Design of integrated inductors through selection from a database created using electromagnetic simulation and neural networks

The design of integrated inductors on silicon has become an important issue to decrease costs and to speed up the fabrication of telecommunication integrated circuits. Nevertheless, the formulae provided by the existing theory still do not show a good result in inductance prediction, specially if one figures that these inductances must fit the measured ones through a wide band of very high frequency values. Electromagnetic simulation is an alternative to cope with this problem, but if done in a non-ordered fashion, it will demand a lot of computer processing time and will also require a very skilled designer to select the inductors to be tested from a large set of possible devices, due to the large amount of variables involved in the device's specification. The purpose of this work is to propose a methodology to perform electromagnetic simulations in a systematic and automatic way. Once finished these electromagnetic simulations, the resulting electrical parameters are stored in a database that correlates these variables to the geometric specification of the inductors. In order to reduce the computation time demanded by the electromagnetic simulations, the possibility of generating the electrical parameters database using neural networks was also verified.