1. Introduction

Mirrors and filters based on alternating dielectric layers (thin film stacks) have been used in optics for many decades, because they can be designed to reflect combinations of incident wavelength, polarization, and angular range with nearly zero loss. In recent years, it was discovered [1–3] that appropriately designed stacks exhibit omnidirectional reflection (near unity reflectance for all incidence angles and polarization states) within one or more bands of wavelength. In the simplest case, the stack is periodic and comprises a pair of alternating dielectric materials. Assume that n_H,d_H, and n_L,d_L are the refractive index and thickness of the high and low index layers, respectively, Λ=d_L+d_H is the period of the stack, and n₀ is the refractive index of the incident medium (often air). As extensively discussed in the literature [1–6], the existence of an omnidirectional band requires that both δ_n0=n_L/n₀ and δn₁=n_H/n_L are relatively large. When these two parameters exceed some minimum threshold, the Brewster line lies outside the acceptance angle for light rays incident on the mirror from the external medium, and there are finite ranges of wavelength for which no propagating modes exist inside the dielectric stack. A common design goal is to maximize the omnidirectional bandwidth, either because the application inherently requires broadband reflection [7] or in order to relax the fabrication tolerance for narrowband applications [8–11]. As with conventional Bragg reflectors, both the omnidirectional bandwidth and the peak reflectivity (in the case of a finite number of layers) scale with δn₁. For experimentalists, a key challenge is the identification of thermo-mechanically compatible materials with maximum index contrast. Given a pair of materials, another parameter of interest is the so-called filling factor η=d_H/Λ. Nearly optimal characteristics are provided by the quarter wave stack design, for which η=η_q=1/(1+δn₁). However, choosing η slightly greater than η_q can result in typically modest improvement in terms of the omnidirectional bandwidth [1–6] and peak reflectivity [12].

Recently, we reported [13] omnidirectional reflectors centered at 1750 nm based on alternating layers of Ge₃₃As₁₂Se₅₅ chalcogenide glass and polyamide-imide (PAI) polymer. The attributes of these mirrors include low-temperature processing (<300°C), the use of commercially available materials, and fabrication on a silicon substrate. The materials provide excellent thermo-mechanical compatibility, evidenced by good planarity and good mechanical integrity over a wide temperature range. Further, the combination is characterized by δn₁~1.55 and δn₀~1.65 (assuming air incidence), which enables a relative omnidirectional bandwidth (Δω/ω₀) on the order of 6–7 % [1]. Here, we report a technique for increasing the index contrast through silver doping of the Ge₃₃As₁₂Se₅₅ layers, resulting in δn₁~1.8 and a doubling of the omnidirectional bandwidth. We compare silver-containing and silver-free mirrors with omnidirectional reflection bands centered near 1550 nm wavelength.

2. Background and fabrication

Metal doping of chalcogenide glasses is a widely studied technique and review articles are available [14–15]. In brief, when deposited in contact with many (perhaps all) chalcogenide glass alloys, certain metals (such as copper, zinc, and especially silver) can be induced to diffuse into the glass under exposure to near bandgap light [16] (this is commonly termed photodoping), ion beams [17], electron beams [18], or upon heating [19]. Up to some maximum concentration, the metal atoms are incorporated into the glass network and the silver-containing alloy retains the usual characteristics of a homogeneous glass (although with modified optical, chemical, thermal, and mechanical properties). The metal infusion tends to shift the glass bandgap towards lower energy [14–15] and increase the linear [20] and nonlinear [21] refractive indices of the glass. The linear refractive index increase can be as high as ~0.5 at wavelengths much longer than the shifted band edge, where the silver-containing glass retains good transparency. Silver doping of chalcogenide glasses has been studied as a means to realize inorganic resists [22], infrared diffraction elements [23–24], microlenses [25], and integrated waveguides [19, 26–27]. Finally, photodoped chalcogenide glasses exhibit reduced tendency for photo-oxidation and photodarkening [28], and therefore have improved stability.

