Through a Window,Brightly: A Review of Selected Nanofabricated Thin-Film Platforms for Spectroscopy,Imaging,and Detection

Introduction

A sample support or window should contribute to a measurement only in the desired manner. An ideal cuvette, for example, would define the path length but not affect the optical transmission. In spectroscopic applications, the usual approach is to minimize material absorption by choice of window material. ¹ In a charged-particle technique such as transmission electron microscopy (TEM), however, the strong interaction between the interrogating beam and matter places stringent limits on the material composition and thickness of any sample support through which the beam must pass. Sample-dependent solutions for TEM have included supports with open areas to leave samples unobscured (but unsupported over that opening), and thin carbon and polymer films that minimize signal loss by reducing support thicknesses. There is considerable “art” in the fabrication of many support films:^2,3 the adoption of nanofabricated thin films has expanded the library of materials and structures to include thin-film silicon nitride, graphene, and nanostructured variants such as porous silicon, among others.^4–15 Both composition and thickness can be tuned in concert to obtain optimal performance. More broadly, new performance capabilities can be engineered (or discovered) when the thin-film sample support or window itself becomes a focus of investigation. Silicon nitride thin-films offer superb mechanical properties—a < 100 nm thick window can withstand over an atmosphere of cross-film pressure differential, for example ¹⁶ —and this can be exploited to create sample supports and containers that overcome severe challenges such as nonlinear window responses on short timescales in ultrafast spectroscopy.^17–19 Thin films with suitable properties and structures can also be used in a more active role in experimental studies: a ∼10 nm diameter channel through a < 100 nm thick thin film—a so-called nanopore—is the basis for a wide range of single molecule sensing and manipulation methods that emerge from the passage of a molecule into, or through, the nanofluidic channel.^20–29 Fig. 1 offers two illustrations to illuminate the rich scope of experimental configurations and possibilities using just these two platforms. We will frame our discussion of the interplay between thin-film design—composition and structure—and application by using these two quite different implementations: a sample cell with thin-film windows bracketing a nanofluidic channel; and a nanopore single molecule sensor platform where the robust thin film supports a nanofluidic through-channel. Fig. 2a shows the common structural foundation of the two nanofluidic devices: a free-standing thin silicon nitride film supported by a silicon frame. ¹⁶ We will use this canonical construct and material to discuss key design and operating considerations, highlight experimental accomplishments, challenges, and opportunities, and touch on alternative materials and architectures. OPEN IN VIEWER

OPEN IN VIEWER

Figure 2.

(a) The basic < 100 nm-thick SiN_x membrane, free-standing over an open aperture in Si can serve as (b) the windows of a nanofluidic cell with additional features such as a wetting layer (green) added through conventional nanofabrication approaches or (c) be used as a single molecule sensor element after formation of a nanochannel through the membrane.

Perhaps the most captivating feature of silicon nitride is its ability to withstand being suspended unsupported, with a film thickness of <100 nm, over millimeter-wide open apertures. A robust, free-standing thin film such as this is a compelling structural component—one that can be further patterned—for use in a host of applications. The freedom from supports is in marked contrast to the usual need for thin carbon film substrates for TEM, for example, to be supported. The success of this unusual configuration hinges on carefully controlling the film stress through the film stoichiometry during fabrication (see below).^16,31 Early examples of applications include their use as vacuum chamber windows, to create a wet cell for X-ray microscopy, to protect the source for scanning transmission ion microscopy (STIM) studies in air, to create a sample cell for scanning electron microscopy (SEM) at atmospheric pressure, and to serve as sample substrates for studies using ultrafast spectroscopy, optical microscopy, TEM, and atomic force microscopy (AFM)—and for studies using such techniques in tandem.^{6–8,16,31–35} The films have the demonstrated ability to withstand even atmospheric pressure differentials. ¹⁶ The impressive mechanical properties of these films have resulted in their increasing use in an ever-wider range of applications that can be further enhanced by additional beneficial properties of the films, such as their transparency to a host of probes across a wide range of wavelength regions and encompassing electrons. The use of silicon nitride windows in creating sample cells and chambers has become increasingly prevalent, as has their use as sample supports; their use as a platform to host structural elements such as nanofluidic channels is a relatively new area of application and investigation.^{16–29,36–58}

Two particularly compelling uses of thin-film windows are shown in Figs. 1 and 2. The windowed nanofluidic sample cell concept has been used for femtosecond mid-infrared (IR) studies of water and for transmission electron microscopy studies on liquid samples. ¹⁷ Alternative configurations exist, including fully sealed sample chambers cut off from liquid control (see the Nanofluidic Sample Cells section, especially Fig. 9, for more detail).^38,40 Such a nanofluidic cell has, for reasons outlined in the article below, been able to deliver molecular-level insights on the fastest timescales and nanometer-length scales relevant for creating molecular movies. The nanopore platform has found principal use in genomics, ⁵⁹ but with increasing use to profile proteins and intermolecular interactions.^21–29 Extension of this single pore design to many pores allows the same thin-film platform to function as a high-throughput filter, with implications also for nanoplasmonics.^60,61 There are excellent reviews focusing on these two particular implementations alone,^{21–29,38,40} and a recent book chapter reviewing both in a unifying context of nanofluidics. ²⁰ We hope the present review will encourage an appreciation of the breadth of application of even an ostensibly prosaic free-standing thin-film structure across a host of spectroscopy, microscopy, and sensing techniques.

