Книги2 / 1993 P._Lloyd,__C._C._McAndrew,__M._J._McLennan,__S._N

customers. Finally, it was anticipated that a framework product could be made available at an affordable price, because the development costs could be distributed over a large customer base.

The problems of implementing and maintaining large software systems are not amenable to solution by relatively small academic groups. Such work has a relatively low academic content, and requires greater resources and more continuity than most academic groups can provide. There is still a place for academic activity in the area of S-TCAD frameworks and systems. Such work can be used to evaluate concepts, and to educate students in the associated issues and techniques. However academic work is not likely to be of direct significance to industry because of concerns about quality, support, and continuity of effort.

From a slightly broader perspective, the problems that are encountered in S-TCAD can be divided into two categories, 'isolatable' problems, and 'systemic' problems. The isolatable problems are aspects of the associated physics, chemistry, and mathematics that can be identified and solved using a divide-and-conquer approach, with particular subproblems becoming the focus of the efforts of a small group. The 'systemic' problems of developing and maintaining a software infrastructure has to be addressed by a different type of organization. Silvaco management saw the opportunity to be the organization that supplies the infrastructure for S-TCAD systems.

Silvaco's model for the future of S-TCAD development is that Silvaco will attack the systemic problems. The infrastructure that is developed will enable Silvaco's academic and industrial collaborators to focus on research into the 'isolatable subproblems'. The solutions that are developed will be fed back to enhance the capabilities of future S-TCAD systems.

3. The General Architecture of The MASTER Framework

The constraint of economic viability has had significant impact on the design and implementation of The MASTER Framework. The need for commercial acceptance means that the design is oriented towards the needs of users, not the needs of developers. An incremental development plan was designed so that revenues from the first phase of implementation could fund a second phase; and performance, simplicity, and maintainability were given precedence over academic elegance and conceptual originality, whenever such choices needed to be made.

The MASTER Framework has two levels (see Figure 1). The first part consists of a standard structure format (SSF) and a set of interactive, graphically-ori- ented 'MASTER Tools'. The SSF and the MASTER Tools provide the tool integration level of the MASTER Framework. This is a high quality, fileand

The MASTER Framework

DeckBuild

TonyPlot

VWF

DevEdit

NetView

MaskViews

Optimizer

Figure 1. The Two-Part Architecture of the MASTER Framework.

utility-based framework with extensive capabilities built in for the tasks of input deck debugging and calibration.

The second level of the architecture builds on the first level, providing additional capabilities that support large-scale simulation-based design and experimentation. This level is referred to as the Virtual Wafer Fab (VWF). The VWF provides the computational analog of the 'split-lot' methodology that forms the basis of empirically-based technology development. The VWF provides the design task integration level of the MASTER Framework.

The Virtual Wafer Fab

MASTER Tools

Simulator

Library

Tool Integration Layer

Design Task Integration Layer

Figure 2. S-TCAD System Structure.

A conceptual view of S-TCAD systems that are based on The MASTER Framework is shown in Figure 2. The heart of each system is the MASTER-conforming process and device simulators that are contained in the system's simulator library. A brief summary of the simulators that are presently available is provided in the appendix.

The architecture of The MASTER Framework allows exceptional flexibility in the construction of S-TCAD systems. Systems can be configured to meet the needs, experience levels, and budgets of a wide range of customers, and systems can be extended as users' needs evolve. It is normally recommended that The Virtual Wafer Fab is added only after a customer has developed familiarity with the simulators and the MASTER Tools.

4. The MASTER Tools

Figure 3. The MASTER Tools.

The MASTER Tools are DeckBuild, TonyPlot, DevEdit, MaskViews, Optimizer, and Manager. The functionality of the MASTER Tools, and the connections between them, are shown in Figure 3. DeckBuild is the central tool. It provides an interactive run time environment, and can invoke and control simulators and other MASTER Tools. The remaining MASTER Tools provide capabilities for: scientific visualization; structure, doping, and mesh specification editing and meshing; specification of layout information; black-box optimization; and dick-and-drop use of MASTER Tools and SSF files. The MASTER tools were written in C and C++ by professional programmers using extremely high standards of software engineering.

4.1. DeckBuild

DeckBuild supplies a flexible environment for generating, editing, and running input decks. It provides a graphical user interface that allows users to produce input decks without knowing simulator-specific input syntax. Information is typed into a series of pop-up windows. When specification is complete DeckBuild can automatically produce a syntactically correct input deck. Existing decks can be read in to DeckBuild, and decks can be edited directly by the user. Multiple simulators can be called from within a single input deck. Simulator invocations and information transfer between simulators are taken care of automatically. Input parameters in decks can be optimized to match known target results. A sophisticated extraction capability allows engineering parameters to be obtained from calculated results.

DeckBuild allows precise user control of how an input deck is run. It provides stop at, pause, restart and single step capabilities. It also supplies a history function that permits a user to backtrack to a previous point in a deck, and then continue computation from this point. This capability is extremely useful during the interactive development of a simulated process flow. Figure 4 shows a DeckBuild window with an edit window, an output window, a simulation control panel, and a process step definition pop-up.

