For large patterned features (>50 μm), the photolithographic process was based on ProLIFT 100–16 applied under standard spin conditions (6,000 rpm/40 s), resulting in a ProLIFT thickness of 350 nm [9
]. Figure shows an SEM micrograph of a sample fabricated by the multilayer patterning and re-anodization process. A layer of 5.65-μm thick high porosity (81%) porous silicon is initially formed and then patterned using CH4
in a RIE. The patterned PS layer shown here is slightly rounded due to minor deformation of the photoresist during RIE etching.
A cross-sectional SEM image (beam voltage of 0.5 kV) of a double-layer PS film. The first high-porosity layer was etched by ICP-RIE followed by low porosity film growth after the etch.
After the patterning of the first PS layer (P1), the sample was anodized again in the HF/ethanol solution with current density of 5 mA/cm2 for 15 min. The resultant second layer PS (P2) had a thickness of 3.6 μm and a porosity of 77%. The top interface of the P2 layer appears lower by around 150 nm (limited by the image resolution in Figure ) in the region where no P1 layer is present, which was attributed to a slight over etch of the RIE into the silicon. A similar step of 150 nm is observed at the interface between the P2 layer and the silicon; however, the P2-layer thickness appears uniform throughout. This step in the P2-layer/Si interface was attributed to the transfer of the RIE surface profile into the underlying layer during anodization.
Previous observations have shown that the etch rate slows down by 2% to 3%/μm, and porosity gradients of around 5%/μm occur during anodization of lightly doped silicon [15
]. These changes in etch rate and porosity occur at high current densities (typically >100 mA/cm2
] due to HF diffusion through the pores. At the low current densities and relatively thin layers used in this work, the change in etch rate and porosity due to HF diffusion is negligible. Another mechanism that could affect uniform porosity and etch rates is spatial current density variation in the P2
layer caused by the patterned P1
layer. Where the P1
layer porosity is low, or the layer very thick, spatial variation of the current density may become significant. However, since the conductivity of the HF is significantly greater than the carrier depleted, high porosity P1
layer shown in Figure , the potential at the Si-electrolyte surface is unaffected by the P1
layer. In the conditions used for anodization in this work, patterning of the P1
layer did not visibly affect the uniformity of the porosity or thickness in the P2
layer. Accurate control of the RIE process was the key in achieving the high level of interface flatness. With due consideration of the issues detailed above, multiple layers of porous films could be created underneath patterned structures with negligible loading effects from the pattern layers above them. Uniform porosity and interface layers are extremely important for high-quality photonic sensors and cannot be achieved by stamping [5
], patterning of silicon followed by anodization [6
], or masking of silicon followed by anodization [17
Patterning of the 2-μm grating features demonstrated in this work initially failed because all the ProLIFT, including that under the photoresist, dissolved during developing due to the small dimensions of the mask. As feature sizes decrease, the allowable undercut must also be reduced to avoid delamination of the photoresist. To overcome this difficulty, ProLIFT 100–16 (16% solid) was diluted using NMP to ProLIFT 100–7.6 (7.6% solid) for the single layer PS grating and to ProLIFT 100–10.6 (10.6% solid) for the double layer PS grating, so as to be just sufficient to fill in all the pores. The significantly reduced thickness of the ProLIFT layer resulted in successful patterning of the P1 layer.
An SEM image of the fabricated PS diffraction gratings is shown in Figure . The single-layer grating is shown in Figure a, having a measured height of 500 nm, close to the design value of 512 nm. The patterned PS features have reproduced the grating mask extremely well. Feature definition for the double-layer grating shown in Figure b is slightly better than the single-layer grating; however, undercut during processing reduced the ridge/groove ratio of the grating from 50% (as designed) to 35%. Layer heights are very close to designed values. The grating optimized height (h1
) is considerably larger than that reported using imprinting techniques [5
]. These SEM images and corresponding height measurements validate the accuracy of the process to achieve the design specification, given that the samples have undergone thermal annealing, up to two anodizations, photolithography, and dry etching processes. All gratings show good quality diffraction of light under visual inspection.
Figure 4 A cross-sectional SEM image (beam voltage of 0.5 kV) of PS diffraction gratings. Gratings were fabricated with a pitch of 4 μm; (a) single-layer grating of height h1 = 500 nm and (b) double-layer grating of heights h1 = 890 nm and h2 = 270 nm. (more ...)
