Picosecond Ultrasonics Provides CoSi Characterization
Guray Tas, Chris Morath, Robert Stoner, Christian Lavoie, Cyril Cabral Jr., James Harper, Yaw-Lin Huang, Donald Huang and Ren Chen -- Semiconductor International, 2/1/2003
|
Silicides are used to form self-aligned source, drain and gate contact regions in MOS devices. Low-resistivity contact silicides include titanium disilicide (TiSi2), cobalt disilicide (CoSi2), and nickel silicide (NiSi). Under typical process conditions, these films exhibit resistivities in the 15-20 µV-cm range.
TiSi2 has been the silicide of choice because of its thermal stability and ability to reduce oxides that form on the silicon surface. However, as design rules shrink below 0.25 µm, it is difficult to form the low-resistance phase of TiSi2 (C54) through thermal annealing alone. It is essential that the transition from the high-resistivity phase (C49) to C54 is complete, because the C49 phase has more than 3× C54's resistivity. In the small lateral dimensions specified by <0.25 µm designs, the transition temperature from the TiSi2 C49 to C54 phase increases.1 Meanwhile, junction depth scaling demands thinner silicide and thinner TiSi2 agglomerates at a lower temperature.2 These conflicting factors result in a very narrow, usable process window. Unlike TiSi2, the formation of the lower-resistivity CoSi2 or NiSi phase is substantially independent of the contact size.3 CoSi2 is projected to be the silicide of choice at the 0.13 and possibly the 0.09 µm technology node.4 In <0.10 µm design rules, NiSi has advantages — it consumes less silicon than CoSi2 and the low-resistivity phase forms at a significantly lower temperature. However, NiSi use is limited because any subsequent processing steps that increase the silicide temperature more than ~600°C can result in formation of higher-resistivity NiSi2.
CoSi2 films require careful processing to obtain a smooth silicide/silicon interface. Rough interfaces can produce high leakage currents through the junction region, affecting performance or resulting in the device's premature breakdown. To prevent this, many process approaches attempt to eliminate oxide layers that would be reduced by titanium. Silicon surfaces generally have a thin native oxide layer. In addition, high oxygen concentrations in cobalt films may lead to the unintentional oxidation of the silicon beneath the metal before or during rapid thermal processing (RTP). Regardless of their origin, oxide layers can impede silicide formation and impact the final silicide/silicon interface. However, cleaning the silicon surface by argon ion sputtering5 or hydrogen passivation to remove the native oxide prior to cobalt deposition can cause faceting of the silicide along the silicon planes, resulting in a rough interface. Similarly, inadequate cleaning can result in a rough silicide adjacent to the silicon/dielectric boundary. Therefore, a thin layer of native oxide may actually prove beneficial in promoting a smooth interface.6
Other process approaches used to promote a smooth CoSi2-Si interface take advantage of titanium's ability to reduce native oxides. Examples include using a thin layer of titanium to scavenge oxygen from the silicon surface,5 depositing a Co-Ti alloy,7 and depositing a thin sacrificial capping layer of titanium or TiN after the cobalt.8 In the latter, the titanium or TiN layers are chemically etched away before the final CoSi2 anneal. Even a brief literature search on cobalt silicide processing reveals a variety of approaches being used to optimize the silicide/silicon interface.
We have carried out studies of cobalt silicide phase formation using sheet resistance, X-ray diffraction, and picosecond ultrasonic measurements. Picosecond ultrasonic technology has been used to successfully monitor TiSi phases9 and provides CoSi phase information that is well correlated with these other metrology techniques.10 It also provides information on the CoSi2 film thickness and the roughness of the cobalt silicide/silicon interface. These capabilities are further demonstrated by comparing the properties of CoSi2 films prepared by two different processes, which may provide a manufacturing solution for creating low-resistivity CoSi2 films with smooth interfaces.8
CoSi2 phase formationTemperature is among the important variables determining the phase of a silicide formed by annealing. As temperature increases, the thin metallic cobalt layer begins to interact with the silicon substrate forming Co2 Si first, then CoSi, and finally the low-resistivity CoSi2. If the temperature continues to increase, the film may experience thermal degradation. To study the phase formation of cobalt silicides, a set of wafers was prepared.10 The single-crystal silicon wafers were cleaned with a HF/H2 O solution to remove surface native oxide. Immediately after, cobalt films were deposited onto the substrates using a magnetron sputtering system. Film thickness was measured by Rutherford backscattering (RBS) to be 85 Å (assuming a bulk density for the cobalt film). A thin oxide film assumed to be Co3O4, consistent with oxidation of the top five monolayers, was also detected. The films were then annealed in a vacuum chamber at a ramp rate of 3°C sec-1 to temperatures of 200-950°C.
