We study the propagation properties of surface plasmon polaritons on a Cu surface by means of photoemission electron microscopy. Use of a CMOS process to fabricate the Cu thin film is shown to enable very high propagation distances (up to 65 μm at 750 nm wavelength), provided that the copper native oxide is removed. A critical review of the optical loss mechanisms is undertaken and shed light on the effect of single grain boundaries in increasing the propagation losses of the plasmon. A microscopic interpretation is provided, relying on groove induced electromagnetic hot spots.
© 2012 Optical Society of America
The use of surface plasmon polaritons (SPP) in metal nanostructures to control light at scales lower than its natural wavelength has emerged in the past few years as a promising way of integrating optical functions in optoelectronics and biochemical sensing applications. The compatibility of SPP device fabrication with complementary metal-oxide-semiconductor (CMOS) processes would enable the development of these applications. However, the metals used in studies of propagating plasmons are generally either silver or gold , which are forbidden contaminants in a CMOS environment. Copper is also a high conductivity metal, widely used in backend CMOS processes, whose use has been surprisingly ignored in plasmon assisted devices, except in one recent example . Investigating this issue, we show here that nanoscale properties of the Cu surface microstructure, namely the topology of the grain boundaries and the oxidation of the copper surface, are crucial to the control of the SPP losses. A fabrication process of Cu films which minimizes the topological effects of grain boundaries on SPP losses is also presented. This work therefore sheds a new light on the use of copper for plasmonic applications, and is key with regard to the integration of plasmonic functions within CMOS foundries [3, 4].
As SPP modes originate from the interaction of light with conduction electrons, they experience what is generally referred to in literature as ohmic losses [5, 6]. These losses are usually accounted for through the use of a dielectric constant of the supporting metal. The ubiquitous model for the latter in the near infrared range of the spectrum is the Drude model [5–7]. However, experiments involving either propagating or localized plasmon modes often exhibit non negligible discrepancies between measured losses and those predicted using dielectric constants measured by ellipsometry [5–7]. This is commonly linked to the imperfections of the sample, eg its surface roughness and material quality. In smooth samples where no losses are expected by both in-plane and radiative out of plane SPP scattering , extra losses are usually taken into account by an additional electronic relaxation time contribution in the Drude model that would originate from the electron scattering on defects of the polycrystalline metal film: eg point defects or grain boundaries (GBs). In particular, the correlation between SPP losses and GB density was evidenced in several works [5–7, 9, 10], using spatial averaging techniques. In contrast, we investigate below the SPP propagation losses at the single GB level, using high resolution imaging of a smooth surface of polycrystalline Cu by photoemission electron microscopy (PEEM). This allows the study of the above electronic effect of GBs on SPP losses.
2. Low optical loss Cu film preparation
Both large grain size and very low roughness are required to observe low loss SPPs and their interaction with a single GB. When using standard thin film preparation techniques such as ion beam sputtering or evaporation, roughness and grain size usually scales with the deposited thickness of the films. Low roughness can then be achieved at low film thickness, but at the expense of small grain size, and consequently higher SPP losses. To circumvent these problems, the use of ultrathin seed layers to control the surface energy of the substrate , or template stripping technique have been proposed . However these techniques can not be compatible with large scale standard CMOS processes used to integrate plasmonic functionalities in devices. We show here an alternative fabrication route for very high optical quality Cu films for plasmonics, employing a typical CMOS foundry process normally used for interconnection fabrication . Large grains are obtained using high thickness (1.4 μm) Cu deposition on a Si wafer, after which samples are annealed at 400 °C for 30 minutes in a vacuum to promote grain growth and good material crystalline quality (4-point probe resistivity ρ = 1.75 μΩ.cm, close to the bulk value). Standard chemical mechanical polishing (CMP) is finally carried out to reduce both thickness down to 200 nm and surface roughness. Figure 1(a) shows a 10 × 10 μm2 AFM image of the Cu sample, where mean