 

Fig. 1. (a). Schematic showing a PAI/IG2/silver layer structure, as deposited and after exposure to heat or light. The initially undissolved silver layers dissolve into adjacent IG2 layers upon exposure to heat or light, and form homogenously silver-doped IG2 layers (Ag:IG2). The layers are not to scale. The targeted layer thickness was ~290 nm for PAI layers, ~125 nm for IG2 layers, and ~20 nm for Ag layers. (b). Normal incidence spectrophotometer scans of sample B (see text) before light exposure (blue curve), and after light exposure (red curve). PAI layers were cured at 200°C in this case, and no significant change in transmission characteristics was observed after light exposure. Presumably, silver doping was thermally induced during the curing of the PAI layers. (c). Normal Incidence spectrophotometer scans of a sample created in a similar fashion to sample B, but with PAI layers cured at 90°C. The blue curve was taken before white light exposure, and the red curve after white light exposure. Prior to light exposure, the sample is essentially opaque due to the presence of un-reacted silver. After photodoping, the spectrum is that expected for a Bragg mirror.

Download Full Size | PDF

The chalcogenide glass Ge₃₃As₁₂Se₅₅ was sourced as IG2 from Vitron AG (France) and the polyamide-imide (PAI) polymer, TORLON AI-10, was sourced from Solvay Advanced Polymers. The processing of these films was as described previously [13], except for some variation in the curing of the PAI layers. 99.99% pure silver was sourced from Kurt J. Lesker Company (Clairton, Pennsylvania). Layers of silver were deposited using a magnetron sputter system (Kurt J. Lesker Company) equipped with a substrate rotation system.

Two prime grade silicon wafers (Silicon Valley Microelectronics) were cleaned in a piranha solution, rinsed in de-ionized (DI) water, and dried in nitrogen. The first sample (sample A), was processed as described in [13], but with slightly reduced period in order to shift the omnidirectional band to shorter wavelength. 13 layers (7 PAI layers and 6 IG2 layers) were deposited on sample A, producing a 6.5 period Bragg reflector. The PAI layers were approximately 285–290 nm thick and were cured at a peak temperature of 200°C in nitrogen after each spin. With our current process, we get inconsistent results when attempting to spin PAI layers thinner than ~285 nm. Given this restriction, the IG2 layer thickness was used as an adjustable parameter to target an omnidirectional stop band centered near 1550 nm. For sample A the IG2 layer thickness was set to ~150 nm.

 

Fig. 2. Normal incidence digital camera picture of (a) sample A (silver-free) and (b) sample B (silver-containing), showing mirror finish of the omnidirectional reflectors. The color of the samples is due to their spectral reflection and absorption features as shown in Fig. 4 below, and described in the text. A reflected image of the digital camera appears in each case.

Download Full Size | PDF

Sample B was processed in the same way as sample A, except that the IG2 layer thickness was set to ~125 nm and ~20 nm of silver was sputtered after each IG2 deposition. Further, a PAI layer was spun-cast and cured after each silver deposition. We note that adhesion to a variety of materials including metals is one of the attributes of PAI [29]. In total, 16 layers were deposited on sample B (6 PAI layers, 5 IG2 layers, and 5 silver layers). We have processed numerous samples of this type, with some variation in the final cure temperature of the PAI layers. As shown in Fig. 1, the silver layers are dissolved into each adjacent glass layer, either during the curing of the subsequent PAI layers at high temperature (~200°C) or upon light exposure if the PAI is cured at lower temperature (~100°C). Light exposure was provided by a Cambridge Instruments fiber-optic illuminator (cat. no. 31-30-50) with output spectrum centered in the 450 nm – 750 nm range. Intensity at the sample was approximately 0.1 W/cm² as measured using a silicon detector through a pinhole. For all silver-containing samples (regardless of PAI cure temperature) the optical characteristics reached a saturated state for exposure times of a few hours, apparently indicating complete dissolution of the silver. It should be noted that some silver is expected to diffuse into the glass during the high energy sputter deposition [18]. After doping, sample B is essentially a 5.5 period Bragg reflector comprising two homogeneous dielectrics (PAI and silver-containing IG2 [Ag:IG2]). As evidenced by the optical characteristics discussed below, this process resulted in homogeneous doping of the IG2 glass layers. The estimated silver concentration (~20 at. %) is probably near the solubility limit for a Ge-As-Se glass [14]. The thickness of these layers is expected to be approximately equal to the sum of the starting IG2 and Ag layer thicknesses [25], but its precise value depends on the way in which the Ag is incorporated into the glass network. Based on our experiments with IG2, we find the doped layer thickness is typically much less than the sum and in fact only slightly larger than the undoped IG2 layer thickness. It should be noted that the Ag layers are very thin and begin to diffuse into the glass even during deposition. This makes it impossible to verify their thickness directly on the samples being processed. For an estimate of silver thickness, microscope slides were placed in the system with the multilayer samples. The microscope slides were patterned with a liftoff technique, and layer thicknesses were measured using a contact profilometer (Tencor Alphastep 200).