ResultsFigure 1. These were obtained with a standard four-point probe (4PP) method. At low temperatures, the metal does not diffuse into the silicon but remains as metallic cobalt with a sheet resistance of ~50 O/sq (resistivity of 36 µO-cm for the 85 Å film). As the annealing temperature increases to ~475°C, Co2 Si is briefly formed, causing a slight increase in resistance. As the temperature continues to rise, the Co2 Si converts to the monosilicide, CoSi, which has a sheet resistance of ~150 O/sq (resistivity of 250 µO-cm). Finally, at ~600°C, the desired disilicide, CoSi2, is formed, and the sheet resistance falls to the desired value of 10 O/sq (resistivity of 26 µO-cm — the bulk value would be lower, corresponding to the 15-20 µO-cm bulk resistivity for a CoSi2 film).
 |
| 2. X-ray diffraction intensity as a function of 2u for an as-deposited cobalt sample, and samples annealed at 200-950°C. |
Starting with the as-deposited film up to an annealing temperature of 475°C, the diffraction peak of 52.7° is indexed to the (002) orientation of the hexagonal cobalt phase. Upon reaching 485°C, the metal-rich Co2 Si phase forms, producing a peak at 54.3°. At 505-570°C the film is in the CoSi phase, as identified by the low-intensity CoSi (110), (111), and (210) peaks at 33.4, 40.9 and 54.1°, respectively. By 620°C, most of the film is in the CoSi2 phase as shown by two strong disilicide peaks at 33.7° (111) and 56.6° (220).
Two features remain unclear. The first is the low-intensity broad peak at ~43° observed for anneal temperatures up to 330°C. One possible explanation for this is that Co-Si interface layers of unknown composition and thickness were created by energetic ions during cobalt sputtering. Another is that a thin surface layer of oxide is present. The second unexplained feature is a small peak below 35° present only at 485°C, which is the only temperature where Co2 Si is measured. This may be a result of a structure variation in this very thin Co2 Si film from a bulk film, or a surface reaction with an oxide present.
10 are shown in Figure 3. As with the X-ray diffraction results, clear qualitative changes can be seen in the reflectivity signals when the silicide formations occur. The desired final phase, CoSi2, is characterized by a sharp peak near t=0. Films annealed at 380 and 475°C also have distinctive features near t=0. The temperature range in which these two anneals fall is bordered at the low temperature by the abnormal grain growth in the cobalt film and at the high temperature by the formation of the metal-rich Co2Si. Further analysis reveals other subtle attributes that could be used for phase identification. For example, the shape of the first peak in the data is significantly different for cobalt and CoSi.
The reflectivity signal contains an acoustic response, measured as an oscillatory component with a period of a few picoseconds. The time-varying elastic strain in the film results in a modulation of the optical parameters of the cobalt film, giving rise to a change in the reflectivity of the structure. For the films in Figure 3, there is only one readily apparent oscillation frequency. The signal's acoustic part is dominated by the lowest thickness mode. Therefore, to analyze the data quantitatively, the measured signal is fitted to a single frequency function:
A cos (2pf t + F) exp(-Gt) + F background (t) (1)
where f is the frequency, and G is the damping rate. A, f, F and G are constant for a given fit and Fbackground (t) is a smoothly varying function designed to fit the background. A standard non-linear least squares fitting technique was used, and data over the time range of 2-30 psec were included in the fit.
The theoretical damping rate, G, of the acoustic oscillations in smooth films may be expressed as:
G = ( (?film/2d film)ln [(Zsub-Zfilm)/(Zsub + Zfilm)] (2)
where nfilm is film sound velocity, d film is film thickness, Zsub is acoustic impedance of the substrate, and Zfilm is acoustic impedance of the film. For practical films, the damping rate may be affected by extrinsic factors such as film roughness and inhomogeneity.
 |
| 4. The frequency shows a decreasing trend dominated by a gradual increase in thickness as the silicide phases form (a). Damping rate varies with intermixed phases, and lateral non-uniformity in either the composition, crystal structure, roughness, or the texture of the film (b). |
Frequency results are shown in Figure 4a. Because the metal film's acoustic impedance is larger than silicon's, the acoustic wavelength in the lowest-order thickness mode should be twice the film thickness. As expected, the frequency shows a decreasing trend dominated by a gradual thickness increase as the silicide phases form. When the film has transformed to CoSi, the thickness should theoretically increase by a factor of two, and the final thickness of the CoSi2 should theoretically be 3.5× that of the starting cobalt film, assuming all the metal reacts with the silicon. This is in good agreement with the picosecond ultrasonic measured thicknesses determined from the frequency and the speed of sound in the silicides. The thickness of the metallic cobalt was measured to be 91 Å (in reasonable agreement with the RBS measurement10), and the final CoSi2 was measured to be 285 Å.
