2.0 Demo of Functions & Programs
2.1 System documentation hyperdek operation demonstration
2.1.1 Corporate Operation Demo Generation
220.127.116.11 Load the company generator with the generic product data sheet.
18.104.22.168. Populate the functional capability of each group, department and section.
22.214.171.124. Initiate start-up sequence of groups, departments and sections.
2.1.2 Engineering Procedures Demo Operation
126.96.36.199 Initiate generic company report system.
188.8.131.52 Initiate generic analytical algorithms.
184.108.40.206 Initiate the Cyber NPI System generic operation.
2.1.3 Cyber NPI System OperationDemonstration
220.127.116.11 Real time, On Line Multi-Agent NPI Operations sequence.
18.104.22.168 Real time, On Line Data Uploading sequence.
22.214.171.124 Real time, On Line Cyber Advisor Interaction a system impliments a generic demo functional capability based on each of the 27 sections having 27 distinct functions.
Appendix - Function
II-A Standard Analytical Methods
II-A(1) Company & NPI Simulation
II-A(2) Circuit Simulation
II-A(3) Manufacturing Simulation
II-B. Advanced Systematic Analytical methods
II-B(1) Spatial Analytical Methods
II-B(2) Temporal Analytical Methods
II-B(3) Conditional Analytical Methods
II-C. Comparative N-dimensional Analytical methods
II-C(1) The Integrated Yield Management Triangle(IYM)
The original IYM Triangle — Figure 1 - was developed by Nick Atchison and Ron Ross as a teaching tool to present a schematic representation of the data analysis techniques that were in use at the time, ~1996. The diagram was used to organize a hierarchy of analytical procedures capable of predicting the FAB yield and performing root cause analysis of process and design problems.
II-C(2) The 6 Triangle Stack (STS)
To correct limitations of the IYM Triangle, a schematic diagram of yield analysis consisting of a stack of identical, repeating analysis diagrams was developed. Before the stack of diagrams could be made, a simple yet general analysis method that could be used at all levels had to be developed. The elements of the analysis had to be hierarchically arranged so that analysis would move sequentially from the top general level to bottom root-cause level.
II-C(3) The N-Dimentional, Holographic Analysis ov Yield Variation and Cause
To correct limitations of the STS Triangle, a schematic diagram of yield analysis consisting of a stack of identical, repeating analysis diagrams that corresponds to the “atomic” production flow chart was developed. Before the stack of diagrams could be made, a revised simple yet general analysis method that could be used at all levels had to be developed. The elements of the analysis had to be hierarchically arranged so that analysis would — at each step of the flow chart — be able to move sequentially from the top general level to bottom root-cause level.
II — Automated NPI Monitoring & Performance Analysis Methods
The third segment of this paper proposes a modular, evenly branching explicit testing as the simplest, low entropy method to initiate and monitor the performance of fabless semiconductor companies. This performance, monitoring and analysis is based on a general performance, monitoring and analysis that is inherent to all fabless startups. The monitoring and analysis methods presented are at the forefront of modern AI capability.
Note that the product data sheet which is expanded by technology, design and operations becomes the “engineering data sheet” that is used to personalize the general specification system. The super data sheet has an entry for each of the 1092 functions that specify any particular restrictions and requirements related to a given operation. It is important to note that the modular specification of the testing is important to the modular data analysis.
The stability of the product is reported as the deviation from the ideal performance stated in the “super product data sheet”. The average wafer FAB has over 2000 SPC data sheets derived from measurements taken on wafers that are updated hourly. The 50% of the 1000 units of equipment used to fabricate the wafer may be generating several gigabytes of information a day via their electronic chart recorders. At e-test 200 to 500 tests are performed on up to 10 sites per wafer on as many as 24 production wafers per lot. At wafer sort all die on all wafers may be tested as many as 5 times. At package test each package may be tested up to 5 times. Quality assurance and Reliability add additional complexity to the testing process and enormous amounts of data. All of the tests mentioned so far need to be cross correlated in a meaningful way. This work must be done with automated data analysis systems that can provide early warnings of subtle, aberrant changes that are indicative of pending disaster. When defects do occur, their root cause must be located and corrected quickly.
