Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
NAVIGATING THE PERFECT STORMA New Playbook for the A.I.-Big Data Era
Gary DickersonPresident and CEO, Applied Materials
EXTERNAL USE
PC + Internet
Mobile + Social Media
A.I. Big Data
A Brief History of Computing…
1998 2008 20181988
Mainframe Computing100,000s of units
100 Millions of units
Billions of units
Trillions of units
“Data is to this century what oil was to the last one: a driver of growth and change”– The Economist
“In the future, every business
will be in the technology business”
AGRICULTURE TRANSPORTATION HEALTHCARE EDUCATION
High bandwidth, high speed connectivityto move data between edge and cloud
1 Trillion
edge devices
added in next
15 years
Explosion of
data generation
Shift from
general purpose
to workload
specific
computing
SOURCES: Arm, Applied Materials
5x
90%
increase in next 5 years
generated by machines(source: Arm)
New Computing Approaches are Needed to
Take Advantage of All the Data
GENERAL PURPOSE COMPUTING(GOOD AT MANY THINGS)
A.I. COMPUTING (CUSTOMIZED FOR SPECIALIZED TASKS)
Deep instruction pipeline,
out-of-order execution, branch
prediction, pre-fetching, …
Extreme ParallelismSimple operations on multiple
chunks of data in parallel
Multi-level on-die cache architectureHigh Speed MemoriesTo improve memory latency
and memory to logic bandwidth
64 bit or higher precision data pathsLower PrecisionTraining and inference at lower floating point
(16 bit) and integer (8 bit) precision
1 Billion
IP CAMERAS
Today’s U.S.
DATACENTERS
Average
POWER PLANT
25TWh 90TWh 10TWhSOURCES: Reolink, EIA
10 to15% of global electricity supply could be used by A.I.-Big Data by 2025
TR
AIN
ING
INF
ER
EN
CE
CLOUD EDGE
NEEDED VS. CURRENT SOLUTION (GPU)
75xthroughput and
1,000xperformance per watt
Limited impact on roadmap
NEEDED VS. CURRENT SOLUTION (CPU)
50x better latency and
75x throughput
NEEDED VS. CURRENT SOLUTION (CPU)
50x better latency and
500x throughput
Major advances
at edge and in
cloud needed to realize potential
of A.I. without
impacting world’s
energy equation
IMPROVEMENTS IN…
For 40 Years…
105
104
103
102
10
1
COMPONENT PER IC
RE
LA
TIV
E C
OS
T/C
OM
PO
NE
NT
1 10 102 103 104 105
1962
1965
1970 PPAC
POWER
PERFORMANCE
AREA-COST
14
me
ters 600kW
$110M
MOORE’S LAW If built in 1980…
PERFORMANCE IMPROVEMENTS OVER TIME (VS. VAX-11/780)
100,000
10,000
1,000
100
10
19
78
19
86
20
03
20
11
20
15
20
18
25%per year
52%per year
23%per year
12%per year
3.5%per year
End of Dennard Scaling
Limits of parallelism of Amdahl’s Law
End of Moore’s Law
VAX-11/780
TIME BETWEEN LOGICNODES IN YEARS
>5
2.5
2
1.8
SOURCE: Computer Architecture: A Quantitative Approach, Sixth Edition, John Hennessy and David Patterson, December 2017 SOURCE: Bernstein
Performance Improvements are Slowing
A New Playbook for Semiconductor
Design and Manufacturing is Needed
ENABLED BY
Advanced packaging
New structures / 3D
New ways to shrink
New architectures
New materialsPPAC
POWER
PERFORMANCE
AREA-COST
Renaissance of Hardware Innovation
INITIAL DEPLOYMENT ADDRESSES
AcceleratorsGPU, TPU, ASICS, FPGAs
NowCloud training and inference
Automotive inference
Near MemoryDDR, SRAM, HBM, NAND, X-point
Now to 2 yearsCloud and edge
Training and inference
New MemoryMRAM, ReRAM / CeRAM, PRAM, FeRAM
2 to 5 yearsInitially edge
May migrate to cloud over time
In-Memory ComputeAnalog, ReRAM, PCM
2 to 5 yearsInitially edge
May migrate to cloud over time
Novel HPCQuantum, Synaptic
5 to 10 years Cloud + TBD
Eco-system Collaboration is Key
TODAY: Serial / compartmentalized interaction between key parts of eco-system
“Von Neumann” mindset “Neuromorphic” mindsetD
ES
IGN
INT
EG
RA
TIO
N
ED
A
MA
NU
FA
CT
UR
ING
EQ
UIP
ME
NT
MA
TE
RIA
LS
OPPORTUNITY: Parallel development to get powerful tools to designers faster
DESIGN
INTEGRATION
MANUFACTURINGEDA
EQUIPMENT
MATERIALS
vs.
Tainan Display Lab
State-of-the-art R&D capability
for advanced displays
OPENED JUNE 2019
APPLIED’S INVESTMENTS TO ACCELERATE INNOVATION
Maydan Technology Center
Advanced 300mm R&D lab
Advanced Packaging
Development Center
Fully-integrated 300mm advanced
wafer level packaging lab
META Center
Expanded R&D capabilities
focused on ‘lab-to-fab’ acceleration
SCHEDULED TO OPEN IN FALL 2019
R&D Investment
~15% of revenues annually
($2.1B in 2019)
Applied Ventures
>$250M assets
under management
of collaboration between LETI and Applied Materials
EXTERNAL USE