Although designed for optical operation at 1,565 μm, initial diffraction measurements for the PS diffraction grating were obtained by reflecting a visible laser (HeNe laser at 632 nm) from the surface of the single-layer grating for comparison. The grating was angled at α = 4.5° relative to the incident laser beam to ensure that the 0th-order power could be measured. A large 0th-order power is expected at λ = 632 nm as the grating was not optimized at this wavelength. The measured reflected visible light diffraction spectrum is shown in Figure . The inset shows the light reflected off the sample and onto a screen (at α = 0°), where the central hole in the screen allowed the incident light to pass. The angular locations of the diffraction peaks obey the diffraction grating equation given in Equation 1. Both the inset and the measured data show considerable energy spread between each diffraction order which is not explained using the diffraction equation. These side lobes are largely attributed to the grating shape, non-optimum grating height (at λ = 632 nm), surface roughness, and interferometric effects from the Si interface. For sensing applications, scattering and optical interference lead to crosstalk between detectors which are designed to detect the power of specific orders as illustrated in Figure .
Reflection measurement of the diffraction orders at λ = 632 nm from the single-layer grating. Inset shows the image of the reflected orders on a screen with the incident light and 0th order passing through the same aperture.
Measurements at 1,565 nm of the diffraction efficiency transmitted through the porous grating and substrate (α = 4.5°) are shown in Figure a,b for the single- and double-layer gratings, respectively. The inset shows the same data on a linear scale. Key features of this data are the relatively few diffraction orders (compared to Figure ), high transmission of >40%, and low scattering resulting in a background noise in the order of 10−5, which is 4 orders of magnitude lower than the 0th-order power. The high transmission is attributed to the optimized grating height which can be achieved using our processing techniques. The scattering losses are much lower at a wavelength of 1,565 nm compared to 632 nm, leading to significantly reduced noise where no diffracted orders are present. Both diffraction gratings produced similar diffraction order efficiencies; however, the background noise level at −50 dB for the single-layer grating showed fine scale structure not present in the double-layer grating. This may have been due to interferometric effects between the grating and the PS/silicon interface.
Figure 6 Measured transmission data (logarithmic scale) of the diffraction orders at λ = 1,565 nm. (a) Single-layer grating and (b) double-layer grating. The numbers next to the transmission peaks indicate the diffraction order. The inset in both plots (more ...)
In our grating, the top layer was optimized to suppress the 0th-order transmission by choosing the top layer height as [8
However, our patterned PS acts as both a phase and amplitude grating, and is formed on a high-index substrate which results in significant perturbation of the optical field through the structure. Such a complex, asymmetrical-layered structure does not obey simple models which predict 0th-order suppression with a 50% duty cycle grating [8
]. The purpose of forming the P2
layer was to reduce reflections from the PS/silicon layer and improve the 0th-order suppression in transmission. This can be achieved using a layer of thickness:
The λ/4 P2 layer introduces a π phase shift in the reflections from the interfaces either side of the P2 layer, suppressing the effect of these reflections in the transmitted beam (similar to an antireflection coating). By designing the layers as described by Equations 2 and 3, the 0th-order transmission through the grating can largely be suppressed.
A detailed understanding of this complex system requires a more thorough analysis. To understand the variation of diffraction efficiency with analyte induced index changes, a model based on the rigorous coupled mode theory [11
] was evaluated by varying the index of the porous layers. The results are shown in Figure , as a function of the grating index layer and assuming no back reflection from the substrate backside. Comparing the model to the measured transmission in Figure a, the diffraction efficiencies of the single-layer grating matches well at the designed grating index layer of n1
= 1.78. For this grating, the diffraction efficiency variation with index shows several discontinuities which were attributed to the interferometric effects. Such changes are undesirable in a sensor where a monotonic response over the measurement range is required. For this grating, the 0th order is not suppressed for either the measurement or the model as predicted by Equation 1; this was due to the initial error in the design of the height (h1
) of the single-layer grating. Nevertheless, good agreement with measurement validated the model and will enable subsequent optimization.
Figure 7 Modeled 0th and 1st-order diffraction efficiencies as the index of the layer(s) is changed. The single layer is drawn with a circle marker and the double layer gratings as a dashed line. Modeling was performed in transmission at a wavelength of λ (more ...)
