Artificial Intelligence The next big growth driver for the semiconductor industry

Size: px

Start display at page:

Download "Artificial Intelligence The next big growth driver for the semiconductor industry"

Felix Ramsey
5 years ago
Views:

1 Prepared for ISS Europe 2018 Artificial Intelligence The next big growth driver for the semiconductor industry Tanjeff Schadt,

2 Welcome Biography: Tanjeff Schadt Principal Munich, Germany 8 years of industry experience in management functions (R&D, product portfolio, strategy) 6 years of strategy consulting experience in Semicon, Electronics and Automotive industry Active member of PwC s semicon and technology strategy practices Lead various projects at semicon clients, esp. with focus on innovation, R&D and operational excellence Prepared for ISS Europe

3 PwC s semiconductor industry consulting experience spans the entire ecosystem and is unparalleled among consultancies Expertise in semiconductor industry Our Semiconductor Sector Consulting Expertise 40+ major semiconductor industry projects in the past two years alone Deep, global bench of >50 global consultants, with experience in all aspects of the semiconductor industry Extensive knowledge base of industry-specific best practices Broad suite of offerings Project experience along the entire value chain Value streams Strategy Product innovation & development Capital project and infrastructure (PMC) Supply chain Marketing & sales Technology consulting Prepared for ISS Europe

4 We see that eight essential technology streams have emerged Artificial Intelligence (AI) is one of them The Essential Eight Technologies Internet of Things Robots Drones 3D Printing PwC Essential Eight Artificial Intelligence Blockchain Augmented Reality Virtual Reality PwC is continuously tracking more than 150 technologies The most impactful technologies emerged as the essential eight Each technology stream at PwC is represented with dedicated teams building a well-grounded foundation of knowledge Artificial Intelligence is among our essential eight technologies Prepared for ISS Europe

5 The global semiconductor market will continue to grow AI is a major growth driver in the upcoming decade Global Semiconductor Market [$ bn] 540 CAGR 4.8% Total market: $ 530 bn AI share: $ 26 bn In the next decade we expect AI share growing to > $ 100 bn 495 CAGR 3.7% +? Total global semicon revenue Total global semicon revenue w/o AI... Source: IC Insights (Global semiconductor market), JP Morgan (AI silicon market foundry revenues, not exhaustive) Prepared for ISS Europe

6 Most attractive growth opportunities for AI are Automotive and Financial Services however, edge devices offer a huge untapped potential Artificial Intelligence Semiconductor Market Overview SELECTION Edge devices for consumer electronics Automotive Sample Use Cases Deep-learning wireless camera Augmented human decisioning ADAS Driver safety systems Financial Services Authentication Portfolio Management Healthcare Tech, Media, and Telecom Retail Industrial Smart Buildings Disease Prevention Diagnosis Network Security Personal Assistants Customer Insights Pricing Analytics Manufacturing Automation Proactive Failure Detection Monitoring & Security Energy Efficiency Market Overview 2021 Market Forecast ($ bn) AI Classification Training System Inference System Cloud Edge Source: IDC, Allied Marker Research, Analysis Prepared for ISS Europe

7 The AI stack consist of multiple building blocks innovation is brought across the stack to various target applications and use-cases The Artificial Intelligence Stack ILLUSTRATIVE Stack Element Description Examples of Solutions and Vendors Applications & Services Software applications leveraging AI for intelligence Alexa AI Platforms AI Frameworks, Tools and Interfaces Ready-to-use building blocks and services that provide a host of AI capabilities (often proprietary Tools and frameworks to leverage underlying ML algorithms to design, build, and train deep learning models for specific applications Watson Current battleground: where will AI be processed? AI Libraries AI Hardware (Accelerator vs. Edge Processing) A set of low-level software functions that help optimize the deployment of an AI framework on a specific target silicon Processor units and semiconductor logic circuits for accelerated execution of AI workloads / computations as well as adaptable AI processing on the edge MKL DL SDK Vision SDK Nervana NNP cudnn Tensor RT Telsa Snapdragon NPE Snapdragon NPE SDK ARM ML ARM NN Loihi Tools to optimize deployment to hardware architecture AI-optimized silicon architectures Prepared for ISS Europe

