PtSi Clustering In Silicon Probed by Transport Spectroscopy

Metal silicides formed by means of thermal annealing processes are employed as contact materials in microelectronics. Control of the structure of silicide/silicon interfaces becomes a critical issue when the device characteristic size is reduced below a few tens of nanometers. Here we report on silicide clustering occurring within the channel of PtSi/Si/PtSi Schottky barrier transistors. This phenomenon is investigated through atomistic simulations and low-temperature resonant tunneling spectroscopy. Our results provide evidence for the segregation of a PtSi cluster with a diameter of a few nanometers from the silicide contact. The cluster acts as metallic quantum dot giving rise to distinct signatures of quantum transport through its discrete energy states.

annealing can result in atomically abrupt silicide/silicon interfaces [10], often this is not the case and various degrees of interface roughness can be found [11][12][13]. In addition, several experimental evidences of unintentional Pt impurities in the channel of silicon MOSFETs have been reported [14][15][16]. Such Pt atoms originate during the silicidation process as a result of a diffusive motion from the PtSi/Silicon interface into the Si channel. Yet little is known about their most favorable arrangement within the silicon crystal and, in particular, about how this arrangement is affected by collective interactions among multiple Pt impurities. In this Letter we address this problem through a combined theoretical and experimental study involving atomistic simulations and transport measurements in short-channel PtSi/Si/PtSi SB transistors.
We begin from the simplest case of a single Pt impurity in a Si lattice. Some experimental studies suggest that isolated Pt impurities can indeed be found in the Si region adjacent to a PtSi/Si interface [17,18]. It has already been shown experimentally and theoretically that in such a case the substitutional position (i.e. a Pt impurity at the place of a Si atom) is the most energetically favorable configuration [19][20][21][22]. Here we consider the problem of multiple Pt impurities in a Si lattice. We intend to evaluate the possibility that nearby impurities out-diffusing from a PtSi contact can aggregate into small clusters. We also intend to find the most stable cluster structure.
In order to tackle the problem of multiple Pt impurities we used numerical calculations based on density-functional theory [23]. Starting from a single substitutional Pt, we identified the most energetically favorable position for a second substitutional Pt. This identification was accomplished by comparing three different options: first, second, and third neighboring lattice site. As a next step, starting from the most stable two-atom configuration, we studied three possible scenarios for the addition of a third Pt atom. Notice that, given the large difference between the formation energies of interstitial and substitutional single Pt, we ruled out aggregates involving interstitial Pt.
A branched diagram summarizing the results of these total-energy calculations is shown in Fig. 1. Two observations can be made: (i) a driving force promoting aggregation exists each time the formation energy per atom decreases following the addition of Pt atom; (ii) given a set of possible configurations promoting aggregation, substitution as a second neighbor of the pre-existing Pt atom(s) is always preferred. Hence, both first-and second-neighbor substitution are favored for the two-impurity aggregate, as the formation energy per atom decreases from 0.92 eV to 0.77 and 0.5 eV, respectively, the latter being the most stable.
When a third Pt atom reaches the second-neighbour two-atom aggregate, on the other hand, the only configuration that leads to an increased stability is the one where all the Pt atom are second neighbors. The other cases considered feature an increase of the formation energy per impurity (0.71 and 0.96 eV) [24].
These results suggest that even a moderate supply of Pt atoms into a Si lattice would lead to the formation of PtSi clusters initially adopting the zincblende structure imposed by the host crystal. In order to test the validity of this conclusion for large numbers of Pt atoms, we considered a PtSi cluster with a diameter of approximately 1 nm, embedded in a 512-atom bulk Si supercell. In Fig. 2 we plot the density of states decomposed in the contributions from bulk Si atoms, and from Pt and Si atoms in the PtSi cluster. It can be seen that a complex structure of peaks appear in the Si band-gap, which reveals the metallic character of the cluster.
As the cluster grows in size, a phase transition to the bulk PtSi structure (favored by 0.55 eV/PtSi pair) is expected to occur. For this reason we have explicitly addressed the structural relaxation of a PtSi cluster of similar size with the thermodynamical stable structure, and we have found that for a cluster diameter of 1 nm relaxation to the host zincblende lattice is still thermodynamically favored.
We have also considered the formation of all-Pt clusters, examining both clusters where Pt adapts to the host zincblende symmetry or clusters where Pt takes the fcc symmetry of its bulk form. Although the formation energies of Pt or PtSi inclusions depend on the chemical potentials of the respective constituent species (which have a large degree of uncertainty), the differences in favor of PtSi clusters are so large that all-Pt clusters can be safely discarded.
In the following we shall present experimental evidence for the existence of PtSi clusters in silicon devices with PtSi contacts. We shall investigate the effect of a single PtSi cluster on the transport properties of a short-channel transistor. PtSi/Si/PtSi SB transistors were fabricated from undoped silicon nanowires (NWs) with diameters in the 20 − 40 nm range.
