(image) (image) [Previous] [Next]

Degradation of Electrical Parameters of Power Semiconductor Devices – Process Influences and Modeling

Chapter 5 Impact of temperature on BTI

The poly-heater described in Chapter 4 allows a thorough investigation of the influence of temperature on the BTI degradation and recovery mechanisms. The two main benefits of the poly-heater, which are unique in the context of BTI research, are temperature switches during the application of bias and the extension of the accessible temperature range. These features give for example the possibility to study the impact of the stress temperature independently of the recovery temperature. This provides extensive insight into the BTI degradation mechanism.

5.1 Measurement results overview

BTI is a temperature activated degradation mechanism, meaning the effect becomes more severe with increasing temperatures. Typically, temperatures at which BTI occurs are around 100 °C to 200 °C [AM05; HDP06; Sch07]. In a conventional BTI experiment the temperature is set by a thermo chuck or a dedicated furnace. There, it is usually not possible to maintain a bias applied to the device while performing a temperature switch. Furthermore, reliable temperature switches require minutes to hours. As a consequence, most (math image) dependent BTI data is measured at the same stress and recovery temperature. This limitation may introduce errors such as inaccurate (math image) values because the recovery temperature impacts the (math image) value [ZCG07]. Furthermore, higher temperatures accelerate recovery from previously created damage [Kat08; ANG08; Ben+09]. The increased recovery with increasing temperature balances the increasing degradation, resulting in an apparently weak temperature dependence of BTI, which is inconsistent with the rather large activation energies of the defect precursors in the Si-SiO2 system [Gra+11b]. In the following Subsections the impact of temperature is strictly separated in an influence on the stress phase and an influence on the recovery phase. This is achieved by a sophisticated use of the poly-heater.

5.1.1 Stress temperature dependence of BTI

The poly-heater can be used to vary only the stress temperature in an MSM experiment while maintaining the recovery temperature always at the same level. That is to say, a temperature and gate bias sequence as illustrated in Fig. 5.1 becomes possible.


Fig. 5.1: Evolution of (math image) and the device temperature in an isothermal MSM experiment.


Fig. 5.2: Stress temperature dependence of NBTI in an MSM experiment. A bake phase at \( \approx \SI {400}{\celsius } \) for 10 s removes all created damage between each measurement. (math image) is the cumulative sum of the stress duration. The 90 °C curves prove the reproducibility of the measurement.

As a particular example, the device temperature is switched between 30 °C for the recovery and 100 °C during the stress. The temperature switch to 30 °C is always performed 10 s before the termination of the stress bias. This assures that the recovery remains unaffected by the choice of the particular stress temperature. The delay between the temperature switch and the bias switch causes a period with stress bias at 30 °C. Due to the temperature activation of the degradation the impact of the stress at 30 °C can be neglected in comparison to the stress at 100 °C. This approach is referred to as degradation quenching [ANG08; Aic10].

The results of such isothermal (constant stress temperature) MSM experiments are shown in Fig. 5.2. The recovery phase is an interruption of the stress bias of about 1 s at 30 °C. Within this period, the drain current measurement is performed 10 ms after the gate bias switch from stress to readout level ( (math image)). The removal of the stress at a low temperature keeps the impact of the recovery small. The data in Fig. 5.2 marked with circles is measured at the same stress temperature but different stress times between each measurement point. The good overlap of all those traces indicates that the influence of the length of an individual stress phase is insignificant considering the sum of the stress time. Furthermore, The characteristics are recorded subsequently on the same device with intermediate bake steps. These bake steps for 10 s at 400 °C assure complete recovery from the NBTS and restore the virgin state of the device [Kat08; BOG08]. The overlap of the characteristics marked with circles therefore also proves that the measurement is reproducible.

The unique feature of the experiment shown in Fig. 5.2 compared to other, similar plots is that the recovery level is always the same, 10 ms at 30 °C, independent of the stress temperature. This gives quite different results compared to the standard approach where the stress temperature equals the recovery temperature as shown in Fig. 5.3.


Fig. 5.3: Comparison of MSM experiments where \( \gls {Tstr}=\gls {Trec} \) and where \( \gls {Tstr}\neq \gls {Trec} \). The recovery temperature for the \( \gls {Tstr}\neq \gls {Trec} \) characteristics is 30 °C.

For a measurement with \( \gls {Tstr}=\gls {Trec} \) the degradation level is much lower than with \( \gls {Trec}<\gls {Tstr} \). This is because in the recovery trace 10 ms after termination of the stress a larger part of the degradation has already recovered at higher (math image). Furthermore, the interruption of the stress for about 1 s at 200 °C causes a larger recovery. As a result, this measurement shows that the impact of the stress temperature is larger than what would be expected from measurements where \( \gls {Tstr}=\gls {Trec} \) and conventional MSM experiments at different temperatures should not be compared with each other.

