The Role of Ru Passivation and Doping on the Barrier and Seed Layer Properties of Ru-Modiﬁed TaN for Copper Interconnects

The Role of Ru Passivation and Doping on the Barrier and Seed Layer Properties of Ru-Modiﬁed TaN for Copper Interconnects. Size reduction of the barrier and liner stack for copper interconnects is a major bottleneck in further down-scaling of transistor devices. The role of the barrier is to prevent diffusion of Cu atoms into the surrounding dielectric, while the liner (also referred to as a seed layer) ensures that a smooth Cu film can be electroplated. Therefore, a combined barrier+liner material that restricts the diffusion of Cu into the dielectric and allows for copper electro-deposition is needed. In this paper, we have explored barrier+liner materials composed of 1 and 2 monolayers (MLs) of Ru-passivated epsilon-TaN and Ru doped epsilon-TaN and focus on their interactions with Cu through the adsorption of small Cu clusters with 1-4 atoms. Moreover, different doping patterns for Ru doping in TaN are investigated to understand how selective doping of the epsilon-TaN surface influences surface stability. We found that an increased concentration of Ru atoms in the outermost Ta layer improves the adhesion of Cu. The strongest binding of the Cu atoms was found on the 100% Ru doped surface followed by 1 ML Ru passivated surface. These two surfaces are recommended for the combined barrier+liner for Cu interconnects. The closely packed arrangements of Cu were found to exhibit weak Cu-slab and strong Cu-Cu interactions, whereas the sparse arrangements of Cu exhibit strong Cu-slab and weak Cu-Cu interactions. The Cu atoms seem to bind more favourably when they are buried in the doped or passivated surface layer due to the increase in their coordination number. This is facilitated by the surface distortion arising from the ionic radius mismatch between Ta and Ru. We also show that the strong Cu-Cu interaction alone cannot predict the association of Cu atoms as a few 2D Cu clusters showed stronger Cu-Cu interaction than the 3D clusters, highlighting the importance of Cu-surface interactions Size reduction of the barrier and liner stack for copper interconnects is a major bottleneck in further down-scaling of transistor devices. The role of the barrier is to prevent diﬀusion of Cu atoms into the surrounding dielectric, while the liner (also referred to as a seed layer) ensures that a smooth Cu ﬁlm can be electroplated. Therefore, a combined barrier+liner material that restricts the diﬀusion of Cu into the dielectric and allows for copper electro-deposition is needed. In this paper, we have explored barrier+liner materials composed of 1 and 2 monolayers (MLs) of Ru-passivated (cid:15) -TaN and Ru doped (cid:15) -TaN and focus on their interactions with Cu through the adsorption of small Cu clusters with 1-4 atoms. Moreover, diﬀerent doping patterns for Ru doping in TaN are investigated to understand how selective doping of the (cid:15) -TaN surface inﬂuences surface stability. We found that an increased concentration of Ru atoms in the outermost Ta layer improves the adhesion of Cu. The strongest binding of the Cu atoms was found on the 100% Ru doped surface followed by 1 ML Ru passivated surface. These two surfaces are recommended for the combined barrier+liner for Cu interconnects. The closely packed arrangements of Cu were found to exhibit weak Cu-slab and strong Cu-Cu interactions, whereas the sparse arrangements of Cu exhibit strong Cu-slab and weak Cu-Cu interactions. The Cu atoms seem to bind more favourably when they are buried in the doped or passivated surface layer due to the increase in their coordination number. This is facilitated by the surface distortion arising from the ionic radius mismatch between Ta and Ru. We also show that the strong Cu-Cu interaction alone cannot predict the association of Cu atoms as a few 2D Cu clusters showed stronger Cu-Cu interaction than the 3D clusters, highlighting the importance of Cu-surface interactions.