The NWs were grown by chemical vapor deposition via a catalytic vapor-liquid-solid method [25]. After growth, the NWs were transferred onto an oxidized silicon substrate and individually contacted by pairs of 80-nm-thick Pt electrodes deposited by sputtering (for more details on NW growth and device fabrication we refer the reader to refs. [26,27]). Each Pt electrode consisted of a 500-nm-wide and 3-µm-long strip whose edges were connected to two Cr(10nm)/Au(65 nm) bonding pads via progressively wider Pt metal lines defined in the same deposition step. In order to promote the formation of PtSi contacts, the Pt strips were annealed one at a time by means of Joule effect. To this aim, an electrical current of ∼10 mA was sequentially applied through each Pt strip causing a local increase of the temperature and hence promoting the silicidation of the contacts. This silicidation technique, which was introduced and extensively discussed in Ref. [27], was applied to obtain PtSi/Si/PtSi NW junctions with controlled Si channel length down to ∼10 nm ( Fig.3(a)).
Each NW junction was capped by a 5-nm-thick aluminium-oxide (Al 2 O 3 ) layer, grown by Atomic Layer Deposition, and a Cr(10nm)/Au(60nm) top-gate electrode defined by e-beam lithography, metal evaporation, and lift-off.
Due to the absence of intentional doping and to the short channel length, the fabricated devices were found to operate as SB transistors [1]. In these transistors, the silicon channel contacts is set by the n-and p-type SBs, respectively. Since the p-type SB (φ p ) is significantly smaller than the n-type SB (φ n ), electrical conduction is dominated by hole-type carriers.. for two different temperatures. At room temperature (black trace) and for V sd << φ B , transport through the silicon region is dominated by the thermionic emission of holes over the reverse-bias p-type SB. At 7 K (red trace), thermionic emission is entirely suppressed, and the residual conduction is due to temperature-independent tunneling through the silicon section, which acts as tunnel barrier [28]. A finite differential conductance, G = dI sd /dV sd , is observed throughout the entire I sd (V sd ), thus including the linear regime around zero bias. The linear conductance, G, decreases with the gate voltage, V gate , as shown by the measurement in Fig.3(d), which was taken at 0.24 K. This p-type transistor behavior is characteristic of hole-dominated conduction. Once again, this follows from the fact that φ p < φ n . An increase of V gate causes a downward band bending in the silicon section leading to a higher tunnel barrier for holes ( Fig. 3 (c)), and hence a lower conductance. Yet, due to a short-channel effect (the nanowire diameter is about four times the channel length), the gate effect is largely screened by the metallic PtSi contact. As a result, the G(V gate ) exhibits a moderate modulation over the accessible gate-voltage range. In particular, G(V gate ) remains finite up to the highest gate voltage applied (no higher voltages could be achieved due to the onset of significant gate leakage). A negative gate voltage produces an upward bending of the valence-band profile. Even for the largest V gate , however, the valence-band edge remains well below the Fermi level of the contacts, such that no hole accumulation is induced in the silicon channel. This is consistent with the absence of Coulomb blockade behavior, which would be expected in concomitance with a gate-induced formation of a hole island in the silicon region (see, e.g., Ref. [29]).  we assume flat-band condition in the silicon section, the cluster electrochemical potential lies above the Fermi levels of the contacts. In this regime, transport is due to mainly hole-like direct tunneling through the silicon band gap. At V gate ≈ 3V , the downward bending of the silicon bands results in a higher tunnel barrier for holes leading to a decrease in the direct-tunneling conductance.
Simultaneously, the cluster electrochemical potential lines up with the Fermi levels of the leads resulting in a resonant-tunneling current. (d) A resonance peak appears over the background differential conductance measured at 230mK with a lock-in excitation voltage of 100µV .
Interestingly, a sharp conductance peak is observed at a positive gate voltage close to 3 V, superimposed on the slowly varying background conductance. As shown in the inset to Fig. 4(a), following the subtraction of this background, the conductance peak can be fitted very well to a Lorentzian function [30,31], revealing an underlying resonant-tunneling transport channel. The fitted peak width w gives a measure of the tunnel coupling between the resonant state and the leads. The temperature dependence of the observed conductance resonance is shown in Fig. 4(a). As expected for resonant-tunneling, the conductance peak gets smaller and wider upon increasing temperature from 0.24 to a ∼ 5 K. Above T ∼ 5K, the peak height increases again as shown in the lower inset of Fig. 4(b). On the contrary, w, exhibits a monotonic temperature dependence as shown in Fig. 4(b)(main panel). To a closer look, however, w is roughly constant below 0.6 K (see upper inset to Fig. 4(b)).
