( DAC 04 Item 32 ) --------------------------------------------- [ 02/09/05 ]

Subject: Nassda HSIM, Cadence Spectre & UltraSim, Mentor Eldo

RESISTANCE IS FUTILE! -- The big news this year at Nassda is they settled
their lawsuit with the Synopsys Borg by simply merging with them.  It's
just like when Cadence sued Cooper & Chyan back in 1996 and they suddenly
merged.  (In fact, the CEO of CCT, Jack Harding, a year later became the
CEO of Cadence, replacing Joe Costello.  Something tells me, though, that
Aart ain't stepping down from Synopsys any time soon...)


    We evaluated Nassda's HSIM in April 2002, and started using it in
    production in October 2002.  Our main evaluation criteria were:

    1. Ability to trade off speed with accuracy
    2. Able to handle all the different kinds of passive elements we used
    3. Accurate enough to simulate PLL's locking mechanism, but still
       has faster simulation time than HSPICE/Spectre.
    4. Ability to read Cadence Spectre models (since we use Spectre as
       our main circuit simulator)
    5. AE support (the AE we have is very knowledgeable)

    Our devices are not large (under 100K gates) but contain many analog
    parts (PLL's, VCO's) and other full custom digital circuits.  A big
    chunk of them are running at multi-GHz.  In general we see any where
    between 3 to 10 times speed up over SPICE-like simulator depending on
    the type of circuits and how accurate we want the result to be.

    HSIM's positives are that it allows us to simulate large mixed signal
    designs, plus complex digital/custom analog interfaces.  Also, it has
    the ability to simulate inductors, s-parameter blocks.  HSIM's area
    for improvement is its accuracy and speed -- it's already pretty good
    but we always want it to be better.  ;)

    Our experience has definitely been a positive one.  We depend on it
    to verify big portions of our design and it is a tapeout gating check.

    Nassda and Mentor jointly developed an enhanced DSPF format to support
    hierachical back annotation in HSIM Plus.  We would like to do fully
    back annotated circuit level simulation on some of our big mixed signal
    designs, and this seems to be a potential solution for us.  We haven't
    evaluate HSIM Plus yet, but would like to do so.

    We are thinking of looking into using the combined Calibre XRC
    extraction flow with HSIM Plus.  We are interested in using it for
    post-layout acceleration and power nets checks.

        - Joe Dao of Aeluros


    We used to have ATS, an event-driven circuit simulator product from
    Cadence.  It was being end-of-life'd and we needed an alternative.  So
    in mid-2003, we evaluated HSIM.  Our main criteria were accuracy and,
    of course, speed.  HSIM's accuracy was fine, and its speed was
    basically on par with another tool we were evaluating at the time.
    HSIM consumed more memory, but it had more going for it overall.  For
    example, HSIM supported Verilog-A and could directly read a Spectre
    netlist.  We use Spectre as our main analog simulator so being able to
    read Spectre netlist was a big plus.  This eliminated another layer of
    translation, which required additional CAD support and is usually
    prone to error.

    We purchased HSIM in 2003; increased our licenses in 2004.

    Benchmark information

    Our designs are not large by any means; they are typically smaller than
    200,000 devices.  For 5000 transistors:

           HSIM speed          Analog         Time
           ----------         ---------      ------
           8 (high)           -1 (off)       <1 min
           4 (med)               0           29 sec
           1                     0          623 sec
           0                     2          too long, stopped test

    Other criteria we had were simulation of set up/hold times on a very
    low power design and support for digital vector inputs, which HSIM was
    capable of.

    HSIM was easy to use, and our engineers came up to speed quickly,
    especially since some of our engineers had prior experience with it.

    Nassda's AE and R&D support was really good, as was Nassda's sales
    rep, who was very helpful through out the negotiation phase.  It was
    important to us to build a good company relationship.

    On the negative side, HSIM's accuracy vs. performance settings was
    initially confusing.  We had to learn this through several conference
    calls with the AE's and I think it is still a challenge for a new user.

