Nanocrystalline Diamond-Glass Platform for the Development of Three-Dimensional Micro- and Nanodevices

Low-cost and robust platforms are key for the development of next-generation 3D microand nanodevices. To fabricate such platforms, nanocrystalline diamond (NCD) is a highly appealing material due to its biocompatibility, robustness, and mechanical, electrical, electrochemical, and optical properties, while glass substrates with through vias are ideal interposers for 3D integration due to the excellent properties of glass. However, developing devices that are comprised of NCD films and through glass vias (TGVs) has rarely been attempted due to a lack of effective process strategies. In this work, a low-cost process, free of photolithography and transfer-printing, for fabricating arrays of TGVs that are sealed with suspended portions of an ultra-thin NCD film on one side is presented. These highly transparent structures may serve as a platform for the development of microwells for single-cell culture and analysis, 3D integrated devices such as microelectrodes, and quantum technologies. The process is demonstrated by fabricating TGVs that are sealed with an NCD film of thickness 175 nm and diameter 60 μm. The technology described can be extended by replacing NCD with nanocrystalline silicon or silicon carbide, allowing for the development of complex heterogenous structures on the small scale.


Introduction
In this work, we promote the potential advantages of an nanocrystalline diamond-glass platform for the development of three-dimensional micro-and nanodevices. A detailed description of the fabrication our a platform is given and several examples of devices that can be developed with it are mentioned.
Nanocrystalline diamond (NCD) is a highly appealing material for a variety of applications due to its biocompatibility, robustness, and mechanical, electrical, electrochemical, and optical properties. At odds with the general misconception that diamond is expensive, diamond films can be grown at low cost with microwave plasma assisted chemical vapor deposition (MWPACVD) [1]. For NCD, this can be done on large area substrates using an inexpensive precursor mixture, consisting usually of methane gas diluted in molecular hydrogen, after the deposition of diamond seeds of diameter below 10 nm [2]. During growth, a boron-containing precursor can also be added to make p-type diamond that exibits metallic properties when heavily doped [3,4]. Boron-doped NCD has a wide potential window which [5], in combination with its chemical inertness and biocompatibility [6], makes it an attractive material for electrodes [7]. During growth, dopants such as nitrogen and silicon can also be incorporated in diamond to form color centers for quantum technologies [8,9]. After growth, NCD films can be processed to form two or three-dimensional suspended structures [10,11], and, the surface of an NCD film can be functionalized with a variety of (bio)molecules for use in biosensors and solar cells [12,13]. Due to these excellent properties, NCD based devices, such as micromechanical resonators of high Q-factor [14,15], pressure sensors for harsh environments [16,17], tunable optical lenses [18], biosensors that, for example, can detect influenza [19,20], optically transparent electrodes [21,22], CO 2 reducing electrodes [23], superconducting quantum interference devices [24], and conducting atomic force microscope tips are being developed [25].
For the fabrication of 3D structures in microdevices, interposing layers with through holes that serve as conduits for fluids or electrical connections to thin films are indispensable [26,27]. In recent years, these interposing layers have been made available by several companies in the form of glass substrates with through glass vias (TGVs). Being inexpensive, transparent, electrically insulating, chemically inert, biocompatible, of high mechanical stiffness, and reusable, glass is a natural choice for the fabrication of interposing layers. Moreover, the properties of glass are strongly tunable. It can, for example, be made with coefficients of thermal expansion similar to those of semiconductor materials such as silicon, and can thus be used to fabricate microdevices with minimal residual stress [28]. Work disclosed by Corning  In step 2, blind holes of approximate diameter 42 µm and approximate depth 40 µm are made by laser ablation. In step 3, an NCD film of thickness less than 180 nm is grown on the growth side. In step 4, HFA etching is done locally to form the NCD sealed TGVs. (b) Schematic of the cross-section of the home-built chemical reactor that is used to etch a glass substrate as described in steps 1 and 4. A detailed description of the reactor and its use is given as supplementary material. (c) Image of a 10 × 10 × 0.2 mm 3 Lotus NXT glass substrate, laying on graph paper, taken immediately after step 1. (d) Scanning electron microscope (SEM) image of a blind hole that is made by laser ablation. The image is taken under a tilt of 25 • . (e) Image of a sample, with the camera on the growth side, immediately after diamond growth. The array of 9 blind holes in the middle of the circular shape are located at the etching side. (f) Image of an array of TGVs that are made by HFA etching in step 4. The image is taken with a reflecting light microscope with the objective on the etching side and the TGVs are sealed with portions of an NCD film. (g) Image of a sample, also taken with the same microscope but with the objective on the growth side of the substrate. The circular shape represents a suspended portion of the NCD film, which seals the center TGV that is depicted in (f).
shows that a viable process for fabricating TGVs is based on wet etching and either laser ablation or the modification of glass with laser light [29,30], while AGC relies on a process that is based on a focused electrical discharging method [31].
Towards the development of devices that rely on the exceptional properties of glass and NCD, we present a low-cost process, free of photolithography and transfer printing, for fabricating arrays of TGVs that are sealed on one side with suspended portions of an ultra-thin NCD film. The resulting platform can be useful for single-cell culture and analysis when the NCD film is made porous, which can be achieved through annealing [32], and thus, allow for nutrient or drug delivery. From work on the fabrication of robust membranes by Salminen et al. [33], it is also clear that our platform has a future in the field of modeling vascular systems. When made with boron-doped NCD films, our platform can be used to construct electrodes for microfluidic channels. Alternatively, if comprised of quantum grade NCD, our platform can also be used for quantum technologies. It is also noteworthy that NCD can be replaced by other materials that are resistant to hydrofluoric acid (HFA), such as silicon carbide or silicon nitride, and that their thermal properties, along with that of the glass we use, allow for the fabrication of MEMS that can operate in air up to temperatures of about 400 • C.

