16th Annual

PDF Solutions
Users Conference

S1.2 – Exensio Platform

October 15, 2019 Said Akar, GM
This presentation contains forward-looking statements
regarding PDF Solutions’ future products and business
prospects that involve risk and uncertainty. Actual results
could differ materially from those discussed. You should
review PDF Solutions’ SEC filings, including its annual
report on Form 10-K and quarterly reports on Form 10-Q,
for more information on these risks and uncertainties.
PDF Solutions does not undertake an obligation to update
any such statements.

© 2019 PDF Solutions, Inc. All rights reserved.

Exensio Platform: Big Data Analytics and Control for Semiconductors
Process Control
Detect and identify process or tool
Process problems in manufacturing in real time
Control
Test Operations
OEE optimization, escape prevention, and
yield recovery
Test
Assembly Manufacturing Analytics
Operations
Operations Higher manufacturing yields via integration
of all front-end and back-end data
Exensio
Platform Assembly Operations
Traceability of wafers, die, and multichip
modules through assembly & packaging
Process Manufacturing
Characterization Analytics Process Characterization
Analytics that support the DFI System and
CV Core System
PDF Solutions, All Rights Reserved 3
Exensio as a Platform

Platform Exensio Platform

Exensio Exensio Exensio Exensio Exensio
Products Fabless IDM Foundry OSAT …

• Manufacturing • Manufacturing • Manufacturing • Test • Future

Analytics Analytics Analytics Operations Configurations

• Test • Test • Process • Assembly

Main Operations Operations Control Operations

Modules • Assembly • Assembly • Foundry • OSAT

Operations Operations Solutions Solutions

• Fabless • Process
Solutions Control

• IDM Solutions

Cloud / Big Data

PDF Solutions, All Rights Reserved 4

Motivation – Fabless Example
Exensio Fabless

Manufacturing Analytics Test Operations Assembly Operations

Advanced Capabilities Advanced Capabilities Advanced Capabilities

Monitoring Monitoring Monitoring

Data collection / Control Data collection / Control

Fabless Solutions
Reporting
Big Data

Targeted configuration - Core capability – Consistent sub-modules – One

business model – Lite versions – Big Data across the modules –Reporting
and automation across the modules – Fabless Solutions
PDF Solutions, All Rights Reserved 5
Early Life Failure Detection (ELF)
Machine Learning based solution to
decide reliability risk – likelihood of Signature Collaborative
“Expert”
Library Learning
Early Life failure in the field

SYL
PCM/WAT Spatial
SBL
Outlier
Final Test Ensemble
STL Manual Review
SPL
Sort
Test Outlier Equipment
ML A Grade
Assy FDC Ensemble Variance

BurnIn Die/Pkg Down Grade

Reliability Grade
Defectivity Process Reliability
Scrap
Exceptions Indicators
Metrology

Multiple data types, Multiple algorithms, Machine Learning, Potentially large data sets,
Collaborative Learning, …
PDF Solutions, All Rights Reserved 6
Point Solutions not Sufficient
FDC / APC
TOOL
FDC
MES

METROLOGY WIP

E-MAIL / FTP / data files

COMSUMABLES FAB Team
ENVIRONMENTAL
YMS
MAINTENANCE
SCHEDULES (PM)

WAT Defect
DESIGN Team
YIELD/QUALITY
EDAS TEST Team
WAFER SORT/
FINAL TEST

PACKAGE PROD
ASSY Team

Multiple Databases – Multiple User Interfaces - Custom integrations - Silos with only part
of the picture available to each team – Local optimization – Many data types/formats

PDF Solutions, All Rights Reserved 7

Point Solutions not Sufficient – Tool Vs. Process Optimization
YieldAware Approach to Tool Control

Traditional FDC/APC
Sensor/Tool Audit:

Approach
Does the toolSensor
achieve tool capability?
Prioritization:

APC/SPC Limits:

Configuration Efficiency

YieldAware
Approach
Exensio Unified Big Data Infrastructure
Does the equipment achieve required process
capability?
Diagnostics

Modeling

PDF Solutions, All Rights Reserved 8

Raw Files in a Data Lake Also Not Sufficient
Intelligence Domain Knowledge

 Solutions  Unique Data

 Machine Learning  Unique Analytics
Efficiency
 Ease of Use  Semantic Models
 Collaboration Quality

ROI
Yield

Infrastructure Data Quality

Time to
 Performance Mkt  End-to-End
 Control  Traceability
 Automation  Consistency
 Security  Completeness
 Cloud

PDF Solutions, All Rights Reserved 9

Semantic Models - A Key Element for Advanced Analytics and Control
• Semantic models allow for automatically cleaning, aligning, and interpreting data

