VWF Automated Design of Experiments (DOE) and Optimization - - PowerPoint PPT Presentation

VWF

Automated Design of Experiments (DOE) and Optimization Framework

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Contents

• Introduction
• Benefit and Advantages
• Architecture
• Case study
• Conclusion
• 2 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Introduction

• 3 -

VWF is software used for performing Design of Experiments (DOE) and Optimization Experiments. Split-lots can be used in various pre-defined analysis methods. Optimization algorithms can be used for automated parameter variation. Scripts can be used to define custom DOE algorithms. Split parameters can be defined for any of Silvaco’s process, device, parasitic extraction and circuit simulators. All simulations can be carried out in parallel either on a cluster of workstations or on a single SMP machine. VWF comes with a GUI, that also enables examination of experimental results.

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Advantages

• Integrates SILVACO’s simulators into one graphical user interface
• Offers wide range of pre-defined DOE strategies

(including: Full-Factorial, Half-Factorial and Box-Behnken)

• Offers a JavaScript interface to implement own DOE strategies
• Allows to import DOE definitions from a comma separated values

(CSV) file

• Offers many Optimization Algorithms

(including: Levenberg-Marquardt, Genetic Algorithm, Parallel Tempering, Simulated annealing)

• Offers a JavaScript interface for scripted optimization target

computation

• Supports Sensitivity Analysis: Study the effects of varying process

input parameters on structural and electrical characteristics

• 4 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Advantages

• Is Time and processor efficient – Supported computing

environments range from single SMP workstations to large grid computing clusters (including Sun Grid Engine and LSF)

• Supports both database (powerful SQL-92 compliant database)

and file mode data storage environments

• Increases productivity by ease of use and automation
• Facilitates effective transfer of technology from development to

manufacturing

• Reduces cycle-time for development of new process technology
• Allows central TCAD groups to generate results which are used by

general engineering groups in a simple to use environment

• Supports the Silvaco SRDB database concept to backup and

restore databases easily

• 5 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Advantages

• Supports multiple databases
• Seamlessly invokes analysis tools (e.g. TonyPlot and SPAYN)
• Allows Response Surface Model (RSM) generation
• Effortlessly exports worksheet data results in common formats

(e.g. CSV)

• Offers powerful import and export capabilities to easily

exchange data via zipped TAR files

• Offers an Interface to the CVS version control system to import

simulation decks

• Supports a JavaScript batch mode to sequentially run

experiments outside the main VWF application.

• Comes with a comprehensive set of representative examples

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture

 Graphical User Interface (GUI) – Main application to define, control,

run and analyze experiments

 VWF Background Module – Runs experiments in the background

when GUI is closed

 VWF Database – Stores data like the simulation deck and split

definitions, as well as simulation results

 VWF File space – Holds all simulation result files like generated

structures and runtime output

 VWF Visualization Module – Offers powerful visualization and post-

processing facilities

 VWF Import/Export – Allows to move an experiment from one

database to another

 VWF Backup/Restore facility – Perform regular automated backups of

data

• 7 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Database Access

• Access multiple databases on several hosts through user name

and password

• 8 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Secure Database Explorer

• Protection of experiments by individual user permissions

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• Input decks can be loaded from text files, pasted from another

directory, imported from deckbuild, or loaded from a CSV repository.

• 10 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• Any parameter from any simulator can be defined graphically as

a variable

• 11 -

QUEST ATLAS SmartSpice

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• External files used in the input deck are loaded in the database

using the Resources menu

• 12 -

External file

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• Parameters can be varied in many ways, ranging from manual

selection over pre-defined DOEs, to custom DOEs in JavaScript script, or one of many optimizer algorithms.

• 13 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• Meaningful description of the experiment can be added within

the Description Tab

• Actions made on each experiment are recorded and can be

retrieved in the Logging pane

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• Properties of the experiment can be comprehensively

configured using the Setup Tab

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

 Open Interface (Java Script) for implementing own DOE

algorithms

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Definition

• Open Interface

(Java Script) for implementing custom

• ptimization target functions
• Complex experiment types

combining DOE and

• ptimization

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Execution

• Sun Grid Engine (SGE) and LSF Compatible
• Can utilize local multi processor machine
• 18 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Execution

• The experiment and its status can be viewed as a Tree, a

Worksheet or as Jobs

• 19 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Execution

• Run-time output and simulation results are attached to each node

and are available in real time

• Can be retrieved from all views
• 20 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Execution

• Summary of extracted data is available in real time within the

Worksheet Tab

• 21 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Execution

• View output of simulation

results and JavaScript target function (cost) for combined DOE and

• ptimization experiments

Target function (cost) Simulation results

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Analysis

• Results can be analyzed directly in the Worksheet in text form or

in graphical form (TonyPlot interface)