Overview
We summarize the fabrication process before addressing the details necessary to understand, reproduce, and extend our work. Fig. 1.a shows a schematic of the cross-section of a glass substrate before and after fabrication steps. In step 1 of our process, we locally etch a 10 × 10 × 0.2 mm 3 Corning Lotus NXT glass substrate to a thickness of approximately 50 µm. This is done in a home-built chemical reactor loaded with HFA. The face of the substrate that is etched lays on the etching side of the substrate and a schematic of a cross-section of the reactor is shown in Fig. 1.b. A result of etching in step 1 is shown in Fig. 1.c. In step 2, blind holes of approximate diameter 42 µm and approximate depth 40 µm are made by laser ablation on the etching side. Fig. 1.d shows a scanning electron microscope (SEM) image of such a blind hole, which is taken under a tilt of 25 • . In step 3, an NCD film of thickness below 180 nm is grown on the face that lays on the growth side, which is the side opposite to the etching side. To illustrate this, Fig. 1.e shows an array of nine blind holes, from the vantage point of the growth side, immediately after step 3. Finally, in step 4, HFA etching of the substrate is performed on the etching side to form TGVs that are sealed with suspended portions of an NCD film. During this step, the thickness of the glass around the TGVs is reduced to approximately 25 µm. The outcome of this step is shown in Fig. 1.f, which shows TGVs that are sealed with suspended portions of an NCD film. This image was taken with a reflected light microscope with the objective on the etching side. Fig. 1.g shows an image of the NCD film taken with the same microscope but with the objective on the growth side. The circular structure represents the center TGV of Fig. 1.f, sealed with a suspended portion of the NCD film, and is visible due to a difference between the refractive index of air and glass.

