Abstract: This document from the MFGM032M project provides guidance for semiconductor equipment
suppliers, device makers, and other implementers of Prognostics and Health Management (PHM)
systems. The requirements for the integration of predictive and non-predictive condition-based
maintenance (CBM), scheduled preventive maintenance (PM), and maintenance optimizer
applications are discussed. The document introduces common language to describe the data
requirements and functionalities of the equipment and factory systems and uses examples to
justify extending existing PM factory systems to create an optimized maintenance system.
Note on Terminology: Predictive preventive maintenance (PPM) as a term is an invention of
ISMI. ISMI has defined PPM as “the integration of predictive maintenance (PdM), CBM, and
scheduled maintenance to optimize factory performance.” The broader industry-accepted term is
PHM, defined as “the discipline that links studies of failure mechanisms to system lifecycle
management.” The broader industry-accepted terms for PdM and CBM, as used by ISMI, are
predictive CBM and non-predictive CBM. In the interest of using broadly accepted terminology,
PHM, predictive CBM, and non-predictive CBM will be used throughout this document.
Keywords:
Table of Contents
List of Figures
List of Tables
Table 1 Data Quality for Input and Output for Scheduled PM Module..................................5
Table 2 Data Quality for Input and Output for Non-Predictive CBM Module .....................11
Table 3 Data Quality for Input and Output for Predictive CBM Module .............................15
Table 4 Data Quality for Input and Output for Decision Application...................................17
Definitions
Advanced Process Control (APC) – Techniques covering both feed-forward and feedback
control and automated fault detection, applied by both the equipment (in situ) and by the
factory (ex situ).
Degradation Rate – Rate at which the operational quality of an equipment system,
subsystem, or component decreases over time.
Equipment Data Acquisition (EDA) – The suite of SEMI Diagnostic Data Acquisition (DDA)
standards; at this writing, the suite includes SEMI E120, E125, E132, and E134. Also
referenced by Interface A are SEMI E128 and E138.
Equipment Performance Index (EPI) – Indicators composed of equipment hardware and/or
factory data. They can be used to calculate system/sub-system/component health metrics.
Fault Detection & Classification (FDC) – A methodology of monitoring statistical variations in
processing tool data and detecting anomalies.
Just-in-Time (JIT) Inventory – A strategy to improve the return on investment by reducing in-
process inventory and its associated carrying costs.
Maintenance Event – Any activity (e.g., tests, measurements, replacements, adjustments
and repairs) intended to retain or restore a functional unit in or to a specified state in which
the unit can perform its required functions.
Mean Time to Failure (MTTF) – The average time between failures with the modeling
assumption that the failed system is run until it fails without preventive maintenance.
Out of Control Action Plan (OCAP) – A controlled document detailing the procedure to
disposition product/processes or equipment repairs or maintenance activities in response to
an equipment fault indication.
Parameter Set – A set of related information that would likely be passed together as input
parameters to an operation request or output parameters in a corresponding response.
Specific examples include a process data set, a context data set, configuration data set,
etc. It is not intended to specify any aspect of the data’s persistent storage.
Prognostics & Health Management (PHM) Information – The discipline that links studies of
failure mechanisms to system lifecycle management. PHM uses information to allow early
detection of impending or incipient faults, remaining useful life calculations, and logistical
decision-making based on predictions.
Raw Data – Unprocessed data.
Remaining Useful Life (RUL) – A forecast of time or operating cycles remaining until a
failure occurs.
Run-To-Failure Methodology – A method whereby an equipment system, subsystem, and
component is fixed or replaced only after it fails.
Sensor – A component that responds to changes in the physical environment and provides
an analog or digital output value.
Subsystem – An intelligent aggregate that behaves as a unit. A subsystem is made up of
sensors and/or actuators and may contain mechanical assemblies. Multiple modules may
share subsystems.
1 EXECUTIVE SUMMARY
This document provides guidance for semiconductor equipment suppliers, device makers, and
other implementers of prognostics and health management (PHM) systems. It introduces the
PHM system data requirements and functionalities to enable integration of the building blocks of
an intelligent maintenance system; predictive CBM, non-predictive CBM, scheduled preventive
maintenance (PM), and the maintenance decision making applications. Functionalities include
gathering equipment and factory data, assessing equipment health, assessing conditional states,
forecasting performance degradation, assessing scheduled PM calendars, optimized decision-
making, messaging, and visualization.
This document complements the 2008 Predictive Preventive Maintenance (PPM) Equipment
Implementation Guideline [1], the 2009 PPM Research White Paper [2], and the 2009 PPM
Implementation Guide [3].
2 INTENDED AUDIENCE
An integrated intelligent maintenance system is complex. It is comprised of multiple
interconnected building blocks, each of which may be developed by an equipment supplier, the
integrated device manufacturer (IDM), or third-party application suppliers. The audience for this
document consists of the building block developers, computer integrated manufacturing (CIM)
system integrators, equipment engineers, and factory management.
