Electrical Properties of Ultrathin Platinum Films by Plasma-Enhanced Atomic Layer Deposition.

The ability to deposit thin and conformal films has become of great importance because of downscaling of devices. However, because of nucleation difficulty, depositing an electrically stable and thin conformal platinum film on an oxide nucleation layer has proven challenging. By using plasma-enhanced atomic layer deposition (PEALD) and TiO2 as a nucleation layer, we achieved electrically continuous PEALD platinum films down to a thickness of 3.7 nm. Results show that for films as thin as 5.7 nm, the Mayadas-Shatzkes (MS) model for electrical conductivity and the Tellier-Tosser model for temperature coefficient of resistance hold. Although the experimental values start to deviate from the MS model below 5.7 nm because of incomplete Pt coverage, the films still show root mean square electrical stability better than 50 ppm over time, indicating that these films are not only electrically continuous but also sufficiently reliable for use in many practical applications.

Introduction

Modern day electronic devices are moving toward 3D architectures with high aspect ratios in order to push the lower limit of device sizes.[1−4] This downscaling of devices necessitates not only thin films but also conformal deposition on complex geometries. Platinum has attracted considerable attention for thin-film applications in micro- and nano-electronic devices because of its chemical inertness, high work function, and excellent electrical properties, including a linear temperature coefficient of resistance (TCR) over a wide temperature range.[5−9] However, the use of Pt in devices has been somewhat limited until recently, both because of cost as well as limitations caused by the directionality of conventional Pt deposition techniques such as physical vapor deposition.[10] Therefore, solving the problem of how to deposit conformal, ultrathin, and electrically stable Pt films is essential for future devices.

The development of thermal atomic layer deposition (t-ALD) allows precise control of thickness and high conformality on complex 3D architectures.[11−15] However, even with t-ALD there are limits to the attainable film thickness because material-dependent nucleation difficulties prevent continuous films below a certain number of t-ALD cycles.[16−19] Pt is one of the materials known to present nucleation challenges with t-ALD.[20] The high surface energy of Pt leads to Volmer–Weber (VW) growth, forming islands at the early stages of nucleation.[21,22] This growth behavior sets a lower limit on the minimum closed-film thickness, as these Pt islands must grow large enough to connect to neighboring islands before the film becomes continuous.[23] Thus, one way of depositing the thinnest possible continuous Pt films is by maximizing the number of nucleation sites, such that adjacent islands are closer and can merge at a lower cycle count.

Previous research has explored two very different approaches to increase the number of nucleation sites. The first approach explores the use of different nucleation layers such as W,[22] Al2O3,[19,24,25] TiO2,[26] or STO[27] with t-ALD. The thinnest resulting Pt is 1.5 nm on tungsten (W). However, as W itself is conductive, the resulting stack is not feasible for use in ultrathin electronic layers because it would degrade the desirable electrical properties of platinum. The other approach uses plasma-enhanced ALD (PEALD), a modified version of t-ALD that replaces the co-reactant with plasma. Plasma mitigates some of the nonideal nucleation behavior because of its higher reactivity.[12,28,29]

In this paper, recognizing the individual benefits of PEALD and different nucleation layers, we combined the two approaches: using PEALD in combination with Al2O3 and TiO2 as a nucleation layer. Our first finding is that metal oxides that are capable of reduction can serve as good nucleation layers to deposit thinner Pt films. The second finding is that PEALD Pt films’ surfaces are partly specular and subject to low grain boundary scattering, which leads to higher conductivity based on fitting the Mayadas–Shatzkes (MS) model.[30] Pt films down to 5.7 nm fit the same MS curve well. The Pt film’s TCR values were then further characterized with the Tellier–Tosser model.[31] Whereas films below 5.7 nm deviated from the MS curve because of incomplete coverage, we successfully deposited electrically continuous Pt films as thin as 3.7 nm with an electrical conductivity of 3.78 × 106 Ω–1 m–1 and TCR of 0.84 × 10–3 1/°C utilizing TiO2 as a nucleation layer. These films had a root mean square (RMS) electrical stability better than 50 ppm, proving their viability for device applications.