Thin-Film Silicon Nitride Membranes

Our references to “silicon nitride” in the context of free-standing SiN_x thin films are to silicon-rich silicon nitride, SiN_x (or, more restrictively Si_xN₄ with x > 3; Si₃N₄ is stoichiometric silicon nitride), prepared by low-pressure chemical vapor deposition (LPCVD). ²⁰ Low-pressure chemical vapor deposition SiN_x is an excellent micro- and nanofabrication material, used widely for surface passivation including as a barrier to electrons, ions, water, and oxygen in microelectronic and microsystem devices, and as an etch stop in some fabrication processes.^62–65 The basic fabrication processes for SiN_x thin films are well-established and can be easily found in standard texts of micro- and nanofabrication, underscoring the prevalent use of this material. ⁶² Dichlorosilane (SiH₂Cl₂) and ammonia (NH₃) source gases are mixed in reduced-pressure furnaces at <1000 ℃ and deposited, in our applications, onto polished Si (100) wafers.^62,66 The film stoichiometry—and consequently its residual stress—is controlled by tuning the two source gas feeds. ⁶² Enriching the relative amount of silicon in fabricating the non-stoichiometric (silicon-rich), amorphous films controls the residual stress and is responsible for what is, to us, the most dramatic property of SiN_x films: their ability to withstand large cross-membrane pressure differences.^16,67 Indeed, vendors may use and report only the film’s residual stress, rather than its stoichiometry.^16,67 The silicon-rich nature of the surface has additional consequences for surface chemical modification (see below). ²⁰

The creation of a free-standing membrane of thin SiN_x follows a conventional route combining photolithography and etching.^16,20 The detailed fabrication process is depicted in Fig. 3. Low-pressure chemical vapor deposition SiN_x films are formed on both sides of a doubly polished Si (100) wafer. Standard photolithography and reactive ion etching (RIE) are used to create the window feature in one of the SiN_x layers (thereby exposing the Si). The exposed Si is anisotropically etched with ∼40% by weight aqueous potassium hydroxide (KOH) at 80 ℃, for which the intact SiN_x on the other face serves as an etch stop. ⁶⁵ The unsupported area of the film is smaller than the opening in the patterned side of the wafer owing to the 54.7° etch profile arising from preferential etching of the Si planes. Openings from micrometers to millimeters can be made,^16,17 and the remaining Si frame can be used for support and manipulation. It is standard to design patterns that accommodate multiple frames on a single Si wafer, and to fabricate snap lines that extend approximately one-third of the depth of the wafer for easy separation of the individual frames. Though KOH etching of Si (100) with a SiN_x mask is by far the most standard approach used to fabricate free-standing structures, unconventional geometrical etch profiles can be created by employing Si wafers with alternative crystallographic orientations and surface miscuts, and by use of different etch stops and etching solutions. ⁶⁸ Moreover, SiN_x thin films can be completely removed from their underlying substrates by following a modified fabrication procedure, although that requires the films to be porous. ⁶⁹ OPEN IN VIEWER

Figure 3.

(top) Fabrication steps for free-standing SiN_x windows: (a) SiN_x is deposited on both sides of a double-side polished (100)-oriented Si wafer by LPCVD; (b) photoresist is sequentially coated onto both sides of the wafer. While photoresist is only needed on one side for photolithography, the photoresist on the back (membrane) side serves as a protective layer against etching of the SiN_x in step d. (c) The window feature is created in the photoresist layer via standard photolithography (UV exposure and development). (d) Reactive ion etching (RIE) is used to transfer the window feature into the SiN_x layer. Note that RIE is isotropic; the vertical profile of the exposed area as shown in the figure does not accurately depict the actual isotropic etch profile, but because the depth of the etched region is much smaller than the lateral dimension of the exposed area, this effect can be ignored. (e) Photoresist is stripped from both sides using a suitable chemical bath. (f) Potassium hydroxide (KOH) is used to etch through the wafer and create the free-standing SiN_x structure. Potassium hydroxide stops at the SiN_x layer. Potassium hydroxide is anisotropic, with etch rates of the (100) and (110) planes much larger than the (111) planes. Thus, the etch front progresses along the (111) planes of the crystal, resulting in the narrowing of the window dimension relative to the dimension of the mask feature by a factor of the thickness of the wafer multiplied by √2. To speed up the onset of the KOH process, it is recommended to dip the patterned wafer into a buffered oxide etch solution (BOE) prior to the KOH etch step. The buffered oxide etch solution removes the native oxide layer that forms naturally on the exposed Si. (bottom) Top-down views of the fabricated membrane and supporting frame. The view from the membrane side shows only the window. The view from the patterned side shows the etch pit whose angles are determined by the silicon crystal planes.

There are various subtleties and practical know-hows to the fabrication process that are beyond the scope of this article. We mention, however, a couple of important ones, both related to the KOH etching process. The fabrication procedure described above assumes that SiN_x is a perfect KOH etch stop. Yet, the selectivity of KOH etching for LPCVD SiN_x versus Si, though extremely high, is not infinite. While the KOH etch rate of SiN_x is not well studied, some sources report selectivity values that range from 1:10 000 to 1:50 000.^70,71 The upper limit on the corresponding KOH etch rate of SiN_x is 0.1 nm/min. Such an etch rate is not negligible when fabricating ultrathin windows (< 10 nm) and should be taken into account when deciding on the thickness of the LPCVD deposited SiN_x layers. The other point relates to the use of doped Si wafers. For certain applications, it is desired to use highly doped Si in order to electrically ground the supporting frame. For instance, this is useful when employing the window as a support structure in electron microscopy, in which the structure is continuously exposed to a high flux beam of electrons. However, using boron-doped Si can significantly retard the KOH etch rate, and, above a certain doping threshold, boron-doped Si can serve as an effective KOH etch stop. ⁷² Therefore, it is recommend to use n-type doped Si (e.g., phosphorus-doped) for applications where high electrical conductivity is desired, without severely impacting the KOH etch rate.