• Examples:
– Aligning events in a fab with wafer data to answer question like “which wafers were processed with
the new batch of resist”?
– Mapping equipment signals across a fleet of tools to account for configuration differences
– Meaningful merging of chip data as the chips flow through wafer sort, assembly, and final test

• Digital Twins require models and harmonized data collection to enable machine
learning

Semantic models allow our customers to

deploy advanced analytics and control to production

PDF Solutions, All Rights Reserved 10

Data Quality – End-to-End

IC Design Fab Sort Assembly Final Test System

Material MES/ WIP Fault Detection

Descriptions and Control Defect & Equipment and Assembly / External Data
Meta Data Equipment Product
History
Metrology Non-Lot System Sources
In Hierarchy (FDC)

Technology Equipment Equipment Indicators Parametric Equipment SPC sCV testchip Die traceability Other
Family Operator TrackIn/Out Summaries Categorical Fab chemical DPCV Location of performance
delivery metric databases
Process Program Recipe Trace Charts Lot / Wafer PCM reel/tube
Product Recipe Operator Summaries Equipment …
Model Wafer Sort Solder paste
Defect counter data
Source Lot Date/Time Chamber prediction Bin Map batch, vendor
Lot Process Flow QueueTime Summary Equipment Event
Exensio real Multi-Bin Equipment
Kill Ratio / PM
Wafer # Stages CycleTime time data parameters
Defect Images Consumables Final Test
Die Steps Wfr Counts collection Operator logs
Reticle
Module Data

>100 Fab Tools Types supported, >20 Tester Types supported, >160
Assembly Tool Types supported, > 50 Data types supported
PDF Solutions, All Rights Reserved 11
Data Quality – Completeness/Consistency – Data Collection/DEX
Data Exchange Network - DEX
FDC Assembly
Data Collection and Data Collection and
Control Control
Analytics
…….. ……..
Data Collection and Data Collection and
Control Control

Test Exensio Characterization

Platform
Data Collection and Data Collection and
Control Control

…….. ……..
Data Collection and Data Collection and
Control Control

> 4 PBs under management

PDF Solutions, All Rights Reserved 12
Edge Deployment - Smart Testing
Foundry testers OSAT testers Event Trigger
Feature
TXT EXN
Generation
GPU/CPU Monitoring
Table Join Execution Service

Parquet Input Prediction Service

Data Directory
Exchange
Platform Decision: Cassandra/Spark
(output spec) Output
EXN Test Directory
OSAT testers No-Test Feature Eng.
Longer Test Script
Shorter Test
Test Program A
Test Program B Python Model
…
Object

Legend: Customer Dashboard

Data Lake Model Generation
Cassandra/Spark
Customer
Model Training
PDF GUI’s Parquet
Deployment Service
(Optional) Customer
ML Model Testing
Model Validation
Training Service
Engine

PDF Solutions, All Rights Reserved 13

 Direct Data Collection and Control

>40 vendors and

> 100 equipment
models for
manufacturing

> 50 vendors and

> 150 equipment
models for
assembly

> 20 vendors and

> 50 tester / prober /
handler models for
test
Continuously Adding to the Supported Platforms
PDF Solutions, All Rights Reserved 14
Data Quality – Traceability - Single Device Tracking without ECID
o Required
– Die Attach Assembly
Exensio Operations
– Record Wafer ID + XY to Strip ID + XY
– Device Mark
– Mark package with unique ID Assembly Equipment Integration
– Upload strip map with ID + XY + Device
ID
– Singulation
– Singulate and sort good packages for
test Die Attach Laser Mark Singulation Final
Wire
– Final Test (singulated or strip test) Test
Bond
– Read ID on package and insert in test Strip
data log
Tray
o Optional
– Wire Bond, AOI Tape & Reel
– Skip FAIL bin locations and update strip Epoxy Gold Wire Package
map with wire bond defects

PDF Solutions, All Rights Reserved 15

Infrastructure – Performance
Exensio Big-Data as measured against Pure RDBMS architecture - Systems with same cost

