r-car gen3: computing platform for autonomous …€¦ · demonstrate renesas concept car for level...

© 2017 Renesas Electronics Corporation. All rights reserved.

R-CAR GEN3: COMPUTING PLATFORM FOR AUTONOMOUS DRIVING ERA

Mitsuhiko Igarashi, Toyokazu Hori, Yoshihiko Hotta, Kazuki

Fukuoka and Hirotaka Hara

AUTOMOTIVE SOLUTION BIZ. UNIT,

RENESAS ELECTRONICS CORPORATION

© 2017 Renesas Electronics Corporation. All rights reserved. BIG IDEAS FOR EVERY SPACE

Outline

Renesas Autonomy : ADAS/Autonomous Driving Platform

Autonomous Driving Challenges and Feature of R-Car H3

Level 4 Autonomous driving Concept Car

Concept of Reliability Management for future auto. driving usage


NEW RENESAS CONCEPT FOR AUTONOMOUS DRIVING

Global No.1 share

in vehicle MCU

No.1

Promote end to end solution in collaboration with many global partners

END TO END SOLUTION FROM CLOUD SERVICE TO SENSING AND VEHICLE CONTROL

CONTRIBUTE TO NEXT AUTONOMOUS DRIVING SOCIETY


SENSING &COGNITIVE SOLUTION FOR AUTONOMOUS DRIVING

Scalable solution for sensing (Camera/Rader) and Cognitive from Level2/3 to Level 4/5


Outline






Automotive

High Reliability

Autonomous Driving Challenges for LSI

HardwarePerformance @ Low power

Functional SafetyAvailability & Fail Operational


Solutions for Performance and Low Power by R-Car

OSOSOS

GFX

processing

Sensor into

processing

Virtua-

lization

PlanningOptimization

problem

Image

recognition

Video

processing

High speed bus network optimized for parallel processing

Image

recognition

engine (IMP)

GPU/

GPGPU

Cortex A57 x 4

Cortex A53 x 4

Cortex R7 x 1

Video

processing

engine

FPU

R-Car H3Power management

Performance at Low power

✓ Allocate each process to suitable automotive engines/processors

✓ Several power management techniques

Real time operation : Parallel processing preferred bus network / HW assist virtualization


Several Power Management Techniques for Low Power

Power gating : 21 power gated domains control, introduce each cluster shutdown for GPU

Dynamic Voltage Frequency Scaling / Adaptive Voltage Scaling to CPU and GPU

CortexA57

CPU0

L1$

Common L2$

CPU1

L1$

CPU2

L1$

CPU3

L1$

cl0

#0 #1

cl1

#2 #3

cl2

#4 #5

Common

GPU

PSW

controller

Voltage

controller

CLK

controller

cpuclk

gpuclk

Delay

Monitor

Thermal

Monitor

variable voltage (VDD_DVFS)

on/off

PD21

PMICOS

AVS

control

DVFS

control

Thermal

control

PSW

control

Delay

Monitor

Thermal

Monitor

PD1-5

PSW : Power Switch

PD6-9

Fixed voltage


Power Management for Functional Safety

Power management with several on chip monitor for several time scale events

Introduce voltage droop prediction and recovery technique with droop monitor [1]

Delay

Monitor

Droop

Monitor

CPU0

VDD_DVFSDelay

Monitor

noncpu

stop

cpu_clkPLL

clk ctrl

N/32

Div

stop_clear

PMIC

dly

ctrl

#count

11vup

vdown

Voltage

Monitor

vcpu_mon

delay_thFuse

clk

clk

Droop

Monitor

CPU1

clk

voltage

reg

Time

Ts=667ps

keyoff 15yr

Level:0

Level:1

Level:2

test

140mV

keyon0yr

Ts=1.6us

[1] C. Takahashi et al., ISSCC, 2016

Level Target Vmin Used monitor Control

2 DC operation Delay

monitor

voltage up

1 AC(>1MHz)

operation

Droop

monitor

Clock stop &

restart

0 Monitor Vmin Voltage

monitor

System

shutdown

VDD_DVFS


Power Management for Functional Safety and Run-time Test

Droop prediction technique can prevent the failure caused by voltage droop

Wear-out faults detection by run-time test

0

5

10

15

20

25

0.0 5.0 10.0 15.0 20.0 25.0 30.0

VC

OD

E

Time [sec]

