Scalable Fabrication of Nanogratings on GaP for Efﬁcient Diffraction of Near-Infrared Pulses and Enhanced Terahertz Generation by Optical Rectiﬁcation

: We present a process ﬂow for wafer-scale fabrication of a surface phase grating with sub-micron feature sizes from a single semiconductor material. We demonstrate this technique using a 110-oriented GaP semiconductor wafer with second-order nonlinearity to obtain a nanostructured device (800 nm lateral feature size and a 245 nm height modulation) with applications relevant to near-infrared optical diffraction and time-resolved terahertz (THz) technologies. The fabrication process involves a plasma-enhanced chemical deposition of a SiO 2 layer on the wafer followed by contact photolithography and inductively coupled plasma reactive ion etching (ICP-RIE). We discuss the required radiation dosage, exposure times, temperatures and other key parameters to achieve high-quality nanogratings in terms of ﬁlling ratio, edge proﬁle, and overall shape. The phase-grating properties, such as the pitch, spatial homogeneity, and phase retardation, are characterized with an atomic force microscope, scanning electron microscope and a non-invasive optical evaluation of the optical diffraction efﬁciency into different orders. We demonstrate an application of this device in a time-domain THz spectroscopy scheme, where an enhanced THz spectral bandwidth is achieved by optical rectiﬁcation of near-infrared laser pulses incident on the grating and efﬁciently diffracted into the ﬁrst orders. Finally, the reported process ﬂow has the potential to be applied to various materials by considering only slight adjustments to the ICP-RIE etching steps, paving the way to scalable fabrication of sub-micron patterns on a large range of substrates. RIE process, while also eliminating the photosensitivity of the PR. This allows for the sample to remain unaffected when exposed to visible light, thus allowing for characterization with an optical microscope. selectivities using a SiO 2 patterned layer on GaAs [42] and Si [43] substrates,


Introduction
The development of cost-effective techniques to produce nanodevices has led to a number of world-changing technologies such as semiconductor-integrated circuits (IC) [1], high-density magnetic storage [2], and energy-harvesting devices [3,4]. Electron beam lithography (EBL) is a common fabrication technique used by researchers to produce unique structures and devices with nanometer precision. Many recent advances in the field of photonics have notably relied on such techniques [5][6][7][8]. However, EBL is a time-consuming process with limited scalability for wafer-scale production [9]. Meanwhile, the photolithography process is a high-throughput fabrication technique suitable for academic and industrial applications requiring large sample sizes or high-volume manufacturing [10,11]. The technique is an image-transferring process using UV exposure through a mask to imprint precise patterns onto a layer of photosensitive polymer, called a photoresist (PR). This technique can be used in projection, proximity or contact modes. The projection mode provides the best pattern-writing accuracy, with a minimum feature size of about 25 nm [12], but at the expense of costly instrumentation which significantly limits its accessibility. Proximity photolithography is a cost-effective nanofabrication method where the photomask We first use a standard three-step cleaning process to remove any surface residues on the GaP wafer: (i) an acetone ultrasonic bath; (ii) an isopropyl alcohol (IPA) ultrasonic bath; and (iii) a deionized (DI) water rinse. After drying the wafer with a nitrogen blower, a 150 nm SiO2 layer is homogeneously deposited on the entire wafer surface using a plasma-enhanced chemical vapor deposition (PECVD) method. As briefly mentioned above, the SiO2 layer is etched and then acts as a mask while inductively coupled plasmareactive ion etching (ICP-RIE) is used to locally remove the GaP material. We adopted this strategy as SiO2 exhibits a high selectivity to standard RIE processes used to etch III-V semiconductors, such as GaP. Using BCl3 gas, we observe a 3:1 etching selectivity for SiO2 on GaP. Before depositing the PR layer, the sample is first exposed to hexa-methyldisilazane (HMDS) inside a dedicated oven for 5 min at 90 °C, which creates an adhesion promoter layer between the SiO2 and the PR. The SiO2 layer forms polar OH bonds when exposed to the atmospheric humidity, which results in a