MEXTRAM 504 convergence A proposal for model modification May 8th, 2006

1
MEXTRAM 504 convergenceA proposal for model
modificationMay 8th, 2006

• Samuel Mertens
• Mary Tolikas

The contents of this presentation are Ansoft
proprietary
2
Current convergence issue
Slow convergence when self-heating is present in
Mextram 504 Raised at March 10th CMC
meeting Nexxim was not part of this original
comparison
3
Proposed model modification
Observation Temperature node (temp_node) can get
significantly higher in absolute value terms than
the other (voltage) nodes
Proposed modification Scale temperature node to
voltages that are more in line to the rest of the
circuit Add model parameter Tscale in model
definition and adjust Rth -gt Rth / Tscale Cth
-gt Cth Tscale dtemp -gt Tscale
V(temp_node) Default Tscale value is 1.0
(corresponds to the old model)
4
Our observations on the impact of the proposed
change

• Test netlist example provided to us by Intersil
• Nexxim DC convergence
• With Tscale1 (old model formulation) 367
  iterations
• With Tscale1000 134 iterations
• Accuracy remains unaffected
• The iteration times quoted were obtained using
  Nexxim

We kindly acknowledge Slobodan Mijalkovic at
TUDelft and Robert Lomenick at Intersil for
their help and for providing the test netlist