No abstract

The design and simulation of an integrated common mode filter (CMF) for differential data lines, like the USB 2.0 interface, is presented in this paper. The device is manufactured in a bipolar semiconductor process for the integration of diodes protecting sensitive CMOS circuits against damage by Electrostatic Discharge (ESD). The filter is formed by planar coupled coils that are processed in copper/polyimide layers applied on top of silicon die. The design process is using 2.5D simulation techniques based on method of moments. Furthermore a lumped model is derived that allows exact and efficient transient simulations in SPICE [1] based simulators. The filter design itself shows strong common mode rejection in the GSM spectrum. The small size (1.34 mm x 0.95 mm) of the device makes it well suited for integration in modern mobile phone applications, to suppress electromagnetic interference (EMI) between a USB transmitter and a GSM receiver. Measurement data demonstrates the EMI protection in the GSM downlink spectrum. The dynamic resistance of the ESD diodes is derived in transmission line puls (TLP) measurements.

In portable electronic devices, like smart phones, coupled planar coils are often used as common mode filters (CMF). The purpose of these CMF is to suppress electromagnetic interference (EMI) between wireless communications systems (e.g. WIFI) and digital highspeed interfaces (e.g. USB 3). A designer of such an electronic device usually carries out a signal integrity (SI) analysis, using models of the system components. There are two alternative ways of modelling the CMF: One is based on matrices (called S-parameters) that describe the behaviour in the frequency domain and are either derived from measurements or simulation tools. The other is using a representation based on lumped circuit networks. In this work, a lumped network is generated manually based on expert knowledge. The advantage of this approach is the reduced number of only passive network components compared to traditional methods that produce much larger networks comprising of many active and passive devices. On the other hand, suitable component values of the lumped network need to be found so that the network exhibits the same frequency response as the physical device. Since there are many interacting parameters to be tuned, this cannot be achieved manually. Hence, a genetic algorithm is applied to this optimisation problem. Two sets of experiments were carried out and a sensitivity analysis has been conducted. It has been shown that the proposed method is capable of finding near optimal solutions within reasonable computation time.