grain size and root mean square (RMS) roughness were measured as ≈ 2 μm and ≈ 0.65 nm, respectively. The surface texture consists of large grains whose roughness is very low (0.4 nm), delimited by 1 nm height steps. Some 1 – 2 nm depth, 50 nm wide grooves are observed along the twin boundaries punctuated by some 3–5 nm depth holes often observed at triple junctions (shallow groove B in Fig. 1(a)), and a low surface density of deep (5 – 15 nm) and large (few hundreds nanometers) grooves (labeled A in Fig. 1(a)) here again pinned at a triple grain junction. This discontinuous surface originates from the different polishing rates of grains of differing crystallographic directions and sample aging. Figure 1(b) shows the dielectric function (εellips = ε′ellips + iε″ellips) of the fabricated Cu layer measured by ellipsometry, together with the Palik data of Cu (εpalik = ε′palik + iε″palik) for comparison . By fitting ellipsometric data in the near-IR range, we derived the Drude model parameters of our Cu thin film ωp = 1.35 × 1016 s−1 and τellips = 2.65 ± × 10−14 s, which are very close to the literature bulk Cu electrical transport data, and particularly for a defect related relaxation time τ0 = 2.7 × 10−14 s . This value is three times higher than that derived from the ellipsometric Palik data, indicating a much better material quality. This can be attributed to the fabrication method used in this work, leading to lower GB density and lower point defect density within grains using annealing , and also presumably lower surface scattering using polished surface .
3. PEEM imaging of SPP propagation
SPP propagation losses at the single GB scale have been investigated by high resolution photoemission microscopy (PEEM). The PEEM imaging principle is based on the collection of electrons produced by a pulsed laser source impinging on the sample (photoelectric effect) [15, 16]. These photoemission processes are non-linear in nature, in the present case the photoemission yield is related to the 6th power of the optical electric field amplitude inside the metallic region (3 photon absorption event). The light source used for PEEM measurements is a pulsed mode-locked Ti:sapphire laser system (Chameleon Ultra II, Coherent Inc.) delivering IR photons in the 690 – 1000 nm range. Experiments are carried out at grazing incidence (θ = 75° from surface normal) in p polarization, by focusing the laser beam on the surface using a low numerical aperture objective. The instrument allows simultaneous topographical investigation using a low energy electron (LEEM) imaging mode (routine resolution 10 nm), which is then spatially correlated to the PEEM image (routine resolution 20 nm).
In the context of GB investigation, PEEM microscopy offers several benefits over the well-established scanning near field optical microscopy SNOM. (i) PEEM is a non-intrusive method, i.e. make no use of a scanning tip. Near field measurements are cleared from any tip-to-sample perturbation and quantitative comparison to analytical model are readily possible. (ii) PEEM opens a convenient route to address multi-length scale investigation. Indeed available fields of view range from tenths of μm (plasmons-polaritons propagation length scale) down to tenths of nm (GB length scale) with high resolution near field mapping capabilities. (iii) PEEM exploits nonlinear photoelectric effect. The acquired signal exhibits a large signal-to-noise ratio and the background removal operation is straightforward. For interested readers, a recent panorama of the real space microscopic imaging techniques devoted to plasmonics is available .
To launch a SPP wave on the planar Cu surface, we used a 150 nm × 1.5 μm dielectric (Si3N4) ridge, which is known for having a high SPP launching efficiency for obliquely incident light . The experimental layout for the dielectric launcher is depicted in Fig. 2. The geometry of this ridge was designed using 2-dimensional boundary element method (BEM) calculations in TM polarization , so as to maximize the PEEM signal which results from the beating wave pattern between the propagating SPP wave and the incident beam. The 75 μm long Si3N4 ridge structure was fabricated on the Cu surface by e-beam lithography and lift-off processes. The temperature budget of the process was limited in order to minimize any roughness increase during the process, after which the RMS roughness was measured to 0.55±0.1 nm. PEEM measurements carried out under laser light excitation at a 700 nm wavelength on this sample are displayed in Fig. 3(a). Indeed, one observes a strong photoemission yield exhibiting a decreasing oscillating signal when the distance from the ridge is increased, as the SPP wave decays. In addition to the main signal, high intensity hot spots of various sizes are present, i.e. localized regions where electromagnetic field is enhanced.
