autonomous driving: need for ai and new architectures€¦ · short-range lidar 0-2 0-4 0-4 adas...

© 2018 IHS Markit. All Rights Reserved.

Autonomous Driving: Need for AI and New Architectures

March 25, 2019

Luca De Ambroggi: Senior Director Transformative Technology, IHS Markit

[email protected]


OUTLINE

• AI pervasiveness

• Vehicle and Semiconductor market forecast

• Role of SoCs and AI-based systems in future cars

• Why AI in Automotive

• New E/E architectures required

• Growth of electronic content and costs by autonomy levels

• Software: a noble gas

• Open challenges and opportunities

2


AI is pervasive: everywhere at the Edge and at the Cloud

Thousands of

Nodes

Millions of

Nodes

Billions of

Nodes

CLOUD

Data Centers

EDGE

End-Point

Devices

EDGE

Network

Computing

La

ten

cy

Pro

ce

ssin

g p

ow

er

En

erg

y c

on

su

mption

Development

requirements

for AI solutions

Priva

cy a

nd

se

cu

rity

Cost in

da

ta c

om

mu

nic

ation

Cybersecurity

Monitoring diagnostic

Predictive maintenance

Quality and testing

Genome & chemistry simulation

Retail & CRM

Autonomous machine


Different Metrics for Investment and Research engagement:

4

Different Metrics

Different Strategy

Different Companies

> Business Case: Investment (Legacy Equipment, Process node and Product Life cycle) and ROI

– Market Volumes (i.e. Shipment) vs Time frame

– Revenue Expectation

– Market Momentum & Growth expectation

– Margin

– Technology Readiness

– Competition analysis

> Regulatory needs/assessment

> Eco-system development and needs (Partner and M&A)

– Start-up and new players

> System Complexity (HW and SW)

– Memory and Interfaces

– Key performance parameter:

• Processing/Data workload, Latency, Power, Safety

– Silicon differentiation in data-processing: GPU, CPU, TPU…up to Heterogeneous SoC;


Vehicle production rises slowly

5

0

20

40

60

80

100

120

2015 2016 2017 2018 2019 2020 2021 2022 2023

Au

tom

oti

ve V

eh

icle

Pro

du

cti

on

(M

illio

ns o

f U

nit

s)

Middle

East/Africa

South America

South Asia

Japan/Korea

North America

Greater China

Europe

1.5% CAGR (2017 – 23)


Electrification, automated driving and connectivity Fueling automotive semiconductor growth

6

8.3% CAGR (2017 – 24)

300

350

400

450

500

550

600

0

10

20

30

40

50

60

70

2016 2017 2018 2019 2020 2021 2022 2023 2024

Au

tom

oti

ve s

em

ico

nd

ucto

r r

even

ue

(B

illio

ns o

f U

S $

)

Powertrain

(ICE+H/EV)

Infotainment &

Telematics

ADAS

Body &

Convenience

Chassis & Safety

Other Automotive,

Trucks, AM

Avg. Semiconductor

Value


Implication of AI and Deep LearningMajor advantages in comparison with traditional machine vision

• Assumptions:

> New silicon solutions will be developed with focus on AI algorithm

> The functional safety aspect will be addressed by the entire supply chain

• Deep learning can:

> Allow detection and recognition of multiple object ➔ improve perception

> Perform semantic analysis of the area surrounding the vehicle

> Reduce development time of ADAS and IVI systems (once DL is in steady-state)

> Reduce the power required compared to the same operation w/ traditional algorithms

• Deep Learning needs help

> Recognition/Prediction of actions and Fusion - Bayesian Net and other stochastic algorithms may complement DL in the run to autonomous cars (L4-L5)

• Required precondition:

> Telematics will be broadly deployed to: 1) enable gathering of “real” patterns and data for training 2) allow over the air system update and security

7


Extra Requirements for Deep Learning in ADAS & AV

• DL in ADAS for Autonomous functions requires in-vehicle HW:

> Latency: for active function system needs to react in less than 70-80ms

> Deep Learning offer deterministic latency also for “noisy” input from sensors

> Performance: TFlop/TOP/TMAC is barely the minimum

• Power:

> Individual sensor subsystems need to stay in the power budget of 4W;

> Sensor Fusion ECUs might allow targets up to 15-20W or more. Some OEMs expect already they need to find a trade off if no silicon is available and performance needed.

• Backhaul and data storage infrastructure:

> Connectivity (IoT) is a need to:

– Store training data and vehicle parameters.

