Adaptive Control of a Multi-Bias S-Parameter Measurement System

1
Adaptive Control of aMulti-Bias S-Parameter
Measurement System

• Dr Cornell van Niekerk
• Microwave Components Group
• University of Stellebosch
• South Africa

2
Presentation Overview

• Introduction Background Information
• Equivalent Circuit Non-Linear Modeling
• Adaptive Algorithm Requirements
• Defining the Safe Operating Area (SOA) of a
  Device
• S-Parameter Driven Adaptive Measurement
  Algorithms
• DC Driven Adaptive Measurement Algorithms
• Results Conclusions

3
Introduction Background

• Interest is in algorithms required for
  construction of device CAD models
• Focus is on small-signal equivalent circuit
  extraction procedures
• Have developed robust multi-bias extraction
  algorithms for GaAs FETs
• Focus is shifting to bulk Si MOSFET devices
• Diagnostic applications for monitoring technology
  development
• Starting point for construction of equivalent
  circuit based nonlinear CAD models
• Local interest is packaged power FETs, especially
  LDMOS devices
• Apply modeling to off-the-shelf devices,
  scalability therefore not an issue
• Do require accurate modeling of extrinsic
  networks
• Model extraction algorithms constrained not to
  use device design information

4
Multi-Bias Decomposition-Based Extraction

• Algorithm is formulated to overcome the
  ill-conditioned nature of problem
• Combines data from multiple bias points into one
  integrated problem solver
• Decomposition-based optimizer used to efficiently
  handle large number of parameters
• Have been hybridized with analytic extraction
  procedures
• Fast, robust and starting value independent

5
Moving to Bulk Si MOSFET Devices
6
Nonlinear Equivalent Circuit Modeling Process
Measure Multi-Bias S-Parameters DC Data
Extract Small-Signal Circuit Models from the
Multi-Bias S-Parameter Data
Construct Nonlinear Circuit Model from Equivalent
Circuit Data and DC Measurements
Verify Nonlinear Model thru Design Nonlinear
Measurements
7
Equivalent Circuit Models
8
Typical Multi-Bias S-parameter DC Measurement
System
9
Why Create an Adaptive Measurement Algorithm?

• Nonlinear measurement-based models require large
  volumes of data
• This implies the use of computer controlled
  measurement setups
• Want more bias points in areas where the device
  characteristics change rapidly
• For larger devices, a high uniform density of
  bias points is not practical
• An adaptive control procedure with following
  qualities is required
• Must ensure equipment device safety
• Must exploit all available measured data (DC
  S-Parameter data)
• Decisions should be based on direct analysis of
  data (technology independence)
• Make provision for finite programming
  measurement resolution of DC sources

10
Who is the competition?

• Most extensive work done by Fan Root (Agilent)
• 1 S. Fan, et. al. Automated Data Acquisition
  System for FET Measurements and its Application,
  ARFTG Conference, pp. 107-119
• 2 D.E. Root, et. al. Measurement-Based
  Large-Signal Diode Modeling Systems for Circuit
  and Device Design, IEEE Transactions on
  Microwave Theory and Techniques, Vol. 41, No. 12,
  Dec. 1993, pp. 2211-2217
• Ref 1 only uses DC data adaptive exploration
  of IDS(VDS) curves
• Ref 2 uses AC data via previously extracted
  diode small-signal model
• Majority of work on adaptive sampling procedures
  is focused on EM analysis procedures to reduce
  the number of time consuming simulations required
• Techniques developed for EM simulations not
  directly applicable to measurement examples due
  to measurement noise

11
Components of an Adaptive Measurement System

• Define a fine measurement grid minimum bias
  point separation
• All bias points to be measured must fall on the
  fine grid
• Fine grid is a square defined by min/max bias
  voltages
• Easy way to handle DC source programming/measureme
  nt uncertainties
• Experimentally determine Save Operating Area
  (SOA) of device
• SOA limits defined by max/min VGS, VDS, IGS, IDS,
  PDS
• Boundaries to be determined experimentally using
  minimum of measurements
• Establish fine grid bias points that fall inside
  the SOA
• S-Parameter Driven Refinement Algorithm
• Start with an initial selection of measurements,
  and refine selection by placing N new bias points
  based on analysis of S-parameter data
• DC Driven Refinement Algorithm

12
Determining the Safe Operating Area (SOA)

• Measure an approximate value of threshold voltage
  VT
• User defined list of VGS bias voltages, with most
  in device active region
• Explore IDS(VDS) curves at each VGS bias using
  large ?VDS to find SOA limits
• Linear extrapolation is used to check if a
  projected measurement will exceed a SOA limit
• Key to procedure is lots of safety checks

13
S-Parameter Driven Refinement Procedure

• SOA procedure provides initial set of
  measurements for refinement procedure
• Adaptive procedure places N new bias points so as
  to best capture nonlinear behavior of device
• Analyze the device S-parameters to determine the
  position of new bias points
• Higher density of bias points in regions where
  any of 4 S-parameters are experiencing large
  variations with bias
• Change in S-parameters signifies change in model
  parameter values
• During measurement phase it is not important to
  know which parameter has changed, just that
  change has occurred

14
Increasing Diversity in Selected S-Parameter Data

• Need to define the differences between
  S-Parameters
• S-Parameter curves change in
• Length
• Position
• Shape Orientation
• Require a geometric abstraction to describe
  S-Parameters
• S-Parameter Centroids

15
S-Parameter Driven Refinement Procedure

• Identify adjacent bias points makes use of
  Delaunay triangulation
• Calculate distance between centroids of adjacent
  bias points
• Place new bias points between bias points with
  largest centroid separation
• Safety checks for duplicate bias points
• Fine measurement grid introduces refinement
  limitations

16
DC Driven Refinement Algorithm

• For complete characterization, both the DC AC
  characteristics must be considered
• Can use existing procedures, such as those
  proposed by Fan Root
• Simple alternative is to use difference between
  linear and spline interpolation models of
  IDS(VGS,VDS)
• Place new measurements where difference between
  interpolation models is largest
• Draw back is that boundaries of SOA needs to be
  well defined

17
Illustration of Adaptive Bias Point Selection (1)

• GaAs HEMT
• 50mV Fine grid
• 9 Initial measurements defining boundaries of the
  SOA
• 100 iterations of the S-parameter refinement
  algorithm
• 463 newly selected bias points

18
Illustration of Adaptive Bias Point Selection (2)

• Bulk Si MOSFET device
• Physical gate length 70 nm
• 20 µm total gate width
• 2 gate fingers
• 50 mV x 100 mV fine grid
• 28 initial measurements, determined with SOA
  exploration algorithm
• 80 iterations of S-parameter refinement algorithm
• 292 newly selected bias points

19
Nonlinear Modeling Verification (GaAs FET)

• Table-based model implemented in Agilent ADS
  circuit simulator
• Table-based model used linear interpolation
• Reference model was constructed using all the
  data, in other words, every point on the fine
  grid
• 2nd model was constructed using adaptively
  sampled data 50 data reduction
• NNMS Nonlinear measurements were performed
• Device biased in class-AB mode
• Fundamental excitation is 5 GHz
• Single tone power sweep driving FET into
  compression

20
Modeled Measured Nonlinear Results
21
Conclusions Future

• Incorporates both S-parameter DC data into
  decision making process
• Captures both VDS and VGS switch-on regions
• Procedure is technology independent
• It has a high emphasis on device and equipment
  safety
• Makes provision for equipment measurement
  limitations
• Future work will focus on characterizing LDMOS
  power devices
• Extensions include the incorporation of designer
  knowledge into the adaptive measurement procedure