hydrophilic surface. This causes low adhesion between the surface and low polar materials such as a photoresist. In order We first use a standard three-step cleaning process to remove any surface residues on the GaP wafer: (i) an acetone ultrasonic bath; (ii) an isopropyl alcohol (IPA) ultrasonic bath; and (iii) a deionized (DI) water rinse. After drying the wafer with a nitrogen blower, a 150 nm SiO 2 layer is homogeneously deposited on the entire wafer surface using a plasmaenhanced chemical vapor deposition (PECVD) method. As briefly mentioned above, the SiO 2 layer is etched and then acts as a mask while inductively coupled plasma-reactive ion etching (ICP-RIE) is used to locally remove the GaP material. We adopted this strategy as SiO 2 exhibits a high selectivity to standard RIE processes used to etch III-V semiconductors, such as GaP. Using BCl 3 gas, we observe a 3:1 etching selectivity for SiO 2 on GaP. Before depositing the PR layer, the sample is first exposed to hexa-methyldisilazane (HMDS) inside a dedicated oven for 5 min at 90 • C, which creates an adhesion promoter layer between the SiO 2 and the PR. The SiO 2 layer forms polar OH bonds when exposed to the atmospheric humidity, which results in a hydrophilic surface. This causes low adhesion between the surface and low polar materials such as a photoresist. In order to make the surface water repellent and PR-attractive, non-polar molecules of an adhesion promoter such as HMDS can be used [39]. Spin coating deposition of the PR (MEGAPOSIT SPR 955) is performed Crystals 2022, 12, 684 4 of 10 at high rotation speed (up to 6000 rpm) to reduce the thickness and therefore optimize the pattern resolution [13,40]. We use five steps to avoid the presence of microbubbles and optimize the thickness uniformity: (1) 15 s at 200 rpm/s, increasing the speed from 0 to 3000 rpm; (2) 30 s at 3000 rpm; (3) 15 s at 200 rpm/s; (4) 30 s at 6000 rpm; (5) 30 s at −200 rpm/s, reducing the speed from 6000 rpm to 0 rpm. We obtain a 600 nm-thick positive PR approaching the minimum thickness of 500 nm specified by the manufacturer [9]. We then let the device rest for 1 min and perform a prebake for 1 min at 100 • C on a hotplate.
An Optical Associate Inc. (OAI) 200IR mask-aligner combined with two i-line filters performs UV exposure at a wavelength of 365 nm with a uniform spatial intensity distribution of 10.15 mW/cm 2 . Direct contact between the photomask and the substrate is used to minimize pattern deformation due to diffraction effects [40]. UV light reflected by the substrate may still induce an optical standing wave resulting from interference effects, which affects the spatial homogeneity of the structure in the vertical direction. To reduce the standing wave effect, a post-exposure bake is performed at 115 • C on a hotplate for 2 min [41]. During the PR development, the exposed regions of the photoresist are chemically etched with CD-26. This step is immediately followed by DI rinsing and nitrogen drying. We first tested a dosage of 90 mJ/cm 2 and a 45 s development time, but we observed over-etching of the grating height while the photoresist did not develop properly in the exposed areas. Figure 2 compares this result to a different process yielding a square PR grating with sharp edges, which consists of a dosage of 145 mJ/cm 2 and a development time of 12 s. A hard bake performed at 135 • C on a hotplate for 10 min is then used to increase the resistance of the photoresist for the up-coming RIE process, while also eliminating the photosensitivity of the PR. This allows for the sample to remain unaffected when exposed to visible light, thus allowing for characterization with an optical microscope.
to make the surface water repellent and PR-attractive, non-polar molecules of an adhesion promoter such as HMDS can be used [39]. Spin coating deposition of the PR (MEGAPOSIT SPR 955) is performed at high rotation speed (up to 6000 rpm) to reduce the thickness and therefore optimize the pattern resolution [13,40]. We use five steps to avoid the presence of microbubbles and optimize the thickness uniformity: (1) 15 s at 200 rpm/s, increasing the speed from 0 to 3000 rpm; (2) 30 s at 3000 rpm; (3) 15 s at 200 rpm/s; (4) 30 s at 6000 rpm; (5) 30 s at −200 rpm/s, reducing the speed from 6000 rpm to 0 rpm. We obtain a 600 nm-thick positive PR approaching the minimum thickness of 500 nm specified by the manufacturer [9]. We then let the device rest for 1 min and perform a prebake for 1 min at 100 °C on a hotplate.