Size reduction of the barrier and liner stack for copper interconnects is a major bottleneck in further down-scaling of transistor devices. The role of the barrier is to prevent diffusion of Cu atoms into the surrounding dielectric, while the liner (also referred to as a seed layer) ensures that a smooth Cu film can be electroplated.
Therefore, a combined barrier+liner material that restricts the diffusion of Cu into the dielectric and allows for copper electro-deposition is needed. In this paper, we The strongest binding of the Cu atoms was found on the 100% Ru doped surface followed by 1 ML Ru passivated surface. These two surfaces are recommended for the combined barrier+liner for Cu interconnects. The closely packed arrangements of Cu were found to exhibit weak Cu-slab and strong Cu-Cu interactions, whereas the sparse arrangements of Cu exhibit strong Cu-slab and weak Cu-Cu interactions.
The Cu atoms seem to bind more favourably when they are buried in the doped or passivated surface layer due to the increase in their coordination number. This is facilitated by the surface distortion arising from the ionic radius mismatch between Ta and Ru. We also show that the strong Cu-Cu interaction alone cannot predict the association of Cu atoms as a few 2D Cu clusters showed stronger Cu-Cu interaction than the 3D clusters, highlighting the importance of Cu-surface interactions.

I. INTRODUCTION
The scaling of interconnects has become the most serious limiting factor in the downsizing of complementary metal oxide semiconductor (CMOS) devices, as shown schematically in Fig. 1. 1,2 While Cu performs extremely well as the interconnect metal, it requires a conductive diffusion barrier to prevent migration of Cu atoms into the dielectric and a liner material (also referred to as a seed or adhesion promoter or glue layer) to allow for successful electroplating of Cu films. [3][4][5][6][7] The first diffusion barriers used were Ta and Si 3 N 4 . 4,8-11 A wide variety of materials has been studied as alternative barrier materials in order to drive scaling and performance. [12][13][14][15][16][17] Most of these materials are binary or ternary materials involving refractory metals, such as Ta, Mo and Os. [18][19][20][21][22][23] Less work has been undertaken to find Ta. 26 To test different barrier as well as liner materials, van der Veen et al studied Ru and Co liners in combination with both TaN and MnN barriers, focusing in particular on examining the gap-fill performance as well as the electromigration and resistivity performance of the different material combinations. 27 Cheng and coworkers carried out a detailed theoretical study of various transition metals as potential liner materials on δ-TaN (1 1 1) and derived criteria based on adhesion energies that allow the prediction of a suitable liner material for a given system. For an ideal liner material on a given diffusion barrier the following three criteria were proposed 28 : 1. Metal agglomeration on the liner material is prevented when the adhesion energy of the metal to the liner is stronger than the adhesion of the metal to the substrate. The substrate in this case would be the barrier material. is stronger than the adhesion of the metal on the substrate. 3. It is important that the liner material does not diffuse into the metal. This is prevented so long as the adhesion of the liner to the substrate is stronger than the adhesion of the metal to the liner material.
For example, using these criteria, they were able to accurately predict that when using Al as the liner material, Al atoms would diffuse into the Cu layer, as it failed to meet the third of their three criteria. The adhesion energy between Cu and Al layers was stronger than that between Al and TaN and thus diffusion occurred.
Currently, a tri-layer stack of TaN/Ta/Cu is being used as the barrier, liner and interconnect layers in industry. 29 While this functions well at larger length scales, it is not practical for smaller sizes at the 14nm node and smaller, as indicated in Fig. 1. The three layers of material become increasingly difficult to grow in the high aspect ratio interconnect vias us-ing standard methods such as physical vapour deposition (PVD), which can cause pinch-off or blocking of the via. In addition, the volume which the barrier/liner stack can occupy, without the volume available for Cu deposition becoming too small, decreases with each technology node. [30][31][32] While there has been some success growing these films using atomic layer deposition (ALD) 33 and atomic layer etching (ALE) 34 , two layers of material take up too much of the available volume in the interconnect via, leaving no room to electroplate Cu. 32 In addition, Cu has a propensity to form 3D nanoclusters 35-37 rather than a thin film and Cu resistivity strongly increases at these length scales. 30,31 This means that an alternative material is needed in order to overcome this interconnect bottleneck and drive scaling below the 14nm node.
In our previous work, we studied the behaviour of Cu atoms and Cu dimers on Rupassivated and low percentage of Ru atom doped -TaN in detail, to examine fundamental aspects of Ru-TaN as a combined barrier+liner material and the fundamental interactions with Cu. 38 The (1 1 0 and adsorption of up to four metal atoms. Such a structure can be fabricated with an ALD process using Ta and Ru precursors and nitrogen plasma. 39,40 We believe that the results of this simulation work will be of value to experimentalists to grow materials with suitable Ru content to prepare a combined barrier-liner material that promotes Cu deposition.