Measured predictedw/ predictionw/o prediction

Large droop

Measurement condition

Droop monitor

Measurement Result

4 cycle prediction

: GPU running

: Run-time test

: Voltage monitoring

: Collect Min. Vcode and Display

667ps


Reliability Aware Droop Mitigation with MIM

Satisfy both voltage droop reduction with MIM and automotive reliability grade by

✓ Dedicated ESD protection for MIM (AEC-Q100 capable)

✓ MIM area optimization of CPU and GPU

MIM cap

Dedicated ESD protection

0.0

2.0

4.0

6.0

8.0

10.0

12.0

14.0

16.0

18.0

20.0

0.0 50.0 100.0 150.0 200.0 250.0 300.0

Voltag

e D

roop

Im

pro

vem

en

t [m

V]

MiM Cap [nF]

Failure rate limit

16mV

improvement

CPUGPU


Reliability Aware Droop Mitigation with MIM

MIM cap failure modelGlobal metal

VSS VDD

MIM

Short current

Introducing MIM IDDQ testing enable early failure rejection of MIM

Emphasize only MIM short current

CPU

VDD

MIM

VSS

Short

current

Large leakage

Conventional IDDQ test

Suppress

Current

MIM IDDQ test

MIM

Leakage

emphasizerCPU

VDD

VSS

Emphasize

Current


Outline






Renesas Concept Car for Level 4 Autonomous Driving

Level 4 driving with redundancy at 25 Watts

Renesas NVIDIA NXP

HAD x 2(H3 x 4

+ MCU x 2)

Drive PX2(Parker x 2 +

Pascal GPU x 2)

BlueBox(QorIQ LS2088A)

25 W 250 W *1 40 W *2

*1 http://www.eetimes.com/author.asp?doc_id=1328609

*2 http://www.nxp.com/jp/products/microcontrollers-and-processors/

arm-processors/s32-arm-processors-microcontrollers/bluebox-autonomous-driving-platform-s32vls2-rdb:S32VLS2-RDB


Renesas Concept Car : Fail Operation

Fail operation tolerant by ASIL-D MCU (RH850) and triple redundant H3 system

Control the car to move to the safety zone when the failure is detected

https://www.youtube.com/watch?v=r59vYREilHY

Failure

https://www.youtube.com/watch?v=eA39bKtbuAs


Outline






Big Waves in the Future Automotive

Owner Driven Car

Level2/3 (Level4)

Reduce Car Accident

Driving pleasure

Convenience for Driver

Economic efficiency

（ EV Range, Function vs cost ）

Level4/5 (Driverless)

Complete Safety

Service pleasure

Mobility convenience

Cloud Service model

Mobility Service CarOwner Driven

CarMobility Service

Car


Connected Car Solution with Cloud Service

Renesas can link all information to Cloud from control (RH850) to car information (R-Car).

One of important Service is “Vehicle quality Enhancement” for Service Car

Service

Data1)


Car Usage Challenges in Future Autonomous Driving Era

What will happen at fully automated driving era?

Increase an annual mileage dramatically

Reliability enhancement will be one of key challenge

*1 Automotive Electronics Council, "Failure Mechanism Based Stress

Test Qualification for Integrated Circuits,” AEC-Q100 Rev-H, 2014

Operation rate

~ 50%

Owner car Ride sharing

etc.

Mobility Service Car

with level 5 auto.

driving + EV

An

nu

al

mileag

e

Operation rate

< 10% *1


Trade-off between Reliability Enhancement and Performance

Handling of operation temperature, voltage and aging variation further optimize reliability

Low temp. limitation and low volt. operation may restrict the performance

Propose concept of reliability management for further LSI reliability enhancement

Volt. Temp.

