1	Changing the Paradigm for Compact Model Integration in Circuit Simulators Using Verilog-A M. Mierzwinski, P. O’Halloran, B. Troyanovsky, R. Dutton* Tiburon Design Automation, Inc. *Stanford University
2	Outline Current state of compact model development Verilog-A for compact modeling Examples of compact models in Verilog-A The paradigm shift Summary and conclusions
3	Typical model flow
4	Model Developer issues Complexity in model development has shifted from model creation to model implementation Fewer models implemented Models become more general to try to handle all processes
5	Model Developer issues Simulator model interfaces that are powerful are complex. Model interfaces promote errors due to complex math and program requirements. Software engineering aspects of implementation are time consuming.
6	Model Developer Issues Existing source-code interfaces are inherently non-portable High burden on the developer to: Manually compute matrix stamp Manually compute derivatives Implement multiple, self-consistent entry points for analysis types Requires close cooperation with the simulator vendor
7	End User Issues Want accurate, robust, tried and tested models. Want to use models confidently across different simulation tools. Want all models to simulate efficiently. Want a wide range of models to be available and easily obtainable. Want to easily modify the underlying equations to suite their particular need.
8	Vendor Issues Standard models, such as BSIM3, are a necessary part of any analog simulator Supporting standard models offers no competitive advantage. Multiple implementations of same standard model increases risk of errors Always requires verification and testing. Potential model differences prevent new customers from evaluating their product.
9	Foundry Issues Standard models often do not provide accurate enough representation of the foundry process Foundries do not have sufficient resources to support and verify all vendors’ tools. This can limit their customer base.
10	Verilog-A Natural language for model development Concise OVI standard (proposed to IEEE) Implemented in many simulators
11	Existing solutions Current Verilog-A solutions Are not sufficiently fast for simulation with typical compact models Are not universally available May not offer IP protection Don’t support all analysis types Do not provide a simple distribution mechanism
12	Proposed Solution Use Verilog-A for analog model definition Develop a Verilog-A model compiler and support tools to create analog models that can be used in a wide range of simulation platforms.
13	Proposed Solution Maintain simulation performance comparable to existing C/C++ level interfaces. Provide robustness better than existing implementations. Support for all analysis types, e.g. transient, harmonic balance, shooting, nonlinear noise.
14	Proposed Architecture A Verilog-A compiler and simulator-specific run-time environment (RTE)
15	Benefits to End Users Solves the problem of model availability and compatibility. Models with source code available can be modified.
16	Benefits to Vendors Removes need to support complicated standard models. Proprietary models can also be developed Verilog-A and distributed in compiled form Provides IP protection
17	Benefits to Foundries Model definition can be part of foundry kit Can be completely independent of simulators and simulator versions. One parameter set to extract, distribute, and support. Models can be modified to fit the foundry process Parameter information and model changes can be hidden to protect IP.
18	Benefits to Model Developers Concentrate on model development, not implementation issues. Models and updates can be made available immediately on all platforms. Model IP can be protected and licensed.
19	Demonstration of Verilog-A based Compact Models Support for popular models is necessary for language acceptance Provides templates for generation of similar models in Verilog-A Demonstration of language capability RTE implemented in two simulators: SPICE 3F5, HBsim.
20	Examples: Gummel-Poon BJT SPICE Gummel-Poon Very old model Yet very popular Users often like to modify model slightly Kull-Nagel Thermal Avalanche ~300 lines in Verilog-A
21	Examples: MEXTRAM BJT Philips MEXTRAM 504 BJT in Verilog-A Colpitts oscillator
22	Examples: Compact models MOS UC Berkeley BSIM3 MOSFET Only CMC “standard” model Typically tens of thousands of lines of C-code Example of simple op-amp using Verilog-A Same results as SPICE built-in
23	Examples Compact models MOS UC Berkeley BSIM4 MOSFET model Next generation BSIM Many more parameters First released March 2000 April 2002 CMC still working on implementation problems between vendors.
24	Examples Compact models MOS Philips MOS Model 9 BSIM3 counterpart Philips MOS Model 11 Next generation MOS model
25	Example of single model library used in multiple simulators HISIM Hiroshima University STARC IGFET Model Popular drift-diffusion/surface potential based model Smaller parameter set size Physically based Uses local Newton-Raphson loop
26	HiSIM Implementation
27	New design capability Circuit designers can control level of abstraction All levels can be combined in one hierarchical implementation
28	Behavioral model example: PLL
29	Performance Simulation speed of Verilog-A based compact models is critical for acceptance Myth of hand-coded equations
30	Outstanding issues Verilog-A is quite capable of describing compact models Addition improvements Parameter attributes Additional program functionality Function parameters Local declarations
31	New Modeling Ecosystem
32	New Paradigm Model interface is standardized on Verilog-A Model definition no longer needs to be static Fit model to process Remove unneeded functionality of standard models Models can be a competitive advantage Models can have revenue potential Users can choose simulators based on analysis algorithms rather than model set.
33	Conclusions Verilog-A has the capability to support complex compact model implementations A Verilog-A compiler provides fast execution and support for all analysis types Virtually all popular compact models have been implemented as a demonstration. Model implementation and distribution can be greatly simplified. Model developers, simulation vendors, end users all benefit.