II. METHODS AND COMPUTATIONAL SETUP
All calculations reported in this paper are based on spin polarized density functional theory as implemented in VASP v5.4. 41 The exchange-correlation contribution to the electronic interaction is approximated by Perdew-Burke-Ernzerhof (PBE) generalized gradient (GGA) exchange-correlation functional. 42 In this paper, the valence electronic configuration of Ta, N, Ru and Cu atoms used are 6s 2 5d 3 , 2s 2 2p 3 , 4d 7 5s 1 and 3d 10 4s 1 , respectively. The valence electrons are expanded in a plane wave basis set with an energy cutoff of 400 eV, while the core electrons are described by projector augmented wave (PAW) potentials. 43,44 The description of the bulk -TaN and the TaN  Cu binding energies reported in this paper are computed per atom from: Here, E int.sys. is the total energy of the relaxed interacting system where n Cu atoms are adsorbed on the respective TaN surface, E bare refers to the relaxed energy of the corresponding bare surface and E Cu−atom is the total energy of an isolated Cu atom. Therefore, E bind also includes the Cu-Cu interaction. nCu is the total number of Cu atoms in the adsorbed nanocluster. For certain doped surfaces, the adsorption of the Cu atoms distorted the surface atomic arrangements. Therefore we compute E surf which refers to the single point energy of the relaxed interacting system with the copper atoms removed. E surf would be almost similar to E bare if the surface distortion is minimal. For some surfaces we found the difference between these two terms, termed as surface rearrangement energy (E SR ), is large. Therefore, to make a fair comparison of the Cu binding energies computed at different surfaces, we removed this E SR contribution from E bind . Thus the above equation becomes We compute E bind * , which is the binding energy of the Cu n cluster relative to a gas-phase Cu n cluster.
Here, E int.sys. and E surf are the same as in Eq. 1, while E Cu−cluster refers to the single point energy of the Cu-cluster isolated from the TaN surface. Now, an estimate of the Cu-Cu interaction energy can be computed from The doping energy per dopant is computed as in which n is the number of Ru dopants in the surface layer. E Ta−atom and E Ru−atom are the energy of a single Ta and Ru atom in vacuum, respectively. E total is the total energy of the relaxed doped surface and E clean is the total energy of the clean TaN(1 1 0) surface.
Using this equation we can obtain an estimate of the relative energy required to create the doped system from pure TaN and compare how this is influenced by the concentration and distribution of Ru dopants.

III. RESULTS
In this section we discuss the different Ru passivated and Ru doped surfaces and the role of the distribution of the dopants in the outermost Ta layer. We then describe the adsorption energetics and structures of Cu nanoclusters with 2, 3 or 4 atoms on these Ru-modified TaN surfaces.

A. Ru Passivated and Ru Doped TaN Surfaces
The (Ru)TaN (1 1 0) surfaces used in this study are shown in Fig. 2  50%, 75% and 100% Ru content in the outermost Ta layer are shown in Fig. 2, while Fig. 3 shows the various dopant distributions studied with 50% doping, which will be discussed in detail in Section III B. For brevity, we will refer to these doped surfaces as Ru X with X=25, 50, 75, 100. The ten different Ru 50 surfaces in Fig. 3 are labelled in the order they appear  From the surface geometries given in Fig. 2 shows an undistorted high symmetry structure in Fig. 2 which is likely a local minimum. A distorted, lower symmetry structure of the Ru 100 surface can be obtained through adsorption of a single Cu atom, which upon relaxation strongly distorts this surface, as shown in Fig. 4.
The surface retained the distortions upon relaxation after the removal of the Cu atom and it is more stable than the high symmetry surface by -0.57 eV per Ru dopant. The smaller ionic radius of the Ru atom ((1.78Å) 46 as compared to the Ta atom (2.00Å) 46 ) is the primary origin of this distortion and we can clearly see recesses (in Fig. 4) caused by this distortion that will potentially trap Cu atoms, which in turn could act as nucleation points for the Cu film growth. However, whether this phenomenon will promote Cu wetting or Cu agglomeration is the subject of a further study that is beyond the scope of the present investigation. We do not discuss the geometries of other doped surfaces with an adsorbed Cu atom as no significant surface distortions, were observed as shown in Fig.1 of SI. However, a similar distortion is also found on the Ru 50−4 surface on adsorbing two or more Cu atoms which will be discussed in a later section. The surface distortion upon Cu adsorption was smaller on the Ru 25 and Ru 75 surfaces as the Ru atoms are well separated in the former and the relaxed bare surface itself is already distorted in the latter. For fair comparison between different arrangements the surface rearrangement energy, which is computed by subtracting the energy of relaxed bare surface from the energy of the resulting distorted system upon Cu adsorption (without Cu atoms), is removed from the Cu binding energy calculation as described in the methods section.

