MCAST: Automatic device modeling in model compiler for efficient and accurate circuit simulation

Manual implementation of a compact device model into a target circuit simulator has become more and more difficult and time-consuming as the use of more accurate and complex device models becomes pervasive. The device model evaluation time has started to dominate the overall simulation time. To address the issue, MCAST: a general M&barbelow;odel C&barbelow;ompiler for automatic device modeling---based on A&barbelow;bstract S&barbelow;yntax T&barbelow;rees---is introduced. MCAST reads compact device models described in high-level behavioral languages VHDL-AMS/Verilog-AMS and automatically generates the device implementation codes in C/C++ that can be directly linked with circuit simulators such as SPICE3, SPECTRE1 and LEADER2, etc.; This dissertation presents four types of techniques developed in MCAST: (1) A set of implementation schemes to implement a general device model automatically, (2) Multiple compiler-based optimization methods to generate device model codes as efficient as possible, (3) Hierarchical low-dimensional table lookup optimization method and (4) Design-adaptive device modeling method utilizing design specific information to generate evaluation-efficient device models. Of particular emphasis are the optimization methods developed in MCAST to generate efficient device model codes for target circuit simulators. The work presented in this dissertation tries to attack the circuit simulation efficiency problem from the device-modeling point of view.; With these techniques, the design automation of device model implementation has been achieved. MCAST can generate accurate and robust device models for a much wide range models, and the MCAST generated code after optimization has found to be very efficient, thus greatly enhancing the performance of the target circuit simulator. Experimental results show that the model evaluation time of a typical generated device model is up to 10 times faster compared with manually coded built-in general device model, and the overall simulation time using the generated device models can be up to 8 times faster than that using the general device models while achieving the same accuracy.; 1Cadence circuit simulator 2IBM in-house circuit simulator...