Log(B

TI lif

etim

e)

~ 2x Lifetime

by - 20 mV

~30x

Log(B

TI lif

etim

e)

Log(B

TI lif

etim

e)

-3σ 3σ

~ 3x Lifetime

by - 15 deg C

TemperatureVoltage Aging variation

Opera

tion f

requen

cyFrequency down

Th

erm

al desig

n p

ow

erPerformance

restriction

BTI : Bias Temperature Instability


Reliability Management Concept ~ Static AVGS

Before Burn-in

Tester

CPU

Fuse

Fuse

Delay

Monitor

Lifetime

Odometer

After Burn-in

F_fresh

P-code

A-code

ΔF_agedGPU

Chip

P-code

(Process)

Burn-in

A-code

(Aging)

One of key reliability item is BTI/HCI degradation

Lifetime odometer (LTO) monitors BTI/HCI degradation per chip

Delay monitor and LTO detect Process variation and Aging variation at testing

Static Adaptive Voltage Guardband Scaling (AVGS)

BTI : Bias Temperature Instability

HCI : Hot Carrier Injection


Reliability Management Concept ~ Static AVGS

Static AVGS optimize Voltage Guardband (VGB) chip by chip with reflecting

✓ Operation Voltage difference (by P-code)

✓ Aging variation difference (by A-code)

20 mV lower operation extend lifetime at SS chip with negligible yield loss by LTO

FFSS

Adjusted

by A-code

Voltage

0Conv. AVS setting

Optimized

by P-codeStatic AVGS

setting

Vmin w/o agingAging

Guardband

TT

SS/TT/FF : Slow/Typical/Fast Chip


Reliability Management Concept ~ Dynamic AVGS

Dynamic-AVGS and Adaptive Thermal Control

Lifetime

odometer ΔVDD

for aging

Lifetime prediction

Adaptive

Thermal ControlPG, DVFS

System

controller

Tj

Volt.

Aging code

Process code

External flash

VGBLogThermal

Sensor

P-code

(Fuse)

A-code

(Fuse)

Dynamic voltage control based on P&A-code and lifetime prediction extend lifetime of

chip with initial low volt. (TT~FF chip)

Thermal

Sensor

ΔF

Deep

learning

VGB

threshold

+30mV

Lifetime extendedFF Chip

Lifetime extension with dynamic-AVGS

Dyn

am

ic A

VG

S Δ

VD

D

Effective stress time Teff @Tj=125C [hrs]

12K 120K1.2K


Reliability Management Concept ~ Dynamic AVGS

Dynamic-AVGS and Adaptive Thermal Control

Lifetime

odometer ΔVDD

for aging

Lifetime prediction

Adaptive

Thermal ControlPG, DVFS

System

controller

Tj

Volt.

Aging code

Process code

External flash

VGBLogThermal

Sensor

P-code

(Fuse)

A-code

(Fuse)

Adaptive thermal control based on P-code extend lifetime of chip with low leakage (SS~TT)

Thermal

Sensor

ΔF

Deep

learning

Tj relaxation of low leakage chip


Reliability Management Concept ~ Result

Burn-in stress

12K hrs (AEC-Q100)

Total lifetime extensionMeasured result of Lifetime Odometer

NBTI sensitive

osc

2.6x

67.2K hrs

Fre

qu

en

cy d

eg

rad

ati

on

[%

]

INV

based osc

Lifetime @Tj=125C [hrs]

2x

-20mV

5.6x

-20mV

&-15C

Power on time Field lifetime Duty ratio

12 K hrs 15 years 9.1%

67.2 K hrs 15 years 51.1%

5.6x lifetime in total without performance degradation realize ~ 50% operation rate


Summary

Renesas Autonomy : Total solution for auto. driving computing platform

R-Car H3 satisfy all of Performance, Low power, Functional safety and

Reliability

Demonstrate Renesas concept car for level 4 autonomous driving

✓ 25 W with triple redundant H3 and ASIL-D MCU system

Introduce concept of reliability management for future auto. driving usage


BACKUP


Definition of Levels of Driving Automation by SAE International

Source : SAE International J3016, Sep. 2016

r-car gen3: computing platform for autonomous …€¦ · demonstrate renesas concept car for level...

Documents