Search

The programming method used to adapt an existing time-domain electrical circuit simulator to the parallel computation is presented. The originality of the simulator results in the semiconductor device numerical physical modeling. Thus, the organization of the existing software, initially developed to be run on a monoprocessor sequential Unix workstation, is firstly detailed. Accounting for specifications at once regarding the effort necessary to modify the software, the wished simulator application field and the constraints resulting from the available computer, two levels of parallelization have been pointed out and implemented by means of the message passing interface parallel programming tool. As an illustration, some results concerning the simulation of a microwave monolithic integrated circuit (MMIC), especially a 2–40 GHz HEMT transistor cascode stage distributed amplifier, are presented. Circuits of increasing complexity have been considered. The evaluation of the sequential/parallel computation ratio demonstrates that significant gains can be expected from the parallel computation opening the way to analysis of the operation of MMICs of mean complexity by means of a numerical physical approach. Copyright © 2008 John Wiley & Sons, Ltd. [ABSTRACT FROM AUTHOR]

A new P-N junction device model is presented. It consists of two numerical un-idimensional macroscopic models. The first is a drift-diffusion model. The second is an energy model accounting for the carrier energy relaxation effects. Because device optimization requires systematic investigations, the computing time is of paramount importance and, so an accurate comparison between these two models was undertaken under both static steady-state and dynamic 94 GHz conditions in order to point out the relative interest of the energy model. This study is mainly devoted to silicon millimetre-wave IMPATFs. This comparison has highlighted the superiority of the energy model for the description of low field transport phenomena. But as IMPA1T RF properties are mainly governed by high field transport, the RF power levels predicted by the two models have been found to be similar. Moreover it has shown that the energy model tends to underestimate both the efficiency and the diode terminal negative resistance level. This parasitic effect is inherent to the description of the carrier generation process by impact ionization which in this model is assumed to be governed by the carrier energy. [ABSTRACT FROM AUTHOR]