Low-loss amorphous silicon waveguides grown by PECVD on indium tin oxide

Sandro Rao sandro.rao@unirc.it Department of Information Science, Mathematics, Electronics and Transportations (DIMET) “Mediterranea” University, Via Graziella Località Feo di Vito, Reggio Calabria, 89060, Italy Francesco G. Della Corte Department of Information Science, Mathematics, Electronics and Transportations (DIMET) “Mediterranea” University, Via Graziella Località Feo di Vito, Reggio Calabria, 89060, Italy Caterina Summonte Institute for Microelectronics and Microsystems, Consiglio Nazionale delle Ricerche Unit of Bologna, Via Gobetti 101, Bologna, 40129, Italy


INTRODUCTION
Crystalline silicon waveguides and photonics devices based on SOI technology have been an active field of research for more than 10 years [1]. Compared to III/V-compound semiconductors, an inherent advantage of this material is the cost and simplicity of processing. The demand for high refractiveindex differences in order to create photonic crystals or highspeed optical modulators [2] have driven these efforts.
For the fabrication of active devices, like switches and modulators, two or more electrical contacts have to be embedded inside the structure to allow the application of an external bias. Although standard microelectronic processes provide plenty of technologies for the formation of metal contacts, their deployment has in general a negative impact on the optical characteristics of the devices, especially in terms of propagation losses. The preferred solution in these cases consists in adopting either highly doped (poly-) silicon layers [2,3], or a transparent conductive oxide (TCO), like Indium Tin Oxide (ITO) or Tin Oxide (TO) [4]- [8]. These last materials, widely deployed in thin film solar cell technology [9], offer some advantages over doped silicon in terms of electrical properties, with a conductivity which can be as high as 1.8 × 10 3 Ωcm −1 [10,11], and optical properties, with a refractive index n as low as 0.5 at 1500 nm wavelength [12], therefore ensuring the high refractive-index difference required in sub-micron waveguiding devices.
Unfortunately, TCO films can only be used for top contacts, because the SOI technology does not allow laying it at the waveguide bottom, i.e. between the SiO 2 cladding and the silicon core. Therefore only highly doped silicon films can be used in this case [2,3].
With the recently growing interest for hydrogenated amorphous silicon (α-Si:H) deposited at relatively low temperatures (T < 300 • ) as a suitable material for integrated passive [13]- [17] and active [8,18,19] photonic devices, and the consequent reduction of the thermal budged required for their fabrication, many otherwise unusable materials and technologies might gain a role in this field. ITO, in particular, can be considered a valid candidate for a simple formation of the buried contacts in multi-layer devices. Unfortunately, some drawbacks might derive in this case from the very high extinction coefficient of ITO (k = 0.6 at λ = 1.55 µm [12]), whose negative consequences are enhanced by the known texturization effect observed in the α-Si:H films grown on TCO [20,21]. In fact, for the extreme light confinement in submicron strip waveguides, propagation losses due to the interaction of the waveguiding modes with the interface roughness are significantly increased and can become prohibitive for dense integrated circuits [22,23].
In this paper we report technological results and compare the loss measurements concerning α-Si:H waveguides, either di-  rectly deposited on ITO, or separated from this by a thin spinon-glass (SOG) layer.
The experimental performances have been compared to those obtained through calculations using an optical simulation package. The results are found to be in agreement with the experimental data.