• 23 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Analysis

• Efficiently select files for plotting in the SplitPlot Worksheet Tab

Cell numbers (3.1, 3.3 and 3.6) are shown in the plot

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Analysis

• Convergence of the optimizer can be graphically visualized

during an optimization experiment

• 25 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Analysis

• A direct link to SPAYN allows statistical analysis and RSM

generation for DOE experiments

• 26 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

VWF Architecture: Experiment Analysis

• RSM generation done in SPAYN can be visualized in TonyPlot
• 27 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Case Study

• Sensitivity Analysis
• Impact of Process Variation on Circuit Performance
• 3D Parasitic Capacitance Optimization
• 3D Stress Simulation for Mobility Enhancement Optimization
• CIGS Solar Cell Optimization
• Ge solar cell External Quantum Efficiency Optimization
• Inductor Performance Optimization
• Inductor PDK Generation
• Comparison between Doe and Optimization approaches
• Managing Circuit Simulation
• 28 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Sensitivity Analysis

• Sensitivity analysis in VWF allows quick and efficient parameter

screening in order to reduce the number of variables to be used in a subsequent DOE

• 29 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Impact of process variation on circuit performance

• CMOS layout driven Ring oscillator simulation was done in VWF
• Key circuit figures of merit (i.e frequency of the ring oscillator) can

be plotted versus process splits

• 30 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Impact of process variation on circuit performance

• Statistical summary, histograms and correlation matrix of

simulated data between and for all variables (process parameters) and corresponding extracted parameter (ring oscillator frequency)

• 31 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Impact of process variation on circuit performance

• Response Surface Model of ring oscillator frequency as a function
• f gate oxidation time and Vt implant dose
• 32 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Impact of process variation on circuit performance

• Ring Oscillator frequency yield analysis, based on user defined

input distribution of each process parameter.

• 33 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Parasitic Capacitance Optimization

• CLEVER 3D process based RC parasitic extraction tool can be used

in VWF to study back end process induced capacitance variation

• 34 -

3D Inverter built in CLEVER 2D cut-line along the red line

IN OUT

The capacitance between IN and substrate will be calculated as a function of PMD_THICK

IN OUT

PMD_THICK

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Parasitic Capacitance Optimization

• PMD thickness is optimized to reduce the overall capacitance
• 35 -

IN OUT

PMD_THICK

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Buried Oxide Fin Gate SiN

3D Stress Simulation for Mobility Enhancement Optimization

• Nominal 50 nm 3D FinFet structure.

Buried Oxide thickness is fixed at 400 nm. Vary : Fin width Fin height Gate length

• 36 -
• 36 -

height/width 50 / 50nm height/width 150 / 50nm Lg = 50nm

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Stress Simulation for Mobility Enhancement Optimization

• 3-D stress contour profiles along the channel in the <100> fin

direction

• 37 -
• 37 -

Channel under tensile stress Channel under compressive stress

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Stress Simulation for Mobility Enhancement Optimization

• 3-D contour profiles showing stress distribution in the

polysilicon gate

• 38 -

FinFET under tensile stress FinFET under compressive stress

• 38 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Stress Simulation for Mobility Enhancement Optimization

• 3D FinFET structures showing gate length and height variations.

Silicon nitride capping layer is not shown

• 39 -
• 39 -

Gate length variation  Gate height variation 

50nm 100nm 250nm 500nm 50nm 75nm 150nm 300nm

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Stress Simulation for Mobility Enhancement Optimization

• Mobility enhancement for <100> nominal structure with various

gate thicknesses when tensile and compressive capping layer (1Gpa) is applied

• 40 -
• 40 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

3D Stress Simulation for Mobility Enhancement Optimization

• Mobility enhancement for <100> nominal structure with various

gate length when tensile and compressive capping layer (1Gpa) is applied

• 41 -
• 41 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• CIGS structure and Photo-generation Rate
• 42 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• Simulate all the variations in VWF
• Integrated statistical package SPAYN loads all the results from

VWF data base

• 43 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• Click “Tools…

Regression” to create a Response Surface Model (R.S.M.)

• Choose a “Target”

and “Input Parameters” to create the model

• 44 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• Click “TonyPlot” to visualize the RSM (Response Surface Model)
• 45 -

Solar Cell Efficiency RSM using two variables ZnO thickness X-Composition Optimum values can be visually

• btained

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• 46 -

Optimized parameters also for 4 variables and more can be

• btained from the RSM

created by VWF through the integrated statistical package SPAYN. (here optimized with respect to a target efficiency of 10%)

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• Effects of ZnO Thickness Variation (process induced)
• 47 -

In this example, we analyze the effects of ZnO thickness variation during production. A “Gaussian” thickness variation is input into the model. We can instantly see the effects on the solar cell

• utput.

