Empirical Study on the Validity of the Modal Separation Approach for Deembedding of Highly Asymmetrical Passive Differential Devices

Accurate on-wafer characterization of differential devices poses the challenge of de-embedding four-port error networks as a multimode problem. We have shown in our previous work that, under the assumption of negligible mode conversion, e.g. less than -30dB, the 4×4 transfer matrix can be separated into two 2×2 matrices representing the differential and common-mode S-Parameters. This enables independent de-embedding of the two modes using classical two-port de-embedding techniques such as Thru-Line (TL) and Thru-Open (TO), which greatly decreases the complexity of the de-embedding problem. In this work, we compare the Multimode-Thru-Reflect-Line (MTRL) de-embedding technique to the classical two-port Thru-Reflect-Line (TRL) procedure including the pure-mode approximation. The MTRL technique performs true multimode de-embedding under consideration of all modal conversion terms and can be viewed as a theoretical gold standard for de-embedding of asymmetrical differential devices. We compare the two methods for the edge case of highly asymmetrical passive error networks to investigate further whether the modal separation approach / pure-mode approximation holds up to true multimode de-embedding. As an example, we characterize an on-chip transformer as the device-under-test (DUT) realized in 28 nm bulk CMOS technology. Several asymmetric de-embedding structures were realized in the same process.