– Update/Upgrade the system

• Data acquisition is a challenge for validation and test: mix Real & Synthetic data (Simulation)

• Safety is the biggest uncertainty to have autonomous car based on AI.

8

Standardisation is a must have


Evolution of human machine interface (HMI)

StoneAge

Buttons & knobs Displays & more displays• Resistive displays

• Basic speech recognition

DisplayEra

DisplaysEverywhere

DigitalAssistance

Displays out, audio in• Capacitive displays

• Advanced speech recognition

• Speech-to-text and text-to-speech

Advanced audio & gestures• Natural voice recognition

• Interpret human speech and gestures

• Enable tasks and services

• HD and >10” displays

What pushed the changes?

• Diverse technologies: from displays to AI

• Lowered costs

• Influence of consumer electronics

• Increased customer needs

Why human machine interface?

• High added value and brand differentiation

• Limited performance and cost

• Steady algorithms

• Available data

• Not safety critical

• Ubiquitous across industries

AI


Cost, Performance and Power for Autonomous Vehicles: still far from high volumes vehicle production

10

• Power consumption is critical in automotive➔ where can compromises be acceptable?

• Performance 50<TOPS<100 looks to be the target in L4-L5 ➔ is silicon today able to cope with it and ensure a long term roadmap?

• Sensor and ECU add up to several thousands of dollars➔ what is the Business Model?➔ where is the ROI?

• ISO26262 is the biggest challenge considering ML deployment


AI in Automotive: Infotainment and ADAS

11

0

50

100

150

200

250

2017 2018 2019 2020 2021 2022 2023 2024 2025

Millions o

f U

nits

Shipment of systems with AI functions


12

ADAS Modules L3 L4 L5

Ultrasonic Sensors 12 12 12

Long-range Radar 1 1-2 1-2

Short/Mid-range Radar 2-4 2-6 4-6

Exterior Camera 5-8 12 8-15

Interior Camera 1 1 1

Night Vision Camera 0-1 0-1 0-1

Long-range Lidar 1 1 0-2

Short-range Lidar 0-2 0-4 0-4

ADAS Domain Controller 0-1 0 0

Autonomous Driving DC 1 1-2 2

TOTAL (without ultrasonic) 14 22 25

Typical ADAS content from level 3 to 5

Radars

Cameras

Lidars

ECUs

~ $2-3k ~ $3-6k ~ $6-9k

*Architectures based on existing pilot car platforms from BMW, Volvo, Audi, Nissan..


Possible architecture for L4/L5 in model year 202x

13

Asymmetric redundancy

© 2019 IHS MarkitSource: IHS Markit

• No Driver - Redundancy for L4/L5 is key

> Two identical (or nearly) Domain Controllers

• Provides complete redundancy.

• Expensive but comprehensive.

• DC2 can either share normal operations with DC1 or act just a back up.

• Redundant network and power supply

> Limited or distributed redundancy

• Cockpit Domain Controller (CDC) and/or Front View Camera are candidates because of their processing capabilities.

• Lower cost than symmetric redundancy but maybe less comprehensive.

*DC= Domain Controller


ADAS classification by passive warning and active control

0 2 4 6 8 10 12

Volvo XC90

Tesla Model S

Mercedes-Benz E-Class

BMW 7-Series

Audi A8

Number of ADAS functions

Active Control Functions Passive Warning Functions

ADAS functional classification on 5 models

© 2017 IHS MarkitSource: IHS Markit

*The results of Tesla Model functions will be updated after investigation completes

14


ADAS system cost to OEM by component value on Audi A8

$0 $100 $200 $300 $400 $500 $600 $700 $800 $900

12x Ultrasonic Parking

5x Basic Camera

Driver Monitoring Camera

Long-range Radar

Mono-camera Module

4x Mid-range Radar

LIDAR

Night Vision System

Domain Controller

Software value ($) SoC Value ($) Other Semiconductor Value ($) Other Component Value ($) Tier-1 Margin ($)

ADAS system cost to OEM by module and components

© 2017 IHS Markit

AD

AS

mo

du

le t

yp

e

ADAS system cost to OEM

Source: IHS Markit

IHS Markit TECHNOLOGY - Advanced ADAS Architecture Strategies

*Price of SOC does not include software value.


Thanks for your attention

[email protected]

Senior Director Transformative Technology - AI

URL:https://technology.ihs.com/Research-by-Market

16

mailto:[email protected]

https://technology.ihs.com/Research-by-Market

autonomous driving: need for ai and new architectures€¦ · short-range lidar 0-2 0-4 0-4 adas...

Documents