Metal diffusion properties of ultra-thin high-k Sc2O3 films

The diffusion barrier properties of Sc2O3 against metal diffusion were studied. Tin and ruthenium were used as probe materials to study the barrier properties of Sc2O3 in thickness ranges that are of relevance for gate materials. Tin deposition and hydrogen radical etching from Sc2O3 layers of 0.5-1.5 nm thickness, deposited on Ru, show that these Sc2O3 layers effectively block the diffusion of Sn into Ru. We show that Sn adhesion and etching depends strongly on the thickness of the Sc2O3 film. The etch-rate is found to be inversely proportional to the Sc2O3 layer thickness, which we attribute to Sc2O3 becoming a more effective charge transfer barrier at larger thicknesses.

possible candidates for high-k dielectric applications, due to their possession of a suitably large optical band gap (5-6 eV).
Moreover, Sc2O3 has a high dielectric constant (ɛ= 13) in comparison to SiO2 (ɛ = 4.5). 3 Scandium oxide, doped with lanthanum, has been investigated as a high-k gate material for silicon-based integrated circuits. 4 However, much of the interest in Sc2O3 has been in applications in AlGaN/GaN devices. Devices, based on AlGaN/GaN without passivation, show significant gate lag effects due to the presence of surface states in the region between the gate and drain contact.
Moreover due to large polarization induced field and large conduction band offset, high current density can be achieved with AlGaN/GaN heterostructures. Scandium oxide has attractive band gap and thermal lattice properties for use on GaN. 5 A Sc2O3 layer was shown to effectively mitigate the collapse in drain current through passivation of the surface traps. 5,6 Additionally, it was demonstrated that Sc2O3 can be used simultaneously as a gate oxide and as a surface passivation layer for AlGaN/GaN high electron mobility transistors. 7 However, the gate oxide material must satisfy several criteria: in addition to the electronic function of a gate dielectric, the gate oxide material must maintain a high dielectric constant, and serve as a diffusion barrier against diffusion of material from the top electrode. Furthermore, the gate dielectric must be as thin as possible, so a high resistance to diffusion is critical.
Thus, in addition to the electronic properties of the gate, it is also of importance to understand the Sc2O3 diffusion-barrier behavior. The thermal stability of the high-k dielectric material is also of importance, thus, a dielectric material should show good thermal stability, at least at the processing temperatures of the device, and have a low coefficient of thermal expansion. 4,8,9 In order to carry out diffusion studies, we use tin (Sn), a highly mobile metallic probe atom, to test if Sc2O3 layers of various thicknesses act as a diffusion barrier. Although Sn is not relevant for high-k dielectric applications we used it due to the fact that Sn has the advantage of forming a volatile hydride, which allows it to be removed from surfaces with hydrogen reactive species. 10 Tin intermixes and/or binds strongly with highly electronegative materials, such as ruthenium (Ru), gold (Au), and silver (Ag). Furthermore, it has been shown that Sn, after it diffuses into the aforementioned materials cannot be etched by atomic hydrogen [11][12][13] , while Sn deposited onto Sc2O3 can be completely removed with atomic hydrogen. 14 In this work, we study Sc2O3 barriers deposited on top of highly electronegative Ru surfaces, with in situ ellipsometry. The deposition and etching results are understood in the context of a metal-insulator-metal tunnel device.

EXPERIMENTAL
Ruthenium and scandium were deposited by direct current (DC) magnetron sputtering from targets with 99.95 % and 99.5 % purity, respectively. First, 4 nm of Ru was deposited onto a silicon wafer. On top of Ru, thin layers of scandium (Sc) with thicknesses of 0.5, 1, and 1.5 nm were deposited (see Table 1).
An additional sample with 4 nm of Ru, deposited on a silicon wafer, was used as a reference for analysis with low-energy ion scattering spectroscopy (LEIS) measurements (see below).

AS DEPOSITED SAMPLE ANALYSIS-LEIS
To ensure that the Sc2O3 layer forms a closed film, several samples were analyzed with LEIS. To remove surface contaminants, the samples were sputtered with a dose of 4·10 15 3 keV He + ions/cm 2 before analysis. The LEIS spectra after sputtering are presented in Figure 1. to the tail signal at lower energy relative to the surface peak.
The tail signal thus provides a depth profile of Ru in the sample. 16 For comparison a typical LEIS spectrum for Ru is also presented in Figure 1  respectively, corresponding to a minor swelling of the nominal thicknesses deposited, due to oxidation.