Glass etching
Fabricating TGVs of small diameter requires a thin glass substrate. However, a thin substrate is difficult to handle. We tackle this challenge in step 1. We use HFA to etch the substrate locally around its center. The remainder part of the substrate serves as a supporting frame for the etched portion of the substrate. More specifically, we use a 10 × 10 × 0.2 mm 3 Corning Lotus NXT glass substrate and etch it with 0.6 ml of 48 m% HFA to a thickness of approximately 50 µm. In step 4, the chemical reactor is loaded with 0.5 ml of 11 m% HFA, with the glass substrate etched on the etching side to form TGVs that are sealed with suspended portions of an NCD film. Although the glass layer of approximately 10 µm thickness between the lowest point of a blind hole and the NCD film should be etched, overetching must be avoided. We therefore choose to etch the 50 µm thick glass neighboring the blind holes to approximately 25 µm. Since an etching rate of 48 m% HFA is too high for controlled etching of 25 µm glass, we use 11 m% HFA. The lower HFA concentration reduces the etching rate and, thus, affords finer control over the etch depth.
We performed systematic studies to determine the etching times for steps 1 and 4. The results for 48 m% and 11m% HFA are presented in Figs. 2.a and b respectively, which show the maximum etch depth d that is reached as a function of time t for each of these concentrations. Each data point on the plots represents one experiment and t and d have respective errors of approximately 10 s and 3 µm. Fig. 2.c shows a typical surface profile taken after etching a substrate. To predict the time t that is necessary to etch a certain depth d, we adapt a model that was first used to describe the oxidation of silicon [34]. The model assumes steady-state diffusion, a first order chemical reaction, and a concentration of HFA c 0 at d = 0 that is constant with t. Under these assumptions, the time-evolution of the depth is with where c g , k, and D respectively denote the product of a stoichiometric constant and the concentration of species in glass that react with HFA at the HFA-glass interface, the reaction constant, and a diffusion coefficient. Since most of these parameters are unknown, we fit the data in Fig. 2.a with Eq. 1 to find that a = −4 ± 10 µm and b = 1054 ± 79 µm 2 min −1 . With this information, we are able to etch a glass substrate of approximately 200 µm locally to a thickness of 50 ± 5 µm in no more than 25 min. During etching, a crust of fluorides forms on the surface of the glass substrate. After etching, that crust is removed by rinsing the substrate with deionized water. For fitting the data in Fig. 2.b with Eq. 1, we find that a and b are equal to 29 ± 19 µm and 49 ± 17 µm 2 min −1 , respectively and that, with these values, we can etch a glass plate of 50 ± 5 µm locally to a thickness of 25 ± 3 µm, in no more than 35 min. We also investigate the effect of HFA etching on the surface roughness R a of a glass substrate. This is done because rough surfaces scatter light, adversely affecting transparency. In this work, R a is defined as the arithmetic mean deviation. Prior to etching, the value of R a for a glass substrate is approximately 4 nm or less. During etching in step 1, in which approximately 150 µm is etched with 48 m% HFA, R a increases up to 6   to approximately 13 nm. This cumulative increase in R a is clearly visualized by Figs. 2.d-f, which depict the surface profile of the glass substrate before step 1, between steps 1 and 2, and after step 4, respectively. It is remarkable that R a increases strongly after step 4, during which only approximately 25 µm is etched. To provide additional proof that etching Lotus NXT glass with 11 m% HFA induces more surface roughness than with 48 m% HFA, we present in Figs. 2.f-i the surface profiles of glass substrates etched with 11 m% HFA to depths of approximately 13 µm, 18 µm, and 30 µm. More surface profiles of etched glass substrates can be found as supplementary material.

Blind holes
In step 2 of our process, we use the front side laser ablation technique to create blind holes, of approximate diameter 42 µm and approximate depth 40 µm, on the etching side of the glass substrate. This is done where the substrate is etched maximally to approximately 50 µm. After laser ablation, the thickness of the substrate at the deepest point of a blind hole is therefore approximately 10 µm. It is important to recognize that the surface opposite to etching side, namely the growth side, does not show damage due to the laser ablation process. The pattern that is written during laser ablation consists of 20 concentric circles each of which is separated from its immediate neighbors by 1 µm. The largest circle is of diameter 40 µm, whilst the smallest circle is of diameter 2 µm, and the sequence of writing is from the circle of largest diameter to the circle of smallest diameter. To create holes, we first place the focal plane F of the laser light in air at f = 0, where f denotes the depth relative to the locally etched surface that is located on the etching side of the substrate. The pattern of concentric circles is then written, and a blind hole is formed. To clear Table 1: The depth of the blind holes that are depicted in Fig. 3.a. The procedure of making the holes is explained in the caption of that figure. The error on the depth is approximately 2 µm. P denotes the laser power scaled with the maximum power of the laser used in the experiment and f denotes the depth measured from the etched surface of the substrate to the focal plane of the laser light in air.  the hole of debris generated during laser ablation, pattern writing is performed 5 times in total. F is then lowered towards the substrate, with f = 5 µm, and the pattern is written 5 times again. The procedure of increasing f in increments of 5 µm and writing the pattern 5 times is repeated until a predefined f is reached. Through repeated experiments, we found using a laser pulse frequency of 500 kHz, a pulse duration of approximately 270 fs, and a write speed of 150 mm s −1 minimized crack formation during ablation. With these parameters, the ablation threshold of the glass surface is at a value of P equal to 0.33, where P denotes the laser power scaled by the maximum laser power. During the process of optimizing the parameters for laser ablation, we observed that crack formation becomes more apparent with increasing laser pulse frequency and P . To create blind holes at the lowest value of P possible and thereby avoid crack formation, we performed a study in which the depth of a blind hole was measured as a function of P for various choices of depth f . Results from this study are shown in Table 1 and an SEM image of the holes, taken under a tilt of 25 • , from which the data is obtained, is shown in Fig. 3.a. From this study, it is clear that for f = 35 µm the lowest value of P at which blind holes of approximately 40 µm deep can be made is 0.38. Fig. 3.b shows a reflected light microscope image of an array of 9 blind holes of approximately 42 µm diameter and approximately 36 µm deep in a substrate that is locally etched to a thickness of 46 µm. These blind holes are made with f = 30 and P = 0.38. The values listed in Table 1 are visualized by a graph as supplementary material. Fig. 1.d shows an SEM image, taken under a tilt of 25 • , of a blind hole similar to those in Fig. 3.b. From the image, it becomes clear that the shape of the blind hole resembles that of a paraboloid. This is most likely a consequence of laser light shading by the glass substrate. Fig. 3.c shows an SEM image of the surface at the deepest point of the blind hole. Cracks caused by laser ablation are not detected from this image.

