David Pellerin Cloud-Accelerated Innovation for Semiconductor Design and Verification Head of Worldwide Business Development, Semiconductor Industry Amazon Web Services
David Pellerin
Cloud-Accelerated Innovation for Semiconductor Design and Verification
Head of Worldwide Business Development, Semiconductor IndustryAmazon Web Services
Amazon is in the semiconductor business
We design our own silicon devices, and we source from a global supply chain
Amazon has multiple, globally distributed silicon teams, for • Datacenter infrastructure• Consumer devices• Robotics and AI• And moreWe benefit from cloud in our own silicon development processes
AWS global Cloud infrastructure
22 geographic regionsA region is a physical location in the world where we have multiple Availability Zones
69 Availability Zones Distinct locations that are engineered to be insulated from failures in other Availability Zones
NetworkAWS offers highly reliable, low latency, and high throughput network connectivity. This is achieved with a fully redundant 100 Gbps network that circles the globe .
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
AWS custom hardwareThe AWS Global Infrastructure is built on Amazon’s own custom hardware By using our own custom hardware, we provide customers with the highest levels of reliability, the fastest pace of innovation, all at the lowest possible cost AWS optimizes this hardware for only one set of requirements: workloads run by AWS customers
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
AWS completes multiple silicon tape-outs each year
Powerful and Efficient for Server for
Modern Applications
Cloud Hypervisor, Network, Storage,
and Security
Machine Learning Inference Hardware
and Software at Scale
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Silicon design workflow
Design verification Synthesis Physical
layoutPhysical verification
Tape out/manufacturing
Power/signal analysis
Design specification
Silicon validation
For advanced node design, the pain is here
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Silicon design workflow
Design verification Synthesis Physical
layoutPhysical verification
Tape out/manufacturing
Power/signal analysis
Design specification
Silicon validation
Requiring ever-larger computing for signoffCloud is becoming the new signoff platform
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Design and verification on CloudOn the Cloud, secure and well-optimized compute and storage chambers can be created, operated, and torn down in just minutes
Machine learning and analytics cloud services
Third-party IP providers and collaborators
Virtual Private Cloud on AWS
Remote desktop
License managersWorkload schedulersDirectory services
Cloud-based, auto-scaling compute clusters
Shared file storage Storage cache
Corporate datacenter
On-premises HPC resources
AWS Direct Connect
AWS Snowball
EDA, CAE, analytics, and machine learning
TSMC Virtual Design Environment (VDE)
Validated by TSMC for…• Security of IP, workflows, and tools• Performance: compute , memory, networks, storage• Automation and cluster/job/license management• Remote graphics for interactive applications• CAE/EDA and IP vendor support
Developed in collaboration with TSMC,Synopsys, Cadence
Virtual Private Cloud on AWS
Remote desktop
License managersWorkload schedulersDirectory services
Cloud-based, auto-scaling compute clusters
Shared file storage Storage cache
Astera Labs Develops Complex SoC 100% on AWS
Industry: Semiconductor and Electronics
Headquarters: San Jose CA
Website: www.asteralabs.com
At Astera Labs, we are intensely focused on delivering high-quality PCIe connect ivity solut ions to our customers, and reduce t ime-to-results.
Our High-Performance Compute (HPC) infrast ructure is hosted ent irely on AWS and we heavily leverage the cloud-scalability enabled by AWS and Synopsys tools to accelerate our development schedule.
Jitendra Mohan, CEO Astera Labs
“
”
About Astera Labs
Our vision is to be the t rusted partners to dist ribute data in intelligent systems. We develop purpose-built connect ivity solut ions to remove performance bot t lenecks in data-centric systems and work-load opt imized plat forms. The company’s product port folio includes system-aware semiconductor integrated circuits, boards and services to enable robust PCIe connect ivity.