8 Most chip vendors are providing AI-specific acceleration to enhance their existing product portfolios Artificial Intelligence Stack Current Status (1/2) Chipmakers IP Licensors EXTRACT Target Applications AL Libraries AI Processing HW / Silicon CPU DSP GPU Datacenter Self-driving cars AR/VR Drones Surveillance MKL DL SDK Vision SDK Myriad Dev Kit Xeon PHI FPGA Arria 10 Custom Nervana NNP Myriad X Loihi NMP Datacenter Self-driving cars Retail Analytics Smart Cities Surveillance cudnn Tensor RT Pascal, Volta Maxwell, Tesla ASIC TPU Computer vision Snapdragon NPE SDK Hexagon 685 DSP (Snapdragon NPE) ADAS Drones Datacenter Medical revision SDAccel Toolkit Zynq MPSoC Vision Processing for ADAS S32 Design Studio IDE Vision SDK S32V Vision Processor Computer Vision Smart Driving STM32 STM32 Cube AI SoC for DCNN Voice assistants Consumer robots Smartphones ARM NN Cortex-A75 Cortex A55 ARM ML Automotive Surveillance Drone Mobile / Wearable Tensilica NN mapper toolkit DSP SDK Tensilica Vision C5 DSP Prepared for ISS Europe

9 however, they face an unexpected threat from hyperscalers and product companies, who are gravitating towards customized chips for AI processing Artificial Intelligence Stack Current Status (2/2) EXTRACT Cloud Player Others (Product Companies) Target Applications Facial recognition Text-to-speech Smart Assistant Digital Transformation Intelligent Assistant Image search Voice search Translate Smart Reply Search Voice Assistant Computer Vision Facial recognition Animated emoji Self-driving cars Face ID Animoji AL Libraries AWS DL AMI Xilinx SDAccel Bing API Face API Analytics API Video API Vision API Speech API NL API Core ML CPU AI Processing HW / Silicon DSP GPU FPGA Custom GPUs AWS EC2 F1 AI chip for Edge (Alexa) GPUs Project Brainwave AI chip for Hololens Cloud TPU TPU Cloud Server Neural Engine (Exynos 9) GPUs In-car chip Neural Engine (A11 Bionic) Prepared for ISS Europe

10 Four main forces will shape the AI opportunity for semiconductor players in the coming years Main Forces shaping AI Opportunities Ever broader accessibility of AI Development of applications is increasingly supported by platforms, frameworks, libraries, sensors Entry costs become increasingly lower, but so is ability to differentiate for application makers Domain-specific architectures Semiconductor node scaling very expensive, and increasingly so Fabs offer standard IP Proliferation of IoT outside of PC and datacenter Proliferation of AI at the edge AI becomes increasingly feasible in small form factors Cost of data transmission to the cloud Latency becomes critical Data privacy concerns Evolution of AI algorithms & technologies Current AI technologies are far away from enabling general intelligence Ability to test and validate AI behavior is a big question mark Evolution in AI algorithms will continue, raising the need to adapt silicon AI is an open battleground AI features: a must in many devices / applications Differentiation for application makers becomes complex, not pure AI-driven Winning horizontal solutions very expensive to develop Pockets of value in increasingly fragmented industry applications Growth in edge devices and applications Related pull in sensors Growth in intelligent device testing and management The capability to understand AI evolution and implications holistically is critical Prepared for ISS Europe

Semiconductor players should define their distinct way to play in AI Pure-Play Archetypes Horizontal solution leader Industry application leader Outsourced solution designer SELECTION

data center Broadly applicable software tools Focus on interoperability and compatibility Customized silicon, based on standard or proprietary IP Application-specific integration and