Experimental Section

Sample Preparation

Prior to depositing oxide nucleation layers, p-doped Si(100) wafers, with resistivity of 0.1–0.9 Ωm, were RCA cleaned and thermal SiO2 layers of ∼180 nm (wet oxidation, 900 °C, 1 h 34 min) were grown to serve as an electrical insulation layer between the Si substrate and Pt. On each wafer, ∼11 nm of one of the two nucleation layers of interest, Al2O3 and TiO2, were deposited on top of the thermal oxide using PEALD. Pt was deposited on samples with both nucleation layers simultaneously, with the number of PEALD Pt cycles varying from 10 to 250 to achieve different thicknesses. All ALD films in this paper were deposited in an ALD reactor (Fiji, Cambridge Nanotech/Veeco) equipped with a remote inductively coupled plasma. The precursors used for the metal oxide deposition were trimethylaluminum and tetrakis(dimethylamido)titanium(IV), for Al2O3 and TiO2, respectively, with O2 plasma as the co-reactant. For Pt depositions (methylcyclopentadienyl)trimethylplatinum (MeCpPtMe3) was used with O2 plasma as a reactant. Before every film deposition, samples went through an in-situ O2 plasma clean at 300 W (for detailed conditions refer to the Supporting Information, Figure S1). To fabricate electrical test structures, Pt beams (4 μm × 996 μm) were patterned by ion milling (Nanoquest Research Ion Beam Milling System, Intlvac), with 50 nm thick Pt bond pads evaporated (Kurt J. Lesker Co.) and patterned by lift-off (Figure a).

Figure 1

(a) Scanning electron microscopy image of electrical measurement test structure; (b) TEM image confirming the thickness of the thinnest continuous Pt film measured with XRR (3.7 nm).

Material Characterization

To study the nucleation phase, transmission electron microscopy (TEM) grids (PELCO Silicon Dioxide Support, 18 nm, Ted Pella, Inc.) attached to a Si wafer with carbon paste (Electrodag 502, Ted Pella, Inc.) were loaded prior to PEALD of oxides and Pt. bright and dark field micrographs were taken using a transmission electron microscope (FEI Tecnai G2 F20 X-TWIN FEG, 200 kV). Pt coverage, defined as Pt coverage (%) = , was calculated using bright field micrographs post-processed with a low-pass Wiener filter (5-pixel averaging) and thresholding. Additionally, cross sections of the samples were prepared following the conventional TEM sample preparation procedure described by Strecker et al.[32] (Figure b). Metal oxide thicknesses were measured with Spectroscopic Ellipsometry (M 2000, J.A.Woollam Co.), and Pt thicknesses were evaluated with X-ray reflectivity (XRR, X’Pert Pro, PANalytical) (see the Supporting Information, Figure S2). Platinum roughness was measured with noncontact mode atomic force microscopy (AFM, XE-100, Park Systems). X-ray photoelectron spectroscopy (XPS, PHI 5000 VersaProbe III, Ulvac-PHI, Inc.) was used to confirm the material composition. Crystallinity of Pt films was confirmed with high-resolution TEM, selected area diffraction patterns, and grazing incident X-ray diffraction (X’Pert Pro, PANalytical) patterns (see the Supporting Information, Figure S3). Surface hydroxyl concentration was assessed using the water contact method on a contact angle goniometer (model 290, ramé-hart).

Electrical Measurements

Electrical conductivity and wafer scale resistivity of as-deposited Pt were measured using a resistivity mapping system (OmniMap RS35c, KLA/Tencor). The TCR was measured from 30 to 100 °C using a semiconductor device analyzer (B100A, Agilent Technologies) in a probe station (CM170-6, Signatone) installed with a high-temperature chuck (S-1060H-4TP, QuieTemp). RMS film stability measurements were made in a temperature-controlled chamber using a four-point probe (Keithley 6221/2182A, Keithley) and delta current reversal techniques to minimize voltage errors from the leads and thermal EMFs. Electrical stability was determined from resistance measurements collected continuously for 30 min, and calculated as RRMS (ppm) = , where each r is an individual resistance measurement, n is the number of measurements collected, and is the mean resistance.

Theoretical Models

Once the grain size and thickness of a metal film drop below the bulk metal’s electron mean free path (MFP), electron scattering at the surface and grain boundaries causes changes in the film’s electrical properties. This effect is known as the size effect. The MS model, an extension of the Boltzmann equation that accounts for surface and grain boundary scattering, predicts reductions in the electrical conductivity of thin metals by treating grain boundaries as a periodic potential barrier.[30] The MS model gives the film conductivity σf aswhere σo is the bulk Pt conductivity, l0 is the electron MFP for bulk Pt (l0 = 22.4 nm),[33]lf is the film thickness, d is the average grain size, R is the grain boundary reflection coefficient, and p is the specular reflection coefficient. The average in-plane grain size was approximated based on dark-field TEM images (see Supporting Information, Figure S4). The constants p and R are fit to the experimental data. The film TCR βf is given by the Tellier–Tosser model[31] aswhere and β0 is the TCR of bulk Pt.