    We also encountered some small bugs like it couldn't read the tilda
    mark at the beginning of a directory path or support for angle bracket
    for bus notation.  However, Nassda was very responsive in fixing the
    bugs.

    We use Assura for physical verification and RCX extraction.  It would
    be great to see HSIM support for hierarchical RCX capability now
    available in Assura RCX.

    So far our overall experience with HSIM has been positive.

        -  Vahid Vafaei of Impinj


    Nassda's main products are Fast SPICE and Power Analyzer on transistor
    level. We are using both tools still now.  Right now we are interested
    in critical path analyzer on gate level with transistor level accuracy.

    Nassda's biggest strength is fast and accurate SPICE engine in each
    tool.  So HSIM is working very well for both analog and memory design.
    And LEXSIM is the only transistor level analyzer for power integrity
    within reasonable TAT.  I think this SPICE engine might be good for
    critical path analyzer on gate level.

    Although Nassda's weakness was outstanding litigation by Synopsys
    against Nassda, it was closed by the acquisition.  On the other hand,
    I'm afraid of price increases and development stagnation after the
    merger.

        - [ An Anon Engineer ]


    Nassda's HSIM is really easy to set up and use.  The ability to easily
    partition the circuit in terms of simulation speed is very useful.

    For sure the ratio between speed & accuracy is excellent.  In some
    cases when high precision is required the charge conservation is not
    respected and the simulation fails.  In these cases it is really hard
    to achieve a good setup (simulation correct) without compromising the
    simulation time

    For sure HSIM is best simulator because the user can tune its
    precision as he wants.  Other simulators can reach the same accuracy
    but lose in performance.  Others have a speed that is comparable but
    are really tricky and cannot reach HSIM's precision.

        - Mauro Giacomini of ST Microelectronics


    The content of analog logic in our mixed signal designs ranges from 5%
    to 100%.  For many of these designs, the best top level simulator is a
    fast MOS simulator.  We used to use Synopsys' StarSim/StarSim-XT and
    Powermill/NanoSim.  Later we added Nassda's HSIM and Mentor's Mach-TA
    to our design flow.  Our experiences in the fast MOS simulator area:

    Avanti StarSim/StarSim-XT

    Before HSIM and Mach-TA, we used StarSim for designs with more analog
    content, because it was accurate and easy to use.  The biggest problem
    with StarSim is convergence & speed.  We later migrated to StarSim-XT,
    as it has better convergence and is faster than StarSim.  The
    performance improvement over StarSim is inconsistent -- overall
    StarSim-XT is faster than StarSim, however we've seen the opposite
    for some designs.

    Synopsys NanoSim

    For large circuits, or circuits with high digital content, NanoSim is
    faster than StarSim/StarSim-XT and has better convergence.  NanoSim's
    weakness is that too many parameters/switches need to be set to achieve
    an optimized simulation (good performance and accuracy).  Also, Nanosim
    is not as accurate as StarSim for designs with high analog content.
    Recently NanoSim became more user friendly with a new GUI and new
    parameters to simplify the set-up.  However due to past experiences and
    the availability of better tools, we have shied away from using these
    new capabilities.

    Our designers got frustrated quite often, due to spending a lot of time
    on setup, and then neither StarSim-XT nor NanoSim working well.  For
    example, we started a simulation with StarSim-XT.  After many trials to
    overcome the convergence, the run time turned out to be unacceptable
    (months).  It took us several weeks to fine tune the setup when we
    switched to NanoSim, and Nanosim still does not always lead to the
    accuracy provided by StarSim-XT.

    Nassda HSIM

    HSIM is known for its high performance for memory simulation, but
    memory is not the main reason why we adopt it in our design flow.
    Our designers like HSIM for the following reasons:

    - Ease of use.  Very minimum setup is required to bring up an HSIM
      simulation with good performance/accuracy.
    - No convergence problem.
    - Better runtime than StarSim-XT and NanoSim.  Overall HSIM provides
      2X performance improvement over StarSim-XT and NanoSim for our
      mixed-signal designs.
    - Supports more types of design such as designs with switch cap,
      embedded flash or Eprom, etc."
    - Rich feature set. Some features are unique.  As an example, Nassda
      is releasing a check for floating node at power down as part of
      their CircuitCheck option.  This unique feature can reduce many
      design respins.
    - Excellent support.  Nassda's AEs are very dedicated with high
      level of expertise.
    - More aggressive in adding new features.
    - Quick turnaround on bug fixes.

