Degradation of Electrical Parameters of Power Semiconductor Devices – Process Influences and Modeling
3.2 Power metal type
In principle, proper encapsulation of a device should impede any influence of BEOL process steps on the NBTS reliability. However, rare observations indicate indeed that particular BEOL process steps may have an impact on NBTI [Nel+05; Ho+06; Aic10]. Also an influence of the type of power metallization, be it aluminium (Al) (usually standard) or copper (Cu) (introduced only recently for power devices [Nel+11]), on the instability was reported [Aic10].
In order to further study this effect a dedicated wafer split with equivalent devices, differing only in the power metallization, has been fabricated and analyzed. The results of this Section are also summarized in [SPN12] and were obtained by the Master Student Roberta Stradiotto [Str13] in cooperation with the author.
3.2.1 Process differences
Considerable differences between the process steps for deposition of aluminium (Al) and copper (Cu) as metal layers to connect power semiconductor devices exist. Al is usually sputtered uniformly over the whole wafer area. A photoresist mask is then used to expose only certain areas of the wafer to an Al etching agent. In contrast, processing Cu, usually only a thin seed layer is sputtered over the whole wafer area. The photoresist mask is then used to impede the electrochemical deposition of Cu where the material is not needed. After removal of the photoresist mask the thin seed layer of Cu is etched away.
Both variants of power metallization share the initial sputtering step over the whole wafer. So independently of whether there is metal above the device or not, it had been covered by the metal during processing. Consequently, devices with or without a power metal plate on top should perform and degrade equivalently. It will be shown in the next Subsection that this is indeed the case.
3.2.2 Electrical characterization
Remarkably, in a comparison of the transfer characteristics of the virgin devices with Al or Cu metallization the device-to-device variability within one metal type is larger than the impact of the power metallization process (not shown). Only with a CP measurement, which is very sensitive to differences in the virgin interface trap density, a small difference as illustrated in Fig. 3.13 is detected.
Thereby, the device with Al power metallization has a larger CP current than the Cu power metallization device, indicating a larger virgin interface trap density [SPN12]. Furthermore, it is irrelevant for the virgin performance of the device whether the metal is kept at the top of the device (traces labeled ‘plate’ in Fig. 3.13) or removed.
Fig. 3.14 shows the result of NBTS for devices with different power metallization. It is observed that the devices with Al power metallization drift more than the devices with Cu power metallization.
By using a repetitive MSM experiment it is shown in Fig. 3.14 that the difference is due to the permanent component P while the recoverable component
R is equivalent for both types of power metallization. It is remarked that this measurement gives only an estimate of P and R. P is calculated from the remaining shift after a 1 ks long recovery phase around the . R is estimated from the recovery between the first and last measurement point, 10 ms and 1 ks after termination of the stress after a recovery phase at 200 °C.
3.2.3 Discussion
The larger drift of the devices with Al power metallization despite the very similar virgin characteristics is challenging to understand. One possible reason could be in principle plasma induced damage. This effect is the charging and associated degradation due to accumulation of ions during plasma processing [Eri+12]. If this is the reason for the different drift behavior, also small differences in the performance of the virgin device should be visible. Furthermore, difficult to understand is the aspect that the virgin interface trap density of the Al device is larger than that of Cu processed devices. If the hydrogen passivation degree, as described in Section 3.1, would be responsible for the different behavior, the result should be the opposite, meaning Al should drift less compared to Cu. However, it is still likely that the behavior is connected to hydrogen because only the quasi-permanent component P is affected by the type of the power metallization. This shows similarities to the devices with different hydrogen passivation described in Section 3.1.3, where also only the quasi-permanent component is affected. This indicates that both topics have most probably a common microscopic origin, the interaction of defects near the interface with hydrogen.
A possible explanation for this discrepancy can be found in [DMC69; Dun89], where it is described that Al acts as a catalyst to split molecular H2 or water (H2 O) into more reactive atomic hydrogen. It is speculated that the high temperature steps during Al deposition cause atomic as well as molecular hydrogen to diffuse towards the Si-SiO2 interface. There, the two species may either create additional interface traps via the reaction (3.4) which would lead to a larger interface trap density measurable in CP, or may passivate existing interface traps with the inverse reaction. For either power metal type, one of the two mechanisms may have a larger impact than the other.