VWF – Automated Design of Experiments (DOE) and Optimization Framework

CIGS Solar Cell

• Efficiency Variation due to Thickness Variation
• 48 -

Here we analyze the effects of thickness variations on Solar Cell Efficiency.

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Ge Solar Cell

• Ge solar cell simulation in VWF

Note that string can be defined as variable

• 49 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Ge Solar Cell

• Ge solar External Quantum Efficiency (EQE) as a function of

– optical model (RTM versus TMM) and – substrate thicknesses

• 50 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Ge Solar Cell

• Ge solar cell EQE using TMM as a function of

– surface velocity recombination and – InGaP thicknesses

• 51 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Inductor Performance Optimization

• The objective is to optimize a Q factor versus Frequency curve as a

function of process parameters

• To do so a scalar target is defined by comparing with a target curve so that a

global optimizer (genetic algorithm) can be used. A relative error and then a least square of that relative error is computed in dbinternal

• The target of the optimization is to minimize this least square error
• 52 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Inductor Performance Optimization

• Visualization of the optimizer convergence as well as the result

after and before optimization

• 53 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Inductor Spice Parameter Extraction versus Layout Variation

• The objective is to vary the inductor geometry (number of turns,

radius, ..) and to extract spice model parameters with UTMOSTIV in VWF to generate an inductor PDK

• 54 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Inductor Spice Parameter Extraction versus Layout Variation

• A parametrized Expert LISA script runs in VWF to create inductors

GDSII layouts as a function of geometrical parameters (number of turns, radius …)

• 55 -

LISA script of Expert

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Inductor Spice Parameter Extraction versus Layout Variation

• Tree view representation of all inductors with different geometries

and corresponding results in the worksheet

• 56 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Inductor Spice Parameter Extraction Versus Layout Variation

• The worksheet is exported to SPAYN where a regression analysis

is done

• 57 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Comparison between DOE and Optimization

• The goal is to optimize a threshold voltage. A DOE based

approach will be used as well as a direct optimization approach and results of both approaches will be compared

• The DOE approach consists of first defining a DOE on the

parameters we want to vary and run the simulation in VWF. Then a regression analysis is performed in SPAYN and a synthesis based

• n the previously generated Response Surface Model is done to

get the desired threshold voltage

• The optimization approach consists of using an optimizer directly

in VWF by selecting a target (threshold voltage) and variables

• 58 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Comparison Between DOE and Optimization

• The DOE approach
• 59 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Comparison Between Doe and Optimization

• 60 -
• The Optimization approach

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Comparison between DOE and Optimization

• 61 -

Gate Oxidation Time Gate Oxidation Temperature Vt Adjust Dose Vt Adjust Energy

DOE Approach 11min 940 C 9.6e11at/cm2 8.15Kev Optimization Approach 11.95min 934 C 1e12at/cm2 9.3Kev

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• A comparator circuit is used for this example where a transition

from 0 to 1 is expected when a sine wave crosses a DC voltage. Variables are Vdd, Temperature, Load capacitance and Spice model (typical, slow and fast)

• 62 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• Input deck, variables definition and tree view representation

defined within the VWF GUI

• 63 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• Simulation results shown in the worksheet. Working specification

defined as rise and fall time are both below 25ns

• Circuit output is defined as “1” in this case “0” else
• 64 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• Corner analysis (impact of spice models: typical, slow and fast).

Circuit output is “1” for all spice models used

• Simulation conditions were Vdd=5V, C=1e-12F, Temperature=25C

and spice model=typical

• 65 -

Zoom

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• Specifications are not achieved when temperature as well as load

capacitance increase

• 66 -

Zoom

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• A ring oscillator instead of a comparator is used in this example
• Impact of Supply Voltage and Output Load Capacitor on Ring

Oscillator Frequency

• 67 -

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Managing Circuit Simulation

• Impact of spice model on Ring Oscillator Frequency
• 68 -

Cmos1 correspond to BSIM3 cmos2 to BSIM4 and cmos3 to BSIM4 with different model parameters.

VWF – Automated Design of Experiments (DOE) and Optimization Framework

Conclusion

• VWF is a feature rich tool for performing DOE and Optimization

Experiments

• Seamlessly integrates SILVACO’s simulators and post-processing

tools into one graphical user interface

• Compatible with workstation clusters for time and processor

efficient calculations

• Incorporates a wide range of DOE strategies and optimization

algorithms

• Highly customizable by offering a powerful scripting interface

(JavaScript)

• 69 -