In this low-temperature regime, the peak width is dominated by the life-time broadening of the resonant state due its tunnel coupling to the source and drain leads. Precisely, w ≈hΓ =h(Γ S + Γ D ), where Γ S and Γ D are the tunnel rates to the source and drain contacts, respectively. Between 0.6 and 4 K, w increases linearly with temperature according to the expectation for tunneling through a single discrete resonant level, i.e. w = 3.52K B T (( Fig. 4(b), green line). This linewidth matches exactly the thermal broadening of the Fermi distribution function in the source and drain leads. A linear temperature dependence is observed also above 8 K, yet with a larger slope corresponding to the expectation for tunneling through an ensemble of closely spaced levels, i.e. w = 4.35K B T (Fig. 4(b), red curve). Therefore, the crossover temperature T ⋆ ≈ 6 K identifies the transition from quantum (single level) to classical (multiple levels) regime [32]. This finding suggests that the observed conductance peak arises from resonant tunneling through a quantum dot with characteristic mean-level spacing δE ≈ k B T ⋆ ≈ 0.5meV .
To further support this conjecture we present in Fig. 5(a) a measurement of dI sd /dV sd as a function of (V gate ,V sd ). The color plot (stability diagram) exhibits an X-shaped pattern which is typical for Coulomb-blockaded transport in quantum-dot systems [33,34]. The crossing point, which corresponds to the conductance peak in the linear regime, represents the boundary between two consecutive Coulomb diamonds. In each diamond, transport through the quantum dot is blocked and the quantum dot hosts a well defined, integer charge state. Additional multiple dI sd /dV sd lines parallel to the diamond edges can be seen  including Pt nanoclusters [38] (we are aware of only one work [39] whose results do not agree with this trend). Therefore, our g-factor measurement provides an experimental indication against the hypothesis of a cluster consisting of pure platinum. The hypothesis of PtSi cluster, as suggested by our ab initio calculations, appears more plausible. With the aid of adequate numerical tools, it would be interesting to perform a calculation of the g-tensor in PtSi. Intuitively, one may indeed expect the g-factor to approach the bare electron value as a result of the considerable silicon content in the cluster. This tendency should be further reinforced by surface effects associated with the leakage of the electron wave-functions of surface atoms into the silicon host matrix [40][41][42].
In summary, through atomistic simulations and electronic transport measurements we have provided evidence of platinum clustering in silicon devices employing PtSi contacts.
Our experiment on a short-channel PtSi/Si/PtSi SB transistor revealed the emergence at low temperature of a single-electron tunneling channel. This transport channel, which causes current resonances in the full-depletion (i.e. off) regime, is ascribed to a Pt-based metallic cluster embedded in the silicon section. Our atomistic simulations suggest the cluster to be most likely composed of PtSi. The cluster has discrete electronic levels with a characteristic energy spacing of 0.5 meV, corresponding to a diameter of a few nm, which is comparable to the characteristic length scales of the device. In addition, the cluster has a charging energy of 50 meV, which explains why single-electron effects survive up to relatively high temperatures. In the perspective of ultrascaled transistor devices, our study shows that the possible formation of PtSi clusters during the silicidation process can have important consequences on device performances. This issue can lead to significant device variability undermining the gain from using metal silicides and doping-free devices. Hence, we expect our results to have an impact in the engineering of a wide class of emerging electronic devices, including fully-depleted nano-transistors and ultra-fast PtSi Schottky barrier photodetectors.

ACKNOWLEDGMENTS
This work was supported by the Agence Nationale de la Recherche and by the EU through the ERC Starting Grant HybridNano . The authors would like to thank Laurent Cagnon Stephane Auffret and Xavier Waintal for technical support ans useful discussions. [17] S. Mantovani, F. Nava, C. Nobili, and G. Ottaviani, "In-diffusion of pt in si from the ptsi/si interface," Phys. Rev. B 33, 5536-5544 (1986).
[18] A. Prabhakar, T. C. McGill, and M-A. Nicolet, "Platinum diffusion into silicon from ptsi," Applied Physics Letters 43, 1118Letters 43, -1120Letters 43, (1983. [23] Theoretical first-principles calculations have been performed within density-functional theory (DFT), as implemented in the Siesta package [43,44]. We have used an optimized double-ζbasis set plus polarization functions for the valence electrons, while core electrons are accounted for by means of norm-conserving pseudopotentials of the Troullier-Martins type.
The exchange-correlation energy is treated within the Generalized Gradient Approximation (GGA) [45]. Pt point defects and 2-and 3-atom aggregates are studied n the 3×3×3 supercell of the 8-atom bulk Si unit cell, while the ∼ 1 nm clusters were created in a larger 4 × 4 × 4 supercell. The Brillouin zone was sampled with a 2 × 2 × 2 grid of k-points, though a finer 8 × 8 × 8 was used for an accurate determination of the densities of states in Fig.2. All the structures have been optimized until the force on the atoms were lower than 0.04 eV/Å.
[24] For the sake of simplicity we are assuming that, in the slow-rate limit, Pt atoms reaches the clustering zone one by one. We believe, however, that this simplified model is enough to capture the physics of the aggregation process.