Visual inspection of both samples reveals mirror quality surfaces, with only a few defect spots attributed to dust particles contaminating the sample surfaces during processing. Digital photographs taken at normal incidence are shown in Fig. 2. The silver-free sample has a distinctly red color, attributed to a narrow 3^rd order stop band in the 650 nm wavelength region as discussed below. The silver-containing sample is green, attributed to a wider 3^rd order stop band and higher absorption and reflection by the Ag:IG2 glass throughout the visible range.

3. Characterization

The samples were cleaved and inspected by scanning electron microscopy (SEM); a typical SEM image of sample B is shown in Fig. 3(a). Both the silver-containing and silver-free samples exhibit good planarity, with nm scale surface roughness similar to the mirrors reported previously [13]. Excellent adhesion is maintained between films and with the silicon substrate, even upon the thermal and mechanical shock of cleaving samples at liquid nitrogen temperatures. The PAI polymer used is a high performance plastic, known for its adhesive properties and retention of mechanical integrity across a wide range of temperature. We believe this is important for the work described here; underlying PAI layers do not soften and wrinkle (due to stresses arising from coefficient of thermal expansion [CTE] mismatches, etc.) when subsequent PAI layers are cured. On some samples (results not shown) we have cured the PAI layers at peak temperatures of ~300°C, with similar results to those described here. Layer thicknesses estimated from SEM were very close to the targeted values for sample A (~290 nm for PAI and ~150 nm for IG2). For sample B, the Ag:IG2 layers were estimated to be ~130 nm thick. These values were used in the simulations below, producing good agreement between the experimental and simulated spectra. Note that the first PAI layer in Fig. 3(a) appears thinner than the others. However, we were able to obtain much better fits to the experimental spectra by assuming all PAI layers are 290 nm thick. It is possible that the bottom PAI layer stretched on cleaving [13] and only appears thinner in the SEM images.

 

Fig. 3 (a): SEM image of a facet of sample B after cleaving in liquid nitrogen. Good adhesion is evident with no evidence of film delamination or cracking. The overlaid ladder is a guide to the eye, with the distance between orange bars ~100 nm. (b). Normal incidence transmittance of sample A. The green curve is experimental data, and the yellow curve is a simulated curve based on a transfer matrix method and neglecting absorption. (c). Normal incidence spectrophotometer scans of sample B. The red curve is experimental data, and the blue curve is simulated assuming Δn=0.33. Sample B has a broader stopband, as is expected from the increased index of refraction of the Ag:IG2 layers as compared with standard IG2 glass.

Download Full Size | PDF

Optical characterization was conducted in both transmission and reflection, and compared with simulation. Transmission scans were measured at normal incidence using a Perkin-Elmer Lambda-900 UV-Vis-NIR dual beam spectrophotometer. Normalized data is shown in Figs. 3 (b) and (c). Since the substrates are single-side polished silicon, transmittance data is restricted to the wavelength region above ~1100 nm and is affected by scattering at the rough back surface. The spectral shape, including the position and width of the normal incidence stop band, is nevertheless apparent. We stress that the absolute transmittance level (~10% at the peaks) was very similar for both samples. This indicates that the Ag doping of the IG2 layers does not significantly increase their absorbance in the wavelength region below the shifted band edge, in agreement with the literature [16]. Normal incidence spectral stop bands lie between ~1535 nm and ~1950 nm for the silver-free sample, and ~1475 nm and ~2000 nm for the silver-containing sample. Simulations were performed using a standard transfer matrix method, and for simplicity absorption (by both the layers and the substrate) was neglected. The dispersion curves used for IG2 glass and PAI polymer were given in reference [13]. For simulations of the Ag:IG2 layers, a wavelength-independent increase of the refractive index (Δn) was assumed for simplicity (ie. n_{silver-containing}(λ)=n_silver-free(λ)+Δn). This approximation does not account for the fact (predicted by Kramers-Kronig relations) that the index increase is largest near the red-shifted band edge, and monotonically decreases for increasing wavelength [20]. Nevertheless, very good agreement between simulated and experimental spectra was possible using this simplified approximation. For the transmittance scan [Fig. 3(c)], the best fit to the data in the 1200 to 2500 nm range was obtained using Δn~0.33.

