Advanced Electromagnetic Modelling of Multilayer Monolithic Microwave Integrated Circuits

Recently, the technique of multilayer structures has been adopted in the development of monolithic microwave integrated circuits (MMICs). With this technique one can employ several layers of metals sandwiched by insulators, allowing both thin film microstrip lines and coplanar techniques to be used. This approach gives microwave engineers great flexibility in design with the advantages of high packing density and improved circuit performance. One example is a stacked inductor, shown in Fig. 1, in which coupling between conductors is maximised by putting adjacent turns on different metal layers. Another one is a stacked patch antenna, shown in Fig. 2, in which operating bandwidth is expanded by having two patches on different metal layers resonating at two different frequencies.

However, there is an urgent need for upgrading the existing CAD technique for the multilayer MMICs. The problems arise from several aspects. Firstly, the present popular EM simulators for MMICs, like em of Sonnet and Momentum of HP, are based on planar surface meshing (2D) and the method of moment. So they are not able to capture the full features of multilayer structures (3D). Secondly, these EM simulators just mentioned are only operating in the frequency domain. So some characteristics of the devices in time domain can not be revealed in the modelling. Thirdly, the conventional optimisation methods currently used are based on the iterative techniques, that is, the EM evaluation process would be run repeatedly until an optimal solution is obtained. This makes the optimisation, even for a planar MMIC device, a very time-consuming and computing intensive task. So it will be painstaking to carry out a design optimisation for the multilayer MMIC based on these methods.

The overall aim of this project is to develop and facilitate an advanced EM modelling technique suitable for the analysis and optimal design of the multilayer MMICs.

The traditional optimisation methods used in MIC design, even including the Aggressive Space Mapping (SM) method proposed recently, tend to be very time-consuming for optimising the design of multilayer MMICs (tens hours of CPU) because of their complexity in structure. Therefore, in order to overcome this difficult an innovative optimiser based on artificial neural networks is proposed to be developed here. This neural network optimiser can operate in an incredible speed (only minutes of CPU), once it being trained and validated with the characteristic information of the component. One of proposed model is a Multilayer Perceptron Neural Network (MLPNN), shown in Fig. 3, which has been proved to be suitable and effective in the CAD of MICs. It is also expected that the neural network simulator/optimiser being developed can be generalised for the design of other type of microwave components, antennas and circuits.

Participants:

Dr X. Chen
N.P. Somasiri

Sponsor:

EPSRC