Mike Mayberry, Senior VPIntel CTO & GM Technology Development

Copyright © 2019 Intel Corporation

inter

FUTURE COMPU[ING MODELS . . • '. ~ ' : 0 . ,;,~ "':',, ··>

TECHNOLOGY IN THE 2030-2040 HORIZON ,

2

Legal Notices

This presentation contains information provided by Intel Corporation (“Intel”), and may refer to Intel’s plans and expectations for the future, which are provided for discussion purposes only and are subject to change without notice. Forward-looking statements involve a number of risks and uncertainties: Refer to Intel’s SEC filings for authoritative discussion of Intel’s results and plans.

This presentation imposes no obligation upon Intel to make any purchase, and Intel accepts no duty to update this presentation based on more current information. Intel is not liable for any damages, direct or indirect, consequential or otherwise, that may arise, directly or indirectly, from the use or misuse of the information in this presentation.

Copyright © 2019 Intel Corporation.Intel and the Intel logo, are trademarks of Intel Corporation in the U.S. and/or other countries. Other names and brands may be claimed as the property of others.

… motivates new optimizationsData at the edge, more unstructured

Distributed compute, more M2M

Unintended programming inefficiencies

The Shifting Nature of Data…

160 ZB2 ZB

Broadcast Media

OTT/Streaming

Broadcasting

Digital TV

PB

M2M

Productivity data

Relational databases,

BI suites

Web data

Social

Search

Mobility

~60%

~45%

~20%

CAGR(17-25)

~20%

~15%

Digita

l Dat

a Gen

erat

ed

15 ZB

Structured UnstructuredBest processed near source

Source: Analyst and Intel Data , Oct 2018

3

PRODUCTIVITY DA TA

C++ Java JS

Deep learning▪ Known answers▪ Generate procedures

with training

Conventional▪ Known

procedures▪ Generate answers

Neuromorphic▪ Many procedures▪ Adapt the answer

with reinforcement

Quantum▪ Answers

superimposed▪ Select and measure

the answer

Graph Analytics▪ Graph edges and

vertices represent relationships

▪ Big, sparse data structures

Compute Architecture progression

4

Co-evolution of Software & Hardware Solutions

2015-2025

2020-2030

2025-2035

sw defined hw AI generated sw

IA++/XPU ISA

DSL + Runtime

(execution model)

Application

SW defined flow

XPUCPU

Domain Accelerator Processing Unit

Central Processing Unit

Domain Specific Arch.

SDH Augments GP Processors

5

AI Driven SW Development

App

Domain Specific Language

Domain Specific Architecture

H/W Product

Design

Intelligent Assist

Code Generation

Manage

Test

Maintain

• Identify patterns• Discover dependencies• Suggest reuse

• Automate code gen• Proactive command gen• Improve productivity

• Automate deployment• Decrease defects• Optimize traffic routing

• Analyze code• Automate test generation• ID bugs/focus testing• Predict outcomes

Program Development

Cycle

i=i== □□ ooo §:§:§:oo:S:S:S §§§00888

Graph Analytics

opportunitiesStatus ChallengesMultiple query languages and processing engines

DARPA Graph Challenge

Real-time decision making

Fraud detection

Social network analysis

Anomaly detection

Sparse and irregular memory accesses

Small data accesses with frequent synchronization

Scaling to very large datasets

6

neuromorphic computing

opportunitiesStatus ChallengesComplex neural network

topologies

~1 billion neurons demonstrated

Constraint Satisfaction

Real-time Learning

Adaptive Control

Complex Systems Modeling

Neuroscience model development

Algorithm development

S/W & Simulation tools

7

Quantum computing

opportunitiesStatus Challenges~100 organizations investing

8 qubit types

~100ms coherence time for SC

Hybrid classic-quantum algo’s in dev’t

Molecular structure

Materials science

Pharmacology

Cryptography

Decoherence & error-correction

Low-latency qubit control

Qubit scale-up

Algorithms/Compilers

Workloads for testing

8

power/thermals drives beyond cmosP

ow

er

De

nsi

ty

R. Schmidt et. al., IBM J R&D (2002)

2030 2050

?Beyond CMOS

9

14

12

10

8

6 Junction Transistor

Integrated Circuit

Bipolar

• • . . . • . . . . . . . . . . • . . . .

CMOS • • • • • • • • • • • • • • • • • • • • • • • • • • • • •

4

2 Vocuum ~

.4,!;_....:,:'u,iib@.,-,~ ---l.--....;;• ;...--=!!!!=::::._ ______ ....,:•_,:•~--.J f - - - - - - -+

Beyond CMOS

10

Source: C. Pan; A. Naeemi. “An Expanded Benchmarking of Beyond-CMOS Devices Based on Boolean and Neuromorphic Representative Circuits” IEEE Jan 2018 10.1109/JXCDC.2018.2793536

Desired state

Improved switching efficiency

Enabled by academic-industry-government

Novel devices improving Example: meso

Better understanding of logic circuits

Magneto-Electric Spin-Orbit

100mV operation

10X-30X better power/perf than CMOS

MESO

10'

CMOS

10 5 • TFET1

0 FefTOIIIKlric

• .,._ 0 --· 10

4

-, ,S 10 3

f w

10 2

10 1

G - 0 Gpn.1-Vo3 Vol

• • CMOS HP MITTET

• 't>FEFET CoMET

r MEMTJ

Ho<JTFET • NCFET • ITFET e_; MEMT-"

'"-'TFET f GeHTFET • YlfNFeT-8P SWO

• TMOTFET ~ (:) MT ........

* CMOSLV

BioFET

Summary

New architectures and new technologies needed to manage massive workloads

The new data era will shift compute toward data and S/W-centric H/W

Power & Thermal drive the need to move beyond CMOS

11

"Any opinions, findings, conclusions or recommendations

expressed in this material are those of the author(s) and do not

necessarily reflect the views of the Networking and Information

Technology Research and Development Program."

The Networking and Information Technology Research and Development

(NITRD) Program

Mailing Address: NCO/NITRD, 2415 Eisenhower Avenue, Alexandria, VA 22314

Physical Address: 490 L'Enfant Plaza SW, Suite 8001, Washington, DC 20024, USA Tel: 202-459-9674,

Fax: 202-459-9673, Email: nco@nitrd.gov, Website: https://www.nitrd.gov