In Fig. 5.4 the impact of the stress temperature on NBTI is depicted.


Fig. 5.4: Stress temperature dependence of NBTI in an MSM experiment. Only the stress temperature is varied while the recovery is always reduced to 10 ms at 30 °C.

Particularly evident is the larger increase of the (math image) at stress temperatures \( >\SI {200}{\celsius } \). Possible reasons for the temperature activation of the degradation will be given in Sections 5.2 to 5.5.

5.1.2 Recovery temperature dependence of BTI

Similarly to the stress temperature, also the dependence of BTI on the recovery temperature can be measured using the poly-heater. However, the sequence of temperature and bias switches is more complex and is sketched in Fig. 5.5.


Fig. 5.5: Sequence of voltage and temperature changes for the experiment of Fig. 5.6.


Fig. 5.6: Impact of the recovery temperature on NBTI [PG13b]. The stress phase is always at the same temperature, has the same duration and the same electric oxide field. Higher recovery temperatures decrease the observed (math image) values.

As for the measurement of the stress temperature dependence, a degradation quenching phase is needed to avoid acceleration of the recovery due to unintentionally high recovery temperatures. The lowest temperature of this experiment was chosen to be −60 °C which is the lowest achievable temperature of the thermo chuck used in the thesis.

Simultaneously with the termination of the stress bias the power supply of the poly-heater is switched to a value corresponding to the desired recovery temperature. The thermal capacitances of the materials surrounding the heater and the device will cause the device temperature to follow the supply switch within a few seconds delay. See Section 4.5 for details on the restrictions of the poly-heater. The recovery in the first seconds after termination of stress will therefore not occur at exactly the desired (math image). However, since the temperature is switched starting from −60 °C, and most of the temperature switch occurs already in the first millisecond (see Fig. 4.17) the recovery is not accelerated. Also, the temporary temperature deviation impacts mostly the dependence of the drain current on the temperature. The small temperature difference to the target (math image) causes a change in the drain current which results in a spurious (math image) shift [ANG09a].

There are two ways to account for this effect and to obtain \( \gls {dVth}(\gls {trec}) \) data which is unaffected by the unstable temperature phase. The first is to record the spurious \( \gls {dVth}(\gls {t}) \) characteristic following a temperature switch on the unstressed device prior to stress and to subtract this characteristic from the resulting \( \gls {dVth}(\gls {trec}) \) characteristic of the stressed device [ANG09a; Aic10]. This is a very convenient approach and sufficient for most experiments. A second, more precise way is to record virgin ID VG characteristics for every recovery temperature prior to stress. This is of course only possible by measuring ID VG s at a few distinct temperatures in the region of e.g. −60 °C to 250 °C, modeling the change of the transfer characteristics, and interpolating the temperature dependence of the parameters of the ID VG model. The latter method is more elaborate but also more accurate because it also accounts for a possible dependence of the (math image) value on the particular readout temperature [ZCG07].

The result of a measurement using this correction is shown in Fig. 5.6. In order to avoid inaccuracies related to device-to-device variability, the measurements were performed consecutively on a single device with intermediate bake steps of 10 s duration at 300 °C at zero gate bias to recover from all previous damage of the device [Kat08; BOG08; Pob+11b]. It is clearly evident that lower recovery temperatures cause a larger measurable degradation level [ANG08; Pob+11b; PG13b]. Particularly important is that the recovery occurs at the target recovery temperature already before the first measurement point 10 ms after termination of the stress. This means at higher (math image) most of the degradation has recovered before the first drain current measurement 10 ms after stress. Conversely, at −60 °C, most charges do not recover and a large (math image) value is measured. Further discussions on the reason for the larger degradation level will be given in the following Sections.

By using the poly-heater it is also possible to switch the temperature during the recovery while keeping the bias applied to the gate. The temperature and gate bias sequence for such an experiment is sketched in Fig. 5.7.


Fig. 5.7: Temperature and (math image) sequence for an experiment where (math image) is switched during the recovery phase.


Fig. 5.8: Change of the threshold voltage drift over recovery time with intermediate (math image) switches. The recovery temperature is switched to higher values during the application of a constant recovery gate bias (math image). The temperature and gate voltage sequence is sketched in Fig. 5.7.

The result of an example experiment is shown in Fig. 5.8. The switch to higher temperatures leads to an abrupt decrease of the remaining (math image) [ANG08; Aic10]. Inversely, a switch to lower temperatures during recovery leads to a freeze of the degradation at the corresponding (math image) value [ANG08; Aic10].