Figure 2. Delta value for 513 nm during Sn deposition and Sn etching. Deposition starts at 5 min and finishes at 25 min. H˙ etching starts at 25 min and proceeds in 5 minutes cycles with the filament ON (solid line) and OFF (dotted line).
It can be seen that the total etching time increases with increasing Sc2O3 thickness. The Sn etching time as a function of Sc2O3 thickness is presented in Figure 3.

Figure 3. Sn etching time (time when filament was OFF is excluded) as a function of Sc2O3 thickness on top of Ru. The point with 3 nm Sc2O3 is taken from ref [14]. For this point, the Sc is deposited on a silicon wafer.
The etch-rate is inversely proportional to the Sc2O3 layer thickness. Moreover, with increasing Sc2O3 thickness, the etching time converges to the value found for bulk Sc2O3, which is presented in Figure 3 as the point at 3 nm Sc2O3 (obtained from ref. [14]). It should be noted that, for this point, the Sc was deposited on a silicon wafer and full oxidation of the layer was achieved. 14 For etching to be successful, tin hydride must be formed, and charge transfer from the Ru layer to the Sn layer must occur. Hot electrons are generated when atomic hydrogen reacts with Sn, and these may tunnel through the oxide barrier, or get trapped in defect states in the oxide. The tunneling rate depends on the thickness of the Sc2O3 layer, and the potential difference between the work functions. Thus, the contribution of tunnelling to the etch rate may be estimated (see below).

BINDING ENERGY OF Sn TO Sc2O3
The binding between Sn and Sc2O3 can be understood in terms of charge transfer between the Ru and Sn layers. In this model, the Ru/Sc2O3/Sn structure acts as a metal-insulator-metal tunnel device (MIM) and we can compare the Sn etch rate from the tested samples. MIMs consist of a metal back electrode, an insulating oxide layer and metal top layer. The thin insulator layer acts as high-pass filter for carrier transport. [18][19][20] Charge transport is limited by the band gap of the insulator, but the small thickness of the oxide layer, and the presence of local defect states in the band gap allow (hot) electrons and holes to tunnel through the oxide barrier. 20 Tunneling from a metal through a thin insulator to another metal has been extensively studied for different material combinations. [18][19][20][21][22] Noting that the Fermi level remains constant across the interfaces, schematic band diagrams for the different thicknesses of Sc2O3 can be constructed (see Figure 4 (a), Figure 4 (b) and Figure 4 (c)). The work function values for Ru and Sn were taken to be 4.71 eV and 4.35 eV, respectively 23 , while the electron affinity and band gap for Sc2O3 are reported to be 0.9 eV 24 and 6.3 eV. 25 The values of the work function, electron affinities and band gap are indicated on a relative scale on the diagrams below. and where we have assumed that > and 2 indicates that the momentum should be evaluated at the Ru/Sc2O3 interface.
Inside the Sc2O3, the electron's momentum becomes complex, leading to an exponentially decaying wave function. As before, the wave function and its derivative should be continuous at the Sc2O3/Sn barrier, giving: where 2 is the reflection coefficient from the Sc2O3/Sn interface, and 2 is the electron momentum, evaluated at the Sc2O3/Sn interface. The bar indicates that the average momentum in the Sc2O3 layer should be used. Furthermore, we have also assumed that the potential difference between the Sn layer and the Ru layer is the same as the work function difference between the two materials. From Equations (2) and (4) the amplitude of the electron wave function in the Sn layer can be estimated as: ̅̅̅̅ t ̅̅̅̅̅ (k 2r +ik 1 )(k 2s +ik 1 ) while the transmission probability can be estimated as In deriving equations (1)-(5), we have assumed that 2 is negligible. We confirm this by combining equations (2) and (3) to calculate the probability density of the 2 term, due to reflection from the Sc2O3/Sn interface. For a layer thickness of 0.7 nm, 2 is two orders of magnitude less than 2 , indicating that tunneling due to reflected electrons is, indeed, negligible.
To estimate if tunneling is significant, the transmission probabilities in Figure 5    Reversing the calculation above, we estimate in Figure

CONCLUSIONS
Scandium oxide diffusion barrier properties were tested using