Results and Discussion

Nucleation Behavior of Platinum on Different Nucleation Layers

By investigating the nucleation behavior of Pt on two different nucleation layers, Al2O3 and TiO2, we found that the Pt film’s morphology differs between the two layers throughout the nucleation process (Figures and 3). Whereas Pt exhibits VW growth behavior on both nucleation layers as reported in the literature[28,34] (and confirmed by TEM, Figure S5), the number of PEALD cycles required to reach crystallite coalescence differs significantly between Al2O3 and TiO2. RMS roughness will be largest around the time individual islands coalesce,[35] as this is when the film height variation is at a maximum as each individual crystallite is tall when the gaps between crystallites remain at nearly the level of the nucleation layer. Using this approximation, we can conclude that the Pt films are nearly at the coalescing point at around 50 and 150 Pt cycles on TiO2 and Al2O3, respectively (Figure ).

Figure 2

Planar view TEM image showing morphology evolution of PEALD Pt on (a) Al2O3 and (b) TiO2 from 10 to 250 cycles. Numbers in the upper left denote the number of cycles. For electrically continuous films, Pt thickness is given in the upper right corner.

Figure 3

AFM images showing morphology evolution of PEALD Pt on (a) Al2O3 and (b) TiO2 from 10 to 250 cycles. Numbers in the upper left denote the number of Pt cycles. For electrically continuous films, Pt thickness is given in the upper right corner.

Figure 4

RMS roughness versus number of Pt PEALD cycles on Al2O3 and TiO2. These values were collected from Figure .

The difference in the coalescing point is a result of more Pt islands nucleating on TiO2 as compared to Al2O3, which suggests that more nucleation sites were available on TiO2. To quantitatively confirm this, the number of nucleation sites was estimated by the percentage of Pt coverage (Figure ) on TEM images, as this is proportional to the size and number of crystalline islands and therefore to the number of reaction sites on the nucleation layer. Whereas there are Pt particles on both nucleation layers after 10 cycles, TiO2 has ∼3 times more Pt coverage than Al2O3. At 25 cycles of Pt, the difference is even more pronounced with TiO2 having ∼50 times more Pt coverage than Al2O3. At this cycle number, the Pt particles on TiO2 are beginning to coalesce and form the worm-like structure characteristic of the initial formation of a continuous film. In contrast, the Pt particles remain isolated on the Al2O3 nucleation layer. This trend continues with increasing numbers of PEALD Pt cycles; at 100 cycles the Pt has reached full coverage on TiO2, whereas on Al2O3 it is still coalescing. This shows how, even though both films start with VW island growth, TiO2 enables a continuous film at such low thicknesses: the higher density of nucleating Pt particles on TiO2 results in particles meeting and coalescing into a continuous film at a lower cycle count.

Figure 5

Pt coverage versus number of PEALD Pt cycles for both nucleation layers. Three 170 nm × 85 nm TEM images were processed to calculate the coverage for each data point.

It is interesting to note that the coverage points can be fit to cycle count to yield an estimate of the amount of the exposed nucleation layer covered by Pt per cycle. If we assume 10 and 25 cycles of nucleation delay for TiO2 and Al2O3, respectively, then with every cycle 5% of exposed TiO2 and 2.8% of exposed Al2O3 are covered with Pt. This means the rate of nucleation is approximately 1.75 times faster on TiO2. This trend in the rate of nucleation is consistent within a major portion of the PEALD temperature window of 200–270 °C.

The significantly different nucleation behavior observed on the two nucleation layers can be explained from the unique half-reactions of the Pt precursor. Previous research on Pt ALD growth atop a continuous Pt surface showed that during the first half-reaction, when MeCpPtMe3 is pulsed into the chamber, the precursor reacts with exposed oxygen atoms on the surface and undergoes dehydrogenation. Oxygen, introduced for the second half-reaction, serves two purposes. First, it removes the carbonaceous layer left from the first half-reaction. In addition, some of the oxygen is chemisorbed by the Pt surface, thereby creating reaction sites for the next cycle. The oxygen pulse can be either O2 gas, in the case of t-ALD, or O2 plasma, for PEALD. Growth per cycle is higher for PEALD because the plasma more readily oxidizes the surface Pt.[28] This importance of surface oxygen atoms for Pt ALD growth on continuous films can also explain growth at the nucleation stage.