3 BACKGROUND
The objective of PHM is to maximize equipment return on investment. A PHM system will
optimize scheduled maintenance, predictive condition based maintenance, and non-predictive
condition based maintenance. Scheduled maintenance is performed on a fixed, calendar-based
schedule. Non-predictive condition-based maintenance is accomplished by instantaneous
monitoring of equipment and by performing maintenance when an equipment health indicator
reaches a predetermined threshold. Predictive condition based maintenance is accomplished
through acquiring relevant equipment and factory data and applying an equipment degradation
model to predict the equipment’s remaining useful life (RUL). A PHM system will combine
scheduled maintenance, condition-based maintenance, and predictive maintenance to enable
effective cost vs. performance decisions.
This document assumes a modular architecture to identify as many potential system interfaces
(and hence, data sources and destinations) as possible. The document scope will not address
other applications outside of the PHM building blocks, like FDC or APC, as either sources of or
destinations for PHM data.
ISMI has defined PPM as “the integration of predictive condition-based maintenance (CBM),
non-predictive CBM, and scheduled maintenance to optimize factory performance.” The broader
industry-accepted term is Prognostics and Health Management (PHM), defined as “the discipline
that links studies of failure mechanisms to system lifecycle management.” In the interest of using
broadly accepted terminology, PHM will be used throughout this document.
4 INTEGRATION NOTE
The optimum configuration for integration of the PHM system into a factory will depend on the
particular factory’s existing infrastructure. Even within the PHM system, the individual
subsystems may stand alone or be combined. It is likely that the decision application and the
scheduled maintenance system can be easily merged, since these two subsystems reside in the
factory and consume similar data. It is also likely that the non-predictive and predictive CBM
systems can be merged, since these two subsystems consume vast amounts of overlapping data
from equipment and factory and may reside closer to the equipment in the factory architecture.
This document will treat each subsystem of the PHM system as independent modules in the
interest of maximizing flexibility in integration. A large portion of this document will discuss the
modules of the complete system, their functions and libraries, and their place in the factory. This
discussion of each module will be followed by a list of the data entering and exiting the module
(i.e., the data flow). This data flow discussion will include a discussion of the data content and
data quality for each data flow.
5.1 Scheduled PM
The scheduled PM system maintains a calendar of all scheduled PMs. Scheduled PMs are those
maintenance events that are scheduled on the basis of elapsed calendar time. In the
semiconductor factory, some of these scheduled PM events are monthly PM, quarterly PM,
semi-annual PM, and annual PM. The content and timing of these events are typically specified
by the OEM, and maintaining tools in accordance with the OEM specification is often a
condition of a warranty. Every predictive CBM and non-predictive CBM related maintenance
event will have a record in the database to which the CBM modules will write updated
information as those modules generate the associated signals.
Changes to the schedule should reseed the future date assignments for the scheduled
maintenance for the subject tool. Event IDs are unique to the event, such that multiples of the
same tool and model that require a common maintenance event have the same event ID. The
scheduled PM system is a factory-level system. In the event that OEM provided scheduled PM
systems are already in use, or elected for use, the output of the OEM systems will require
integration into the factory-level system. The OEM system, therefore, must output data in the
format required by the factory-level system. A pictorial representation of the scheduled PM
system is shown in Figure 2. In the figure, a manually rescheduled event is shown in the
scheduled maintenance calendar as m→m to denote that the monthly PM for the specified tool
was manually delayed. Unique events invoked by the predictive and non-predictive CBM
modules are also shown on the schedule. The required data from the CBM systems is the same as
for manual data input for the tool ID and event ID, but in place of a frequency the CBM system
passes a specific date.
Note: Showing information flow from scheduled and non-scheduled CBM systems through the scheduled maintenance system
and displayed on the scheduled maintenance calendar.