Nanocrystalline diamond
In step 3 of our process, an NCD film is grown on the surface of a glass substrate with blind holes. This is done on the growth side, which is the side opposite to the etching side. During growth, the etched part of the glass substrate is not in contact with the cooled substrate holder of the CVD system in which growth occurs. X-ray crystallography confirms that the film is crystalline. The asymmetric grazing angle diffractogram in Fig. 4.a shows the most intense peak that is generated by copper (Cu) K α X-rays, which are scattered by the film, after background subtraction and scaling. A Voigt function is fitted to the data with the least-squares method from which the center of the peak is estimated at 2θ = 44.03 ± 0.01 • , where θ denotes the Bragg angle. From the lattice parameters of diamond [35], we deduce that the peak corresponds to constructive interference of the X-rays scattered by the (111) crystal planes of the diamond crystallites in the NCD film that are compressively strained by 0.1%. Results from a Raman spectroscopy study that is included in the supplementary material support these findings. Fig. 1.e shows a photograph of a substrate after NCD growth that is taken with the camera located on the growth side. The substrate was placed on a mirror and therefore, a large part of the light recorded by the camera is scattered, making it possible to observe the array of 9 blind holes. Fig. 4.b shows an SEM image of the NCD film, which is closed and exhibits a nanocrystalline structure. SEM images of this film taken with different magnifications are provided as supplementary material. Fig. 4.c shows the thickness of the film together with the thickness of the glass substrate. At the thinnest part of the glass substrate, in the vicinity of the blind holes, the film is 175 ± 5 nm and reduces from that area radially towards the edges of the diamond film to approximately 120 ± 5 nm. Due to thin-film interference [36], the variations in film thickness can also be observed from difference in the apparent color of the NCD film, as shown in Fig. 1.e.  Figure 4: (a) Asymmetric grazing incidence X-ray diffractogram of the NCD film grown in step 3 of our process with a Voigt function fitted to the data. The intensity of the scattered X-rays is scaled and θ denotes the Bragg angle. The center of the peak is at 2θ = 44.03 ± 0.01 • , which corresponds to constructive interference of copper Kα X-rays that are scattered by the (111) crystal planes of diamond [35]. (b) SEM image of the NCD film. The image shows that the film is closed and consists of nanoscopic diamonds. (c) Thickness of the NCD film and the thickness of the Lotus NXT glass substrate on which the film is grown. Both thicknesses are given as a function of a lateral position. The glass substrate is etched on the etching side and NCD is grown on the growth side, which is the side opposite to the etching side. In the vicinity of the thinnest part of the glass substrate, which is of approximate thickness 50 µm, the NCD film is approximately of thickness 175 nm, and on the thickest part of the glass substrate, which is of approximate thickness 200 µm, the NCD film is of approximate thickness 120 nm.

Through glass vias
In step 4 of our process, the glass substrate is further etched by HFA to form through glass vias (TGVs) that are sealed with suspended portions of an NCD film of approximate diameter 60 µm, which are the structures intended to serve as a NCD-glass platform for the development of 3D micro-and nanodevices. The substrate is etched locally from approximately 50 µm to 25 µm and thinning is performed on the etching side of the substrate. Fig. 1.f shows a micrograph, which is also taken on the etching side, of nine TGVs that are sealed with suspended NCD of thickness 175 ± 5 nm. In the vicinity of these TGVs, the substrate is of approximate thickness 23 µm and light scattering causes the walls of the TGVs to appear as black circles. Fig. 5.a is a dark-field optical microscope image, taken on the growth side of the substrate, that captures the same NCD-glass platform as Fig. 1.f. From this image, we infer that, despite the walls of the TGVs, the platform barely scatters light. Fig. 1.g shows a micrograph taken on the growth side. The circular shape represents the suspended portion of the NCD film and is of approximate diameter 60 µm. Fig. 5.b depicts an SEM image of the center TGV, which is taken on the etching side of the substrate, under a tilt of 25 • . The wall of the TGV is clearly rougher than the rest of the structure. All TGVs are tapered from the etching side towards the growth side and have minimum and maximum diameters of approximately 50 µm and 80 µm, respectively. Since the diameter of the suspended film is approximately 60 µm, the NCD film is underetched by approximately 5 µm. Fig. 5.c depicts the surface profile of a portion of the NCD film that seals the central TGV. The profile is taken on the growth side and the suspended portion of the NCD film is slightly buckled towards the glass substrate with a maximum deflection of approximately 1.25 µm. This behavior is seen for all suspended portions of the NCD film that seal TGVs, as can be seen in the supplementary material. To show that the bond between the NCD film and the glass substrate is sufficiently strong for practical use, we applied a gauge pressure of 300 kPa to the suspended NCD on the etching side, the result of which was that the structure of the NCD-glass platform was able to withstand such pressures.