WAVEGUIDE FABRICATION
In an active α-Si:H-based photonic device, thin films of this semiconductor may be deposited on top of an ITO layer in a RF or DC glow-discharge plasma of SiH 4 , pure or mixed with other gases. At the time of the deposition of the first α-Si:H layer, the ITO film is primarily exposed to a plasma hydrogen containing. This results in a degradation of the TCO caused by the reduction of the oxide surface to the corresponding elemental metals [21].
As a consequence, a memory of the surface roughness of the ITO substrate is retained in the growing α-Si:H film [24]. It should be mentioned that this effect is somehow useful in thin film solar cells because it helps improving their efficiency by favouring light trapping [25].
On the other hand, this behaviour leads to serious problems in waveguiding photonic devices, and primarily to high transparency loss in the interface region of the amorphous silicon films at the IR wavelengths. Moreover, TCO films deposited on α-Si:H show an increasing roughness with the deposition temperature [11].
The impact of the ITO film morphology on the propagation losses in α-Si:H-based waveguides, and a new technological solution to overcome this issue are outlined in the following.
In particular, the α-Si:H was uncoupled from the ITO surface by inserting a SOG layer in between. Three waveguides were realized. For all samples, the substrate is a doped c-Si wafer (ρ = 0.001 Ωcm) and the waveguide core is about 1.5 µmthick α-Si:H layer. The substrate and the stack are separated by a 310 nm-thick low refractive index LPCVD SiO 2 cladding layer, so that a good vertical confinement is achieved.
The geometrical data of the waveguides are summarised in Table 1.
The fabrication of the first waveguide begins with a surface cleaning treatment, using H 2 SO 4 + H 2 O 2 and HF solution on the n-type doped c-Si substrate. The substrate was than loaded into the four-chamber RF (13.56 MHz) system. The SiO 2 cladding layer was grown by wet thermal oxidation. Subsequently the 80 nm-thick ITO layer was deposited by RF magnetron sputtering using a In 2 O 3 :SnO 2 ceramic target, at 250 • C substrate temperature and a base pressure adjusted to 10 −5 torr [26]. The SOG spinning on the ITO layer was done at 4000 rpm, followed by three pre-baking steps of 60 s each, at 80 • C, 150 • C, and 200 • C respectively, and a furnace annealing at 425 • C for 30 min in flowing nitrogen. Finally the α-Si:H layer was deposited in SiH 4 atmosphere by an RF Plasma Enhanced Chemical Vapour deposition (PECVD) at the frequency of 13.56 MHz, 220 • C substrate temperature, 0.55 mbar gas pressure and 20 sccm (standard cubic centimetres per minute) silane flow rate. The base vacuum was 4 × 10 −8 mbar. The deposition time was adjusted to obtain films thickness of about 1.5 µm. The fabrication of the second and third waveguides followed the same procedure, although SOG and ITO/SOG layer depositions were skipped for sample 2 and sample 3 respectively.
Note that in sample 1 the active α-Si:H layer is indeed insulated from the buried ITO contact. However, this is not really a problem, as small via-holes can be locally introduced across the SOG layer, e.g. laterally to the waveguide core, for warranting electrical continuity between the buried contact and the active device.

Optical Characterization
The samples were optically characterized using specular spectral reflectance R. Data were obtained using an optical fiber UV-visible spectrophotometer (Avantes). Figure 1 shows the experimental and computed R spectra for samples 1 and 2.
For both samples, at wavelengths longer than 650 nm, an interference pattern is visible, which reflects the contribution of all interfaces down to the silicon surface. For shorter wavelengths, no fringe pattern is visible. In fact, due to film absorption, there is no contribution to R coming from the interface of α-Si:H with SOG. In other words, the α-Si:H film behaves as a bulk. Therefore, the R curves should coincide for the two samples. Yet, for sample 2 a severe damping of the signal, stronger at shorter wavelengths, is observed. This indicates the presence of roughness at the α-Si:H surface, that causes light scattering and consequent decrease of the specularly reflected light intensity.
Simulations were obtained using the OPTICAL computer code, which makes use of a Generalized Scattering Matrix Method to treat multilayers with any distribution of coherent or incoherent behaviour [27], associated to χ 2 minimization. The simulation of sample 1 was performed using flat surfaces (thin black line in Figure 1), while simulation of sample 2 required the introduction of 25 nm roughness at the α-Si:H surface. The roughness was accounted for by using modified Fresnel coefficients, according to the exponential decay model of the rough interface [28].

Mechanical profiling
The surface roughness was measured on all samples. Figure 2 shows the profiles taken using a mechanical profilometer (Veeco Dektak 6M). The curves where vertically shifted each other for the sake of clarity, the shift having no correlation with the actual film thickness. The roughness of amorphous silicon, when grown on top of ITO, is evident. In this case a statistic on surface profiling gives a standard deviation of ±29 nm, which correlates very well with the value obtained by optical simulation. When a layer of SOG is inserted between the two, the roughness is greatly reduced. The roughness of ITO itself, SOG, and amorphous silicon deposited on SiO 2 /c-Si is below the detection limit of this technique.