The damping rate results for the uncapped wafer set are shown in Figure 4b. The damping rate reaches many local maxima throughout the full annealing temperature range. These are a consequence of intermixed phases, and lateral non-uniformity in the composition, the crystal structure, the roughness, or the texture of the film. What is of interest is the theoretical damping rate for a perfect CoSi2-Si interface. This theoretical value can be compared with the measured value, providing information about interface quality. For this CoSi2 film the theoretical damping rate was determined to be 0.144 psec-1. The measured damping rate for films annealed at 700 and 800°C is ~0.18 psec-1, indicating a rougher film. For an anneal temperature of 950°C the experimentally measured damping rate increased by >28%, which is probably attributable to the beginning of agglomeration that further roughens both the surface and interface of the film.
Picosecond ultrasonic measurements of phase formation were well correlated to both the resistivity measurements and the X-ray diffraction results. Picosecond ultrasonic metrology was also shown to provide information about film thickness and interface quality, which is difficult to extract using other methods.
CoSi2 process examplePicosecond ultrasonic measurements are well-suited for monitoring production processes of CoSi2 . Small-spot (10 × 15 µm) non-destructive measurements can be made on product wafers at production throughputs. One technique that has shown promise in creating a reproducible and manufacturable process involves depositing a thin sacrificial cap layer of titanium or TiN on top of the cobalt. The layer is removed after a first RTP anneal step in which the cobalt layer is converted mainly into CoSi. The film is then re-annealed at a somewhat higher temperature to form CoSi2.
To investigate the effects of different cap layers, cobalt films were deposited at a 120 Å nominal thickness on 23 different silicon wafers.8 Within the same deposition system, a layer of TiN, nominal thickness of 150 Å, was then reactively sputtered onto 11 of these samples, and a 100 Å titanium layer was deposited onto the remaining 12. Both capped groups were annealed for 60 sec over a 460-700°C temperature range. The capping layers were then removed. All samples were then exposed to a final anneal at 800°C for 30 sec.
The picosecond ultrasonic CoSi2 thickness measurements obtained for the TiN- and titanium-capped films are plotted in Figure 5a as a function of the first anneal temperature. The TiN-capped samples gave thicknesses close to 400 Å, almost independent of the first anneal temperature. Based on the nominal starting cobalt thickness of 120 Å, the ratio of CoSi2 to cobalt thickness is ~3.3. This is in close agreement with the theoretical ratio indicating that the starting cobalt was almost completely converted into disilicide. Conversely, the titanium-capped samples exhibited a marked decrease in thickness with decreasing temperature, with only the 660°C sample reaching a thickness comparable to the TiN-capped series. The thinner-than-expected CoSi2 films for the titanium-capped samples indicate that a substantial amount of unreacted cobalt or mix of Ti-Co must have been removed along with the titanium cap.
Resistivity results for the titanium- and TiN-capped samples are shown in Figure 5b. As with thickness, TiN-capped samples exhibit no obvious first anneal temperature dependence. CoSi2 resistivities in the neighborhood of 17 µO-cm are consistent with values reported in the literature. For the titanium-capped samples annealed at lower temperatures, resistivity values are also in the expected range. However, they become abnormally large for temperatures higher than ~600°C. For the 660°C sample, for example, the resistivity was nearly 25% higher than the normal value. Because previous results suggest that this sample is relatively smooth and homogeneous, its higher resistivity may be due to chemical poisoning, possibly by titanium incorporated during the first anneal.
In this comparison of potential production process, picosecond ultrasonics combined with sheet resistance measurements provided information on the thickness, interface quality and resistivity of CoSi2 films. Thickness measurement precision was >0.2 Å, and total measurement time was ~2 sec/site. This information could help manufacturers characterize their own processes as they switch over to CoSi2. Similar measurements could also be used to monitor the process in production, allowing excursions to be quickly detected and high yields to be maintained.
As design rules shrink, the familiar TiSi2 used for self-aligned source, drain and gate contact regions reaches physical limitations. CoSi2 is expected to provide a solution until design rules shrink below 0.09 µm. As new processes required to manufacture CoSi2 are developed and ramped up to production, metrology solutions will be critical.
Picosecond ultrasonics can monitor phase formation as metallic cobalt interacts with the silicon substrate to create silicides. This information is in good agreement with resistivity and X-ray diffraction results. In addition, the non-contact, non-destructive picosecond ultrasonic method can determine the thickness of a CoSi2 film and provide information about the silicide/silicon interface's quality. These capabilities combined with >0.2 Å precision, a small measurement spot, and a high throughput will benefit manufacturers that wish to characterize and ramp up production of CoSi2.
| Author Information |
| Guray Tas is manager of advanced applications at Rudolph Technologies. He is responsible for coordination of R&D activities on characterizing new materials and processes in semiconductor devices and thin-film head fabrication using ultrafast optical techniques. He received a Ph.D. in condensed matter physics at Brown University . |
| References |
|