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
MediaTek7nm tapeout
• Proven results for EDA running on Cloud• Static Timing Analysis (STA) for 7nm process SoC• 1000 AWS instances (32,000 physical cores)• 12 million core -hours of computing for STA
• Successfully eliminated IT compute resource bottleneck• World’s 1st 5G SoCannounced at Computex 2019 (May 29th)
Samsung 7LPP Synopsys reference flow on AWS
ARM Cortex A53_CPUFrequency 2.16 GHz Physical area 232427 (um2)
Performance on AWS (32 CPU, 480G RAM)Full Reference Flow Runtime 71 HoursPeak Memory 24 GB
Cadence reference flows for Samsung Foundry nodesIncluding 28FDS, 14/11, 10/8, 7/5/4 EUV
• Full Cadence Digital Reference flow synthesis-through -DFM on Samsung Foundry nodes• Proven and validated for customer use on the cloud
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Silicon design workflow
Design verification Synthesis Physical
layoutPhysical verification
Tape out/manufacturing
Power/signal analysis
Design specification
Silicon validation
Advanced node design requires better use of dataincluding analytics and machine learning
Design for manufacturability
Materials and processes, yields and failure rates
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Technology partners for silicon design: examples
Design verification Synthesis Physical
layoutPhysical Verification
Tape out/manufacturing
Power/signal analysis
Design specification
Silicon validation
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Autonomous Robots
Simulat ion and Digital Twin
Vert ical & Horizontal Integrat ion
Indust rial Internet of Things
CybersecurityCloud
Addit iveManufacturing
AugmentedReality
Big Data & Analyt ics
Cloud is one of the 9 standalone enablers
Cloud is the cent ral innovat ion enabler across all disciplines and indust ries
Industry 4.0 – technical enablers
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Cloud enables secure partner collaboration
EDA/IPcollaboration cloud chamber
Yieldanalytics cloudchamber
FoundryIP merge cloud chamber
Digitaltwin cloudchambers
EDA vendorsecure cloud chamber
IP provider secure cloud chamber
Fablesssemico secure cloud chamber
Fabless semiconductor
Equipmentprovider secure cloud chambers
Foundry/OSATsecure cloud chamber
Foundry/OSAT
PDK
GDSII
Tools
IP
Curated OT and machine data
Updated ML models
Col
labo
ratio
nC
ham
bers
Sing
le T
enan
t Cha
mbe
rs
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Machine Learning and IoTfor SemiconductorsApplications throughout design and production
Design and verification …• Intelligent local and
global routing• Timing analysis and DRC• Simulation parameter selection• Design flow optimization• And more
Manufacturing and supply chain …• Lithography optimization• Fault detection and classification• Yield diagnostics and failure prediction• Predictive maintenance and OEE• Excursion prevention• And more
Engineering & operational DATA
Ingest ConsumeStore Analyze
Engineering & operational INSIGHTS
Wafer defect classificationCustomer seeks to reduce the need for human inspection to classify wafer defects (scratches, spots, bubbles)SolutionUse SageMakermachine learning on AWS Cloud, in combination with API Gateway, DynamoDBdatabase, AWS Lambda, and cloud storage to create an environment for model training and image classification..