Glass etching
Isotropic etching of silicon dioxide-based glass is typically done with HFA. The overall chemical reaction for etching silicon dioxide glass with HFA is described by the chemical reaction This, however, represents a simplification of what occurs during the etching steps of our process. Moreover, no clear mechanism for glass etching is provided in literature [37]. The etching rate of silicon dioxide-based glass with HFA typically increases with temperature, hydrogen fluoride concentration, and the concentration of species other than silicon dioxide [37,38,39]. Using Eq. 1, it can be found that d ∝ t for a reaction limited process and that d ∝ √ t for a diffusion limited process [34]. When fitting the general relation for a power law to the data for 48 m% HFA etching, we find that the power of T s is 0.48 ± 0.2, from which we infer that the etching process is diffusion limited. For 11 m% HFA etching, we find that the power of T s is 0.72 ± 0.08, which indicates that the etch process is on the border between reaction limited and diffusion limited. To understand the effect of HFA concentration and duration on the etching process, we exploit the Damköhler number Da, which in our setting is given by This dimensionless quantity relates the reaction rate to the diffusion rate as follows: for Da 1, the etching process is diffusion limited; for Da 1, the etching process is reaction limited. From Eq. 4, we find that the glass etching process is reaction limited for relatively small values of t but becomes diffusion limited for relatively large values of t. We also infer that, consistent with our observations, diffusion limited behavior is achieved more rapidly for higher initial concentrations of HFA.
To synthesize a silicon dioxide-based glass with specific properties, well-defined amounts of oxides such as Al 2 O 3 , As 2 O 3 , B 2 O 3 , CaO, K 2 O, MnO, Na 2 O, and P 2 O 5 are mixed with SiO 2 [37]. According to Tay et al. [39], these oxides react with HFA to form fluorides insoluble in HFA that sediment onto portions of the etching surface leading to roughening. Iliescu et al. found that adding hydrochloric acid to HFA provides an effective way to dissolve these fluorides [40]. In addition, Ceyssens and Puers noted that the sedimentation of these fluorides can be avoided by vertically orienting the glass surface that is being etched [41]. They reasoned that gravity drags the fluorides away from the surface, preventing masking, and were surprised that the roughness of the glass surface barely increases even for deep etching. Even without placing a Lotus NXT glass substrate vertically during etching with 48 m% HFA, we found that the surface roughness barely increases while etching to a depth of approximately 150 µm. However, we also found that etching with 11 m% HFA roughens the glass surface more noticeably. The mechanism underlying this observation is not yet understood. Assuming that masking of insoluble fluorides, which are sedimented at the HFA-glass interface, causes roughening during etching, we hypothesize that for etching with 48 m% HFA, the velocity of the HFA-glass interface is larger than the sedimentation rate of fluorides. Since step 4 requires shallow etching with 11 m% HFA, the surface roughness of the glass substrate after etching is relatively low, as can be verified by consulting the dark-field optical microscope image in Fig. 5.a.

Blind holes
Small surface cracks that are introduced by polishing and sand blasting glass can lead to surface roughening during etching with HFA. Spierings showed that these cracks develop into spherical depressions that widen during etching. Where two of these structures meet, a cusp-like structure forms [37]. After laser ablation in step 2 of our process, we do not observe cracks in the blind holes; however, cusp-like structures are clearly evident on the walls of a TGV, as shown in Fig. 5.b, after HFA etching in step 4. This leads us to suggest that although cracks are formed during the laser ablation process, any such cracks are too small to be detected by our imaging techniques. Still, due to the optimization of the laser ablation process, we achieve a feature size that allows the TGVs to be fairly circular. It strikes us as highly likely that activating the glass with a laser instead of performing laser ablation would further reduce the roughness of the TGVs after etching [30]. More details on modifying glass with laser light can be found in the literature [42].