Mechanical Properties of Free-Standing Silicon Nitride

Low-pressure chemical vapor deposition SiN_x deposited on Si possesses an amount of residual stress from two sources: intrinsic stress due to the deposition conditions and thermal stress due to the mismatch in the coefficients of thermal expansion between film and substrate. ⁷³ Even though Si₃N₄ has a slightly lower coefficient of thermal expansion than Si, resulting in compressive residual stress, the residual stress in LPCVD SiN_x tends to be tensile in nature (i.e., the deposited film is elongated relative to a stress-free film). ⁷⁴ The deposition parameters (temperature, pressure, and gas flow rate) offer a certain degree of control over the amount of residual stress. Moreover, non-standard deposition techniques can be employed to control the residual stress within a wide range of values. For example, Shi et al. demonstrated using ion implantation to tailor the residual stress of SiN_x over a range of values extending from 1.2 GPa (tensile) to −0.1 GPa (compressive). ⁷⁵ In micro-electro-mechanical systems (MEMS)-type devices, however, low residual stress films are highly desirable. While there are reports in literature on fabricating stress-free films, ⁷⁶ the standard processes that are commonly accessible in nanofabrication facilities produce films with tensile stress as low as ∼200 MPa.

For SiN_x windows used as support structures, a mechanical property of interest is the amount of bulging (also referred to as buckling or bowing) due to differential pressure (e.g., pressure applied from one side of the membrane normal to the membrane surface). For example, in nanofluidic devices such as the one shown in Fig. 2b, two SiN_x windows and a spacer are used to define a sample path length. ⁷⁷ In such an application, the effective increase in the path length due to bulging of the membranes could severely compromise the functionality of the device. Thus, it is essential both to characterize bulging and to understand which design features can be used to keep bulging to a minimal acceptable level.

A schematic of the bulging geometry for a square membrane is shown in Fig. 4. Characterization of the bulging profile is typically done with load-deflection measurements.^74,78–84 In these measurements, pressure is applied to one side of the membrane and deflection is measured optically under a microscope—calibrated to measure depth through changes in the focal plane. For a square membrane, the load-deflection behavior is found to follow the relationship

( Et / a 4 ) d 3 + ( 1 . 66 t σ o / a 2 ) d = 0 . 547 p

(1)

where d is the maximum membrane deflection, t is the membrane thickness, 2 a is the side of the square membrane, E is Young’s modulus, σ o is the residual stress, and p is the external differential pressure. ⁸⁵ The equation above can be also used to model deflection of a rectangular membrane, with 2 a representing the smaller lateral dimension of the rectangle. Fig. 4 also shows the results of calculations of the maximum membrane deflection, d, as a function of differential pressure. The calculation assumes a SiN_x membrane with residual tensile stress of 200 MPa and a Young’s modulus of 300 GPa.^86,87 OPEN IN VIEWER

Figure 4.

(top) Cross-sectional view depicting bulging of a square-shaped silicon nitride membrane of side 2 a and thickness t, due to net differential pressure p. The parameter of interest is the maximum deflection d at the center of the membrane. Maximum deflection of a 100 µm square membrane (center panel) and a 1 mm square membrane (bottom panel) are plotted as a function of differential pressure up to ∼1 atmosphere and for different membrane thicknesses (10, 100, and 1000 nm). The cross marks represent fracture points calculated according to the model in Eq. 2. The fracture model assumes that the membrane has a pre-existing crack line that is 10 µm in length. Absence of a fracture point indicates that the membrane is able to withstand the pressure at least up to the top end of the calculated range.

In these calculations of the maximum deflection under differential pressure, it was assumed that the membrane is able to withstand the pressure up to the maximum amount displayed in the plots. In reality, the membrane may fracture at some point as the pressure is increased. The fracture toughness of a membrane is an important materials property that defines the resistance of a membrane to fracture. It is directly related to the lateral stress at which a membrane that contains a crack fails. This property of the membrane is measured using the same load-deflection experiments used to characterize the bulging profile. Previous measurements found the fracture toughness to be in the 6–7 MPa m^1/2 range independent of the membrane thickness.^81,88 The fracture toughness ( K Ic ) and failure differential pressure ( p failure ) are related by

K Ic = ( a 2 p failure / 2 dt ) π c

(2)

where 2 c is the length of the pre-existing crack line. ⁸¹ Thus, to calculate the failure pressure, an assumption must be made about the length of the crack. Fig. 4 shows the fracture points for a crack of 10 µm length. As Eq. 2 shows, the failure pressure has an inverse square dependence on membrane dimension and linear dependence on membrane thickness. Thus, it is advantageous from the standpoint of avoiding fracture to use thicker membranes with smaller lateral dimensions. Nevertheless, the calculations above also demonstrate the extraordinary toughness of the SiN_x free-standing structure: a 100 × 100 µm² membrane that is 10 nm-thick is able to withstand differential pressures up to an impressive ∼0.5 atmospheres.

The calculations above give an impression of the sensitivity of the expected amount of deflection and fracture point to window thickness, window dimension, and applied pressure. For devices that use irregular window shapes or where it is desired to map out the exact profile of the membrane deformation, elaborate models of the deformation were developed and tested by Yang et al. ⁸² Lastly, measurements specific to bulging of nanofluidic cells under liquid flow conditions were carried out in situ using a scanning transmission electron microscope (STEM). ⁸⁹ Beyond the design and handling considerations imposed by these calculations and measurements, one should be cautious in using ultrasonication as part of a cleaning process, since it can easily rupture the membranes.^46,90

Optical Properties of Free-Standing Silicon Nitride

Together with the mechanical stability and ease of fabrication, transparency of SiN_x windows is one of the key features that makes them appealing for use as sample support structures in various spectroscopic applications. In applications where light (in the IR, visible, and ultraviolet [UV] range of the spectrum) interacts with SiN_x windows, it is important from the viewpoint of experimental design to understand losses due to reflection at interfaces and due to absorption within the window material. In addition to informing design of the experiment, knowledge of the optical properties of the window is key to performing proper data analysis (e.g., in terms of determining the deposited energy within the sample volume in pump–probe experiments). Measurements of the frequency dependent dielectric function of SiN_x in the 1–24 eV range were previously performed by Philipp. ⁹¹ The dielectric function ε = ε 1 + i ε 2 is related to the complex index of refraction according to

( n + i κ ) 2 = ε 1 + i ε 2

(3)

where n is the real part of the refractive index and κ is the extinction coefficient. A summary of the measurements of the dielectric function is shown in Fig. 5. The left panel suggests that there is no significant absorption below a photon energy of ∼5 eV (corresponding to a wavelength of ∼250 nm). The observed absorption threshold is consistent with SiN_x being an insulator with a band gap of around 5.3 eV. ⁹² The right panel shows the refractive index for a range of wavelengths spanning the spectrum of interest in optical spectroscopy. Notice that there is little change in the refractive index from ∼2 in the 750–2500 nm wavelength region. OPEN IN VIEWER

Figure 5.

