We present a new Short-Open-Load-Thru (SOLT) calibration method for on-wafer S-parameter measurements. The new calibration method is based on a 10-term error model which is a simplified version of the 16-term error model. Compared with the latter, the former ignores all signal leakages except the ones between the probes. Experimental results show that this is valid for modern vector network analyzers (VNA). The advantage of using this 10-term error model is that the exact values of all error terms can be obtained by using the same calibration standards as the conventional SOLT method. This avoids not only the singularity problem with approximate methods, such as least squares, but also the usage of additional calibration standards. In this paper, we first demonstrate how the 10-term error model is developed and then the experimental verification of the theory is given. Finally, a practical application of the error model using a 10 dB attenuator from 140 GHz to 220 GHz is presented. Compared with the conventional SOLT calibration method without crosstalk corrections, the new method shows approximately 1 dB improvement in the transmission coefficients of the attenuator at 220 GHz.

In this paper, we present a two-port on-wafer scattering parameter measurement method to tackle the issue of crosstalk between probes. The proposed method treats the crosstalk separately during the system calibration and the device measurement stages, because the crosstalk during these stages is often different due to changes in the measurement conditions after the probes have been calibrated. For example, devices under test (DUTs) and calibration standards are often situated on different substrates, or, the distance between probes during calibration is different from that during DUT measurement. Based on this concept, we develop a new error model in which the crosstalk is treated as a standalone two-port error network in parallel with the two-port calibration standards or DUTs. The two-port crosstalk error generated during probing, ECT, is removed in the system calibration and corrected during the measurement of the DUT by using a dummy pair of open-circuit standards that are fabricated on the same substrate as the DUT. Since the crosstalk is corrected while measuring the DUT, rather than during system calibration, we call this method "calibration on the fly" (COF). The method is demonstrated using measurements of a 10-dB attenuator between 140 GHz and 220 GHz.

-We present the development of a verification technique for on-wafer noise figure (NF) measurement systems. As the key element of the technique, a verification device consisting of a mismatched attenuator and a low noise amplifier (LNA) has been developed. The attenuator and the LNA are fabricated on two separate chips but joined with a bondwire. The verification procedure based on the device has also been developed and tested on an on-wafer vector network analyzer system with a noise measurement option across the frequency range from 2 GHz to 20 GHz. It has also been found that the bondwire contributes to negligible effect on the system when NF is high e.g. 3 dB but slightly higher when NF is smaller e.g. 1 dB.

This article presents an advanced calibration method for solving the error terms due to probe-probe leakage in an on-wafer test system. A new 12-term error model for the on-wafer test system including vector network analyzer (VNA), frequency extenders (if there are any), cables/waveguides, probes, probe contact pads and probe-probe leakage is introduced. A two-step calibration process and an algorithm with four on-chip calibration standards including one undefined Thru, two pairs of undefined symmetrical Reflects such as Open-Open and Short-Short pairs and a pair of known Match loads has been developed. In addition, an improved circuit model for the Match load is proposed for enhanced accuracy. The calibration method has been tested on a mismatched attenuator for the frequency range between 0.2 GHz and 110 GHz and the results are compared with numerical simulation and existing calibration methods. It's shown that the attenuator's |S11| is more consecutive and |S21| has been improved by up-to 1.7 dB. It is evident that the proposed calibration method has a simpler calibration process and less stringent requirements on calibration standards which are key for on-wafer system calibration at millimeter-wave and terahertz frequencies. More importantly, the new calibration method is more suitable for measurements in which DUTs have variable lengths.

We present details of on-wafer-level 16-term error model calibration kits used for the characterization of W-band circuits based on a grounded coplanar waveguide (GCPW). These circuits were fabricated on a thin gallium arsenide (GaAs) substrate, and via holes, were utilized to ensure single mode propagation (i.e., eliminating the parallel-plate mode or surface mode). To ensure the accuracy of the definition for the calibration kits, multi-line thru-reflect-line (MTRL) assistant standards were also fabricated on the same wafer and measured. The same wafer also contained passive and active devices, which were measured subject to both 16-term and conventional line-reflect-reflect-match calibrations. Measurement results show that 16-term calibration kits are capable of determining the cross-talk more accurately. Other typical calibration techniques were also implemented using the standards on the GCPW calibration kits, and were compared with the MTRL calibration using a passive device under test. This revealed that the proposed GCPW GaAs calibration substrate could be a feasible alternative to conventional CPW impedance standard substrates, for on-wafer measurements at W-band and above.