    We have simulated many designs with HSIM since we integrated it into
    our design flow.  We consider HSIM to be the best fast MOS simulator.
    Once our users started to use HSIM, they did not go back to NanoSim
    or StarSim-XT.

    The announcement of Nassda's acquisition by Synopsys has brought us
    some concerns regarding HSIM's future, such as:

    - Synopsys' ability to maintain the level of innovation and fast
      pace of adding new features to HSIM the way Nassda did.
    - Quality of the support.
    - Effectiveness in capturing new features needed by customers after
      sales

    And the interface with non-Synopsys tools (Analog Artist interface,
    NC-Sim co-simulation, etc.)

    Mentor Mach-TA

    Our limited experience with Mach-TA shows that Mach-TA's performance is
    not consistent.  For most designs, Mach-TA is not as fast as HSIM and
    NanoSim.  In a few instances, it does provide the best run time with
    reasonable accuracy.

    Cadence UltraSim

    We considered UltraSim before Celestry was acquired by Cadence.  At
    that time, UltraSim didn't meet our requirements.  We have not
    revisited it since Cadence acquired Celestry.

        - [ An Anon Engineer ]


    We use Nassda's HSIM.  HSIM's notable features are:

    - Faster than the other fast-SPICE products
    - Co-sim with other commercial tools such as NC-Verilog/Modelsim
    - Compatibility with HSPICE

    We are currently evaluating Cadence's UltraSim, but still no success.

    Nassda's HSIM's biggest strength is its ease of use and compatibility
    with other SPICE tools.

        - Tamir Ofer of Tower Semiconductor


    We use UltraSim/HSPICE

        - [ An Anon Engineer ]


    Cadence has UltraSim, a table driven SPICE simulator that fills the gap
    between their Spectre SPICE simulator and their NC-Sim logic simulator.
    It doesn't do hierarchical simulations like Nassda, but for flat
    simulation (which is probably more common) they say it is competitive
    with Nassda and Synopsys.

    Mentor sells a new table-driven simulator called FastSPICE, which ties
    to their Eldo SPICE simulator and Modelsim -- although one AE told me
    that if you had to do something tomorrow he might use Nassda instead.
    (Wow!  An honest AE!)

    Mentor's Eldo analog simulator now has an Eldo RF add-on.

        - John Weiland of Intrinsix Corp.


    For the fast SPICE sims, Nassda seems to be the way to go.

    With the accurate simulators the choice is really Eldo vs. Spectre.
    If you do a full Cadence flow Spectre is nice, otherwise Eldo fits
    better in a mixed vendor or Mentor-based flow.

        - [ An Anon Engineer ]


    Of the high accuracy simulators, I'd say the choice is between Eldo
    or Spectre.

        - [ An Anon Engineer ]


    We have chose HSPICE over Eldo.  Based mainly on the fact that HSPICE
    is pretty much standard where as Eldo would have required special 
    models from some of our vendors.

        - [ An Anon Engineer ]

Index    Next->Item







   
 Sign up for the DeepChip newsletter.
Email
 Read what EDA tool users really think.


Feedback About Wiretaps ESNUGs SIGN UP! Downloads Trip Reports Advertise

"Relax. This is a discussion. Anything said here is just one engineer's opinion. Email in your dissenting letter and it'll be published, too."
This Web Site Is Modified Every 2-3 Days
Copyright 1991-2024 John Cooley.  All Rights Reserved.
| Contact John Cooley | Webmaster | Legal | Feedback Form |

   !!!     "It's not a BUG,
  /o o\  /  it's a FEATURE!"
 (  >  )
  \ - / 
  _] [_     (jcooley 1991)