The addition of thin-film alloy magnetic materials to integrated circuit processing enables a number of novel monolithic circuits. Examples include fully integrated magnetic field sensors, signal and power galvanic isolation circuits, and power supplies on chip (PwrSoC). There are several benefits to integrating these traditionally off-chip devices, including improved manufacturability, reduced cost, increased power density, better thermal management, and increase board space efficiency. Accurate circuit models are needed to enable successful circuit simulation of novel on-chip magnetic devices.
This thesis focuses on physically-motivated models extracted from terminal characteristics for three categories of devices benefiting from the inclusion of ferromagnetic alloy thin films: a microinductor, a microtransformer, and a microfluxgate. All models are well-behaved asymptotically, are guaranteed passive, and have physical significance to elements and circuit quantities. Measurements used to generate the models are wafer-level compatible and easily obtained with an impedance analyzer or vector network analyzer; no knowledge of the internal physics of the device (e.g. the magnetic material's magnetization behavior) is necessary. For each device an extraction algorithm is presented which obtains all model parameters within a few minutes on a 3.2 GHz Pentium 4 workstation.
The resulting models not only match the measurements used to generate the model, but accurately reproduce specific determinative performance figures in their respective applications. Microinductors in buck converters accurately reproduce saturation dynamics with errors less than 2.5% and power efficiency within 0.2% when compared to field solvers. A fabricated microinductor matched the 10 MHz pulse response of the model to within 10%. A microtransformer used as a signal isolation transformer is shown to match the voltage gain to within an error < 0.5%. Additionally, a microtransformer used in an isolated flyback converter captures the current with an error of 3.3% and power efficiency to with 5.6% error when compared to field solvers. The microfluxgate model exhibited accurate performance giving an error of 6.6% when compared to field solvers, and errors less than 7.5% when compared to measurements in the magnetic field sensitivity curve well beyond the linear range. Ultimately, this research enables accurate circuit simulations of novel on-chip magnetic devices potentially realizing first-pass design success.