Machine learning in the fab
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Cloud Manufacturing Reference Architecture: Edge-Cloud-Edge
IoT GreengrassEdge/GW
S3Manufacturing
Data Lake
Kinesis
MES
Factory Equipment
ML Inference
IoT Core
SageMakerML
QuickSightBusiness
Intelligence
Athena
Data Historian
Storage Gateway
EMR
EBS EC2 Batch AppStreamEBS EC2
E&D Workloads(PLM/EDA/CAE)
Enterprise Workloads(SAP ERP/CRM)
DMS RDS
Local Servers
RedshiftData Warehouse
Dat
a In
gest
ion
API
IoT SiteWise
Snowball Edge
Connected Products
DynamoDB Lambda
IoT Core
Amazon Forecast
IoT GreengrassConnectors
IoT Analyt icsAnalytics
Timest ream
Outposts
IoT Events
EC2
LambdaBusiness Logic
SNS
Connected Vehicles
Foundry/OSAT AWS Cloud
Smart Product
Manufacturing Applications
OPC-UA
Modbus
Customer Connector
EDGE CLOUD EDGE
Partnership with OPTIMAL+Cloud
Large data sets, powerful analytics
• Data aggregation and storage• Analytics tools and massive computing• AI/ML modeling and simulation tools• Edge deployment tools• IoT device management and security
EdgeReal-t ime, low latency, inferencing
• Real-t ime data collect ion/ harmonizat ion• Real-t ime analyt ics• Real-t ime AI/ ML inferencing• Real-t ime data feed-forward/ feed-backward• Machine cont rol and adapt ive manufacturing
Cloud infrast ructure – System architecture expert ise – Edge infrast ructure – Domain expert ise
AssemblyData
OPTIMAL+ and AWS: Edge-Cloud-Edge
Data Collection
DataTransformation
DataCleansing
O+ Data Acquisition
O+ analytics engine
O+ Edge Analytics
O+ Sequoia analytics designer
O+ Central
Portal+Rules+
O+ DataLoading
SageMaker
IoT Core S3
Lambda
GreenGrass
1. Select Training Population
2. Pull Data
3. Build Model
4. BuildInferenceService
5. Deploy to Greengrass
7. Publish Rule (Actions Instructions)
6. Define Rule
9. Inference results
10. A
ctio
n 0. Data Highway
8. Micro -batchData
Factory(internal or outsourced)
O+ data platform
Tester X Other equipment
& data sourcesO+ Proxy
Tester A
O+ Proxy
FabData
MES
Machine Learning use case – example*Use Case Skip/optimize test operation
Skip/optimize downstream operation (e.g. SLT/BI) leveraging upstream operation data (e.g. WS and FT)
Customer Value Significant OPEX and CAPEX savings and increased throughput
Why Machine Learning?1. High dimensional data2. Multi -variant data
relations3. Non-linear data
relations
* Example – this machine learning use case is just one example for many use cases that can be addressed by using the O+ and AWS integrated solution
Test operation
WS
Out
PassTest
operation FT
Out
PassPredicted
failTest
operation BI
Out
MLModel
Skip BI
Pass
Sampling for validation
Example: Burn-in• ML model shows
opportunity of reducing Burn-in by 20% with 0 impact to DPPM
• This allows to define the threshold for the manufacturing Reliability -IndexRI threshold
• Drilldown to most important features
• In this use case, the top 30 predict ive features cover ~ 95% of overall predict ive power of the ML model
© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Unlocking relevant, actionable information buried in wafer fabrication, process control and test data
Enabling secure collaboration without compromising sensitive data
PDFS Exensio® on Cloud
Customer 1SecureCloud
Environment
Customer 2SecureCloud
Environment
PDFSSecureCloud
Environment
Customer-1 HQ
Customer-2HQ
PDF Solutions and AWS
© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Example: Early Life Failure Detection (ELF)Machine Learning to predict likelihood of Early Life Failure in the field
PCM/WAT
Final Test
Sort
AssyFDC
BurnIn
Signature Library
Die/ PkgReliability Grade
A Grade
Down Grade
Manual Review
“Expert”Collaborative Learning
Defectivity
Metrology
Spatial Outlier
Ensemble
SYLSBLSTLSPL
Test OutlierEnsemble
EquipmentVariance
Process Exceptions
Reliability Indicators
ML
Scrap
Multiple data types, multiple algorithms, machine earning, potentially large data sets, collaborative learning …
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Semiconductor innovation enabled by Cloud
Flexible configuration and virtually unlimited scalability to grow and shrink your EDA infrastructure
Secure collaboration and supply chain analytics for semiconductor development and manufacturing
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Thank youITPC 2019
David PellerinHead of Worldwide Business Development, Semiconductor IndustryAmazon Web Services