(left) Experimentally determined complex dielectric function of silicon nitride in the 1–25 eV range (reproduced from measurements by Philipp). ⁹¹ (right) The real part of the refractive index as extracted from the dielectric constant. The y-axis scale on the right side shows the reflectance of SiN_x at normal incidence.

Reflectance of p- and s-polarized light at an air–material interface is calculated from the refractive index according to Fresnel’s equations:

R p = | - n 2 cos θ + n 2 - sin 2 θ n 2 cos θ + n 2 - sin 2 θ | 2 and R s = | cos θ - n 2 - sin 2 θ cos θ + n 2 - sin 2 θ | 2

(4)

In the above, n is the index of refraction and θ is the angle of incidence. At normal incidence, both expressions collapse to

R p = R s = ( n - 1 n + 1 ) 2

(5)

Fig. 5 shows that the reflectance at normal incidence can be up to 20% for frequency ranges below the absorption threshold. There are several strategies for reducing the reflectance to near zero. If the experiment does not constrain the angle of incidence and light polarization, reflections can be mitigated by tuning these parameters: the most straightforward approach would be to use p-polarized light at Brewster’s angle (∼63° for SiN_x) to totally suppress reflection. Other strategies take advantage of multiple beam interference effects—as in a Fabry–Perot etalon—to suppress reflection for a narrow range of frequencies and angles of incidence. Under this scheme, the thickness of the window itself could be a key parameter to control in suppressing reflections, thereby improving signal quality. The theory of multiple beam interference is well developed and can be found in introductory optics text books. ⁹³ To demonstrate the benefits of tailoring the window thickness according to the application of interest, consider an air–SiN_x–water interface. At normal incidence, the reflectivity of the structure varies between

R min = ( 1 - n 2 1 + n 2 ) 2 and R max = ( n 2 - n 1 2 n 2 + n 1 2 ) 2

(6)

where n₁ and n₂ are the respective indices of refraction of SiN_x and water. ⁹³ In the visible range of the spectrum, the variation in reflectivity is in the range of 2–25%. The former can be obtained if the film thickness is made to be an integer multiple of half the wavelength of light. This example, while rudimentary, demonstrates the importance of both considering and taking advantage of interference effects in the structure.

A study by Poenar et al. measured the index of refraction and extinction coefficient of LPCVD SiN_x in the 400–800 nm wavelength range using spectroscopic ellipsometry and direct reflectivity measurements. ⁹⁴ Because of the narrower measurement range compared to the earlier work of Philipp, ⁹¹ this study provides better estimates of the extinction coefficient in the visible range of the spectrum. As shown in Fig. 6, the extinction coefficient, κ , is ∼10⁻³ across the 600–800 nm range and increases up to ∼10⁻¹ towards the lower end of the spectrum (i.e., at 400 nm). The corresponding attenuation depth, given by l = λ / 4 π κ , is ∼50 µm, and decreases to ∼0.3 µm at 400 nm. Therefore, thin SiN_x windows can usually be treated as transparent in the visible range. Optical absorption can be significant in the UV range of the spectrum, though, and should be taken into account: a study of the transmission properties of 100 nm SiN_x windows in the UV range demonstrates that, owing to absorption, window transmission is reduced from 90% at 400 nm to 10% at 200 nm. ¹⁶ OPEN IN VIEWER

Figure 6.

Extinction coefficients of several thin-film materials including low stress SiN_x (solid lines). Data based on ellipsometry and reflectivity measurements by Poenar et al. ⁹⁴ BSG and PSG refer to borosilicate glass and phosphosilicate glass, respectively. Reproduced with permission from the Optical Society of America.

Lastly, a recent theoretical study calculated the dielectric function of silicon nitride using density functional theory. ⁹⁵ The results were consistent with the measurements of Philipp for β-SiN—a crystalline phase of silicon nitride. The same study also examined both the effect of defects due to excess Si and the effect of hydrogen termination on the optical properties. In both cases, changes in the energy loss spectrum below 10 eV were found, suggesting the possibility of tuning optical properties by controlling the composition and surface chemistry of the SiN_x films.

X-ray and Electron Scattering Properties of Free-Standing Silicon Nitride

In applications where SiN_x windows are used as X-ray or electron transparent support structures, the main concern is reducing scattering of the probe from the window material. The objective here is twofold: to reduce background signal resulting from scattering off the window; and to avoid severely attenuating the probe beam (be it an X-ray or an electron beam). Obviously, using as thin a membrane as possible would be the ideal strategy. However, there are often other constraints that come into play when designing a suitable support structure. The most important limitation to reducing window thickness is the reduced mechanical stability and, subsequently, the susceptibility of the window to bowing and fracture. Other constraints may include thermal and charge transport properties in applications where large amounts of energy or charge are deposited into the window-sample structure. Fig. 7 shows the X-ray transmission of SiN_x windows of various thicknesses as a function of photon energy. In the higher energy range (e.g., above ∼5 keV), transmission is close to unity even when using 1 µm thick windows. In the low energy range (below ∼2 keV), windows with thicknesses above 100 nm have severely reduced transmission. It is therefore recommended for soft X-ray applications to keep the window thickness on the order of a few tens of nanometers. OPEN IN VIEWER

Figure 7.

