Commercial interest for millimeter-wave wireless communication has increased significantly over the past years due to the appearance of large volume applications such as automotive radar and telecom applications, i.e. as 5G systems operating in the FR2 band, ISM applications in the 60 GHz E-band, and beyond 100 GHz in the W-band and D-band. Developments in this frequency range are enabled by the continuous improvements in silicon-based technologies over the past decades and are driven by frequency congestion in low-GHz bands. To support development for these applications there is a demand for new millimeter-wave test equipment, de-embedding/calibration methods and device characterization approaches. Three topics related to challenges in millimeter-wave on-wafer characterization are addressed in this work. A novel characterization method is developed for in-band linearity estimation that overcomes the limitation of harmonically generated spectral content by the frequency extension modules in the millimeter-wave active load-pull setups. The method is able to estimate trends in the EVM across input drive level, frequency and loading condition from vector gain measurements of the active device. On-wafer measurements were performed to benchmark the method against true EVM measurements in two sub-30 GHz bands. A proof of concept demonstration at 165 GHz showed the ability of the method to estimate trends for in-band linearity across drive level and loading condition at millimeter-wave. A new parametrized 16-term error model and calibration algorithm was developed to overcome probe-to-probe cross-coupling effects during millimeter-wave on-wafer characterizations. Conventional 16-term calibrations are not practical as these are only effective if the probe-to-probe spacing is not changed after calibration. This new approach allows for interpolation of the error terms for devices under test that have a pad spacing not included in the set of calibration standards. This makes the method powerful and practical as it would only require one set of calibration structures for large-scale on-wafer measurements of devices with a different pad spacing. Finally, a preliminary investigation was performed into active device behaviour when operating close to the maximum oscillation frequency of the technology. This investigation is conducted using on-wafer millimeter-wave large signal measurements on a 22-nm CMOS FD-SOI technology. A new experimental method is proposed that could show different operating behaviours when moving towards the maximum oscillation frequency. This approach uses large-signal (load-pull) measurements of the device across power and supply voltage. Under the hypothesis that a slew rate like operation is obtained when the maximum oscillation frequency is approached, this characterization may show the transition point in frequency for the onset of such device limiting behaviour.