Table 1 Data Quality for Input and Output for Scheduled PM Module
Data Source Type Purpose Form Format Items Special Notes
Manual Input Input Set-up Association Text strings, Tool ID, Event New strings
GUI table numerical ID, Frequency, need consistent
data, dates Entry Entity definition and
use (e.g., Entry
entity - CBM,
Event ID =
LamESC1)
Manual Input Input Maintenance Maintenance Text string, Tool ID, Event Removes event
GUI tracking performed date ID, Date or reseeds next
history table event timing
Manual Input Input Manual Maintenance Date Select Tool ID, Re-seeds next
GUI reschedule lookup table Select Event occurrence of
ID, enter date event
Manual Input Input Messaging set Messaging Text strings Distribution Associates
GUI up lookup table lists, events events to
messaging DLs
for consumption
by messaging
engine
User Request Input visualization Job guild input Menu Select Tool ID, Generates user
GUI Job stepwise selection, input start/end output report
Generation dates dates, select
visualization
type
Non-predictive Input Record update Association Text strings, Tool ID, Event For display in
CBM table numerical ID, Confidence schedule
data, dates RUL
generated
Date
Process Equipment
USER
CBM Libraries
Visualization Job Builder
• Health Indicators
1. Select Health Indicators • Visualizations
Equipment Data 2. Select Visualization Type • Message Distribution Lists
3. Define time period
• Equipment Raw Data
4. Build Data Fetch, execute Fetch
• OEM defined Health Indicators
5. Process Data with Health Indicator Function
6. Build Graphic & Display
Non-Predictive CBM
System Envelope
Metrology Equipment
Factory/Metrology Data
User Requests:
Adjust Specific Thresholds
Specific Virtualization Requests
functions
Calculate Health Indicators libraries
Compare to Threshold Health Indicator Library with Source Data List
Health Indicator Equations/Functions
Visualization Engine
Health Indicator Threshold & Event Association
Message or Invoke Message Visualization Library
Message Association to Maintenance Event
Output to Schedule PM System Message Distribution Lists
Table 2 Data Quality for Input and Output for Non-Predictive CBM Module
Data Source Type Purpose Form Format Items Special Notes
User Input GUI Input Set-up Association Text strings, Tool/component Unique
table SVIDs ID, Health equipment
indicator ID, health indicators,
source data IDs, the data
equation/function elements
required to
generate the
health indicator,
and the function
to use from the
function library
User Input GUI Input Threshold Association Text strings, Health indicator User sets health
setting and table numerical data ID, threshold, indicator
event maintenance threshold for
association event ID maintenance
set-up event trigger
and/or
messaging
trigger
Manual Input Input Messaging Messaging Text strings Distribution lists, Associates
GUI set-up lookup table events events to
messaging DLs
for consumption
by messaging
engine
User Request Input Visualization Job build input Manu Select Tool ID, Generates user
GUI job generation stepwise selection, health indicators, output report
dates input start/end
dates, select
visualization type
Maintenance Output Record update Association Text strings, Tool ID, Event For display in
trigger to table update dates ID, date schedule
scheduled PM
module
Visualizations Output Health Graphic or File File For display or
dashboard table or list output
The large overlap in data requirements and functionality between non-predictive CBM and
predictive CBM leads to a conclusion that the modules may be merged into one CBM system.
Although this is logical, and time may prove that this is the preferred integration option, this
document will assume separate non-predictive and predictive CBM modules. This allows for
maximum flexibility, and addresses the significant differences in complexity (and thus required
expertise to develop) of health indicator development in non-predictive CBM and prediction of
RUL in predictive CBM.
Predictive CBM implementation requires tool-specific development. Predictive model
construction will be unique to each OEM’s tool. The source data within the equipment that is
consumed in the predictive models are unique to each OEM’s hardware, naming convention, and
command and control software. The development work to establish these models may be
performed by the OEM, an IDM, a third party, or any combination of these parties. Health
indicators, raw equipment data, and factory data will be consumed by the predictive CBM
application. In the interest of efficient data transfer and storage, it is recommended that data be
acquired once from the equipment and then stored in an equipment data repository for
consumption by applications, such as CBM and FDC. This “collect data once” principle may
require either a “collect all equipment data” strategy or a data collection plan manager in the
factory.
This document assumes that data collection for consumption by the CBM modules will be
managed outside of the application. The collected data that is needed by predictive CBM
modules will be “fetched” periodically as instructed manually or by a programmed periodic
refresh. The form and format of the health indicators, and any source data from the equipment
should comply with SEMI E120, E125, and E134. Any derived health indicators, calculated
within the equipment envelope and exported to the equipment data repository for consumption
by CBM applications, must also be put in context consistent with SEMI E120, E125, and E134.
Factory data may be required to calculate RULs. The association of factory data to equipment
data is accomplished as discussed in Section 5.2.
The predictive CBM application will reside outside the equipment envelope. This requirement is
necessary for the following reasons:
The predictive CMB module will require factory data as input to the calculation of RUL
for some cases
The predictive CBM module will include a manual modification of models at the user’s
discretion
Factory-based applications are easier to propagate common elements of the application
throughout the factory
Figure 6 shows a summary of the predictive CBM module functions and libraries. Figure 7
shows user set-up and visualization job building example.