Nanocrystalline diamond
To explain variations in the thickness of the NCD film, which is depicted in Fig. 4.c, we note that, during NCD growth, it is only the glass substrate surface on the etching side, where no thinning is performed, that is in contact with the water-cooled substrate holder whilst the glass substrate surface on the opposite growth side is in contact with a plasma. It is therefore reasonable to assume that during growth, the temperature T s of the etched part of the glass The temperature at which the viscosity of the glass is 10 13 Pa s. ‡ Linear coefficient of thermal expansion averaged from 0 to 300 • C. For diamond, the value for this property is approximately 2×10 −6 • C −1 [45].
substrate is greater than that of the remainder of the substrate. Additionally, from the work of Tsugawa et al. [43], we learn that the growth rate r of diamond increases monotonically with T s , following an Arrhenius-type equation where A, E a , and k B respectively denote a pre-exponential factor, the activation energy, and the Boltzmann constant.
On that basis, we suggest that the growth rate of the NCD film is spatially nonuniform. Since the growth time at all locations is the same, a nonuniform film thickness is expected. By growing the NCD film before etching, variations in film thickness can be avoided. However, the question of whether or not the heat generated during the laser ablation process affects the NCD film remains to be investigated.

Glass substrate
During the chemical vapor deposition of an NCD film, T s typically ranges from 500 • C to 900 • C [44]. For our purposes, glasses with annealing points above 500 • C are therefore preferable. To avoid stress due to thermal mismatch between the glass and the NCD, the glass should also be designed so that its coefficient of thermal expansion is equal to that of NCD for the complete range of temperatures achieved during the growth process, a criterion that can be difficult to meet. However, such glass is not currently available commercially. Still, several types of silicon dioxide-based glass on which NCD can be grown have annealing points above 500 • C. Glasses that can be used for NCD growth are listed in Table 2, together with some of their properties. Except for fused silica, the glasses listed in this table are designed to be used in conjunction with silicon.
During preliminary experiments, we found that NCD films, grown on fused silica substrates during step 3 of our process, completely delaminate during step 4. We suspect that this is caused by tensile stress, which is to be expected since the coefficient of thermal expansion of fused silica is less than that of diamond over the entire range of operating temperatures used during our preliminary experiments [45,46]. Our experiments and previous work [17,47,48], show that NCD films grown on substrates made of any glass listed in Table 2, except fused silica, are compressively stressed, which leads us to suspect that the coefficients of thermal expansion of those types of glass might be greater than that of diamond for significant portions of the temperature ranges used in the relevant growth processes. It is noteworthy that Iliescu et al. and Ceyssens and Puers found that compressive stress acting on films, which mask parts of glass substrates during HFA etching, is preferable over tensile stress [27,41]. Our findings agree with their results. A typical consequence of compressive stress is buckling [49], which is illustrated by Fig. 5.c. For several applications, buckling should be minimized by properly tuning the coefficients of thermal expansion of the film and the substrate. Since the coefficient of thermal expansion of diamond is greater than that of fused silica, but less than those of many other glasses, we are optimistic that this can be done.