The key to the advanced AI capability is derived from a built in “Holographic” production numbering system. In the past I have gone to great lengths to convert widely disparate numbering systems into one “tag” system that allowed me to program data analysis routines using “C” pointer math. This essentially like machine language — the fastest way to do things on a computer. The manufacturing numbering system and customer ordering numbering system is based on and congruent to the modular organizational format discussed so far. The long manufacturing number is has two main parts. The first part is a specification index number of a list of the specifications and date code of each specification that was active during the fabrication of the part. The second portion of the manufacturing part number is a serial number of an index of all of the vendor part numbers applied to the manufacturing lot. For parts that have wafers and die identification, the index number is distinct for each die.
All vendors now publish their foundry data on EXCEL data sheets that are available on the Web. Simple AWK or Perl program ingest the vendor supplied into tables that are digested by more complex data analysis programs.
IIA. Non-Systematic Data Analysis Tools:
Non Systematic data analysis tools Such as SPC charts are in common use. Alarms occur after the fact of a single measurement value excursion or tendency is detected. Standard procedures will not be discussed here. In Appendix I there is a list of articles and example programs related to non-systematic yield analysis
IIB. Systematic Yield Management Tools:
In the semiconductor industry today it is common to use off the shelf commercial data analysis and new product introduction packages which can cost over a million dollars a year to support. The articles presented in this web page provide access to some of the enabling statistical techniques on which the commercial tools are based. The tools can be programmed and maintained by the individual engineer. Many of the tools can be used with EXCEL. Before going into the actual data analysis methods, it is first important to understand the fundamental structure of Systematic Yield Management as a comprehensive analytical system. Next is a discussion of spatial, Temporal and Relational Yield Tools
IIB(1) Spatial Yield Tools:
There are actually a wide range of spatial yield tools such as die map, stepper field map, wafer map, boat map, etc.The next picture shows a wafer zone analysis plot. I developed this method after reviewing 4,000 wafer maps from one product line. The original analysis method used EXCEL spread sheets. In later renditions I converted the analysis to RS1, C and C++. The yield of each machine in FAB and Test were monitored by zone. This was done because the different machines in FAB tend to have lower yields in specific regions of the wafer. Taking a simple average of the data values by wafer will decrease the accuracy of the analysis. Most FAB data analysis tools now do this kind of analysis automatically.
WAFER ZONE MAP
This is a wafer map that compares two values to each other in wafer space.
This could be a comparison of one wafer sort value to another or an E-test value vs. a Wafer Sort value. When this type of plot is used with more complex where an e-test value is plotted against a wafer sort value and against a post burn-in leakage test the basis for whole system holistic data analysis is developed.
IIB(2) Temporal Yield Tools:
Yield as a function of time. ( Quad Plot To Be Added. ) Although the Quad Plot contains four different plots of which only one plots results with a time X axis, they all actually occurrences that are time dependant.
IIB(3) Relational Yield Tools:
Note that each element of the production flow char is reviewed and classified as to the type (occurrence pattern) of faults they can cause. In this case the relationship between the values observed in terms of there distribution — their relation to one another. Different equipment units or processes can create different types of distributions.
The review of the type of occurrence pattern observed begins to focus on possible causes.
Chart #1 Diagrammatic Analysis of distribution Types
Chart #2 Weighted Fish Diagram Classifying Problem Affect by Distribution Effect
Chart #3 Table of Diagrammatic Analysis Results
1) Initial Bond Strength
Process Settings Governed by specification — would cause problems to the whole lot
Polyimide trapping Not seen as a problem at ASE incoming inspect
Pad Metal Damage Under process control — all die processed pass incoming inspect
Contamination Under process control — all die processed pass incoming inspect
2) Reduction of Bond Strength
Glue On BP Before Bonding Under process control — all die processed pass incoming inspect
High Dt during Assembly DOE testing shows that this is not a problem
Glue Polymer on BP Under process control — all die processed pass incoming inspect
Purple Plague DOE testing shows that this is not a problem
3) Increased Snap Forces
Over Temp @ B.I. Under process control — all die processed pass outgoing inspect
Package Warping Checked a failing package — was not warped
MSL 3/5 Con B/C Checked several failing lots
Assembly Damage Causes gross leakage
4) Freak Catastrophes
Probe Pad Cratering Fail rate is independent of the number of probes
Glue Touches Ball Bond Proven to be capable of being a cause that is occurring at this time
Glue Touches Bond Wire Proven to be capable of being a cause that is occurring at this time
Glue/Die Edge Over Flow Proven to be capable of being a cause that is occurring at this time
Systematic Yield Methodology: (SYM)
The basic analytical procedure that underlies all SYM methods is an everything to everything comparison program using that originally used the RS1 language which was known as “V0D0”. In the early RS1 format, analysis took 6 days to run and brought the Engineering Alpha VAX work schedule to a standstill. I rewrote the program in C using 3 level pointer math and the run time dropped to 6 seconds. Once this tool was up and running, many complex SYM tools were developed including Product Wafer Sort Yield Sensitivity to E-test Parameter Value Analysis. This tool became known as Product Sensitivity Analysis — PSA. TI licensed PSA to DataPower which was bought out by PDF Solutions. The first complex WEB based version of V0D0 used a web page as the GUI which called CGI tools to activate Unix shell programs which in turn called the C code. The graphic output of this system was accomplished when the C code finished the data table and returned control the the Unix shell program which then called a PERL program using a GD graphics module to generate and return the graphs to the users web page.