11 Semiconductor players should define their distinct way to play in AI Pure-Play Archetypes Horizontal solution leader Industry application leader Outsourced solution designer SELECTION Examples Product portfolio Standard silicon for the largest cross-industry application segments e.g. data center Broadly applicable software tools Focus on interoperability and compatibility Customized silicon, based on standard or proprietary IP Application-specific integration and testing tools Possibly proprietary software and algorithms Design and fabrication services Integration of customer requirements or standard IP, as required Multi-purpose packaging, assembly and testing services Core differentiating capabilities Ecosystem, partnership and alliance management Breakthrough innovation in R&D and fabrication Channel management Deep customer intimacy In-depth AI application stack understanding Solution integration & selling Application-specific customer support Customer requirements understanding & relationship management at scale Customer segmentation and selection External R&D integration Prepared for ISS Europe

12 In recent years the AI start-up landscape gained momentum funding of semicon start-ups is back again Semiconductor AI Start-up Landscape EXTRACT Start-up Founded HQ (GEO) Stage Funding to Date ($ m) Strategic Investors Technology Cambricon Technologies n/a China Series A 101 Alibaba Deep learning processor CyberSwarm n/a San Mateo, CA Seed 1 None AI-assisted cybersecurity CPU Graphcore n/a UK Series C 110 Samsung, Dell Deep learning processor Horizon Robotics 2015 Beijing, China Series A 100 Intel Vision DSP KnuEdge n/a San Diego, CA n/a 47 None Neuromorphic processor LightOn 2016 Paris, France Seed 0 n/a Optical/quantum AI computing Movidius n/a San Mateo, CA Series E 56 Intel Neural Compute Engine Accelerator (Appl: Vision DSP) Mythic n/a Redwood City, CA Series A 9 n/a Neuromorphic processor Nervana n/a San Diego, CA Series A 25 Intel Deep learning processor Reduced Energy Microsystems 2014 San Francisco, CA n/a 2 n/a Deep learning processor Rigetti Computing 2013 Berkeley, CA Series B 70 n/a Optical/quantum AI computing Tenstorrent 2016 Toronto, Canada Seed 0 None Deep learning processor Vayyar 2011 Yehud, Israel Series C 80 n/a Vision DSP Vicarious 2010 San Francisco, CA Series C 137 Samsung Neuromorphic processor Wave Computing 2010 Campbell, CA Series D 117 Samsung Deep learning processor Xanadu 2016 Toronto, Canada Seed 3 n/a Optical/quantum AI computing Cerebras 2016 Los Altos, CA Series B 112 n/a Deep learning processor ThinkCI n/a n/a n/a 0 n/a n/a Knowm 2015 Santa Fe, NM n/a 0 n/a Neuro-memristive processors (Thermodynamic RAM) ThinkForce 2017 Shanghai, China n/a 0 No AI Acceleration Engine Groq 2016 Palo Alto, CA n/a 0 No n/a Gyrfalcon n/a n/a n/a 0 n/a n/a Source: Strategy& research, Crunchbase Prepared for ISS Europe

13 The silicon required for Level 5 autonomous driving is likely already available power consumption and form factors still evolving Evolution of relevant IC Alternatives for in-car AI Inference Tera-operations per second (TOPS) in 32 bit floating point precision (fp32) IBM TrueNorth EXAMPLE: AUTONOMOUS DRIVING Intel Nervana Lake Crest 2.0 ** ~25 fp32 Approximate computing power required for inner-city autonomous driving with current algorithms* Inference only AI training in the cloud Nvidia Tesla K40 Google TPU 2.0 Nvidia Tesla P40 Nvidia Tesla P100 Intel Xeon Phi 7250 Intel Nervana Lake Crest ** MobilEye EyeQ5 Nvidia Tesla V100 Claimed to be able to support SAE L5 by MobilEye EyeQ4 1 MobilEye EyeQ3 Intel Xeon E v4*** present 2020 * Based on Google estimates (2016) estimate of 50 TOPS at floating point 16 bit precision, i.e. approx. 25 TOPS at floating point 32 bit precision ** Illustrative based on current Intel press releases. Exact performance and power consumption not announced. *** Representative example of Intel Xeon family Source: Strategy& desktop research June 2017; some devices incorporate multiple dies, e.g. Google TPU 2.0 Specialized CPU GPU GPU/VPU Circle size indicative of relative power consumption Specialized AI processor / Neuromorphic chip Prepared for ISS Europe