X-ray transmission of 10 nm (red line), 100 nm (green line), and 1000 nm (blue line) thick SiN_x windows. Data obtained from the CXRO database, based on the work of Henke et al. ⁹⁶ The calculation assumes a density of SiN_x of 3.44 g/cm³. The left panel highlights transmission at energies below 2 keV, whereas the right panel is for energies in the 2–10 keV range.

At even lower photon energies, working with thicker windows is not an option and the trade-off between mechanical stability and X-ray transmission becomes more restrictive. To circumvent this problem, certain designs employ a porous membrane structure. For instance, several commercial suppliers of SiN_x windows offer what is referred to as “holey membranes.” These membranes are thick enough (∼200 nm) to allow for sufficient mechanical stability, have a large enough window area (∼1 mm²), but contain a number of holes for X-rays or electrons to pass through. The downside of this approach is the limited transmission as defined by the porosity of the membrane (typically below 25%). Alternative approaches implement grid-like support structures to enhance the stability of the film. For example, Törmä et al. fabricated ultrathin SiN_x X-ray windows (40 nm thick) with an area as large as 30 mm². ⁹⁷ Mechanical stability was achieved by embedding in the window design a grid structure made of polycrystalline Si for additional support. The open free-standing area of the window was 77%.

Electrons interact more strongly with matter than X-rays. Therefore, scattering effects are more severe when employing SiN_x windows as support structures in a TEM. There are few data in the literature on the electron transmission of SiN_x, but this property can be calculated from the theoretical elastic scattering cross-sections developed by Riley et al. for the 1–256 keV energy range. ⁹⁸ In order to calculate the electron transmission of the windows, the inelastic contribution to scattering also has to be considered. The inelastic contribution to scattering is derived from the empirical law σ inel = 18 σ el / Z , where σ is the scattering cross-section (integrated over the unit sphere; subscripts el and inel, respectively, refer to elastic and inelastic scattering), and Z is the atomic number. ¹⁰⁰ The total electron mean free path is then given by

Λ tot = 1 N σ tot ≈ 1 N σ el ( 1 + 18 / Z )

(6)

with N the number density of atoms. For a molecule, the total electron mean free path is obtained by averaging the scattering cross-sections of the composite atoms, weighted by the number densities, as follows

Λ tot = 1 ∑ N j σ tot , j

(7)

In the above, the summation j is over the type of atom. Finally, electron transmission, T, of the SiN_x window is calculated from the mean free path by

T = exp ( - t / Λ tot )

(8)

where t is the window thickness.

Fig. 8 shows the electron transmission of SiN_x windows of select thicknesses in the 1–256 keV range as calculated from Eq. 8. Note that these calculations are inherently conservative as they assume that any electron that undergoes a scattering event is “lost.” The derivation above can be readily modified to accommodate a more realistic scenario by excluding from the scattering cross-section the angular range within which scattered electrons remain useful to the experimental measurement. This is done by replacing the elastic scattering cross-section, σ el , by the partial elastic scattering cross-section integrated over the unit sphere minus some defined “useful” solid angle—as dictated by the experiment geometry. To this end, parameterized electron scattering cross-sections as a function of scattering angle can be found in the literature. ⁹⁸ Nevertheless, if we just consider the conservative results of Fig. 8, it is then apparent that at the standard TEM energy of ∼100 keV the windows should be < 10 nm thick. This conclusion is supported by the fact that vendors of SiN windows marketed as TEM support structures offer thickness starting at below 10 nm. OPEN IN VIEWER

Figure 8.

(left) Electron transmission of 5 nm (blue filled circles), 10 nm (red filled circles), 20 nm (green filled circles), and 40 nm (purple filled circles) thick SiN_x windows, 5 nm a-C (green open circles), 40 nm a-C (red open circles), and bilayer graphene (purple open circles). Data obtained from theoretical electron scattering cross-sections. ⁹⁸ The calculation assumes a density of SiN_x of 3.44 g/cm³ and a 1:1.1 Si:N stoichiometry. This composition is used to calculate the average scattering cross-section weighted by the number densities of the composite atoms. a-C and bilayer graphene were assumed to have a density of 2.0 g/cm³. The solid lines are trend lines (the original work modeled energies satisfying 2 ^m keV, with m being an integer between 0 and 8). (right) Selected area electron diffraction (SAED) images of an α-Si₃N₄ nanorod. The image on the bottom left is for a single crystalline section taken from the body of the rod. The image on the top right is for an amorphous structure corresponding to the tip of the nanorod. Image reproduced from Huang et al. ⁹⁹ with permission from Elsevier.

OPEN IN VIEWER

Figure 9.

(a) Vesicles offer nanoscale liquid confinement in soft phospholipid envelopes, with well-established formation protocols. (b) Robust graphene membranes can form nanoscale containers around liquid droplets. (c) Micrometer-diameter openings in otherwise intact vacuum windows provide access to underlying liquid-phase samples (see discussion of SALVI in text). Silicon nitride thin films spanning openings in silicon frames can be used to form a range of nanoscale liquid containers. Cell formats include: (d) those with integral spacers that require sample addition before sealing; (e) those with, e.g., nanoparticle spacers that can be loaded with liquid before or after closing; (f) those with integral spacers allowing (g) liquid loading after closing; and (h) those with an integral spacer but (i) more sophisticated fluidic connections allowing (j) structured flow channels for control over fluid flow useful for, e.g., TEM. Object sizes are not intended to convey scaling information. Figures in (d)–(j) are reproduced from Dwyer et al. ²⁰ with permission of The Royal Society of Chemistry.