Reflection scans were obtained for various incidence angles (20, 34, 48, 62, 76 degrees) and both polarization states (TE, TM) using a variable angle spectroscopic ellipsometer (VASE) instrument (J.A. Woollam Co. VB-250 Ellipsometer Control Module coupled with HS-190 High Speed Monochromator System). This instrument can collect spectral data between 300 nm and 1700nm, with the exception of the window between 1350 nm and 1450 nm. Experimental and simulated curves for sample A are shown in 4(a) and for sample B in Fig. 4(b). Excellent agreement is evident in both cases, except below ~700 nm where the mismatch can be attributed to the neglect of absorption in the simulations. The best fit to the experimental reflection data in Fig. 4(b) was obtained using Δn~0.40. This is slightly larger than the Δn used to fit the transmission scans above, probably because the transmittance data is centered at longer wavelengths. In any case, an index increase of 0.3–0.4 for the atomic silver concentration estimated here (~20 %) is in line with other reports [14–16 ].

A first-order omnidirectional stop band was confirmed for both samples. For sample A, the omnidirectional bandwidth is ~100 nm (~1535 to 1635 nm). For sample B, silver doping increases the omnidirectional bandwidth to ~ 200 nm (~1475 to 1675 nm). These results are in good agreement with theoretical predictions [1]. Sample A has δn₀~1.65 and δn₁~1.55, and the omnidirectional bandwidth (for optimized filling factor) centered at 1550 nm is predicted to be ~120 nm. Sample B has δn₀~1.65 and δn₁~1.76, resulting in an optimized omnidirectional bandwidth of ~250 nm at 1550 nm. Note that because of the restriction on PAI thickness mentioned above, neither structure has an optimal filling factor. For sample A, η=0.34, η_q=0.39 and the optimal filling factor is η0.45 [3]. Similarly, for sample B, η=0.31, η_q=0.36 and the optimal filling factor is η0.40. This suggests that there is room for improvement, assuming a process to spin thinner polymer layers can be developed.

A third order stop band is predicted and was partially verified in some of the reflectance scans of Fig. 4. The silver-free sample A exhibits a peak at η625 nm for 20 degrees incidence angle, in good agreement with simulation. The predicted 3^rd order stop band is relatively suppressed for sample B, likely due to the increased absorption in the visible region produced by photodoping. These spectral features lead to the coloring of the mirrors seen in Fig. 2.

 

Fig. 4. (a). TM (left column) and TE (right column) reflectance curves for the silver-free sample A, for incidence angles of (from top to bottom) 0, 20, 34, 48, 62, and 76 degrees from normal. Red curves are experimental VASE data and blue curves were simulated based on a transfer matrix approach without absorption. See text for layer thicknesses used in simulation. The data agrees very well above the IG2 electronic absorption edge, although experimental data in the NIR stop bands is noisy due to the instability of the VASE light source in that region. An omnidirectional reflection band between ~1535 nm and ~1635 nm is indicated. (b). Same as (a), except for the silver-containing sample B. The Ag:IG2 layers were assumed to have Δn=0.40 in the simulations. The omnidirectional band in this case lies between ~1475 nm and ~1675 nm as indicated by the shading.

Download Full Size | PDF

Finally, we note that such close agreement between experimental and simulated spectra over such a wide range of wavelength and incidence angle is only possible for a structure that exhibits excellent layer uniformity, homogeneity, and planarity. This supports our earlier statement regarding the good thermo-mechanical compatibility of the materials used.

4. Summary and Conclusions

We have demonstrated that the silver doping technique in chalcogenide glasses can be used to increase the omnidirectional stop band of multilayer reflectors based on chalcogenide glass and polymer. These mirrors have the attributes of low-temperature processing, fabrication on a silicon platform, good planarity, and excellent thermal and mechanical stability. The broadened omnidirectional stop band (~1475 nm to 1675 nm) covers much of the spectrum used for fiber optic communications. These mirrors could prove to be useful in the construction of novel waveguides and cavities.