4. Grain boundaries related losses
What is the origin of those hot spots and could they be an additional loss source for the propagation of the SPP? Given the roughness of the sample, observed localization of light must be related to either topology , or material modification , but not to multiple SPP scattering. A closer examination of the PEEM images (Fig. 3(b)) reveals that hot spots are found to be distributed along line shapes forming a network reminiscent of the grained surface texture. A proof of this correlation is readily obtained by a local comparison between the topology of the Cu surface recorded in the LEEM mode and the hot spot map obtained by the PEEM mode (Fig. 3(c)): the positions of GBs and triple boundaries (TBs) identified by triangles in the topographic LEEM picture translate into hot spot positions in the optical near field mapping. High intensity hot spots are frequently associated with GB triple junctions. Material effects can definitely be ruled out by observing that, firstly, the lateral size of hot spots varies from 50 nm to 1 μm (i.e. at least twice that of the PEEM resolution 25 nm) whereas the thickness of the GB in such copper films is expected to be below 1 nm, and secondly that the experimental hot spot size and density is more consistent with the typical sizes of the topographical defects of the Cu surface observed by AFM (Fig. 1(a)). Indeed, in Fig. 3(b), a few large hot spots (A’) are observed in correlation with the low density of large and deep A-type grooves (Fig. 1(a)). Similarly a much larger number of smaller hot spots (B’) are observed, in correlation with a larger density of smaller 3–5 nm deep, 50 nm width B-type holes (Fig. 1(a)). These observations offer direct evidence of topology-induced hot spots. The propagating SPP energy is stored within nano-grooves owing to cavity effects, resulting in a higher optical field in the grooves . As the electromagnetic field is enhanced within the metal at the level of the hot spots, we surmise that they induce an additional power loss, which would further impact the SPP propagation length. We discuss below the magnitude of this effect in our experiment and literature reports. Power losses within the metal can be estimated according to:23]. As roughness is negligible compared to hot spot sizes, we assume that the internal electric field in both fringe and hot spot regions extends in the metal over a typical thickness δ. The photoemission yield is presumed to be not affected by the electronic material effects at the GBs. Using image processing, we can extract the PEEM signal component originating from the hot spots, ie the third power of the corresponding optical losses , and compare those losses to the overall losses obtained by integration of the full PEEM signal over the same surface. We find so that, in our sample, the grooved GB induced hot spot contribution to the optical power losses is negligible. However, given the spatial correlation between GB and hot spots evidenced above, one can infer that hot spot density nHS will increase proportionally to the GB density, hence nHS ∝ 1/d2, d being the grain size. Following this line of thought, the power loss ratio calculated above for a grain size of 2 μm will increase to a non-negligible value of 0.3 for a grain size ten times smaller (d = 200 nm). This is consistent with the several-fold decrease of the SPP propagation length observed for a high density of GB in the case of gold film with d = 20 – 80 nm , a silver film with d = 100 – 200 nm , and also increased losses of the resonance of gold nanoantennae observed with d = 20 – 40 nm . The nHS ∝ 1/d2 scaling law may be valid for low GB density, but for higher GB densities, the contribution of SPP scattered by surrounding GBs will no longer be negligible: indeed a nHS ∝ 1/d scaling law has been experimentally observed in the d = 10 – 100 nm regime . Our data showing electric field enhancement at the single GB grooves is therefore supporting a topology-induced excess optical loss picture, in contrast with some previous explanations relying on the electronic effect of GBs .