14 Innovation in AI compute is increasingly accessible to startups, who focus on novel chip architectures The Start-up Scene in AI Silicon VC Funding in Semiconductor AI Start-ups $ m ( ) Key Technologies of AI Semiconductor Start-ups* ( ) 3X 640 Deep learning processor 8 Neuromorphic processor x 212 < Vision DSP Optical/quantum AI computing AI-assisted cybersecurity Source: Strategy& research, Crunchbase * Startups which received funding in Prepared for ISS Europe

15 AI is THE opportunity for European semicons The opportunity is big! There are plenty of growth options! Don t forget the ecosystem! What is your strategy for Artificial Intelligence? We are in an early phase core is still an opportunity European semicons can build core AI in Asia Edge is core capability of European semicons What is your answer on how to play in the ecosystem? European semiconductor companies have the know-how and a right to win you better have a strategy! Prepared for ISS Europe

16 European semicon players shall take advantage of AI in cooperative mode Way to play? You better have a strategy! Where to play? AI is application driven what are the most relevant applications for you? How to play? What is the right spot in the AI ecosystem?? Cooperate Forget what you can do on your own: What can you achieve together in the European semicon industry? Prepared for ISS Europe

Outlook: Global Semiconductor Report 2018 coming soon PwC Semiconductor Report Series The PwC Semiconductor Report Series provides an overview of market developments,

The reports also covers highlight topics and their implications on the future of the industry past topics included the Internet of Things (2015) and a spotlight on

17 Outlook: Global Semiconductor Report 2018 coming soon PwC Semiconductor Report Series The PwC Semiconductor Report Series provides an overview of market developments, growth opportunities and success factors of the global semiconductor market It includes a forecast on global semiconductor billings by component, region and application The reports also covers highlight topics and their implications on the future of the industry past topics included the Internet of Things (2015) and a spotlight on Automotive (2013) The two highlight topics in 2018 will be: Artificial intelligence the next big growth driver Digitization of semiconductor companies Prepared for ISS Europe

18 Prepared for ISS Europe

19 Contact Phone: Mobile: Tanjeff Schadt Principal Semicon expert (Germany) GmbH Bernhard-Wicki-Straße München Prepared for ISS Europe

20 Thank you! 2018 PwC. All rights reserved. Not for further distribution without the permission of PwC. PwC refers to the network of member firms of PricewaterhouseCoopers International Limited (PwCIL), or, as the context requires, individual member firms of the PwC network. Each member firm is a separate legal entity and does not act as agent of PwCIL or any other member firm. PwCIL does not provide any services to clients. PwCIL is not responsible or liable for the acts or omissions of any of its member firms nor can it control the exercise of their professional judgment or bind them in any way. No member firm is responsible or liable for the acts or omissions of any other member firm nor can it control the exercise of another member firm s professional judgment or bind another member firm or PwCIL in any way. Prepared for ISS Europe

Artificial Intelligence The next big growth driver for the semiconductor industry Tanjeff Schadt, PwC Strategy&

Artificial Intelligence The next big growth driver for the semiconductor industry Tanjeff Schadt, www.strategyand.pwc.com Prepared for SEMICON Europa, TechARENA November 13, 2018 Welcome Biography: 8 years