In Situ Time-dependent Dielectric Breakdown in the Transmission Electron Microscope: A Possibility to Understand the Failure Mechanism in Microelectronic Devices

The prerequisite of success in the TDDB experiment is good sample preparation, especially in the FIB milling process in the SEM. Firstly, a thick Pt layer on top of the “tip-to-tip” structure has to be deposited. The thickness and the size of the Pt layer can be adjusted by the SEM operator, but have to follow three principles: (1) The thickness and the size are enough to protect the target area from possible ion beam damage during the whole milling process; (2) There is still a relatively thick Pt layer (≥ 400 nm) on top of the sample left after the milling, it protects the delicate sample from internal and external stresses and minimizes the contribution of stress to the dielectric breakdown in the next TDDB experiment; (3) The size should not be too big, otherwise a conductive path may form between the two pads which are used to apply the voltage to the test structure. Furthermore, the key step is when to stop the ion beam of the final cut. The ion milling must be interrupted immediately once a specifically designed Cu “dummy” interconnect structure in front of the test structure disappears, because the central Cu interconnects consist of the “tip-to-tip” structure and there is only ~ 60 nm space between them. It will be too late if the “tip-to-tip” structure appears in the live SEM image. The thickness of the H-bar TEM lamella is targeted to be about 150 to 180 nm. This thickness allows electron transparency in the TEM at 200 kV acceleration voltage and also keeps a relatively thick dielectric on the sides which encapsulate the “tip-to-tip” structure. On the other hand, the sample thicknesses of 150–180 nm will cause significant multiple inelastic scattering during the ESI chemical analysis, therefore these effects need to be considered into the result analysis as well.

When handling or transferring the sample, wear an antistatic wrist strap. This is very important, since electrostatic discharge damage on several samples was observed in our experiment if the antistatic wrist strap was not worn. The most crucial transfer step is the transfer from the SEM to the TEM. The transfer time has to be strictly restricted within 15 min or less. A long time exposure of the sample in ambient air may absorb moisture and damage the “tip-to-tip” structure, an example is shown in Figure 4A. This sample was stored in ambient air for two weeks before the test. The breached TaN/Ta barrier influences the intrinsic failure mechanism and shortens the lifetime of the test structure. Massive Cu diffusion is then enabled.

One concern for the in situ TDDB study in the TEM is the beam damage on the ULK dielectrics. Therefore, it is very crucial to minimize the influence of the beam on the experiment ¹⁸. Several strategies can be chosen to reduce this influence, but it cannot be eliminated completely. The options could be categorized into three approaches. One possibility is using a small condenser aperture to reduce the total amount of energy deposited in the H-bar type sample ¹⁸. The other option is operating the TEM at low voltage (≤ 80 kV) ^19-21 and/or low electron dose ^22-25. This option is expected to directly reduce the beam damage on the sample. Furthermore, the scanning TEM (STEM) mode can be a low-dose microscopy technique as well, if the experimental parameters are chosen correctly. Thus, the STEM mode should be prioritized if it is a possible option in the TEM used. Choosing low illumination brightness and recording the TEM images with a chosen, reasonable interval time (low dose) are also recommended ¹⁸ to result in further reduced beam damage.

Apart from the intrinsic TDDB, the TEM sample preparation and the TEM observation could theoretically affect the final breakdown. Nevertheless, the TDDB damage mechanism observed is believed to be valid because: (1) with much less TEM beam irradiation (low-dose STEM imaging, low illumination step and recording images every 30 min/1 h), the test sample showed similar failure mechanisms as in our previous TEM observation (recording images continuously, relatively high-dose TEM mode) ^16-18; (2) the electrical field was confirmed as the driving force and the origin of the migration of metal particles ¹⁷ (Figure 5B and 6A) by reversing the electrical connection; (3) migration of metal particles and dielectric breakdown were both observed at specific locations where the tip-to-tip spacing is relatively small and the Ta/TaN barrier is relatively thin, not everywhere inside the illumination area of the TEM beam; (4) a thick layer of Pt deposition on top of the sample prevents most of the contamination from the vertical implantation of Ga ions — the test structure is believed to be mainly contamination free even if there is a slight amount of contamination on the surface of the side walls (about 60 nm) from the lateral damage of the Ga ions. Therefore, the sample preparation and the TEM observation should not affect the interpretation of the intrinsic failure mechanism to a significant amount.

The need of sophisticated procedures for sample preparation and the experimental setup is probably the main disadvantage. This methodology is only applicable for the specifically designed test structure. Therefore, the design and the complicated fabrication process for the dedicated test structure lead to quite more efforts when compared to conventional testing methodologies. Finally, it is worth pointing out that the modification of the sample by beam irradiation in the TEM is inevitable if the electron beam illuminates the delicate sample for a very long time. Nevertheless, we believe that this methodology can enable the study of TDDB failure mechanisms and degradation kinetics.

Further development on the experiment may be able to provide quantitative data for the Cu migration in the dielectrics as a function of applied voltage and/or time and help to develop a more appropriate model for the Cu/ULK on-chip interconnect stacks. In our study, the ESI signal of a Cu bridge at the bottom of the SiCN layer as shown in Figure 5C clearly points to the fact that Cu most likely diffused along the main dielectric/SiCN interface. This top surface of the main dielectric is affected by the planarization process and is expected to have the highest amount of imperfections/defects, then leading to a relatively weak interface with the SiCN layer. Diffusional processes, enabling significant Cu movement, should occur there. The electronic conduction mechanism, preceding the Cu diffusion and leading to dielectric damage, should follow Poole-Frenkel behavior, therefore favoring the √E-model. A deviation from this model cannot be deduced with the here proposed experimental method yet due to the large biases needed to enable reasonable testing times. It should be noted though that lowering the applied voltages and resulting electrical fields should be one of the future tasks in refining the here described experimental methodology. Actual chip operating biases are on the order of 1 to 3 V. The here applied voltages are quite higher, therefore other effects may play a more dominant role at the elevated voltages. To this end, new test structures were designed which have a significantly reduced spacing on the order of 20 to 50 nm. Then, smaller voltages can be applied and experimental data can still be acquired in a reasonable amount of time. The movement of Cu, possibly eliminated at low biases due to the existence of a threshold for dielectric damage to occur, could then be characterized as a function of applied bias and time. These efforts will be part of a forthcoming study and could prove or disprove the impact damage model, which from a physical damage mechanism perspective is currently the most likely model to describe TDDB effects at low biases ¹⁰.

Transmission X-Ray Microscopy (TXM) could be adopted for this experiment as well if the spatial resolution could be improved to much less than 10 nm. More importantly, better transmission ability and lower radiation dose than in the TEM may boost its application on other active microelectronic devices.