An Optical Associate Inc. (OAI) 200IR mask-aligner combined with two i-line filters performs UV exposure at a wavelength of 365 nm with a uniform spatial intensity distribution of 10.15 mW/cm 2 . Direct contact between the photomask and the substrate is used to minimize pattern deformation due to diffraction effects [40]. UV light reflected by the substrate may still induce an optical standing wave resulting from interference effects, which affects the spatial homogeneity of the structure in the vertical direction. To reduce the standing wave effect, a post-exposure bake is performed at 115 °C on a hotplate for 2 min [41]. During the PR development, the exposed regions of the photoresist are chemically etched with CD-26. This step is immediately followed by DI rinsing and nitrogen drying. We first tested a dosage of 90 mJ/cm 2 and a 45 s development time, but we observed over-etching of the grating height while the photoresist did not develop properly in the exposed areas. Figure 2 compares this result to a different process yielding a square PR grating with sharp edges, which consists of a dosage of 145 mJ/cm 2 and a development time of 12 s. A hard bake performed at 135 °C on a hotplate for 10 min is then used to increase the resistance of the photoresist for the up-coming RIE process, while also eliminating the photosensitivity of the PR. This allows for the sample to remain unaffected when exposed to visible light, thus allowing for characterization with an optical microscope. In the next series of steps, the exposed areas of the SiO2 layer are etched using a SAMCO 10-NR RIE system. Hydrofluoric acid (HF) is often used to chemically etch SiO2, but here we implement the dry RIE method to avoid any potential reactions between HF and the substrate. We use a RIE power of 100 W, a CF4 gas flow of 40 standard cubic centimeters per minute (SCCM), and a chamber pressure of 3 Pa resulting in an etch rate of 30 nm/min for SiO2 and 110 nm/min for the SPR 955. A 5 min etching time is required to expose selected regions of the GaP substrate. Any leftover PR is removed with a <1 min In the next series of steps, the exposed areas of the SiO 2 layer are etched using a SAMCO 10-NR RIE system. Hydrofluoric acid (HF) is often used to chemically etch SiO 2 , but here we implement the dry RIE method to avoid any potential reactions between HF and the substrate. We use a RIE power of 100 W, a CF 4 gas flow of 40 standard cubic centimeters per minute (SCCM), and a chamber pressure of 3 Pa resulting in an etch rate of 30 nm/min for SiO 2 and 110 nm/min for the SPR 955. A 5 min etching time is required to expose selected regions of the GaP substrate. Any leftover PR is removed with a <1 min RIE step using O 2 . The fabricated structure is characterized using atomic force microscopy (AFM) and scanning electron microscopy (SEM) to verify the successful transfer of the photomask pattern onto the SiO 2 layer with high accuracy (Figure 3a,b).
RIE step using O2. The fabricated structure is characterized using atomic force microscopy (AFM) and scanning electron microscopy (SEM) to verify the successful transfer of the photomask pattern onto the SiO2 layer with high accuracy (Figure 3a,b). The ICP-RIE method has been used to achieve isotropic and highly selective etching processes on semiconductors. Previous work has been reported on 8:51 and 4:10 etching selectivities using a SiO2 patterned layer on GaAs [42] and Si [43] substrates, respectively. Here, we apply this etching method with the SAMCO 110ip system using the following parameters: ICP power of 50 W, bias power of 25 W, a BCl3 gas flow of 2.5 SCCM and a chamber pressure of 0.15 Pa, resulting in a 56.5 nm/min etch rate for GaP. After a 4.2 min etch, the remaining SiO2 mask is removed by inserting the sample inside of a SAMCO 10-NR system for 3 min using a power of 100 W, a CF4 gas flow of 40 SCCM and a chamber pressure of 3 Pa. An SEM image of the resulting GaP nanogratings with periodicity of 1600 nm, width of 800 nm, length of 8 mm and height of 245 nm is shown in Figure 4. Finally the wafer is diced to 1 cm × 1 cm samples using a MTI corporation SYJ-800 dicing system allowing for easy integration into a time-resolved THz spectroscopy system.  