Materials and Methods
A plate of Corning Lotus NXT glass, which is an alkaline earth boro-aluminosilicate, was purchased from the company Kuramoto, which, on request, etched the glass plate by HFA to a thickness of approximately 200 µm. The glass plate was next laminated to a plastic sheet and diced with a Disco DAD322 dicer to yield 10 × 10 × 0.2 mm 3 substrates. After dicing, residues of the sheet were found on the substrates and were removed by cleaning for 20 min in acetone with an ultrasonicator. Acetone was rinsed from the substrates with deionized water.
HFA etching was done in a class 1000 cleanroom at a temperature of 23 • C and a humidity of 60%. Due to the toxicity of HFA and hydrogen fluoride gas, all appropriate safety measures were taken.
Blind holes were made with a LightFab system designed for selective laser-induced etching of fused silica [50,51]. The LightFab system was equipped with a 4 W laser. The laser light, of wavelength 1030 nm, was focused with a microscope objective to a spot of approximate diameter 1 µm. The objective had a numerical aperture of 0.4, magnified 20 times, and had a working distance of 10 mm. The galvo scanner head of the system allowed the focal plane F to be moved in three dimensions and with a stepper motor, the stage of the system was also moved in three dimensions. Due to software restrictions, the depth f , measured relative to the surface located at the etching side of the substrate, was changed with the stepper motor rather than with the galvo.
Preceding NCD growth, the substrate was degreased for 20 min in acetone with an ultrasonicator. The substrate then underwent RCA cleaning, omitting the optional oxide strip for bare silicon wafers. After cleaning, the surface on the growth side was covered with detonation nanodiamonds of diameter below 10 nm, a process also known as seeding [2]. Although nanodiamonds are chemically bonded to a matrix of sp 2 hybridised carbon after synthesis, it is noteworthy that they can be separated by bead milling [52]. With a powder of such separated nanodiamonds, purchased from NanoCarbon Research Institute Co., Ltd, we next made a stable colloidal suspension using a method similar to that reported by Ozawa and coworkers [53]. We did this by first mixing 0.1 g of the powder in 0.2 l water (≈ 0.05 m%) and then ultrasonicating the mixture using an ultrasonic probe with a tip of diameter 3.2 mm and length 4.5 cm. The probe was connected to a transducer of power 100 W and frequency 20 kHz that was set to cycle on and off every second for 90 minutes. After the process, the tip was slightly damaged. The obtained suspension was turbid but cleared after a week with the settling of particles. The brownish suspension was extracted by a pipette. After mounting the substrate on a spin coater, 40 µl of the suspension was drop casted on the surface of the substrate until complete coverage was achieved. Alternatives to drop casting involve submerging the substrate in the suspension or squirting the suspension on the substrate [2,54]. One minute after drop casting, the surface of the substrate was flushed for 10 s with deionized water while the substrate was spinning at 4000 rpm. This strategy was used to avoid any aggregation of nanodiamonds. Simply dipping the substrate in an abundance of deionized water also leads to satisfying results; this, however, can lead to unintentional seeding of substrate surfaces that should remain unseeded. After flushing, the substrate was dried by spinning for an additional 15 s without changing the spin frequency. Touching the seeded surface was avoided to prevent removing the nanodiamonds that are bonded by Van de Waals forces. The seeded substrate was placed in the reactor of an SDS6500X microwave plasma-enhanced chemical vapor deposition on a molybdenum substrate holder of 58 mm diameter and 5.5 mm thickness. Subsequently, the gasses in the reactor were evacuated by a Kashiyama SDE90X dry pump to base pressure 8.5 Pa. Hydrogen gas and methane gas were next introduced into the reactor at respective flow rates of 294 sccm and 6 sccm. After reaching a stable pressure of 2 kPa, the gasses were ignited into a plasma with 1.5 kW of 2.45 GHz microwaves. At these conditions, the temperature T s of the substrate during growth remained below the annealing point of Corning Lotus NXT glass, which is 722 • C, and the growth rate g was on the order of 1 nm min −1 . The thicknesses of the NCD films were measured during growth with a home-built interferometer. During growth, a light gray film, consisting likely of hydrogenated carbon, was unintentionally deposited on the etching side of the substrate. This film was removed in the reactor by a Yamato PR200 system. Specifically, after first evacuating the reactor with an oil rotary pump, air was introduced into the reactor at a flow rate of approximately 100 ml min −1 . Subsequently, the air was ignited into a plasma with 170 W of 13.6 MHz radio waves at a pressure below 1 kPa. The plasma removed the light gray film within a minute.
All photographic images were taken with a Canon EOS 5D camera and a Canon MP-E 65 mm lens. Surface profile features with lengths greater than 1 mm were taken with a DektakXT stylus profilometer and surface profile features with lengths less than 1 mm were taken with a Keyence VK-X150 laser microscope. From surface profiles that were taken with this microscope, surface roughness R a was deduced. The VK-X150 was also used to measure the thicknesses of the glass substrates, taking into account the refractive index of the glass which, for the Lotus NXT glass, is 1.53. Moreover, the VK-X150 was also used as a reflected light microscope and a Meiji MT9930L was used to perform dark-field microscopy. The thicknesses of the NCD films were measured with higher resolution than during growth using a Hamamatsu C13027 optical nano gauge. A FEI Quanta 250 FEG and a JEOL JSM-7900F SEM were used to examine the blind holes, the NCD films, and the TGVs. To avoid substrate charging, a conductive film consisting of platinum and palladium was deposited preceding scanning electron microscopy on the blind holes. The film was of approximate thickness 2 nm and was deposited with a Hitachi MC1000 ion sputter coater. Grazing incidence X-ray diffraction measurements were carried out on the NCD films with a Bruker D8 Discover diffractometer, equipped with a copper line focus tube and Göbel mirror that produced a beam of Cu K α X-rays of 0.15418 nm average wavelength. A 0.6 mm divergence slit was used on the primary beam side and a 0.2 • receiving Soller slit was mounted in front of the scintillation counter detector. The angle of incidence β of the X-ray beam with the samples was 0.5 • , which is slightly above the critical angle β c of total external reflection for an air-diamond interface. Raman spectra were measured on a Tokyo Instruments Nanofinder 30 system. Specifically, a 5 mW 532 nm laser was focused on the sample to a spot of a few micrometers in diameter by means of an objective lens, which magnifies 100 times and has a numerical aperture of 0.95, of an upright microscope. The light emitted by the sample was collected in an 180 • scattering geometry using a grating of 1200 grooves per mm, a 0.2 mm pinhole, and an Andor DU920P BR-DD charge-coupled device (CCD) operating at −60 • C. The value of β c was calculated to be 0.27 • . Functions were fitted to data using the lmfit Python library and figures were made with Gnuplot, Inkscape, Blender, and the matplotlib Python library.

