New Features in SPAYN Version 1.6.0

Introduction

This article describes the new features which will be available in the new release of SPAYN. The last official release of SPAYN was SPAYN Version 1.3.2 and the new release will be SPAYN Version 1.6.0. SPAYN will be available on the same platforms as UTMOST. Please refer to the article entitled "New Features in UTMOST Version 12.03.0" in this issue of the Simulation Standard for more details on these supported platforms and operating systems.

Bug Fixes

The following bugs were fixed since the release of SPAYN Version 1.3.2:

  • Importing CSV (comma separated variable) format files which contained blank lines used to cause a problem for SPAYN. In the new version of SPAYN blank lines in the input file will simply be ignored.
  • The introduction of a SPICE simulation job queue manager in SPAYN means that SPAYN will no longer hang up when a SPICE simulation, launched by SPAYN, is interrupted or fails.

New Features

  • File Related

The list of external data formats understood by SPAYN for data import is increased to include RS/1TM EZ_WRITEFILE and Keithley data formats.

A database of parameter values can now be generated provided that parameter names, mean values, standard deviations and correlations are supplied as input. The generated database will contain a user-specified number of data sites, and can be analyzed in exactly the same way as a measured database.

  • Data Search

There has been a redesign of the Parameter Search GUI offering a significant speed increase in selecting or searching required parameters. Parameter names and units can be changed after a database search.

The list of mathematical transforms available during a Parameter Search has been extended.

User-defined filter limits may now be specified in addition to automatically generated filter limits, with the following options possible on a per-parameter basis:

  • User-limits can override automatically generated limits
  • Automatically generated limits can override user-limits
  • Choice of whichever limits produces the narrowest (finest) filter
  • Individual parameters can be specified as being unfiltered during the search.
  • Figure 1

Figure 1. The new SPAYN parameter search window.

A full summary is now available for attribute and filtered parameter searches. This summary indicates sources of bad data. Variable attributes and parameters can be listed in order of appearance, in order of overall frequency, or in order of frequency in rejected (bad) parameter sets.

  • Analysis Features

Data sets can be highlighted based on values of attributes in histogram and scattergram plots, allowing (for example) trends from wafer-to-wafer and lot-to-lot to be easily visualized (see Figure 2 and Figure 3).

Figure 2

Figure 2. SPAYN histogram with wafer high-lighting enabled.

Figure 3

Figure 3. SPAYN scatter plot with wafer high-lighting enabled.

The previous multilinear regression feature has been replaced with a new regression feature allowing regression or response surface analysis with up to (and including) third order terms. A single mouse click plots the regression model values (estimates) versus the measured target parameter values.

Parameter Matching quantities such as absolute, relative differences, and ratios may be calculated for any pair of parameters (e.g. VTO measured for adjacent transistors), with distribution means and standard deviations instantly available. Mean values which are significantly different from zero ( given a user-specified confidence interval) are highlighted, and confidence limits are specified for the standard deviation of the matching distribution.

New parameters may be generated to "match" existing parameters to a user-specified tolerance. This is useful for simulating the effects of device parameter matching on circuit performances, where the matching tolerance is known, but no actual measured data is available.

There has been a redesign of the graphical user interface meaning that the SPAYN user can now work comfortably with large numbers of parameters.

The list of distributions that can be fitted to measured data has been extended to include the Weibull distribution, a "negative" distributions for fitting data that is negatively skewed, and the "Automatic" option allows the best fit to be found automatically.

Cumulative plots may now be plotted with either a linear or inverse-normal Y-scale.

The list of x-y curve fitting options has been extended to include third order polynomial fitting, and the "Automatic" option allows a rapid, automatic choice of the best fit function.

Scattergram zoom/unzoom operations have been implemented. User-selected points on scattergram are now highlighted on the plot for better visualization. Groups of data can be selected from scattergram for export.

  • Parameter Grouping/Equations

Weights may be specified on a per-parameter basis when performing a Principal Components Analysis (thus ranking certain parameters as being more "important" than others). In addition, the sensitivities of certain semiconductor device characteristics (see relevant section below) to each parameter can be calculated, and used to replace or augment the user-specified parameter weights. This allows, for example, a PCA model to be obtained that better represents the variance in certain device characteristics such as terminal current or output conductance. The range over which the sensitivities are calculated is user-definable.

Parameter groups and dominant parameters/factors of variance can now be completely specified by the user without first performing a PCA/PFA analysis. Where a PCA/PFA analysis has been previously performed, the results can be used and changed by the user.

