Why we need to rethink copper low-k reliability issues
01/01/2003
With the introduction of copper and low-k dielectric interconnects for advanced ICs, the semiconductor industry faces tremendous new challenges because reliability performance and failure mechanisms are not well understood (see figure). Moreover, the speed of introduction can exceed conventional capabilities to explore reliability performance.
The industry must therefore fundamentally rethink copper low-k reliability issues to guarantee sufficient lifetime for ICs manufactured with these advanced technologies. We see the associated challenges falling into five major areas.
Low-k dielectric reliability
Dielectric reliability was never an issue with silicon dioxide because it has excellent robustness; it was not even a subject of qualification or reliability testing. The intrinsic reliability of new dielectrics, however, is much less in terms of leakage current, maximum breakdown voltage, etc. Moreover, with ever-scaling dimensions, dielectric spacing between wires becomes so narrow that interactions from processing cannot be avoided.
This is a logical consequence of the porosity of dielectrics used, since all low-k dielectrics are porous and thus permeable to moisture and chemicals. Also, minute dielectric changes at the sidewalls of trenches play an increasingly large role as spacing narrows. What we need are the definitions of failure criteria for these materials; more fundamental work needs to be done to understand conduction and failure mechanisms.
Barrier integrity
The task of a diffusion barrier is not only to protect devices from copper diffusion, but also to protect the dielectric from process gas and chemical penetration. It is a fact that deposition of a thin film on a porous material leads to a porous thin film. This is the case for a metallic or dielectric diffusion barrier. As a result, pore sealing has become the critical issue for the implementation of low-k dielectrics, since defective sealing will lead to holes in the barrier or a leaky barrier for copper diffusion. The quality of the barrier is not related to the choice of the barrier material only, but to the whole process module in which it is implemented.
 A cross-sectional micrograph showing extensive sidewall hillocking formed from thermal stressing of a copper line in a compliant low-k dielectric. |
null
While new methodologies have been put in place that allow at least comparative data on the integrity of barriers within copper low-k modules, we still need a more systematic approach to data collection to translate these comparative data into actual predictive curves for lifetime. Since the push for thinner barriers is enormous (to free up space for lower-resistivity copper), quantitative data for the evaluation of the barrier toward its function and its performance are a must and are the subject of intensive research.
Electromigration
The methodology of electromigration testing for copper interconnects has been based on extensive experience with aluminum oxide modules. With the introduction of copper, accelerated test temperatures have increased from typically 180–200°C to 350–400°. Copper-oxide modules can be tested under these conditions, but this is no longer the case for copper low-k modules. Elevated test temperatures are complemented with huge current densities for test. Because of the lower robustness of low-k materials to temperature, the structures do not pass the tests due to phenomena different from electromigration.
New test structures are being investigated that can give valuable data about the electromigration phenomenon without imposing unrealistic test conditions. Again, we need a systematic study and new models for extrapolation of lifetimes need to be developed.
Stress migration
Stress migration is a very important issue in copper low-k interconnects. Indeed low-k dielectrics are compliant materials with limited rigidity, especially when compared to silicon dioxide. This compliance does have an impact on electromigration, on the thermal stability of structures, and also on their integrity over time. Compliance can lead to creep within the copper low-k module, and again new phenomena show up that need to be tested during process development and packaging.
The impact on front-end reliability
An important yet quite unexplored problem is the impact of copper low-k interconnects on front-end device reliability. To explore this problem, new test structures are needed to evaluate, for example, the possible impact of diffusion of copper and other materials on the reliability of front-end devices, such as MOSFETs, capacitors, bipolar devices, and flash memory cells.
New test methodologies will have to be developed to quantify and characterize this impact. Test conditions and acceleration models have to be reconsidered for use in calculating the reliability under real operating conditions. This requires a fundamental understanding of the failure mechanisms governing these reliability problems.
Also, the problem of plasma processing-induced damage, which is well known from the classical aluminum-oxide backend, might show up again, but probably at different levels and in a different manner due to the specific dual damascene process architecture. Phenomena such as electron shading can become important again and will have to be quantified.
It is clear that the reliability community, which for years has been running tests and extrapolating data using more or less standardized and well-established test methodologies, acceleration models, and activation energies, must step back and rethink these issues for newly developed backend modules. Extrapolations can only be valid if the right phenomena are described and the right activation energies are known. This is the case for the reliability of the dielectric itself, for barrier integrity, for electromigration and stress migration, and — last but not least — for the impact of the backend on frontend reliability.
The successful introduction of copper low-k modules will, to a large extent, depend on the capabilities of the research community to cope with these problems within a reasonable time.
Karen Maex received her PhD from the Department of Electronic Engineering, Katholieke Universiteit (KU). She is a professor at KU and a Fellow at IMEC, Kapeldreef 75, B-3001 Leuven, Belgium; ph 00/32/16-28-13-58, fax 00/32/16-28-12-14, [email protected].
Guido Groeseneken received his PhD in electronic engineering from KU. He is head of technology reliability research at IMEC.
±0.3Å repeatability on TaN under Cu
While tantalum nitride (TaN) forms an excellent barrier layer in copper interconnect schemes, with low resistivity and excellent adhesion to provide good contact to the copper, characterizing TaN films is still an issue.
For example, because the TaN films are buried below copper, it is difficult to get meaningful results from techniques such as 4-point probes.
Now, work done at done at PANalytical (formerly Philips Analytical, Almelo, The Netherlands) using x-ray fluorescent spectrometry (XRF) has demonstrated a repeatability of within 0.3 angstroms for characterizing TaN barrier-adhesion films buried under copper interconnects.
The measurements were taken on a research and development instrument that can be easily converted to production applications by a load port and automated handler.
Excellent selectivity and precision make XRF well suited to this type of measurement.
Results were obtained with a 4kW Rh anode Super Sharp tube, set at 32 kV/125mA, and with fixed measuring channels for N, Cu and Ta. A specially designed fixed channel to measure Cu was used to avoid interference.
The Ta and Cu signals gave the TaN and Cu layer thicknesses, respectively, while the N signal determined the layer stoichiometry.
With XRF analysers, it is normally difficult to produce reliable results for the stoichiometry of the TaN under copper films that are relatively thick (less than 500 angstrom). This is because the copper absorbs almost all the N radiation.
However, with the cited method, fundamental parameter software was able to analyze TaN film stoichiometry and thickness by calculating the absorption in the different layers.