5. Native oxide losses
As we use an interferometric PEEM configuration, accurate comparison between experimental and theoretical calculation of the SPP modes complex wavevector can also be achieved. Here also lies a material issue, as the existence of a very thin native copper oxide, is shown to have a significant impact on the SPP propagation losses. Parallel wavevector mismatch between the SPP kSPP and the plane wave k// = k0 sin(θ), θ being the incidence angle, is responsible for the beating fringe period observed in PEEM λbeat = 2π/(kSPP – k//). From the optical constant measured by ellipsometry, one predicts λbeat =11 μm at a 700 nm wavelength. However we measure λbeat =10 ± 0.3 μm (Fig. 4(a)). In general, a bilayer of thin native oxides Cu2O (1.3 nm) and CuO (2 nm) grows on surface by exposition of Cu films to air . These layers modify the complex SPP wavevector , which we calculated using a finite-differential time-domain (FDTD Lumerical) mode analysis. From a tabulated optical index of copper oxides , we found , allowing a fit of the fringe period from the experimental PEEM data (Fig. 4(a)). The fit was performed by adjusting the amplitudes of the SPP wave and the incident field, and their respective phase difference. We used optical constants and thicknesses of copper oxide, as well as the optical constants of Cu measured by ellipsometry as fixed inputs of the model. CuO and Cu2O being both semiconductors whose gap lies in the near infrared range, they are optically absorbing, and therefore substantially impact the amount of optical losses experienced by the SPP (Fig. 4(b)). In Fig. 4(b), the SPP propagation length is calculated by using the formula in the pristine Cu surface case (red curve), where εm is the dielectric constant of Cu measured by ellipsometry in section 2; or the formula in the case a native oxide is present (blue curve). The SPP propagation length is shown to decrease by about 50% in the latter case. Despite its very low thickness, the oxide layers experience a much higher electric field than the metal surface, leading to a significant optical loss.
We have provided microscopic evidence that GB induced losses of SPP modes originate from surface topology. GBs lead to absorbing electromagnetic hot spots through grain grooving effects which occur during deposition , annealing , or polishing/aging (this work). These field enhancements increase the local optical absorption. Using a CMOS process to fabricate the Cu thin film we were able to significantly limit the amount of excess losses from these electromagnetic hot spots. Cu is also shown to enable very high SPP propagation lengths (up to 65 μm at 750 nm wavelength), provided that surface oxidation is avoided.
This work has been supported by CEA internal Programme Transverse Nanoscience, the French National Agency (ANR) through LETI Carnot Funding, and by ANR Program in Nanosciences and Nanotechnologies (PEEMPlasmon project ANR-08-NANO-034). We acknowledge the PTA for access to clean room facilities and process technologies team for Cu film preparation, S. Maitrejean for discussions, and D. Fowler for his careful reading of the manuscript.
References and links
1. J.-C. Weeber, A. Dereux, C. Girard, J. R. Krenn, and J.-P. Goudonnet, “Plasmon polaritons of metallic nanowires for controlling submicron propagation of light,” Phys. Rev. B 60, 9061–9068 (1999). [CrossRef]
2. C. Delacour, S. Blaize, P. Grosse, J. M. Fedeli, A. Bruyant, R. Salas-Montiel, G. Lerondel, and A. Chelnokov, “Efficient directional coupling between silicon and copper plasmonic nanoslot waveguides: toward metaloxidesilicon nanophotonics,” Nano Lett. 10, 2922–2926 (2010). [CrossRef] [PubMed]
4. J. L. Perchec, R. E. de Lamaestre, M. Brun, N. Rochat, O. Gravrand, G. Badano, J. Hazart, and S. Nicoletti, “High rejection bandpass optical filters based on sub-wavelength metal patch arrays,” Opt. Express 19, 15720–15731 (2011). [CrossRef] [PubMed]
5. M. Kuttge, E. J. R. Vesseur, J. Verhoeven, H. J. Lezec, H. A. Atwater, and A. Polman, “Loss mechanisms of surface plasmon polaritons on gold probed by cathodoluminescence imaging spectroscopy.” Appl. Phys. Lett. 93, 113110 (2008). [CrossRef]
8. D. L. Mills, “Attenuation of surface polaritons by surface roughness,” Phys. Rev. B 12, 4036–4046 (1975). [CrossRef]
9. D. Canchal-Arias and P. Dawson, “Measurement and interpretation of the mid-infrared properties of single crystal and polycrystalline gold,” Surface Science 577, 95 – 111 (2005). [CrossRef]
10. V. J. Logeeswaran, N. P. Kobayashi, M. S. Islam, W. Wu, P. Chaturvedi, N. X. Fang, S. Y. Wang, and R. S. Williams, “Ultrasmooth silver thin films deposited with a germanium nucleation layer,” Nano Lett. 9, 178–182 (2009). [CrossRef]