Another important aspect to the use of SiN_x windows as support structures for TEM samples is the background signal resulting from the scattering of the electrons off of the window material. Even in the single scattering regime, where the window transmission is sufficiently high, part of the electron beam will undergo forward scattering at angles other than 0° (relative to the unscattered beam direction). Because SiN_x is amorphous, the pattern originating from the window has a diffuse character rather than the well-defined diffraction patterns generated by scattering from single crystalline structures. Fig. 8 is an example contrasting the electron diffraction patterns of single crystalline α-Si₃N₄ and amorphous α-Si₃N₄. The amorphous pattern is diffuse and contains broad ring-like patterns that correspond to the nearest neighboring distances in the amorphous material. In order to account for the contribution from the window, measurements are typically done on a blank window (i.e., without sample) and the background pattern is subtracted from the diffraction obtained for a sample-on-window structure. Scattering, however, is not strictly additive in this sense, but the approach does not require complex beam propagation calculations. Accounting for background signal is especially important when the sample of interest is not crystalline and will therefore not have the benefit of spatially well-defined diffraction peaks.

When operating the TEM in imaging mode, another concern is the reduced spatial resolution due to inelastic scattering of the electron beam from its interaction with the window material. De Jonge and Ross modeled this effect in nanofluidic cells. ¹⁰¹ The main conclusion was that for ∼1 µm liquid thickness, the window material itself does not contribute significantly to inelastic scattering. Instead, the resolution-limiting factor comes from inelastic scattering of the electrons with the liquid constrained between the cell windows.

Further reduction of electron scattering from the support structure can be achieved with use of lighter membrane material. The most commonly available option for TEM applications are amorphous carbon (a-C) films that are mounted on standard TEM grids. The thicknesses of these films can be < 10 nm. Fig. 8 shows that the transmission of a 5 nm a-C membrane is slightly higher than that of a SiN_x membrane of the same thickness. Even higher transmission can be achieved using graphene, few layer graphene (FLG), and graphene oxide support membranes—all of which are commercially available. In Fig. 8, the transmission of bilayer graphene at 100 keV is ∼98%. Yet, even with the higher electron transmission, whether it is advantageous to use SiN_x versus a carbon-based support structure depends on the specific application. The SiN_x fabrication process is scalable, whereas with graphene and a-C, the support film needs to be prepared separately then transferred onto a suitable TEM grid. Consequently, the transferred film tends to have folds and wrinkles that limit its usable area. More importantly, when it comes to building of devices, SiN_x is still more attractive because its deposition and processing are part of the highly standardized nanofabrication processes developed by the semi-conductor and MEMS industries. On the other hand, an important distinguishing feature of graphene and a-C films is their high electrical conductivity—in contrast with SiN_x, which is an insulator. It is expected that as progress continues in the fabrication technology of graphene and carbon-based materials, such films will be considered more seriously as viable alternatives for SiN_x.

Chemical Properties of Silicon Nitride Surfaces

The surface of silicon-rich SiN_x exhibits a rich blend of silicon-, oxygen-, and nitrogen-bearing surface species, whose particular mixture depends strongly on sample preparation and history (including “aging” and, in the case of commercial sources, vendor).^{20,64,66,102–106} This history dependence can perhaps be appreciated most easily through a survey of the silicon oxynitride (SiO_xN_y) literature, or through the variability of experimental outcomes when proper care is not taken.^66,107,108 A surface oxide coating readily forms in air (and can lead to an oxide gradient into the film),^102,108,109 and the susceptibility of Si₃N₄ to evolve ammonia by hydrolysis in the presence of liquid- and vapor-phase water should also be considered.^102,103 In spite of this, SiN_x shows very strong chemical resistance to a wide range of etchants and conditions, and indeed, etch resistance is the mechanism by which the film is made free-standing.^16,17,65 These films will, however, yield to etching in hot phosphoric acid (which requires quartz or other suitable vessel material). ¹¹⁰ The consequences of this rich tangle of surface chemistry are both fearsome and tantalizing. In aqueous solution, multiple surface acid–base equilibria exist, and one must be attentive to unintended effects arising from the surface charge (and surface charge screening) that can change as a function of solution pH and ionic strength.^{20,104,109,111–113} Under most conditions, the total surface charge density will reduce to

σ = | e | ( Γ Si 2 - NH 2 + + Γ Si - NH 3 + - Γ Si - O - )

(9)

Here, e is the electron charge, and Γ X is the surface concentration of the specified species X, and the surface will have an isoelectric point determined by solution conditions (especially pH), and of course by the detailed fabrication and processing history that determine the particular surface chemical composition. The use of literature values to calculate the isoelectric point is complicated by process history differences and by variability even in the reported values. ²⁰ The challenges are underscored by the range of values reported for the isoelectric point (IEP) and point of zero charge (pH_pzc) of SiN_x, such as 4.2–7.6^104,109 and ∼5. ¹¹¹ The surface can be generally conditioned using standard approaches such as exposure to a suitable oxidizing or reducing environment, or wet etching including hydrofluoric acid (and analogues) or piranha solution (and analogues).^{20,47,64,102,106,108,114–123} Specialized extensions of these methods can bring additional chemical flexibility.^20,64,104 Silane chemistry is a common route to surface chemical functionalization, but one that can be undermined by insufficient attention to either the SiN_x process history or to the demands of the silane reaction itself.^{64,106,124,125} While the silica surface chemistry literature thus offers strategies by which to modify SiN_x,^126,127 the fact that the films of interest to us are silicon-rich opens up the extensive, and well-established silicon surface chemistry literature for consideration. ¹²⁸ Hydrosilylation reactions are of particular interest: terminal alkenes and alkynes can be covalently linked with excess Si in the SiN_x films, either thermally or photochemically.^{105,106,117,120,121,129} The first step in these hydrosilylation reactions is to etch the SiN_x surface with dilute hydrofluoric acid; while the use of this processing step to remove the surface oxide coating (here, to permit subsequent covalent surface functionalization) is long-established, the exact chemical species left on the treated surface remains an area of active investigation. ¹³⁰ What remains undisputed, however, is that hydrosilylation is a potent and flexible method to install a wide range of different chemical moieties—from small molecules with passivation or reactive surface terminal functional groups, to biomacromolecules—on silicon-rich SiN_x surfaces.^{105,106,117,120,121,129} The approach offers straightforward and robust chemical tuning to prepare the SiN_x surface for compatibility with a diversity of applications, as well as to install additional functional elements.