For the case of the double layer grating modeled in Figure , the P2 layer index was changed by the same amount as the grating P1 layer, which would occur when an analyte infiltrates both layers. The suppression of the 0th order is evident in the model at a refractive index of n1 = 1.78. However, the measured 0th-order diffraction efficiency for the double-layer grating shown in Figure b indicates that the 0th-order efficiency is higher than the 1st-order mode. The model in Figure indicates this occurs at low-grating index values, suggesting that our estimated index for the fabricated layers is lower than expected. Nevertheless, the diffraction efficiency for the double-layer grating has a deeper 0th-order extinction and has a smoother transmission as index changes compared with the single-layer grating. Over an index range of 1.67 to 1.92, a minimum of 1 dB change in 0th-order diffraction efficiency occurs for a 1% change in normalize index change (Δn/n) - near the minimum of the transmission, up to 6 dB change for a 1% change in Δn/n occurs. These results indicate that a high extinction of the 0th order is important in achieving high sensitivity to changes in the refractive index. Separate modeling indicated that at a ridge/groove ratio of 39%, the minimum 0th-order diffraction efficiency for the double-layer grating, is reduced to more than 10−4, showing the importance of accurate patterning.
To test the performance of our double-layer grating as a sensor, we simultaneously measured the 0th-order and 1st-order powers in transmission through the grating and substrate using the setup illustrated in Figure at a wavelength of λ = 1,565 nm. The change in diffraction efficiency is shown in Figure , as isopropanol vapor was introduced to the surface at a concentration of 1,000 g/m3. Significant and rapid change in the diffraction efficiency was recorded as the vapor infiltrated the pores, resulting in the 1st-order diffraction efficiency becoming dominant, while the 0th order was suppressed. We were able to repeat these results many times, which demonstrated the reproducibility of the measurement. The results are consistent with the model for the double-layer grating shown by the shaded region in Figure , assuming the initial grating index was n1 = 1.53, and the index of the grating increased by Δn = 0.11 ± 0.04. This grating index is 14% lower than the design value of n1 = 1.78. The second anodization step is believed to be responsible for a 5% reduction in the index of the grating P1 layer, while the rest may have resulted from processing issues or film characterization errors requiring further investigation. The expected increase in grating index from the vapor, assuming pores saturated with isopropanol, is Δn = 0.3, suggesting only 35% saturation of the film by the vapor. The change in 0th-order and 1st-order efficiencies occurred within 30 s, largely due to the thin PS layers which the vapor needed to diffuse through to affect the refractive index within the pores. As the vapor dissipated (7 to 10 min), the diffraction efficiencies returned to their original values, indicating that the grating response was reversible and predictable. The steps observed in the data are a result of the sampled data quantization noise. An important feature of this highly sensitive nano-porous sensor is the complementary change in the 0th-order and 1st-order diffraction efficiencies. By measuring the difference between 0th-order and 1st-order diffraction efficiencies, the measurement sensitivity is improved, while common mode laser source noise is eliminated.
Measured 0th-order and 1st-order diffraction efficiencies at λ = 1,565 nm after isopropanol vapor was introduced.
Improvements to the performance are expected from further modeling and design optimization. For the double-layer PS gratings with a high-index under-layer (n2
), an antireflection layer could be formed under the grating [18
] to increase the depth of observed null in the 0th order (Figure ). High Q
resonant waveguides could be fabricated using our methods to significantly enhance detection in PS sensors. For example, the fabrication of patterned features over layers of uniform index and thickness is a key to enable the formation of low loss layers required for resonant grating waveguides [19
] and grating coupler waveguides [20
]. While operating in the IR provides many advantages, one issue to contend with is the coherent interference that results when using a polished backside substrate. This interference is most predominant when the transmitted or reflected light is near an intensity minimum. Our modeling results indicate that the intensity dip in the transmitted 0th order shown in Figure can vary from 10−2
as a result of coherent substrate reflections if the sample angle changes by as little as 0.1° relative to the incident light. Such uncontrolled variation could lead to significant errors in detection. These effects can be mitigated by either using broadband incoherent light sources or backside PS antireflection coatings which we have previously demonstrated [14
In the sensors demonstrated here, the nanoscale pores within the films have been engineered in 2D, with the addition of highly uniform, nanometer-thick layers, and high-resolution microscale patterning of the films. These capabilities allow large, micrometer-sized cell and particle trapping between the gratings, while smaller nanometer-sized proteins and analytes could be captured and detected within the pores. The techniques described provide a path to combine both chemical [21
] and physical [22
] sensing in a single platform. Our process is capable of producing submicron features given suitable processing tools.