The image below is a screen shot of a graphic output developed using “G” a public domain 3D Stereo graphic language I developed with H. Wolderidge, which uses a PC compatible version of Cal Tech Intermediate Format (CIF )language. G will be presented later in the appendix.
PSA - Stochastic Graphical Analysis Using “G”
Articles related to Systematic Semiconductor Data Mining Methods:
Patents related to Systematic Semiconductor Data Mining Algorethems:
Patented by Nick Atchison and Nick Atchison & Ron Ross or Nick Atchison & Brad Ferrell. You can use the CROSS-REFERENCES TO RELATED APPLICATONS and Referenced By sections of the patents to locate other related patents by other individuals.
IIC. Holographic Yield Management Tools:
HYM is the next step up from Systematic Yield Analysis (SYM). This set of tools moves beyond the Systematic Yield Methodology (SYM) in that it does not simply compare two or three variable at once. HYM looks at new product development as an integrated whole. The IYM Triangle is the first attempt to diagram the analytical process as a whole. It presents a hierarchical schematic of FAB/E-Test/W-Sort data and related analysis methods.
A clear understanding of the IYM Triangle is necessary because the choice of the analytical sequence used to find the root cause of a given problem is not simple and varies according to type of problem. Very often, the power of the tools can and do overwhelm the non-expert user. On the other hand, an expert yield analyst can do the required extractions and analyzes using a simple SQL data base or — in the case of a start-up - file based system. In fact, all of 42 basic analytical methods now in commercial and home grown data analysis packages that were developed by expert data analysts working for major semiconductor companies. The commercial data analysis package vendors have pulled the diverse methods used by the experts together into more accessible and usable tools.
In-Line Statistical Process Control gathers the data used at the bottom of the IYM Triangle. At each tier of the IYM Triangle more data are added. This means that data from each operation can be analyzed laterally (within tier) and vertically (tier to tier) during root cause analysis. Non-Systematic Statistical Process Control looks at data at each operation. Systematic Yield Analysis looks for correlations between within tier and tier to tier data. Holographic Yield Analysis uses Hypergraphic methods to look at all cross correlations as holistic phenomena. The IYM Triangle provides a schematic for top down diagnoses of the root cause of a yield problem.
The IYM Triangle
The original IYM Triangle — Figure 1 - was developed by Nick Atchison and Ron Ross as a teaching tool to present a schematic representation of the data analysis techniques that were in use at the time, ~1996. The diagram was used to organize a hierarchy of analytical procedures capable of predicting the FAB yield and performing root cause analysis of process and design problems. Note that the IYM Triangle is a essentially a binomial hierarchy
< o:p> o:p>
The 6 Triangle Stack
To correct limitations of the IYM, a schematic diagram of yield analysis consisting of a stack of identical, repeating analysis diagrams was developed. Before the stack of diagrams could be made, a simple yet general analysis method that could be used at all levels had to be developed. The elements of the analysis had to be hierarchically arranged so that analysis would move sequentially from the top general level to bottom root-cause level. Figure 2 shows the Six Stack Analytic Triangle that was developed to eliminate the limitations of the IYM method.
A new systematic root cause analysis method has been developed at that has proved to be instrumental in finding the root cause of low yield at package test, assembly, wafer sort, E-test and all FAB process stages. In this paper, simple six level diagram is presented and compared to the older single triangle IYM Triangle diagram to explain the enhancements related to the new stacked triangle method. The sequence of operations described in the hierarchical structure of the 6 level schematic diagram describes a sequential analytical method that limits the number of analysis that need to be done. By working through the levels correctly, only the critical analyses are done.
< o:p> o:p>
< o:p> o:p>
< o:p> o:p>
(Each of the following files is very large. You may have to down load the file to disk then view it. They require Open Office.)