Nanofluidic Sample Cells

Experimental efforts to gain ever-greater insight into the molecular world—on ever-faster timescales, with ever-greater spatial resolution, with ever-greater departure from tractable model systems, and with ever-greater diversity of samples—can encounter extraordinary barriers.^{17,38,39,131–133} In some cases, the challenges arise from the properties of the sample, itself—high optical absorption or scattering cross-sections that require extremely short path lengths, for example.^17,38,39 In others, experimental conditions can present challenges as simple, but unyielding, as the necessity for vacuum conditions, or as pernicious as probe-induced sample damage.^{38,39,134,135} Challenges can be additive and mutually amplifying in severity. The use of a sample container can play a similarly complex role. On the one hand, it may be used to ensure the requisite conditions are satisfied, but on the other hand, it introduces additional matter into the experiment’s path of interrogation. As a concrete example, attempts to use femtosecond multi dimensional IR spectroscopy to study energy redistribution of the hydrogen bonding network in pure water, rather than in isotopically substituted model systems, had to deal with the high optical absorption at the energies of interest that restricted the water layer thickness to ∼500 nm. Attempts to compress the water to the desired thickness between thin diamond windows could easily place at least one of the “sample” challenges in stark relief: expensive cracked windows without successful spectroscopic measurements (personal communication, ca. 2004). An additional challenge of using windows to physically constrain the water path length to submicron dimensions was that even optically transparent windows could generate a non-resonant contribution to the measured signal at short times, thereby potentially obscuring the short-time dynamics of the water. The joint consideration—the need for a well-defined submicron path length in a configuration that did not add on an (appreciable) non-resonant contribution—prompted the design of a sample cell constructed from thin-film SiN_x windows around a 500 nm-high fluid channel. ¹⁷ The 800 nm-thick windows did not themselves contribute to the signal, which was vital to reveal the ∼50 fs timescale for the decay of the memory of persistent correlations in the hydrogen-bonded network of pure water. The utility of this SiN_x material—or its thin-film brethren—for sample cell windows extends across the spectrum and into charged-particle techniques.^{17–20,35–45,53–58} We note, of course, that there is a tradition of alternative nanoscale sample chambers for spectroscopic studies, including vesicles, micelles, zeolites, and supramolecular entities.^136–140 Vesicles, for example, have been used in both ultrafast and single-molecule studies and use well-established formation protocols to use thin, but fragile, lipid bilayers to provide nanoscale confinement and low window signal (Fig. 9a).^136,138,139 Pores can be formed through the bilayer to permit limited access to the vesicle interior.^136–138 More recently, ∼1 µm thick, window-free liquid samples in vacuum have been achieved and used for transmission soft X-ray studies using a sophisticated arrangement of liquid jets. ¹⁴¹ The performance of this elegant solution is determined by sample properties such as the surface tension, viscosity, temperature, and vapor pressure, where solvent-optimization may lend generality to the study of solution samples using this approach. Nanofabricated equivalents of conventional sample cells and cuvettes,^{17–20,35–45,53–58} have a number of advantages relating, primarily, to their generality of application. Their construction provides for straightforward manipulation, even by hand, and one has access to materials that can offer a greater range of sample compatibilities—either directly or through modifications—than the materials in specialized implementations such as phospholipid-based vesicles. The basic structure (Fig. 9d–9e) can be modified by a host of conventional and unconventional nanofabrication approaches, allowing the integration of functional and structural elements such as electrodes to allow for more in-depth or flexible investigations.^42,142–144 Two additional key advantages emerge. First, the design provides for a convenient macroscopic-to-microscopic interface (Fig. 9f–i). This is seen, on the one hand, by the natural integration of the thin-walled channel with a convenient Si frame. By opening the nanochannel to the edge of the Si frames to create a through-channel, sample could be loaded into the preassembled cell by capillary action or, through appropriate design of a sample chamber holder, through forced flow.^{17,36,41,42,145–147} Second, the dimensions of structural and functional elements can be readily tuned within the existing fabrication infrastructure in order to adapt to different experimental needs. An exceptionally stringent test for thin-walled nanofluidic cells comes in the context of TEM (Fig. 9j).

Liquid (or “wet”) samples have long held allure for practitioners of (transmission) electron microscopy, but the vacuum environment needed for electron microscopy and the need for < 100 nm-thick samples left the field with only a handful of painstakingly executed early investigations.^{38,40,143,148–159} Both challenges similarly afflict a broader class of samples than liquids. Sophisticated and elegant solutions have involved reconfiguring the sample-containing region of the electron microscope itself and using low-vapor-pressure ionic liquids to cope with the vacuum conditions.^{149,156,160–165} Cryomicroscopy rather enviably protects the sample against the vacuum, but suffers from a number of severe limitations: the possibility to observe structural dynamics—such as aggregation, nucleation, electrodeposition, and self-assembly driven processes^{37,38,40,143,166–169}—is essentially eliminated, as is the ability to rapidly exchange samples on-the-fly. ²⁰ A liquid cell for TEM thus holds much appeal. The minimal requirement for such a cell is that it adequately preserve the integrity of a sample, with a thickness suitable for the desired spatial resolution (≤100 nm for higher resolution).^36,40,170 This thickness limit includes sample and window thicknesses, weighted by the scattering properties determined by their structure and composition. The channel height (but not window thickness) restrictions are less stringent on gas cells for TEM. ¹⁷¹ Sample cells for SEM on liquid samples face similar challenges in terms of the effective window thickness (as measured by its contribution to signal attenuation), but without the accompanying requirement to also control the sample thickness which will, instead, by determined by the electron beam penetration depth.^{101,172–175} To fully maximize the potential of in-liquid TEM for exploring nanoscale structure, it is necessary that the nanofluidic cell support through-flow.^{17,36,41,42,145–147,176} Once such a sample cell is loaded into the electron microscope, with connecting fluid lines external to the microscope, the sample can be easily introduced for examination without the delays of waiting for vacuum to reestablish during conventional sample loading. A sample damaged by the electron beam—either directly or through solvent-mediated damage—can be quickly purged and replaced. Surface chemical treatment (or use of alternative window materials) may be required to suppress unwanted interactions between the sample and window—or to encourage favorable interactions.^{20,36,177–180} Perhaps the most exciting research frontier is the pursuit of structural dynamics.^{36,37,41,132,143,167–169,179,181–185} Sample structural evolution can be triggered by changing the liquid composition (e.g., the solution pH, or the solvent polarity, or injecting different constituents such as cross-linkers and denaturants), and then observed with the spatial resolution of the electron microscope.

