The characterization of semiconductor-based electronic devices is a deceptively simple task and more often than not lacking sufficient attention. In fact, even during the measurement of a simple ID-VG curve of a MOS transistor, the state of the device is altered through the charging and discharging of traps close to the interface. This results for instance in a hysteresis in the ID-VG curve, meaning that different drain currents are measured during the up- and down-sweeps. Additionally, if the sweep range is too large, the device will be stressed and defects could potentially be created. Similar issues need to be considered during charge pumping and capacitance-voltage measurements. In order to avoid misinterpretation of the data, all measurements must be carefully controlled and the experimental conditions must be replicated in the models used to interpret the obtained data. This is the primary reason why we supplement our modeling work with experimental results obtained in our own characterization lab, to make sure we know exactly what has been measured and how.
The problem is exacerbated during the characterization of reliability issues, where the device is initially characterized, then stressed, and finally measured once more to assess the impact of the stress. These measurements need to be conducted as quickly as possible (in microseconds, to minimize unwanted stress recovery) and under perfectly defined conditions (to understand the unavoidable recovery behavior), in order to make this data useful for our reliability modeling efforts.
Our workhorse for the characterization of single defects is time-dependent defect spectroscopy (TDDS), an extension of deep level transient spectroscopy (DLTS). The TDDS uses nano-scale devices (e.g. 100nm x 100nm) for conventional silicon technologies, where the impact of individual traps is clearly visible in the device characteristics as discrete steps occurring every time the charge state of a particular defect changes. During a TDDS experiment, devices are repeatedly subjected to stress and recovery phases, where the recovery data is analyzed using statistical methods. By analyzing the time and step-heights visible in the recovery traces (Fig. 1), spectral maps can be built which clearly show the impact of individual defects (Fig. 2). Analyzing the bias and temperature dependence of the defect-related clusters provides a wealth of information on the defect kinetics, including metastable-states and volatility.