Waveguide characterization
Propagation loss measurements were made for the three different waveguides with the cut-back technique [29]. This method is based on measuring the light transmission through waveguides of different lengths and fitting the experimental data length dependence, assuming identical coupling conditions and constant facet quality.
In our experiments, the probe beam was a 15 mW laser diode  radiation (λ = 1.55 µm), coupled to the waveguide core via a 5 µm-core single-mode fiber, without polarization control. The transmitted light was collected at the chip output by a multimode fiber and detected by an InGaAs photodiode. Great care was taken to prevent stray light passing above the sample.
Several samples from different wafers with lengths up to 2 cm were cut by cleavage. Sample facets did not receive polishing treatment, although their good quality was verified by means of micro-optical inspections before measurements. Figure 3 shows the measured transmitted power through type 1 waveguides with different lengths. The loss coefficient was extracted by linear regression. From the slope shown in the figure, we calculated an absorption coefficient of ∼ 2.5 ± 0.1 dB/cm at λ = 1550 nm. Note that this experimental attenuation for unpolarised light is intermediate between the theoretically estimated ones for the TE and TM modes, as reported in Table 2. The overall insertion losses are 10.4 dB, with the main contribution due to the coupling losses, a drawback of the high refractive index of silicon used as the core material.
Using the same technique, the loss coefficient of type 3 waveguide, consisting of the stack of 1.44 µm-thick α-Si:H on SiO 2 , was measured to be ∼ 1.5 dB/cm, a value comparable to standard SOI waveguide losses in α-Si:H [15]. As the two devices differ only in the SOG/ITO layers of waveguide 1, we can conclude that this introduction is responsible for a loss increase of ∼ 1.0 dB/cm. It should be noted, finally, that the introduction of a bare ITO layer drove to totally opaque waveguides (type 2), due both to the high roughness of amorphous silicon grown on top of the ITO layer and its high extinction coefficient.
The experimental propagation losses are summarised in Table 2, which also lists the theoretical results, for both TE and TM polarizations, discussed in the next section. ), or separated from this by a SOG layer (sample 1, see Figure 4(a)). In Figure 4(c) we report a standard SOI waveguide in α-Si:H on SiO 2 (sample 3).

Waveguide simulations and discussion
Optical simulations were performed using Beamprop, the device simulation package from RSoft [30]. Most of the optical data of the materials used for simulations were determined from the analysis of the experimental transmittance and reflectance spectra, in the wavelength range 0.5 µm to 2 µm, of test films deposited on corning glass. The reflectancetransmittance analysis tool OPTICAL [27] was used for this analysis, which also allowed determining the layer thicknesses. Photothermal deflection spectroscopy (PDS) [31] was used to determine the absorption coefficient of α-Si:H at λ = 1.55 µm. The material characteristics, at this wavelength, are listed in Table 3.
The waveguides have been designed in order to ensure only the propagation of the fundamental mode and for this purpose the optical mode spectrum was also modeled. The waveguide geometries were chosen according to [32], to satisfy the single mode condition. Simulations demonstrated that the designed waveguiding structures exhibit a dominant single-mode operation for TE and TM polarization modes once a 5 µm wide rib is defined by photolithographic patterning and etching process. The shallow etch depth is 60 nm. In Figure 5 we report the fundamental optical mode-field profiles of samples 1, 2, 3, for both, TE and TM polarizations.
We calculated the propagation losses for the three realized structures. The corresponding results, for both polarization conditions, are again reported in Table 2   These results allow to conclude that the optical propagation in the waveguides with no interface roughness between the buried ITO contact and the α-Si:H waveguide core is also limited by the ITO absorption. As a consequence, an optimum  design of the structure and a careful control of the light signal confinement inside the α-Si:H core layer would permit the realization of waveguides with losses lower than 1.5 dB/cm, which is considered a sort of threshold for integrated optic links [33].
For this purpose we simulated a new waveguide which differs from sample 1 only in the SOG thickness. If we consider a SOG thickness of 210 nm, instead of 160 nm, the waveguide propagation losses, for the TE mode, decrease from 1.8 dB/cm to 1.1 dB/cm, which is equal to that estimated for sample 3.

CONCLUSIONS
The negative impact of a buried transparent conductive oxide layer on the propagation losses in α-Si:H-based waveguides, grown on an ITO layer useful as a buried contact, was studied. All the fabricated samples were optically characterized and the surface roughness was accurately measured using a mechanical profilometer. We observed that, for an α-Si:H corelayer directly deposited on ITO, the surface roughness, due to the interaction of ITO with hydrogen during the PECVD deposition, is of the order of 100 nm. This leads to prohibitive propagation losses for the waveguide, which on the other hand drop to 1.5 dB/cm in waveguides not including the ITO buried layer. The strong reduction in the optical transmittance was recovered by depositing a spin-on-glass layer between the 1.44 µm-tick α-Si:H core-layer and the ITO layer. In this case, propagation losses of 2.5 dB/cm at 1550 nm were measured.
The experimental results have been compared to those obtained through calculations using an optical simulation package. The results were found to be in agreement with the experimental data.
Technologically, the waveguide fabrication process involves temperatures lower than 450 • C, which are compatible with standard microelectronic processes and in particular suitable for the realization of a photonic layer on top of an integrated circuit.