Conclusion
We present a low-cost and robust nanocrystalline diamond-glass platform for single-cell culture and analysis, ondemand drug delivery systems, the modeling of vascular system, microelectrodes, quantum technologies, and hightemperature MEMS. Our platform is comprised of a glass substrate with through glass vias (TGVs) that are sealed on one side with suspended portions of an ultra-thin nanocrystalline diamond (NCD) film. Our fabrication process is free of photolithography and transfer printing and is delineated in detail sufficient to allow easy replication by others. In this process, hydrofluoric acid (HFA) is used to first etch one side of a 10 × 10 × 0.2 mm 3 Lotus NXT glass substrate to a thickness of approximately 50 µm. On the same side that is etched, blind holes of approximate diameter 40 µm and approximate depth 40 µm are subsequently formed by laser ablation. After growing an NCD film of approximate thickness 175 nm on the surface opposite to the etched side, the etched side of the substrate is further etched by HFA to approximately 25 µm to produce TGVs that are sealed on one side by suspended portions of the NCD film. Our resulting platform is highly transparent and can handle applied pressures of at least 300 kPa.  The profiles are formed after reaching various etch depths d as indicated and, for each profile, the surface roughness R a is given. We conclude that 11 m% HFA roughens the surface of Lotus NXT glass substrates more than 48 m% HFA does.

Blind holes
Supplementary Figure 3 shows the depth of 25 blind holes of 42 µm diameter that are made using the front side laser ablation technique in a 10 × 10 × 0.2 mm 3 Corning Lotus NXT glass substrate on the etching side of the substrate, with the etching side denoting the side where the substrate is etched in steps 1 and 4 of our process. The holes are made in the vicinity where the glass substrate is etched most and is of approximate thickness 50 µm. To create these holes, the focal plane F of the laser light in air is placed at the surface of the substrate and a pattern of concentric circles is written 5 times. F is then lowered by 5 µm, and the pattern is again written 5 times. This procedure is repeated till F reaches a predefined target depth f , with f denoting the vertical distance from F to the etched surface. The symbol P denotes the laser power scaled with the maximum power of the laser. The depth of each blind hole is taken from Table 1. It is clear that the depth of the blind hole that is made with f = 35 and P = 40 deviates from the expected value. So far, the cause of this is unclear. However, we expect that using the stage of the LightFab for changing f rather than the galvo might be the source of this deviation.

Nanocrystalline diamond: scanning electron microscopy
Supplementary Figures 4A-B show scanning electron microscope (SEM) images at different magnifications of the nanocrystalline diamond (NCD) film that is depicted in Fig. 4.b. The image is taken at a location where the film is of approximate thickness 175 nm. The film is grown with microwave plasma-enhanced chemical vapor deposition in the reactor of an SDS6500X system on a Lotus NXT glass substrate after seeding the substrate with nanodiamonds.
A B Supplementary Figure 4: A) SEM image of an NCD film that is grown on a Lotus NXT glass substrate. The film is of approximate thickness 175 nm and magnified 10 4 times. B) SEM image of the same NCD film but magnified 10 5 times.

Nanocrystalline diamond: Raman spectroscopy
Supplementary Figures 5A-B depict scaled Raman spectra of a (111)-oriented 3 × 3 × 0.3 mm 3 single crystal diamond and of the NCD film depicted in Fig. 4.b that is grown on a Lotus NXT glass substrate. In each of these spectra, the most intense feature forms a peak that originates from the first-order Raman line of diamond, which is expected at the approximate Raman shift 1332 cm −1 . 1 Supplementary Figure 5C shows a zoom of those peaks. It can be observed that the center of the diamond peak extracted from the NCD film is positioned at a shift of approximate value 1334 cm −1 , indicating that the film is strained due to a compressive stress of approximately 0.5 GPa acting on the film. 1 The compressive stress is mainly caused by a thermal mismatch between NCD and the glass since a portion of unstrained freestanding film generates a diamond peak approximately at 1332 cm −1 . The large band located at shifts larger than 1332 cm −1 and the large photoluminescence background observed in Supplementary Figure 5B are mainly ascribed to sp 2 carbon and hydrogen that are present in the NCD film, respectively. 2 The presence of impurities as well as point and line defects are also a known factor for the broadening of the diamond peak, which explains the considerably wider peak obtained for NCD compared to the peak obtained for single crystal diamond. 3 Finally, it is noteworthy that this analysis is consistent with the X-ray diffraction study reported in the main text.