Diffusional back flows during electromigration
Introduction
In the course of devising an experimental technique for measuring the electromigration drift velocity in thin films, current was passed through aluminum strips of various lengths deposited onto conducting TiN[1]. Most of the current was carried by the low resistivity aluminum causing it to migrate on top of the TiN in the direction of electron flow leaving a metal free area near the negative end of the strip and metal accumulation near the positive end.
It was discovered that the metal free area and hence, the drift velocity, was a function of the strip length. Longer strips showed a much higher drift velocity than shorter ones. In fact, at a given current density very short strips below a critical length did not drift at all. It was suggested that atoms transported by the electromigration are creating a stress and/or concentration differences between the strip ends causing a back flow counteracting the electromigration. The counterflow in short strips is higher since the stress and/or concentration gradients are higher leading to the slower observed drift velocity.
Both stress gradients and counterflows have been observed experimentally1, 2, 3. The separation between the effects of concentration and stress gradients remained elusive. The counterflow is in general a beneficial phenomenon as it tends to reduce the electromigration damage and in extreme cases prevent it altogether. For example, the increased lifetime at lower current densities was suggested to result from critical length effects4, 5. In fact any geometrical microstructural or temperature change that will lead to increased electromigration flow will immediately create a local opposing back flow.
The concept of critical length, counterflows as well as suggestions for further studies are discussed in the present paper.
Section snippets
Critical length
The unequal drifts of aluminum film deposited on TiN are clearly seen in Fig. 1, a micrograph first published in 1976[1]. Electrical current density of 3.7×105 A/cm2 was passed for 15 h at 350°C through a series of strips, the direction of electron (and atom) flow was from right to left. The strip on the right, 10 μm long, did not drift at all while the strip on the left, 90 μm long, drifted as much as 20 μm under the same conditions. The drift as a function of the inverse strip length is seen in
Local stress measurements
Changes in stress normal to the film grain boundaries were used to explain the observed drift behavior[2]. The free energy (or chemical potential) difference between the strip ends was assumed to originate from differences in the normal stress,where μa and μv are the chemical potentials of the atoms and vacancies, respectively, μ0 is a constant, Ω is the atomic volume, and σnn the stress normal to the grain boundaries7, 8.
z* can be inferred by measuring the stress gradient along a
Suggestions for future work
Since the first report on critical length, a number of publications appeared which deal with drift velocity measurements9, 10. However, the relative roles of stress and vacancy/interstitial super saturation is still not clear.
If a free atom/vacancy exchange occurs on the grain boundaries than Eq. (7)is valid and the stress gradient is solely responsible for the back flow. The stress is in quasi equilibrium with the local vacancy concentration. When electrical current flows through a strip, some
Conclusions
The back flow observations are commensurate with stress as a driving force. The value of the effective charge for aluminum electromigration was found, by analyzing the back flow from a pad, to be about unity. The agreement with experimental results does not constitute a proof of the validity of the theory. The energy needed for driving the atoms back can also be provided chemically by super saturation of atoms (interstitials). Analyzing the back flow for this case yields a much higher effective
References (10)
- et al.
Thin Solid Films
(1975) Scripta metall.
(1971)J. Appl. Phys.
(1976)- et al.
Appl. Phys. Lett.
(1976) - et al.
Appl. Phys. Lett.
(1977)
Cited by (87)
Peridynamic modeling of void nucleation and growth in metal lines due to electromigration in a finite element framework
2023, Computer Methods in Applied Mechanics and EngineeringCoupling model of electromigration and experimental verification – Part I: Effect of atomic concentration gradient
2023, Journal of the Mechanics and Physics of SolidsPressureless two-step sintering of ultrafine-grained tungsten
2020, Acta MaterialiaCitation Excerpt :The involvement of activation entropy S(T) in Eq. (5) seems necessary, so the cooperative motion of many atoms and/or multiple physical processes are likely to be involved. Recently, there have been strong evidence showing large stress generation during grain growth [44] as well as back stress generation associated with diffusive processes in general [45,46]. It is possible that low-temperature grain boundary motion is controlled by stress general/relaxation, especially at grain junctions where stress concentrates [47]; it involves many atoms due to long-range elasticity, thus has a large contribution to the pre-exponential term.
Electromigration in Metallic Materials and Its Role in Whiskering
2017, Handbook of Solid State DiffusionReal-time study of electromigration in Sn Blech structure
2016, Applied Surface ScienceCitation Excerpt :Fig. 3(c) shows a magnified image of the L0 = 20 μm case shown in Fig. 3(a), confirming that no Sn depletion or whisker was created in the short strip case. According to Eq. (1), the effect of the backflow generated by the stress gradient dσ/dx becomes increasingly important in short strips because dσ/dx inversely increases with the strip length [1,17]. Our observations provide a clear manifestation of the backflow effect on the short strips; specifically, the electromigration flux is balanced by the opposing backflow generated by dσ/dx during the testing.