The ICP-RIE method has been used to achieve isotropic and highly selective etching processes on semiconductors. Previous work has been reported on 8:51 and 4:10 etching selectivities using a SiO 2 patterned layer on GaAs [42] and Si [43] substrates, respectively. Here, we apply this etching method with the SAMCO 110ip system using the following parameters: ICP power of 50 W, bias power of 25 W, a BCl 3 gas flow of 2.5 SCCM and a chamber pressure of 0.15 Pa, resulting in a 56.5 nm/min etch rate for GaP. After a 4.2 min etch, the remaining SiO 2 mask is removed by inserting the sample inside of a SAMCO 10-NR system for 3 min using a power of 100 W, a CF 4 gas flow of 40 SCCM and a chamber pressure of 3 Pa. An SEM image of the resulting GaP nanogratings with periodicity of 1600 nm, width of 800 nm, length of 8 mm and height of 245 nm is shown in Figure 4. Finally the wafer is diced to 1 cm × 1 cm samples using a MTI corporation SYJ-800 dicing system allowing for easy integration into a time-resolved THz spectroscopy system. RIE step using O2. The fabricated structure is characterized using atomic force microscopy (AFM) and scanning electron microscopy (SEM) to verify the successful transfer of the photomask pattern onto the SiO2 layer with high accuracy (Figure 3a,b). The ICP-RIE method has been used to achieve isotropic and highly selective etching processes on semiconductors. Previous work has been reported on 8:51 and 4:10 etching selectivities using a SiO2 patterned layer on GaAs [42] and Si [43] substrates, respectively. Here, we apply this etching method with the SAMCO 110ip system using the following parameters: ICP power of 50 W, bias power of 25 W, a BCl3 gas flow of 2.5 SCCM and a chamber pressure of 0.15 Pa, resulting in a 56.5 nm/min etch rate for GaP. After a 4.2 min etch, the remaining SiO2 mask is removed by inserting the sample inside of a SAMCO 10-NR system for 3 min using a power of 100 W, a CF4 gas flow of 40 SCCM and a chamber pressure of 3 Pa. An SEM image of the resulting GaP nanogratings with periodicity of 1600 nm, width of 800 nm, length of 8 mm and height of 245 nm is shown in Figure 4. Finally the wafer is diced to 1 cm × 1 cm samples using a MTI corporation SYJ-800 dicing system allowing for easy integration into a time-resolved THz spectroscopy system.   Table 1 summarizes the details of our phase-grating fabrication process on a single GaP crystal with sub-micron features. Since the photomask pattern is first transferred to a SiO 2 layer deposited on GaP before ICP-RIE, the steps presented in this work could be Crystals 2022, 12, 684 6 of 10 applied to a range of different substrates. Only modifications to the ICP-RIE parameters would then be required.

Optical and Terahertz Measurement
Phase-locked THz transients can be generated by optical rectification of ultrafast NIR pulses propagating inside of a second-order nonlinear crystal when phase-matching conditions are satisfied [44]. Generally, experimentalists must rely on a limited number of crystals displaying proper linear and nonlinear properties for efficient THz generation. Recently, configurations relying on a grating at the surface of the nonlinear crystal were proposed to achieve a non-collinear configuration between a diffracted NIR pulse and a generated THz wave [45][46][47]. In this scheme, the grating pitch, which determines the angles of diffraction, can be adjusted to optimize phase-matching conditions, which are otherwise fixed by the material optical properties. Recent experiments have used this configuration to demonstrate a higher THz generation efficiency at specific THz frequencies and in some cases a larger THz spectral bandwidth [6,48,49]. Those studies relied on e-beam lithography for the fabrication of the phase grating, or used a periodically patterned oxide layer as a diffractive element on the surface of the nonlinear crystal. A related optical geometry proposed by reference [50,51] relies on a simple optical configuration where the incoming NIR pulse is at normal incidence to the crystal while both positive and negative diffracted orders contribute to the generation of THz radiation. However, this scheme depends on a null zero-order phase grating made from a single crystal material to take advantage of a symmetrical redistribution of the transmitted optical energy into the two first orders of diffraction (m = −1, +1). The phase grating described in this work is designed for such a scheme and is fabricated on the surface of a GaP wafer, since this material offers advantageous phase matching conditions and a relatively high second-order nonlinear coefficient for efficient THz generation. In this section, we evaluate the grating performance using a standard time-resolved THz configuration using 60 fs NIR pulses at a repetition rate of 1.1 MHz and centered at 1030 nm to generate the THz wave by optical rectification [37]. When such an optical source is focused into a 110-oriented 450 µm-thick GaP crystal, THz is emitted with an effective spectral bandwidth limited to <3 THz frequencies due to phase matching conditions. Using the grating parameters of a 1600 nm period and a filling factor of 1 / 2 , the resulting non-collinear geometry can allow for efficient THz generation around 4 THz [6].