B. Effect of Dopant Distribution on Surface Stability
We chose 50% surface doping to study the effect of dopant distribution in the surface. A total of 10 surfaces were created, including both symmetric distributions, and less ordered, asymmetric structures, as shown in Fig  Whether Ru atoms are clustered or more distributed over the surface layer appears to Consequently, it is to be expected that Cu atoms will be more prone to bury into a recess on those surfaces where they are larger or more frequent. Examples of some of the recess structures formed are shown in Fig. 6.
In our previous work, 38 we found that there were two different doping sites on the surface, due to the two different types of Ta atoms in TaN. F-type sites are 3-fold coordinated with N and S-type sites are 6-fold coordinated with N. Ru doping at the F-site is more favourable than doping at the S-site. 38 We can see this effect in Fig the Cu-Cu bond axis is oriented in either a vertical (Cu c−v 2 ) or horizontal (Cu c−h 2 ) direction relative to the surface are chosen. The 'c' in the notation represent geometries where every Cu atom in the arrangement is bonded to at least one other Cu atom. In the third structure, the two Cu atoms are isolated from one another (Cu far 2 ), with ca. 12Å between them.
In the case of Cu 3 , we have considered 4 templates. In two of these templates the Cu  overall energy gain and the increasing Z-distance of the migrating Cu atom. It is interesting to see that the migrating Cu atom was initially bound to a surface Ru atom (Cu-Ru bond), which is a stronger bond as compared to the Cu-Ta bond on the clean surface. Therefore, the preference to a 2 layered 3D Cu cluster on the Ru 25 surface suggests that this surface, with no Ru-Ru surface bonds, will promote aggregation of Cu into 3D clusters rather than wetting, indicating that the Cu-Ru interaction at low Ru concentration is not strong enough to promote the stability of this arrangement.
The binding energies (E bind ) for all the arrangements on all the surfaces are plotted in Fig. 10 and are also listed in Tables II to V in the SI. The binding energies including the contribution from Cu-Cu interaction are plotted in Fig. 10(a) and 10(b), and those without the Cu-Cu interaction are plotted in Fig. 10(c) and 10(d). As expected, one could find  The average Cu binding energies (E avg bind and E avg bind * ) and the average Cu-Cu interaction energies (E avg Cu−Cu ) extracted from all 13 cluster adsorptions are listed in Table I. The strongest overall binding of Cu and Cu n (E avg bind and E avg bind * ) is observed on the Ru 100 surface and the weakest overall binding is found on the Ru 75 and the clean TaN surfaces. We noted in our previous publication 38 that the Ru-Cu interaction is stronger than the Ta-Cu interaction. Therefore, the Ru 100 surface, with all Ta atoms in the surface layer replaced with Ru, should enhance Cu binding compared to the clean TaN. While the average Cu-Cu binding energies were very similar on all the doped surfaces, the strongest E avg Cu−Cu was found for the Ru 100 surface closely followed by Ru 75 , Ru 50 and Ru 25 surfaces. This lets us conclude that the Cu-Cu interaction increases, albeit by a very small amount, with an increase in surface doping. Therefore, the Ru 100 surface offers the strongest Cu-Cu interaction as well as the strongest Cu-slab interaction. For all surfaces, including the clean surface, the Cu n -slab interaction (E avg bind * ) is much stronger compared to the Cu-Cu interaction (E avg Cu−Cu ). Consequently, we might speculate that all surfaces should promote wetting over association to an extent. But we know that the clean TaN surface promotes Cu association, so the above cannot be the only criteria to predict Cu association.
The surface rearrangement energy, E SR , due to the distortion caused by Cu adsorption is positive and small (absolute value < 1.5 eV per surface supercell) for all geometries of the clean, 1 ML Ru passivated, Ru 25 and Ru 75 surfaces, while it is negative for all geometries of the 2 ML Ru passivated and Ru 100 surfaces (individual energies shown in Table VI of SI and the maximum and minimum values, depending on Cu n binding mode, given in Table I).
A positive E SR indicates that the undistorted surface is more favourable than the distorted surface and vice-versa. A small |E SR | indicates that the adsorption of Cu leads to minimal distortion and vice-versa. The average rearrangement energy is less than 1 eV per supercell for the first set of surfaces where the distorted surface is less favourable, but they are much larger for the second set where the bare surface model is only a local minimum. Particularly, note that the E avg SR value for Ru 100 in Table I is -24.66 eV. This is the main reason to exclude this contribution in the E bind calculations. Ru 50 is not included in the above discussion because it showed negative E SR for some arrangements and positive values for the others (Table VI of