Factors of variation can be specified as the Principal Components/Factors found as a result of a PCA/PFA analysis, as the default "Dominant Parameters" associated with these, or as any combination of the above. These can be skewed or varied independently during a worst-case or Monte Carlo analysis.

The system of equations relating parameters to the significant factors of variation or dominant parameters now includes regression up to, and including, third order terms.

The percentage variance explained for each parameter is immediately displayed whenever the model is changed (e.g. from expressing parameters in terms of their single dominant parameter to expressing parameters in terms of all factors).

  • Simulation Interface

Possible choices for process "Corner" Model generation now include the following DOE's (Designs of Experiment) to cover the parameter space using less simulations:

  • Single Point
  • Two level full and half factorial designs
  • Three level full factorial
  • Box-Behnken fractional factorial three-level designs
  • Circumscribed Central Composite (CCC)
  • Face-Centered Cubic (FCC)

The Spayn Monte Carlo Analysis now includes a "Monte Carlo for Matching" feature, in which individual parameters can be assigned random variance terms (in accordance with measured differences between individual parameters and the values predicted using the background system of equations) in addition to the values obtained using the regression equations. This allows parameters to "track" the dominant parameters according to the observed correlations, but to also vary independently in a mismatch fashion.

Advantage is taken of the SmartSpice MODIF/DATA feature to rapidly simulate all process corner and Monte Carlo models (both with and without matching) using the SmartSpice circuit simulator.

Input deck titles are now generated automatically (using VYPER), making it very easy to identify which simulation corresponds to which process corner or Monte Carlo instance.

Spice Simulations now take place asynchronously with SPAYN operation, freeing SPAYN for further analysis while external Spice simulations are taking place.

A Simulation Queue manager indicates progress during simulation, and the job queue can be aborted at any time.

  • Semiconductor Device Characteristics

A device characteristic setup window eliminates the need to hand-create a netlist file specifying device dimension and bias points for the simulation or generation of device characteristics.

Automatic generation of the netlist file (and associated post-processing) allows easy measurement/simulation of:

  • MOSFET Ids, Gm, Gds, and Rds characteristics for Berkeley MOS level 1,2,3 BSIM1, BSIM2, BSIM3 and Philips Level 9 models.
  • Bipolar Ic, Ib, Gm, Go and Ro characteristics for BJT Gummel Poon, MEXTRAM, Quasi-RC, IGBT and QBBJT models
  • JFET or MESFET Ids, Gm, Gds and Rds characteristics
  • Diode Id, Gd and Rd characteristics for DIO1, DIO2 and DIO3 models.
  • TFT Ids, Gm, Gds, and Rds characteristics for SmartSpice amorphous or polysilicon TFT models.

In addition to the pre-supplied library of models outlined above, user-written dynamically-linked models are also supported for the simulation of device characteristics in a manner similar to proprietary models in UTMOST.

Distributions of the above characteristics can be generated for any device geometry and bias point using the measured parameter values. A progress indicator provides feedback which is useful when a large number of data sites are analyzed.

Distributions of the above characteristics can be generated for any device geometry and bias point using the parameter values predicted by the SPAYN system of equations. This enables the verification and evaluation of the system of equations in terms of how well the generated models will predict important device characteristics, by comparing the "predicted" device characteristics to the "measured" device characteristics.

Figure 4

Figure 4. New SPAYN parameter matching window.

Figure 5

Figure 5. New PCA parameter weighting window in SPAYN.

The effects of variations in any of the factors or dominant parameters (representing process variation) on any of the above device characteristics can be seen immediately using interactive graphics.

  • Miscellaneous Setup Options

Plot palettes (allowing customization of plot colors or line styles) have been moved to an easy-to-access location under the new main Setup menu, so that all histogram or scattergram plots can be updated simultaneously.

Parameter value Clip-policy controls have been moved to the main Setup menu so that they are accessible from Scattergram and PM Chart sample point windows.

Figure 6

Figure 6. SPAYN device characteristic setup window.

The Model Setup window has undergone changes and is now accessible from a generic Setup menu, eliminating the need to pre-perform a PCA/PFA analysis so that the models can be accessed from the main simulation interface window.

The Model Setup window's graphical user interface has changed for better ease-of-use.

Histogram, Scatter-gram and Control Chart hard copy options now provide support for a large range of non-postscript printers.