IBIS Macromodel Task Group

Meeting date: 22 aug 2006

Members (asterisk for those attending):
*Arpad Muranyi, Intel Corp.
*Barry Katz, SiSoft
*Bob Ross, Teraspeed Consulting Group
Doug White, Cisco Systems
*Ian Dodd, Mentor Graphics
*Joe Abler, IBM
John Shields, Mentor Graphics
Ken Willis, Cadence Design Systems
Kumar, Cadence Design Systems
*Lance Wang, Cadence Design Systems
*Michael Mirmak, Intel Corp.
*Mike LaBonte, Cisco Systems
Paul Fernando, NCSU
*Randy Wolff, Micron Technology
Richard Ward, Texas Instruments
*Sanjeev Gupta, Agilent
*Todd Westerhoff, Cisco Systems
*Walter Katz, SiSoft

-------------
Review of ARs:

None

-------------
New Discussion:

AR Ian contact SystemC expert
- Contacted, but the guy is unable to present to our group
- SystemC topic tabled for now

Name of subcommittee
  - SerDes
  - Advanced Technology
  - Advanced Modeling
  - Advanced Technology Modeling
- Chair can change group name any time
- Could create new subcommittee, but some work for that
- These are Task Groups because the rules for those are simple.
  - No spec produced, for example.
- We voted to change name to ATM

API
- Arpad's outline:
  o  API for compiled vs. ASCII readable model
     - compiled models not desirable because separate compilations
       would have to be generated for each OS and compiler switch
       settings can make things incompatible very quickly
     - binary format not sufficient for IP protection
     - encrypted (and/or obfuscated) ASCII models may be the best
  (
    - Mirmak:
      - John Shields proposal has been incorporated into VHDL.
      - No relationship between that and C.
    - Ian proposed supporting only C
      - API would have to specify parameter passing style
      - EDA tool could decrypt and pipe to compiler.
      - Not a proprietary solution.
  )

  o  C vs. *-AMS or linear vs. non-linear simulation engine?
     - the 10 million bits per hour capability is due to using
       a linear simulation engine, not because of the choice of
       the C language
     - no linear simulators exist today which can use *-AMS
     - the assumption of linearity may not be sufficient
     - existing infrastructure: it is too big of a burden to
       rewrite existing libraries in another language or request
       model makers to learn a new language.  Should both C and
       *-AMS be supported?     
  (
    Mirmak presentation:
    - AMS has been called insufficient
    - Who is the IBIS audience?
      - Tool vendors
      - Model makers
      - Model users
    - Best fit:
      - C
      - AMS
    - IBIS supports what level(s) of system design:
      - Architecture/data processing
      - Buffer/circuit
      - Package and physical PCB
    - Can IBIS satisfy all of them? Probably not
    - Who would have the greatest burden for AMS?

    Arpad would like to allow anything
    - results in proprietary solutions
    - already have too many language variants

    Ian: cost of adding AMS to linear sim would be high
      - man-months vs. man-weeks for C

    Joe: C is easier than AMS

    Arpad: article by Jeffrey Cordon
      C:        2 weeks and buggy
      VerilogA: 2 days and solid

    Arpad: Engineers not familiar with software
    Joe knows a hardware guy who wrot ehis own simulator.

    If we support all languages will EDA companies keep up?

    Ian: could be delivered as socketed executable
    - don't have to worry about compiler switches
    - MSoft Visual Studio has lots of switches

    IC vendors would have to buy compilers

    Visual Basic on windows, VBScript more widespread
    Java is compatible, but interpreted (slow)

    Verilog and VHDL used for architecture
    - Matlab at Intel

    Is only IBM using C?

    What are SerDes vendors willing to provide?
    - IBM: C
    - Intel: concerned about C

    Models for linear and non-linear are completely separate.
    When drivers change impedance linear model might change.

  )

  o  Is there a need for a new API in IBIS?
     - there are numerous (standard) API-s on the market already
     - the existing [External Model] or [Extern Circuit] keywords
       may be used for this purpose (with small modifications)

  (
    Todd: Do we need AMS for characterization and C for linear sim?
    - We could add linear-only addition to IBIS
    - IBIS should be a "brand"
      - brands stand for things

    Arpad: linear models can be implementd in AMS digital, wouldn't
      this be fast enough?
      - VHDL is compiled anyway
      - Runs fast if no analog functions are called.
      where does receiver model get data input
      could link DLL into AMS simulator

    Todd: best to do what is easy for EDA (C)
  )

  o  Specification related questions
     - do not standardize methodologies, it prevents evolution
     - prefer flexible definitions which are extendable later
     - should this be part of the IBIS specification, or a new,
       independent specification?


Next meeting: Tuesday 29 Aug 2006 12:00pm PT