~ Deckbuild V3.3 -l'md/cplu5/daveh/Workideckbuild/anex02.ln. dir. /md/cplus/dav

~ ~ ~ ~ (Maincontrolv) (Commands v) ~

~Y4k-.G dJ"I~

Figure 4. A DeckBuild window showing edit and output windows, with a process step definition pop-up in the foreground.

tion based experimentation was previously restricted to the varying of process flow parameters only. MaskViews supports simulation-based experimentation with layout variations. MaskViews enables users to investigate critical dimensions, polygon reshaping, misalignment tolerances, global shrinks, and phase shift mask parameters. It is fully interfaced to GDS2 Stream formats so that complete IC layouts may be imported and exported, and small subregions can be selected for detailed analysis.

The Optimizer provides black-box optimization, calibration, and tuning capabilities. Control of the Optimizer is integrated into DeckBuild. The Optimizer can be used across multiple simulators, i.e. it is possible to tune parameters of process simulators to obtain specified electrical characteristics from a device simulator. Optimization targets may include structural dimensions, device parameters after a complicated electrical test, and any intermediate outputs.

Manager is a simple application manager that supports interactive point-and- shoot and drag-and-drop use of a variet of files and MASTER tools. The use of this tool is very intuitve. For example, a structure file can be selected using the mouse, and dragged to to the TonyPlot icon. It will then be plotted on the screen.

5. The Virtual Wafer Fab

The VWF automates large-scale, simulation-based, design, calibration and optimization. The underlying paradigm of the VWF is the computational split lot. This mirrors the split lot methodology used for experimental development of semiconductor technologies. The differences are that simulated experiments substitute for real experiments, and workstations and software substitute for operators and equipment.

5.1. Maximizing The Economic Benefits of Simulation

Simulation provides several benefits. These include: predictive 'what-if' capabilities; physical insight; and knowledge encapsulation and reuse. Predictive capabilities allow the substitution of simulation for experiments. Eliminating some of the experiments that would otherwise be performed during technology development lowers costs and saves time. The physical insight provided by simulation helps engineers develop experience more quickly, i.e. it improves the productivity of individuals by giving them additional knowledge. The knowledge encapsulation and reuse provided by simulators improves the productivity of individuals by making it easy for them to access the knowledge of othp.rs.

All of these benefits have significant economic value, but the value is not always easy to measure. The most directly quantifiable benefits are the 'costs

of experiments avoided' when the use of simulation reduces the number of experiments required for technology development. This suggests targeting the capabilities of S-TCAD systems at maximizing the extent to which simulation is substituted for experiment. The VWF is designed to achieve this goal.

The VWF mirrors experimental split-lot methodology very closely. There are several reasons for this. First, the approach is probably very close to optimum: if more effective methodologies existed, they would already be used for experimental development. Second, the simulation methodology can be learned easily by engineers, since the underlying paradigm is already familiar. Third, when simulation mirrors experimental procedures very closely there is an increased probability that simulation will be used as a direct replacement of some experiments.

It may seem surprising that the use of simulation has not previously mirrored experimental procedures. There are several reasons for this. The most important reason is that using simulation in this way has been too time consuming and tedious. The user has been responsible for designing the 'computational experiment', generating all the associated input decks, submitting each individual simulation run, transferring data between simulators, handling storage for the generated data, and extracting useful information from the results. Some of these activities (e.g. automated deck generation, simulator invocation, and inter-simulator information transfer) are automated by DeckBuild. The VWF automates the other activities to a very high degree.

5.2. VWF Capabilities

The VWF employs a database, rather than files, for storing information; and it uses a worksheet as the primary representation of a computational experiment. The database and worksheets are central concepts of the VWF and are reviewed first in this section. The other features of the VWF will be described in the context of describing how a computational experiment is performed within the VWF.

Information storage and handling in the VWF is database-oriented rather than file-oriented. All the information associated with computational experiments is stored in an object oriented database and is accessed through an intuitive icon-oriented and menu-driven interface. The MASTER database supports libraries and workspaces. Libraries are used to store persistent objects that are useful across a project or organization. For example, standardized process recipes and electrical tests can be stored in libraries. Other possible objects are mask layouts and cross-sections. Workspaces typically contain shorter lived information used by an individual engineer, including computational-experi- ments-in-progress.

Figure 6. A portion of an interactive VWF worksheet.

A computer generated worksheet is a major focus of the computational split-lot methodology implemented within the VWF. The worksheet allows users to view input parameters and output results. The worksheet provides a way of simultaneously presenting the definition of a computational experiment, and the results that have been obtained. This representation may be manipulated very naturally through a graphical user interface. The worksheet of a computational experiment can be edited by the user at any time. The worksheet supports filtering of values, and allows columns to be specified as functions of other columns. Experiments can be added or deleted at any time. Results are logged and appear automatically as simulations are completed. A portion of a VWF worksheet is shown in Figure 6.

A computational experiment starts from an input deck developed for a baseline process flow. This deck will often have been tuned to match experimental measurements. Splits may be defined at three levels: Ie layout cross-sections; processing flow split points; and device tests. This computational procedure mirrors the specification of test structures, the fabrication of various process flow wafer splits, and the testing of individual devices.

If desired, a sensitivity analysis on specified output values with respect to selected input parameters may be performed prior to the definition of the computational split-lot experiment. The preliminary sensitivity analysis is