MEXTRAM

1
MEXTRAM
Aliaksandra Kautaniuk Hsien-Chang Wu Koen
Buisman Slobodan Mijalkovic
2
Outline

• Current status.
• Plans for the next release.
• Comparison of Mextram in Verilog-A and Philips
  SiMKit model library.
• Raised issues.
• Action items.

3
Mextram in simulators - overview
4
Mextram Release 504.6 vs 504.5(March 2005)

• The version number has been increased because we
  have added a new parameter DAIS for fine tuning
  of the T-dependence of Ic (Mextram wishlist
  priority 1). The expression 4 - AB - AQB0 is
  replaced by 4 - AB - AQB0 DAIS. The name DAIS
  comes from D Deviation from the expected
  parameter value. A Related to temperature
  scaling IS Related to transistor saturation
  current.
• The clipping of KAVL (KAVLlt1) in order to
  guarantee that the noise correlation coefficient
  is lt1 will be introduced.
• GEM0 added to the equation of the avalanche
  current.

5
Mextram in Verilog-A Current Status

• Compatible with the SiMKit-Mextram naming
  conventionbjt504t_va (full 5-port model)
  bjt504_va (4-port model without thermal node)
  bjtd504t_va (4-port model without substrate
  node) bjtd504_va (3-port model without thermal
  and substrate nodes)
• Independent main modules sharing the code
  modulated by pre-processor commands.
• The code is now distributed as a zipped archive
  instead of the single Verilog-A file.

6
Improved Model Evaluation Time

• The usage of limexp function eliminated as much
  as possible.
• Modularization- Temperature scaling performed
  only in the initial step for bjt504_va and
  bjtd504_va. - define used to adapt parameters,
  variables and model evaluation statements for
  different model complexity.
• The complete code has been revisited from the
  computational efficiency point of view.

7
Implementation of the correlation between base
and collector current noise due to the avalanche

• The Accellera Verilog-A(MS) LRM assign
  correlated noise contribution to a single real
  variable.Sufficient for Mextram but still not
  allowed by some compilers.
• In SiMKit an additional noise source between
  nodes B2 and C2 is added.
• TU Delft Verilog-A implementation is currently
  following the approach proposed by Geoffrey Coram
  and Colin McAndrew. It is based on the
  introduction of an additional electrical node
  that serves to evaluate the correlation noise
  contribution.

8
Compilers and Translators

• Tiburon Design Automation Compiler Simulator
  hpeesofsim
• Synopsys Verilog-A CompilerSimulator hspice
• ADMS TranslatorSimulator hpeesofsim User
  Compiled Model Inteface spectre
  - CMI interface
• Cadence Verilog-A Interpreter (obsolete)
  Simulator spectre

9
Outline

• Current status.
• Plans for the next release.
• Comparison of Mextram in Verilog-A and Philips
  SiMKit model library.
• Raised issues.
• Action items.

10
Plans for Mextram Release 504.7

• Adding base-emitter tunneling current source.
• Adding interfacial oxide noise contribution.
• Updating operating point information with
  terminal voltages.To appear in Verilog-A code
  before Q4 2005 CMC meeting.

11
Implementation of BE Tunneling Current

• Parameter extraction performed using reverse
  Early set-up for Vbc0 and Vsc0.
• Implemented as a CVS branch. The idea is to merge
  it into the next release of the mainline
  Verilog-A code .

12
Implementation of IFO Noise
IFO noise model from U. Texas Arlington
Suggestion

• Backwards compatible
• Implemented as a CVS branch. The idea is to merge
  it into the next release of the mainline
  Verilog-A code .

13
Outline

• Current status.
• Plans for the next release.
• Comparison of Mextram in Verilog-A and Philips
  SiMKit model library.
• Raised issues.
• Action items.

14
Verification conditions

• Parameters sweep
• Cje 73 pF 153 pF
• ßf 215 315
• Cjc 78 pF 158 pF
• ßri 7 14
• Cjs 315 pF 495 pF
• IK 0.1 A 1 A
• VEF 44 V 68 V
• tE 2 ps 14 ps
• VER 2.5 V 4.9 V
• tB 4.2 ps 8.4 ps

• Model Mextram 504.6 (with self heating)
• Verilog-A refer to Mextram Bipolar Transistor
• Model level 504.6 unclassified
  report
• SiMKit version 2.1.1 and 2.2 (C-code)
• Simulator ADS 2004 A
• Test setups 19 setups
• Temperature range 25 C 150 C
• Number of significant digits 12
• Parameters setup
• DIMES 04 with thin/thick epilayer, extracted at
  TU Delft
• Default

15
Verification setups

• Capacitances
• Early forward/reverse
• Gummel forward/reverse
• Output
• S-parameters (frequency/ bias sweep)
• 1/f noise
• High frequency noise
• Transient analysis
• Harmonic balance
• Amplifier

• Thermal influence
• Dta parameter check
• Clipping
• AC
• Noise behavior (1 device)
• Noise behavior (10 devices)
• Transient behavior at T25 C
• Transient behavior at T300 C
• Operating points

16
Error (test setups)
Typical error for setups 11 - 19 10-3
17
1/f noise
2e-4
18
High frequency noise
1
19
The correlation between base and collector
current noise due to the avalanche II
With KAVL1 a clear difference is seen between
the two approaches
20
Transient analysis
0.01
21
Amplifier
4e-3
Parameters DIMES 04 thin epilayer Model bjt
504t Type NPN Temp 25 C
22
Oscillator
0.5
23
Oscillator (Gilbert Cell)
1e-2
24
Outline

• Current status.
• Plans for the next release.
• Comparison of Mextram in Verilog-A and Philips
  SiMKit model library.
• Raised issues.
• Action items.

25
Raised issues

• Convergence
• Reported convergence issues tested in other
  simulator
• Under investigation (test with Verilog-A model)
• SiMKit appears to have two Gmin
• Under investigation
• Small difference between SiMKit and Verilog-A
  implementation in Y-parameters at low
  temperatures (-100C)
• Under investigation
• No operating point variables defined in Verilog-A
  implementation
• Will be implemented
• KBdivQQ losses accuracy in direct calculation of
  KB/QQ
• Corrected

26
Convergence in Simulators (Open Issue)

• Issue
• Intersil reported convergence problems in
  combination SimKit-Commercial Simulator (simA)
  and demonstrated the problem providing the test
  netlist example .
• TUD Actions
• The test netlist has been tested using an older
  versions of the simA.
• The netlist is converted and tested in simB with
  Verilog-A code.
• Conclusions
• Although the test 2 in principle provides
  convergence, the number of iterations using the
  bjt504t_va (200) is still quite high in
  comparison with simpler model versions (10) .
• The problem still exist and will be the subject
  of further investigation.

27
Outline

• Current status.
• Plans for the next release.
• Comparison of Mextram in Verilog-A and Philips
  SiMKit model library.
• Raised issues.
• Action items.

28
Action Items

• Focusing on the raised issues regarding
  convergence in simulators.
• New features to be introduced in Mextram 504.7
  Verilog-A release.
• BE Zener, IFO noise, terminal voltages to OP
  information.
• Extended verification of Verilog-A
  implementation
• Investigate the difference in the correlated
  noise implementation.