x1536 (1280x1024)

2012 Q2 2012 Q3 2012 Q2 2011 Q4 2012 Q2

* : www.bodnara.co.kr

3 AP CPU,

- , , , AR(Augmented Reality) , OS

[ 3-22] AP

1 AP AMP+DSP()

2010 2015 2020

2 AP

3 AP Smart AP( )

- - , , - Sensing, AR, Recognition

MPU+DSP+MMP+GPU(, MID)

: Hummmingbird(+ ) Qualcomm: Snapdragon NVIDIA: Tegra 1, 2 (GPU )

6.

86 87

ISSUE 3 CPU

6.

CPU

IT CPU ,

20 , , , 1 20 [13]

- (Heterogeneous core),

- , , SW

vs ARM CPU

CPU ARM

- ARM v8 big.LITTLE , IDM(Integrated Device Manufacturer) / PC SW

CPU AP OS , x86 CPU ARM

- AP 82$ 2015 362$ , PC CPU

- ARM AP , , TI, ,

88 89

PD ISSUE REPORT JULY 2012 VOL 12-6

[]

1. Chipmakers ARM for Battle in Traditional Computing Market, Sixto Ortiz Jr., 2011 4, IEEE Computer

2. The high stakes of low power, Rachel Courtland, 2012 5, IEEE Spectrum

3. White Paper: Introduction to Intel Architecture, Todd Langley and Rob Kowalczyk, 2009 1, Intel Press

4. Online, http://en.wikipedia.org/wiki/IA-32

5. Online, http://en.wikipedia.org/wiki/X86

6. Online, http://en.wikipedia.org/wiki/Intel_atom

7. White Paper: The ARM v8 Architecture, John Goodacre, 2011 11, ARM Processor Division

8. , /, 2010 10, 25 5

9. White Paper: Variable SMP (4-PLUS-1TM) - A Multi-Core CPU Architecture for Low Power and High Performance, NVidia, 2011, NVdia

10. White Paper: Snapdragon S4 Processors: System on Chip Solutions for a New Mobile Age, Qualcomm, 2011, Qualcomm

11. White Paper: Going beyond a faster horse to transform mobile devices, Brian Carlson, 2011 5, Texas Instruments

12. AP , , 2011 9,

13. The future of Microprocessors, Shekhar Borkar/ Andrew A. Chien, 2011 5, Communications of the ACM

88 89

ISSUE 3 CPU

[ ]

Multi-core

Multi-core Application algorithm architecture Target , RTOS

2008. 06-2011. 12

()

OpenGL|ES 2.0, OpenVG 1.1, OpenCL 1.1 GPU

SoC

Multi-threading Processor OpenGL|ES 2.0, OpenVG, OpenCL SoC

2011. 11-2014. 09

Shader GPU (Fusion)

Unified Shader Shader CPU -Shader Architecture SoC

2012. 05-2016. 04

16GOPS SDK

2010. 03-2014. 02

MPCore SoC

ASIP (SVC, H.264, MPEG-4/2, VC-1, AVS ) C/SystemC Virtual Platform SDK

2007. 03-2011. 02

/ DSP

DSP architecture / C/C++ compiler, Hardware debugger SDK

2006. 03-2010. 02