Sub-mm circuit design requires accurate on-wafer characterization of passive and active devices. In industry, characterization of these devices is often performed with off-wafer SOLT calibration. In this work, validity of this characterization procedure above 110 GHz is investigated by an exhaustive study of on-wafer and alumina off-wafer calibration using measurement and electromagnetic (EM) simulation up to 500 GHz. The EM simulation is performed at two different levels, first at the intrinsic level of the devices under test for reference and afterward up to the probe level to simulate different standards used in the off-wafer calibration or in the on-wafer calibration in presence of the probe. Further, EM simulation data is calibrated with the same procedures and tools that is used in the measurement; therefore, it includes the probe-to-substrate coupling. In addition, precise EM model of a commercial impedance standard substrate (ISS) is developed and used to perform the SOLT calibration. A good agreement is observed between measurement and EM modelling for the off-wafer calibration as well as for the on-wafer calibration. Results clearly highlights a limitation of alumina off-wafer methodology above 200 GHz for characterization of Silicon based technologies. Finally a discussion is given on the pros and cons of the off-wafer and on-wafer methodologies.

This paper investigates on-wafer characterization of SiGe HBTs up to 500 GHz. Test structures for on-wafer thru-reflect-line (TRL) calibration have been designed and are presented. The TRL calibration method with silicon standards has first been benchmarked through electromagnetic simulation. Passive and active components are then characterized up to 500 GHz. The slight discontinuities between the frequency bands are explored. A specific focus was placed on incorrect horizontal probe positioning as well as on probe deformation, resulting in a better assessment of possible measurement errors.

In this paper, we present on-silicon structures on-wafer measurements up to 500 GHz and a comprehensive electromagnetic (EM) simulation analysis to understand non-ideal behaviour in the measured data. The EM simulations are performed in such a way that the simulation setup remains very close to the physical measurement environment where a faithful true EM model of the RF probes is an essential requirement. In this process, four different RF probes used during measurements in the frequency bands 1 GHz - 110 GHz, 140 GHz - 220 GHz, 220 GHz - 325 GHz and 325 GHz - 500 GHz are designed in the EM simulator. We also highlight the importance of the frequency band-specific probe models to develop a deep understanding of the problems encountered in the sub-THz and THz measurements.

The measurement of the figures of merit (FOMs) of an advanced and miniaturized transistor becomes a challenge when its ${f}_{\text {MAX}}$ goes above many hundreds of GHz. In fact, the quantities to be measured become smaller and smaller and thus the influence of the measurement environment becomes less and less negligible. Indeed, when measuring the same test structures using two different renowned commercial probes having a different topology, a “signature” of each probe can been observed, in particular, when plotting $\sqrt {U} \times \text {freq}$ ( ${U}$ is the Mason gain) as a function of frequency, which is usually carried out for ${f}_{\text {MAX}}$ estimation. For millimeter wave (mmW) technologies, ${f}_{\text {MAX}}$ is the key FOM for benchmarking technologies, thus it becomes urgent to clarify this measurement. In this work, we give a proof that measurement becomes probe dependent. We constructed an accurate electromagnetic (EM) model of each probe using X-ray tomography and thus simulated the measurement environment at close proximity of the wafer using EM and SPICE simulation. Hence, each signature of the probe is clearly reproduced by the simulation, highlighting that the unexpected result is not due to an inaccuracy done by the user such as probe positioning or to a limitation of the vector network analyzer (VNA) but is the result of the unwanted coupling between the probe and the substrate or between probes and the inability of the short open load thru (SOLT) and on-wafer thru reflect line (TRL) calibration algorithms to completely remove these couplings.

Terahertz (THz) silicon-based electronics is undergoing rapid developments. In order to keep this momentum high, an accurate and optimized on-wafer characterization procedure needs to be developed. While evaluating passive elements, the measured s-parameter data can be verified by a direct use of EM simulation tools. However, this verification requires to precisely introduce part of the measurement environment such as the probes, pads, and access lines to accurately predict the impact of calibration and layout for on-wafer measurements. Unfortunately, this procedure is limited to passive elements. Hence, in this work, we propose a new procedure to emulate the measurement of active devices using an electromagnetic SPICE cosimulation. By this method, one can clearly highlight that a measurement artifact that was observed for the transistor measurement can be reproduced. One of the most representative examples of measurement artifact involves the measurement and estimation of fMAX which is not constant over all frequency bands. Also, the measurement is difficult to perform above 40 GHz. This typical problem is now undoubtedly attributed to the probe-to-substrate coupling and probe-to-probe coupling which are strongly dependent on the probe geometry. Finally, this cosimulation procedure evidently underlines the need for an optimized deembedding procedure above 200 GHz.