Influence of Cu-Cu and Cu-Slab Distances on the Stability of Cu n adsorption
In Fig. 10b and 10d Fig. 12a, with the results plotted in Fig. 12b. The distance is measured between the top most atom in the slab (which can be a on Ru 50 and Cu far 2 on Ru 100 ) and these are the same two arrangements marked in Fig. 10b and 10d. In these structures at least one of the Cu atoms is buried in the surface (indicated by the negative d(Cu-slab) in Fig. 12b) resulting in a higher coordination and increased binding. Even though there exists a data point below -1.0Å in the case of the Ru 100 surface (Cu c−v 2 arrangement), it still has a less favourable binding energy than the marked arrangements. The reason for lower d (Cu-slab) in the Cu c−v 2 arrangement when compared to the Cu far 2 is that in the former the distance is between the Cu and a surface N, while in the latter it is the distance between the Cu and a surface Ru atom. Since the N atoms are the top-most along the surface normal on the clean and doped surfaces (see Fig. 2), the distance between the surface Ru/Ta and Cu is a better indicator of whether the Cu atom is buried in the surface or not. Therefore, we computed the minimum distance along the Z direction between all the element pairs, namely d(Cu-Ta), d(Cu-Ru) and d(Cu-N) as shown in Fig. 13a, 13b and 13c, respectively.
As expected, the trend in these plots shows that the binding energy becomes more negative with a decrease in the distances. This figure also includes correlation plots comparing the d(Cu-X) (X = Ta, Ru, N) with the overall d(Cu-slab) given in Fig. 12b. From the correlation plots, we find that Ta is the farthest atom (along Z) from Cu in all the arrange- However, the minimum distances on the passivated surfaces plotted in Fig. 12b are between Cu and Ru atoms. Among the two passivated surfaces, shorter distances are reported for the 1 ML surface owing to the fact that the Ru layer on this surface is not closely packed as in the 2 ML surface and a trough is clearly visible in the 1ML structure in Fig. 2 that could trap the Cu atoms. On some Cu arrangements, the d(Cu-Ru) value is near zero on the 1 ML passivated surface. Furthermore, the average d(Cu-slab) are also listed in Table I  When analysing the Cu-Cu distances, we find that they are in a similar range on different surfaces. It is particularly clear, for example on the Ru 100 surface, and similarly on the 1 ML Ru passivated surface, that a longer Cu-Cu distance (sparse arrangements) and a shorter Cu-slab distance (Cu sinking into the surface recesses) lead to a more favourable binding energy. In contrast, only on the bare surface, a short Cu-slab distance and a short Cu-Cu distance are associated with favourable adsorption. The former surfaces are expected to promote wetting of the surfaces with the Cu, while the latter is expected to promote the formation of 3D structures.