11. E. Palik, Handbook of Optical Constants of Solids (Academic Press, Orlando, 1985).
12. P. B. Johnson and R. W. Christy, “Optical constants of the noble metals,” Phys. Rev. B 6, 4370–4379 (1972). [CrossRef]
13. V. Jousseaume, M. Assous, A. Zenasni, S. Maitrejean, B. Remiat, P. Leduc, H. Trouve, C. Le Cornec, M. Fayolle, A. Roule, F. Ciaramella, D. Bouchu, T. David, A. Roman, D. Scevola, T. Morel, D. Rebiscoul, G. Prokopowicz, M. Jackman, C. Guedj, D. Louis, M. Gallagher, and G. Passemard, “Cu/ulk (k=2.0) integration for 45 nm node and below using an improved hybrid material with conventional beol processing and a late porogen removal,” in “Interconnect Technology Conference, 2005. Proceedings of the IEEE 2005 International,” (2005), pp. 60–62.
14. N. Ashcroft and N. Mermin, Solid State Physics (Orlando, 1976).
15. L. Douillard, F. Charra, Z. Korczak, R. Bachelot, S. Kostcheev, G. Lerondel, P.-M. Adam, and P. Royer, “Short range plasmon resonators probed by photoemission electron microscopy,” Nano Lett. 8, 935–940 (2008). [CrossRef] [PubMed]
16. L. Douillard and F. Charra, “High-resolution mapping of plasmonic modes: photoemission and scanning tunnelling luminescence microscopies,” J. of Phys. D 44, 464002 (2011). [CrossRef]
18. H. Liu, H. Liu, P. Lalanne, X. Yang, and J.-P. Hugonin, “Surface plasmon generation by subwavelength isolated objects,” IEEE J. Sel. Top. Quantum Electron. 14, 1522–1529 (2008). [CrossRef]
19. S. Boutami and J. Hazart, “Calculation of a point source radiation in a flat or non-flat stratified background: an alternative to sommerfeld integrals,” Euro. Phys. J.: Appl. Phys. 52, 23305 (2010). [CrossRef]
20. J. A. Sánchez-Gil, “Localized surface-plasmon polaritons in disordered nanostructured metal surfaces: Shape versus anderson-localized resonances,” Phys. Rev. B 68, 113410 (2003). [CrossRef]
21. P. Dawson and M. G. Boyle, “Light emission from scanning tunnelling microscope on polycrystalline au films -what is happening at the single-grain level?” J. Opt. A, Pure Appl. Opt. 8, S219 (2006). [CrossRef]
22. J. Le Perchec, P. Quémerais, A. Barbara, and T. López-Rios, “Why metallic surfaces with grooves a few nanometers deep and wide may strongly absorb visible light,” Phys. Rev. Lett. 100, 066408 (2008). [CrossRef] [PubMed]
23. S. J. Orfanidis, Electromagnetic waves and antennas (Piscataway, NJ, 2004).
24. P. Keil, R. Frahm, and D. Ltzenkirchen-Hecht, “Native oxidation of sputter deposited polycrystalline copper thin films during short and long exposure times: Comparative investigation by specular and non-specular grazing incidence x-ray absorption spectroscopy,” Corrosion Science 52, 1305–1316 (2010). [CrossRef]