User set up of the predictive CBM module requires:
Model list with output RUL association to equipment/module/subsystem/component
List of required data (equipment and factory) for each predictive model
Definition of the predictive models (modular code input to model library)
User Requests:
Adjust Specific Thresholds
Specific Virtualization Requests
functions
Calculate RULs libraries
Compare to Thresholds RUL with Source Data List
Predictive Models
Visualization Engine
RUL Thresholds & Event Association
Message or Invoke Message Visualization Library
Message Association to Maintenance Event
Output to Schedule PM System Message Distribution Lists
Process Equipment
USER
Factory/Metrology Data
Figure 7 Predictive CBM User Set-Up and Manual Visualization Job Build and
Execution Example
Table 3 Data Quality for Input and Output for Predictive CBM Module
Data Source Type Purpose Form Format Items Special Notes
User Input GUI Input Set-up Association Text strings, Tool/component Unique RULs,
table SVIDs ID, RULs, source the data
data IDs, Model elements
ID required to
generate the
RULs, and the
model to use
from the model
library
User Input GUI Input Threshold Association Text strings, RUL ID, User sets RUL
setting and table numerical data threshold, thresholds for
event maintenance maintenance
association event ID event trigger
set-up and/or
messaging
trigger
Manual Input Input Messaging Messaging Text strings Distribution lists, Associates
GUI set-up lookup table events events to
messaging DLs
for consumption
by messaging
engine
User Request Input Visualization Job build input Manu Select Tool ID, Generates user
GUI job generation stepwise selection, RUL, input output report
dates start/end dates,
select
visualization type
Maintenance Output Record update Association Text strings, Tool ID, Event For display in
trigger to table update date matrix ID, date-recipe schedule and
scheduled PM sets access by
module decision
application
Visualizations Output RUL Graphic or File File For display or
dashboard table or list output
scenarios may be run in session (or regularly scheduled to run), and the calendar views from of
the optimized and baseline schedules reported out for management. The IDM may elect to allow
the decision application to set maintenance schedules or may elect to retain manual control.
Maintenance schedules generated will highlights any changes with the source of the change
noted. Thus, if the CBM application output is that a specific maintenance event existing on the
schedule in the future is required immediately, and the decision application schedules that
maintenance immediately, then the recommended maintenance schedule should highlight that the
change was invoked by the CBM output and recommended by the decision application. Once the
maintenance event is performed, the “date performed” field in the maintenance record is updated
and the event record in the scheduled PM association table is reset to the post-tool-qualification
RUL.
User Requests:
Scenario Requests
Automation Set-Up
functions
libraries
Calculate Recommended Actions
Visualization Engine Rules Library
Optimization Scenarios
Messaging Visualization Library
Message Association to Decision Actions
Output to Schedule PM System
Message Distribution Lists
Table 4 Data Quality for Input and Output for Decision Application
Data Source Type Purpose Form Format Items Special Notes
User Input GUI Input Rules set-up Rules table Text strings, Input unique rule Define rules
logicals, ID, select data
values ID, select logical
argument, select
value
User Input GUI Input Scenario Association Text strings Input scenario Define sets of
definitions table name, Select rules
rules from rules
tables
User Input GUI Input Automation set Auto Run table Text strings, Select scenarios, Runs selected
up data input frequency scenarios
(frequency) automatically at
frequency.
Extend to
automate output
Manual Input Input Messaging Messaging Text strings Distribution lists, Associates
GUI set-up lookup table events events to
messaging DLs
for consumption
by messaging
engine
User Request Input Visualization Job build input Manu Select Tool IDs, Generates user
GUI job generation stepwise selection, scenarios, time output report –
dates period, select calendar views
visualization type
Recommended Output Record update Scheduled PM Text strings, Tool ID, Event For display in
changes to association date matrix ID, date-recipe schedule output
events on table update sets
maintenance
schedule
Visualizations Output Scenarios Graphic or File File For display or
calendars table or list output
7 SUMMARY
PHM calls for a paradigm shift from scheduled preventive maintenance plus unscheduled
maintenance towards continuous, data-driven, condition-based monitoring using CBM strategies.
Proper application of these CBM strategies in a PHM system with joint production and
maintenance scheduling will allow the IDM optimize factory performance. Integration will
necessarily be a complicated IDM-specific challenge that can be eased only partially through
standardization or modular development strategies. Health assessment, degradation forecasting,
maintenance and production scheduling can be merged in a PHM system to raise maintenance
activities to next generation standards.
8 REFERENCES
8.1 Documents
SEMI E120: Specification for the Common Equipment Model (CEM)
SEMI E128: Provisional Specification for XML Message Structures
SEMI E125: Specification for Equipment Self Description (EqSD)
SEMI E132: Specification for Equipment Client Authentication and Authorization
SEMI E134: Specification for Data Collection Management
SEMI E138: XML Semiconductor Common Components
SEMI E139: Specification for Recipe and Parameter Management (RaP)
Various AEC/APC Conference Proceedings. (Asia, Europe, U.S., 2005-2008)
W3C: XML Schema Part 2: Datatypes Second Edition
http://ismi.sematech.org
e-mail: info@sematech.org