D. Effect of Doping on the Cu Binding Energies
Increased surface doping with Ru does not always result in enhanced Cu binding as the Ru 75 surface did not produce more favourable binding energies compared to the Ru 50 surface (see Fig. 10 and Table I). To understand this observation better, the individual binding energy differences between the doped surfaces and the clean surface for E bind , E bind * and E Cu−Cu ) are plotted in Fig. 14

IV. DISCUSSION
The first objective of this paper is to compare and characterize models of the TaN(1 1 0 to Ta. This distortion causes the formation of recesses on the surface, with a large distance between atoms, which arises from the increased Ru-Ru attraction in the surface layer and the smaller Ru radius compared to Ta. Such a recess could act as a trap/anchor point for Cu seed layer formation. We also find such a recess in the 1 ML Ru passivated surface where the adsorbed Cu atoms could be trapped. However, the trapped Cu atoms will not diffuse through the film due to the excellent barrier properties of the underlying TaN. 38 Studying the effect of different dopant arrangements using the Ru 50 surface showed us that the surface stability is largely dependant on the ratio of Ru in S and F sites in each surface. We find that surfaces with a larger content of F-site dopants have more favourable doping energies. It is however possible that in a kinetically driven process such as ALD, surfaces with a high concentration of S-site Ru in the surface can be formed. Further, we found that all of the tested arrangements formed recesses in the surface layer due to the difference in atomic radii between Ru and Ta. These surface recesses are formed in areas on the surface where at least 3 Ru atoms were able to associate and thus cause a distortion of the regular (1 1 0) layer structure. As with the surface recesses formed on the surfaces with different doping percentages, further study with a larger number of Cu atoms will be necessary to understand how these recesses might act as a trap for Cu adatoms and how this will affect the growth of a Cu film. We also observed, that on doped and passivated surfaces, configurations with longer Cu-Cu distances and shorter Cu-slab distances (arising from partially buried Cu atoms) were associated with a more favourable binding energy, indicating that Cu atoms prefer to be separated on these surfaces. The same was not true on the clean surface, where favourable binding was associated with shorter Cu-Cu distances, showing a possible preference towards association of atoms.
If we consider the competition between Cu-slab interactions and Cu-Cu interactions as an indicator of whether Cu associates or wets on the surface, we could form two possible conditions. 1. Association is preferred over wetting if the Cu-Cu interaction energy is more negative and thus more favourable than the Cu-slab binding energy, (E avg Cu−Cu < E avg bind * ) and if E avg Cu−Cu reaches the cohesive energy of bulk copper (-3.49 eV/atom 47 ).
2. For wetting to be preferred over association, the Cu-slab binding energy must be more negative than the Cu-Cu interaction energy (E avg bind * < E avg Cu−Cu ) and E avg bind < -3.49 eV/atom.
On all surfaces, including the clean surface where we know that Cu associates, we found E avg Cu−Cu > E avg bind * and the value of E avg Cu−Cu was much more positive than the bulk cohesive energy of Cu ( binding and the lowest Cu-slab distance. Based on the above observation, these two surfaces are strong possibilities for a combined barrier+liner material for copper interconnects which can be produced by atomic layer deposition. The advantage of using either of these surfaces as the combined barrier and liner material, is that it significantly decreases the volume of the interconnect via that is occupied by the barrier+liner material, see Fig. 1, compared to the multi-layer stack that is currently in use. 29 We found that the Cu-slab interaction is stronger when the Cu atoms are separated from the cluster, as in the 'far' arrangement, where the Cu-Cu interactions are weaker. The opposite is true for the 'close' arrangements. The Cu-slab interaction strengthens even more when the Cu atoms are buried into the surface recesses, as is evident from the Cu-slab distances vs binding energy plots. To understand the importance of the trapped atom in the Cu seed layer formation, wetting/association of Cu, a Cu cluster/film larger than used in this paper will be considered and this is the focus of ongoing work.

V. CONCLUSION
A first principles investigation of the stability and reactivity of Ru passivated and Rudoped surfaces of -TaN(1 1 0) with respect to adsorption of 2-4 Cu clusters was conducted.
The strongest Cu-surface and Cu-Cu interactions were observed on the 100% doped surface.
On the other hand, the 1 ML Ru passivated surface offered weaker Cu-Cu interaction than the clean surface and exhibited the second most favourable Cu-surface interaction. An increased Cu-slab interaction was possible by trapping Cu atoms in the recess formed on the higher % of Ru doped and 1 ML of Ru passivated surfaces. A condition to favour wetting of Cu atoms can be achieved when the Cu-slab interaction is stronger than the cohesive energy of bulk Cu. The above condition was satisfied for the 100% Ru doped surface and the 1 ML Ru passivated surface, which makes them top contenders for the combined barrier+liner material. Going forward, our next step is to study the stability of large 2D Cu films on these Ru passivated and Ru-doped surfaces.