The target height of the phase grating in GaP is 245 nm to achieve a π phase shift between the sections of the beam passing through the crests and valleys of the grating, Crystals 2022, 12, 684 7 of 10 therefore resulting in destructive interference in the transmitted 0th order of diffraction. We first characterize the performance of the diffractive optical device by observing the transmitted diffraction orders. As illustrated in Figure 5, the total power in the transmitted 0th order was measured to be approximately 4% of the incident light, while 44% was found in the 1st diffraction orders (equally distributed among m = −1 and +1). This value approaches the theoretical maximum [52] considering that 46% of the incident power is reflected at both GaP interfaces and 9% of the incident power is distributed to the 3rd diffraction orders. Our finite-difference time-domain (FDTD) simulations indicate that at the GaP front surface, the phase grating suppresses the NIR reflection by 5% when compared to a flat incident surface. Additionally, at the back surface of the crystal, there are power losses due to the Fresnel reflection. Considering the reflections of the diffracted orders inside of the crystal, we calculate the component of the transmitted power in the pure 0th order to be approximately 0.4 mW, indicating a high-quality null zero-order phase grating. We measured the first diffracted orders at an angle of 40 • relative to the normal, which confirms the period of the grating to be 1600 nm. The grating is oriented along the 110 crystal axis, or perpendicular to the 110 and 001 axes. When the incident pulse is polarized along this direction, the diffraction does not affect the projection of the optical field components onto the non-zero second-order coefficients of the crystal. This orientation ensures the same THz generation efficiency with and without the presence of the phase grating.
tification [37]. When such an optical source is focused into a 110-oriented 450 µm-thick GaP crystal, THz is emitted with an effective spectral bandwidth limited to <3 THz frequencies due to phase matching conditions. Using the grating parameters of a 1600 nm period and a filling factor of ½, the resulting non-collinear geometry can allow for efficient THz generation around 4 THz [6].
The target height of the phase grating in GaP is 245 nm to achieve a π phase shift between the sections of the beam passing through the crests and valleys of the grating, therefore resulting in destructive interference in the transmitted 0th order of diffraction. We first characterize the performance of the diffractive optical device by observing the transmitted diffraction orders. As illustrated in Figure 5, the total power in the transmitted 0th order was measured to be approximately 4% of the incident light, while 44% was found in the 1st diffraction orders (equally distributed among m = −1 and +1). This value approaches the theoretical maximum [52] considering that 46% of the incident power is reflected at both GaP interfaces and 9% of the incident power is distributed to the 3 rd diffraction orders. Our finite-difference time-domain (FDTD) simulations indicate that at the GaP front surface, the phase grating suppresses the NIR reflection by 5% when compared to a flat incident surface. Additionally, at the back surface of the crystal, there are power losses due to the Fresnel reflection. Considering the reflections of the diffracted orders inside of the crystal, we calculate the component of the transmitted power in the pure 0th order to be approximately 0.4 mW, indicating a high-quality null zero-order phase grating. We measured the first diffracted orders at an angle of 40° relative to the normal, which confirms the period of the grating to be 1600 nm. The grating is oriented along the 11 0 crystal axis, or perpendicular to the 110 and 001 axes. When the incident pulse is polarized along this direction, the diffraction does not affect the projection of the optical field components onto the non-zero second-order coefficients of the crystal. This orientation ensures the same THz generation efficiency with and without the presence of the phase grating. The time-resolved THz field amplitude generated with our grating device was measured and compared to the THz field amplitude generated inside of an unpatterned window from the same GaP wafer. In both cases, THz transients are detected with a 200 µm thick GaP crystal by standard electro-optic sampling. As shown in Figure 6, both signals The time-resolved THz field amplitude generated with our grating device was measured and compared to the THz field amplitude generated inside of an unpatterned window from the same GaP wafer. In both cases, THz transients are detected with a 200 µm thick GaP crystal by standard electro-optic sampling. As shown in Figure 6, both signals have comparable peak field values in the time