Design, Modeling And Parameter Extraction Analysis For On-Chip Spiral Inductors In High-Performance Silicon Rfics

With the high-speed growth of radio communication technology and the improvement of CMOS process design level, RFICs have a more extensive market demand in the field of communication, the inductance of such passive components are used in the key sub-unit circuits of RFICs, and its performance will directly affect the overall performance of the entire circuit. The traditional inductor coils with applying only to specific functional circuits, the fixed structural parameters and a large occupied area of the circuit have difficulty in adapting to the development goals of broadband and miniaturization for RFICs with the rising operating frequency. Meanwhile, various parasitic effects will cause serious performance loss of the inductors. Therefore, it becomes extremely urgent to improve the accuracy of 3D inductor models, which also causes on-chip spiral inductors become far more complicated in 3D modeling, and increases larger difficulties for the extraction, optimization and other processes of component parameters. In addition, since most of the existing technology libraries can’t provide continuously adjustable parameterized inductor models at present, which leads to spend plenty of time and energy on the structural design and model optimization of on-chip spiral inductors, even that do not achieve the desired precision request, and cause great inconvenience for the flexible use of inductors in different functional circuits, the large-scale application of on-chip inductors is also limited greatly. Based on the above difficulties and shortcomings, this paper makes a deep theoretical analysis and carries out a further improvement and perfection for the traditional single-π model of inductors combined with simulation experiment.Based on the influence of high-frequency parasitic effects（the skin, proximity and the substrate coupling effects） on the performance of inductors are not considered for the traditional single-π model varying with frequency. In order to improve the accuracy of models, firstly, an improved single-π model is proposed. Structurally, we connect a L_s1-R_s1 parallel branch on the top of traditional single-π equivalent circuit model, which is used to simulate the skin and proximity effects. In consideration with the influence of the substrate coupling effect on the performance of inductors, a capacitance C_sub is introduced to the substrate branch and characterize this effect. The extraction of the model circuit component parameters are fulfilled using the method of two-port network analysis, quasi-linear functions and supplemented by linear fitting for the improved single-π model. Then, the EM simulation data in HFSS of inductors and the simulation data of multiple performance parameters(quality factor Q, equivalent series inductance L_eff and resistance R_eff, coupling coefficient k, S parameters and its error rate) for the traditional and the improved single-π models are compared, the simulation results show that the accuracy of the improved single-π model appears to increase compared with the traditional single-π model. However, the improved single-π model also have some shortcomings, such as low self-resonant frequency f_res, the poor model equivalence and fitting degree, and can’t meet the requirements of the practical application for inductors. Based on the above shortage, the model structure is optimized design and puts forward an enhanced single-π model. The enhanced circuit structure used to represent the skin and proximity effects is unchanged, uses a parallel branch C₂-R₂ to characterize the substrate coupling effect, and carries out the derivation, extraction and fitting of parameter expressions by using the mentioned method for this model. Finally, the simulation results of several performance parameters for these four models including the EM simulation data in HFSS of inductors, the traditional, the improved and the enhanced single-π models are compared, it fully proves that the enhanced single-π model after structural optimization have a higher precision, fitting degree and equivalence compared with the traditional and the improved single-π models within a given frequency range, which also more in line with and closer to the working state of inductors in practical application.