Because hydroxyl bonds serve as nucleation sites for Pt,[36] Pt nucleation is higher when more hydroxyl bonds are available on the nucleation surface.[26] The relative concentration of surface hydroxyl bonds on TiO2 and Al2O3 was assessed by measuring the water contact angle (see the Supporting Information, Figure S6). Both measured contact angles are below the resolution limit of 4.5°. This minimal difference in angle between two films is not enough to explain the significant difference in nucleation rate.

We speculate that the nucleation difference can be further explained by the fact that TiO2 is more easily reduced than Al2O3.[37] This allows TiO2 to act as an oxygen reservoir, creating many more nucleation sites for MeCpPtMe3 than Al2O3.

Electrical Conductivity, TCR, and Stability of Platinum Films

The electrical properties of these films are subject to the size effect as the film thicknesses are significantly smaller than the electron MFP of Pt. As a result, the electrical conductivity is reduced by as much as 83% for the 4 nm films as compared to bulk values, as shown in Figure .

Figure 6

Experimental value and theoretical curves (with p = 0.5) for electrical conductivity as a function of Pt thickness. Films with complete Pt coverage fit the R = 0.1 curve well, whereas electrically continuous films with incomplete coverage fit along the R = 0.25 and R = 0.44 curves. Each conductivity value is an average from 49 points measured across the 4″ wafer.

Because we are using one equation, the MS model, to fit two unknowns, p and R, additional insight is needed to ensure the fitted values make physical sense. We measured the RMS roughness of the Pt surface (Figure ) to be between 0.4 and 0.7 nm. This low level of roughness indicates that the Pt surface will not be completely diffusive, meaning that the platinum surface is partly specular. We expect R to be between 0.1 and 0.9, which is typical for polycrystalline materials.[7,24,33,38−40] Given these bounds, we found p = 0.5 and R = 0.1 best fit the MS model. These values correspond to low-surface and grain-boundary scattering, suggesting that our films are of high quality. To verify that there are no residual impurities from the precursor in our Pt films we employed XPS (see the Supporting Information, Figure S7). Hence, it is safe to assume electron scattering from carbon impurity is negligible.

The electrical conductivity of all full-coverage films falls on the same R = 0.1 curve (cf. Figure ), suggesting that all films with complete coverage, regardless of the nucleation layer, have a comparable number of defects and differ only in their thicknesses. Thinner films, on both nucleation layers, do not fall on this same curve. As Pt surface roughness is similar for all contiguous films, this deviation on thinner films is a result of gaps between adjacent Pt islands. The gaps can be regarded as “grain boundaries,” but with a higher potential barrier than that found between adjoining grains. Thus, by increasing R to account for this higher barrier, the electrical conductivities can be well explained.

Using the same R values with the Tellier–Tosser model results in a deviation between the theoretical and experimental TCR of up to 24% (Figure ). Increasing R to 0.21 substantially improves the quality of the fit. This discrepancy has been shown previously, and is likely a result of model simplifications which break down for ultrathin films as the dominant mechanism changes.[33]

Figure 7

Experimental values and theoretical curves (with p = 0.5) for TCR vs Pt thickness. Measurements from three electrical structures were averaged for each point.

The electrical stability of the films is crucial for any real device or sensing application.[41−44] To confirm the feasibility of these films for such applications, the RMS voltage stability of each film was measured (see the Supporting Information, Figure S8). All films were found to have an RMS film stability within 50 ppm. These results are comparable to those reported previously for 50 nm PEALD Pt films.[45] Hence, the use of these films in devices is very promising.

Conclusions

In this work, we have shown that electrically continuous Pt films thinner than 4 nm can be deposited using PEALD with TiO2 as a nucleation layer. We compared the nucleation behavior of Pt on two different nucleation layers, Al2O3 and TiO2. Our results suggest that reducible metal oxides promote Pt nucleation by acting as an oxygen reservoir, thereby allowing the formation of thinner, electrically continuous films. The electrical conductivity and TCR of these ultrathin Pt films were reduced by up to 83 and 79% of the bulk values, respectively. The experimental electrical values were compared to the MS and Tellier–Tosser models. Both models matched the experimental data well for the films thicker than 6 nm, indicating that PEALD Pt film properties are influenced primarily by the size effect. Additionally, electrical measurements showed good stability for all deposited films, including those which were electrically continuous but with incomplete Pt coverage. These results indicate that Pt films as thin as 4 nm can reliably be used in many device applications.