45
RDBMS
40
(All Data)
35
40 X
Performance Multipliers

30

25
Cluster
20
RDBMS
15 Cassandra (Meta Data)
10
(Data)

5 8X 8X
Spark
0
Data Ingestion Data Retrieval Server Side Analytics
Parallel Storage + Parallel Analytics
Average results from four pilots in 2019
PDF Solutions, All Rights Reserved 16
Example Benefits of Exensio Cloud Deployment
o Reduced costs compared to on-premise
o For deployed projects range is between 18% and 34%
o 24/7 support and maintenance by expert resources
o Faster access to latest releases and faster resolution of issues
o Dynamic scaling to fit future use scenarios
o Organic growth (or reduction), Temporary needs, ML model training
o Tiered storage offered in future releases
o Reduced costs and longer online data availability benefit
Acquired company in 2015 that was one of the first to
deploy web-native analytics capabilities for the
semiconductor industry on the cloud and integrated
its capabilities within Exensio
Constantly expanding percentage of accounts
deploying Exensio on the cloud
Currently ~ 15% on the cloud with clear acceleration in
this direction
AWS and AliCloud as primary cloud providers
Traffic from/to Cloud to Customer HQ is limited Traffic from/to Cloud to Customer HQ is

IPSec VPN Tunnel

IPSec VPN Tunnel
Customer-1 by agreed on policies. limited by agreed on policies. Customer-2
Remote users HQ HQ Remote users

VPN VPN

TLS 1.2
• Consolidated environment for the enterprise Load
Balanc
er
Remote VCN peering

Compartment = PDF
Load
Balanc
er
Object
Storage
PDF Management VCN

Customer Jump Customer Jump

Host Host

• Continuous Delivery – Cloud releases and more feature/bug

App Server SF Server App Server
App Server SF Server SFTP Server App Server

Cust1 Private Subnet 1

File Storage
Cust2 Private Subnet 1

Data Encryption Data Encryption

SFTP Server

leading to continuous improvement in our solutions

Management Host
LogR Server

Oracle DB
ETL Servers Block Storage Oracle DB
Cassandra/Spark Cluster ETL Servers Block Storage
Update Server Monitoring Host Cassandra/Spark Cluster
Cust1 Private Subnet 2 Firewall
Cust2 Private Subnet 2

IPSec VPN Tunnel

Standby DB

o 10% and 40% projected cost savings depending on retention and

Standby DB Cust1 Private Subnet 3 AD2 Cust2 Private Subnet 3 AD2

VPN
Customer1 VCN Customer2 VCN
Compartment = Customer-01 Compartment = Customer-02

performance requirements
Virtual Virtual
Cloud Cloud
Network PDF HQ Network

PDF Solutions, All Rights Reserved 17

Domain Knowledge – Unique Data

Inline E-test Wafer Sort Final Test / Burn-in In-System

DFI CV Core
CVi

Front End Optimization+Control Back End Optimization+Control In-Field Optimization+Control

Exensio Enterprise Analytics

Semantic model and AI-Enabled analytics

Better Reliability (dppm, predictive fail) - Lower cost ($$) -

Improved performance (speed, power)
PDF Solutions, All Rights Reserved 18
 Ease of Use - Guided and Linked Analyses
L0 L1 Guided Guided Analysis
Dashboard Analysis Drilldowns always be available form high level
Top Level report. Bin Paretos, trends, etc.
Dashboards

What are top signals

to be concerned L2 Guided
about? Analysis
Full Drill Down
As needed, user can enter into Exensio full
client for detailed analytic capabilities

 All Guided Analysis (top level and drilldowns) relies on in-

database analytics, precomputed/summarized data, and our
big data architecture  results are fast and ready for user.

 Hierarchical structure enable users to always know where

there are in the guided analysis tree on left hand side with
easy to use navigation tools.

 Ability to enter into “ad hoc mode” into full client Exensio-
Yield capability, from any analysis as desired.

PDF Solutions, All Rights Reserved 19

Solutions: Combining Products & Expertise to Deliver Results
Advanced Insights for Manufacturing (AIM) Process & Test
– PDF’s primary Solutions infrastructure Hardware

– Exensio SW Platform + ML + Change management

Software
– Drive specific outcomes
IP
DATA Users

User System
Decision
Configuration Learning AIM solutions are not reports:
Engine
– AUTOMATIC
AIM Decision – INTELLIGENT AI + ML
– CONFIGURABLE

PDF Solutions, All Rights Reserved 20

Embedding Best-In-Class Technologies for a Complete Solution

o PDF Focuses its development efforts on the areas where it can deliver value
o We license and OEM best in class components to leverage the expertise of others, including
the open source community
o We handle all of the licensing from our suppliers, so our customers have only to reach
agreement with PDF and get all sublicenses

PDF Solutions, All Rights Reserved 21

Exensio – The Organizing Principles

PDF Domain
Knowledge
Fully scalable, big Automated Machine
data system / analytics Learning Solutions

End-to-End Data Unique data types End-to-End Analytics & Control

For IC manufacturing

IC Design Fab Sort Assembly Test System

PDF Solutions, All Rights Reserved 22

Thank You
pdf-solutions pdfsolutionsinc pdfs.inc pdf_solutions pdf_cn