Root cause analysis of parametric failures in mixed-signal IC designs has been a challenging topic due to the marginality of failure modes. This work presents two case studies of offset voltage (Vos) failures which are commonly seen in mixed-signal IC designs. Nanoprobing combined with Cadence simulation becomes a powerful methodology in fault isolation. Large Vos is typically caused by the mismatch of electrical properties of the components on two balanced rails. In our first case, we present a case-study of nanoprobing combined with bench test and Cadence simulation to debug the root cause of a class-D amplifier voltage offset related yield loss from mixedsignal design sensitivity. Bench electrical measurements confirm the dependency of offset voltage (Vos) on boost voltage (VBST) and amplifier gain settings, which isolates the root cause from mismatch in amplifier gain resistors. The bench measurements match extremely well when an extra parasitic resistance is added to the input of the amplifier in the Cadence simulation. Kelvin 4 points nanoprobing on the amplifier input matching resistors confirmed a 40% mismatch as a result of both layout sensitivity and fabrication. This case highlights that the role of nanoprobing combined with Cadence simulation is not only valuable in physical failure root cause analysis but also in providing guidance to a potential process fix for current and future designs. In our second case, a decrease in offset voltage (Vos) is found through bench validation by reducing the supply voltage (VDD), suggesting a new mismatch mechanism related to the body-source bias. Nanoprobing of the input PMOS transistors clearly shows humps in the subthreshold region of IV characteristics, and the severity of humps increases with body-source bias. Vos derived from the nanoprobing results aligns well with the bench data, suggesting hump effect to be the root cause of Vos deviation. This study suggests that by combining Cadence simulation and nanoprobing in the failure analysis process of parametric failures, suspicious problematic devices can be identified more easily, greatly reducing the need for trial and error.