domain. However, their respective spectral power, obtained by squaring the Fourier transform of the time-domain data, show significant differences. The geometry based on a standard unpatterned crystal (black line) shows a THz spectrum peaking around 2 THz with a bandwidth of 2 THz while the geometry which implements the phase-grating device peaks at 3.5 THz and has a bandwidth of 3 THz. Most notably, we observe that almost half of the THz power is found at frequencies above 3 THz when the non-collinear geometry is used. The resulting spectral broadening is due to the diffraction of the incident optical pulse on the grating, which allows for the phase matching conditions to be satisfied at 4 THz [5,6]. Finally, we note that the focused NIR pulse in our experiment, with a relatively high peak intensity of 3500 GW/cm 2 , did not damage or degrade the performance of the surface grating. ometry which implements the phase-grating device peaks at 3.5 THz and has a bandwidth of 3 THz. Most notably, we observe that almost half of the THz power is found at frequencies above 3 THz when the non-collinear geometry is used. The resulting spectral broadening is due to the diffraction of the incident optical pulse on the grating, which allows for the phase matching conditions to be satisfied at 4 THz [5,6]. Finally, we note that the focused NIR pulse in our experiment, with a relatively high peak intensity of 3500 GW/cm 2 , did not damage or degrade the performance of the surface grating.

Conclusions
We report on a process flow for wafer-scale fabrication of phase gratings in singlecrystal GaP with sub-micron features. The concept is based on PECVD deposition of a SiO2 layer, i-line contact photolithography and RIE to transfer the pattern of a photomask to this layer, and finally ICP-RIE to etch the semiconductor. The contact i-Line photolithography method is used to transfer the nanograting image onto a GaP substrate. This method is well known for low-cost microfabrication processes and we have optimized the process parameters to reach nanoscale dimensions on a 2-inch wafer. SEM images confirm that the fabricated grating has the specified dimensions: a period of 1600 nm, a width of 800 nm and a height of 245 nm. These dimensions are also confirmed by optical diffraction experiments. We notably measure 44% of the incident optical power in the first orders of diffraction while less than 1% is in the 0th order. Furthermore, we observe that gratingassisted non-collinear optical rectification in GaP results in efficient THz generation at frequencies above 3 THz, a region generally difficult to access with time-resolved THz systems. We believe that our fabrication technique is directly applicable to a large range of materials by simply fine-tuning the last ICP-RIE procedure. Our scalable and cost-effective fabrication process relying on contact photolithography could be used for mass production of photonics and electronic devices with sub-micron features. The adoption of these types of photonic devices in the field of THz photonics will enable various nonlinear crystals to be used in a non-collinear geometry for efficient generation of THz waves at selected frequencies.

Conclusions
We report on a process flow for wafer-scale fabrication of phase gratings in singlecrystal GaP with sub-micron features. The concept is based on PECVD deposition of a SiO 2 layer, i-line contact photolithography and RIE to transfer the pattern of a photomask to this layer, and finally ICP-RIE to etch the semiconductor. The contact i-Line photolithography method is used to transfer the nanograting image onto a GaP substrate. This method is well known for low-cost microfabrication processes and we have optimized the process parameters to reach nanoscale dimensions on a 2-inch wafer. SEM images confirm that the fabricated grating has the specified dimensions: a period of 1600 nm, a width of 800 nm and a height of 245 nm. These dimensions are also confirmed by optical diffraction experiments. We notably measure 44% of the incident optical power in the first orders of diffraction while less than 1% is in the 0th order. Furthermore, we observe that grating-assisted non-collinear optical rectification in GaP results in efficient THz generation at frequencies above 3 THz, a region generally difficult to access with time-resolved THz systems. We believe that our fabrication technique is directly applicable to a large range of materials by simply fine-tuning the last ICP-RIE procedure. Our scalable and cost-effective fabrication process relying on contact photolithography could be used for mass production of photonics and electronic devices with sub-micron features. The adoption of these types of photonic devices in the field of THz photonics will enable various nonlinear crystals to be used in a non-collinear geometry for efficient generation of THz waves at selected frequencies.