HiCUM Model Status in Spectre

1
HiCUM Model Status in Spectre

• Dec, 2004

2
HiCUM Development History in Spectre

• Dec,1998 First model implemented for version
  1.0
• Oct, 2001 Updated to version 2.1 from equation.
  No reference is available
• Sept, 2002
• Optimized code
• Minor changes to match DEVICE results
• Jan, 2004
• Implemented temperature dependence for substrate
  current. (ST request). It is invoked by
  updatelevel1.
• Feb, 2004 June 2004
• Modified original diffusion capacitance to
  improve convergence.
• Minor bug fixes for the substrate diode

3
Current Status in Spectre

• Improvements
• Added complete derivatives for transconductance,
  diffusion capacitance
• Added temperature derivatives for
  external/intrinsic diodes
• Added missing stamping for internal base
  resistance
• Provide a new fast model for self-heating
• Demonstrated great performance on both
  convergence and speed.
• More than 50 speed up for transient simulation
  on some customer circuits.
• Available for beta testing now.

4
Why HiCUM being the most problematic model?

• Simplified equations for transconductance and
  diffusion capacitance
• Diffusion capacitance between base and collector
• C_dci T_rS_rb T_fS_fc
• Transconductances for base
• Sfb Itf1/Vaf d/( 1dTf,tItf1/Qp.t)
  
• All these simplified forms are inconsistent with
  charge and current computed from Newton
  iterations. and cause convergence issue during
  transient analysis
• Not unique for Spectre, it is problematic for all
  simulators !

5
Other issues in Original HiCUM model

• Missing temperature derivatives for diodes
• Missing part Jacobian for internal base
  resistance function
• dRbi_dVbe, dRbi_dVbc, dRbi_dVce
• Some smoothing functions are not continuous at
  boundary

6
Suggestion to CMC

• HiCUM needs to have a complete implementation
• Solution from each EDA vendor on their own may
  results inconsistent results, especially for AC
  analysis. This loses the purpose of standard
  model CMC try to promote
• Should start another benchmark tests among
  different simulators to check HiCUM model
  performance in general.
•  

7
HiCUM Level 0 implementation status

• Implement through VerilogA -gt ADMS -gt Spectre CMI
  
• Changes on Verilog-A code from Prof.
• Rename model parameters to be consistent with
  HiCUM version 2.1
• Added attributes for description of input
  parameters
• Reformulate analog functions to macros
• Added attributes for source probes to generate
  more stable initial conditions for model
• Added noise sources including shot, thermal and
  flicker noise.
• Plan to release in Spectre, Q2, 2005