The promise of this new generation of conventional nanofabricated fluid cells for TEM was first demonstrated in 2003. ³⁷ In brief, the cell was hand-assembled using vacuum-compatible epoxy to seal two 100 nm-thick SiN_x windows around a 0.5–1 µm-high ring of SiO₂. A particularly exciting element of this particular cell was the inclusion of electrodes, a configuration that allowed for directly triggering the structural dynamics: electrodeposition of copper within the cell. The two key structural advances of this cell were the use of thin SiN_x windows and the incorporation of a hard spacer to define the liquid channel height. More streamlined fabrication approaches have tackled the limitations of using hand-applied epoxy to seal the cell: thin-film indium can serve to seal the sample cell and define the liquid reservoir height; direct wafer bonding has also been used for cell sealing and is compatible with the use of well-defined hard spacers to define the liquid channel height.^42,167 There is considerable variability and flexibility—depending on acceptable performance specifications—in the construction of devices using thin-film windows in a Si frame (e.g., SiN_x, SiO₂^{4,6–10,186,187}), including using hand-applied epoxy as spacer and for bonding (for a ∼5 µm gap), hand-applied nanoparticles as spacers, and patternable photoresist spacers.^{20,36–38,40,42,43,179,188}

Recent advances in materials science have broadened the choice of window materials beyond traditional nanofabrication materials and, in a sense, have resurrected the use of thin, carbon-based films. The principal advantage of such a material choice is of course that the low (average) atomic number and low atomic density have salutary effects on image contrast and resolution. ¹⁰¹ Of course, the materials—true for all such windows—must offer sufficient mechanical robustness, electron-beam resistance (including conductivity when charging is a concern), and chemical compatibility with the sample. The limitations of using conventional carbon-based polymer thin films as windows for TEM fluid cells were seen quite clearly in early developments in building cells for using SEM to examine wet samples.^101,175 Of a number of carbon-based films tested by Moses et al., only polyimide was found to be suitable, and then with a window thickness of 145 nm (50 nm could be achieved if a small leak was introduced into the microscope chamber), and without being used to constrain a thin liquid layer.^101,175 Carbon nanomembranes, a recent, ∼1.5 nm-thick, addition to the TEM support film toolkit, may hold promise for designing fluidic cells.^5,11,189,190 Large-area carbon nanomembrane films can be readily manufactured, but their more compelling characteristics stem from their synthesis by electron-beam-induced cross-linking of molecular precursors. One can choose from a variety of molecular precursors that will all generate ultrathin carbon-based films that can subsequently be chemically functionalized with relative ease. ¹⁹⁰ Thus, the film properties of these ultrathin, carbon-containing films can be determined with molecular-level control. ¹⁹⁰ In contrast, graphene offers a set of prospects determined by its unique composition and structure. Graphene, though, has received considerable attention in the field of TEM of liquid samples.^{38,179,183,191,192} Liquid-containing “nanobubbles” (Fig. 9b) can be formed between two graphene sheets laid together, reminiscent of earlier work using ∼30 nm-thick “hard” carbon films. ¹⁵⁰ The formation and size of these nanochambers is not directed and so sample-containing regions suitable for study must be found. In a sense, the use of this cutting-edge material has brought the “sample question” of TEM—and of TEM on liquids—full-circle: the nanocontainers are closed (aside from leaks; the contents are not added or exchanged after container formation) and the assembly is essentially by hand. As the fabrication footprint for graphene matures, however, assembly should become more deterministic and more sophisticated graphene-windowed fluid cells for TEM will undoubtedly emerge. Nevertheless, these graphene-bounded liquid cells have set the standard for imaging resolution, through the use of windows a single atom thick, where that atom is of low atomic number, and with achievable liquid path lengths of tens of nanometers.^{38,179,183,191} Nanofluidic through-flow grants incredible capabilities and for both breadth of study and depth of discovery, must remain a focus of effort, regardless of the particular window material or assembly process.^{17,36,41,42,145–147,176}

Many of the most impressive advances in thin-walled nanofluidics are driven by the demands of TEM, but have tremendous cross-over potential for other investigations. The stringent requirements of the sample cells for TEM may lead to designs that have base capabilities greater than those required by a different technique—SEM, for example, needs not have the thin sample path lengths of TEM—but the additional capabilities may nevertheless be beneficial. For example, when a particular sample cell limits the sample to, for example, a 100 nm-thickness, and a particular technique’s penetration depth exceeds that depth, the sample (cell) itself will provide ∼100 nm vertical resolution without the need for additional mechanisms. Thin-film SiN_x windows find use also in XPS studies under ambient, or reactive, environments.^193–196 If one is willing to accept (and in fact, design for) the otherwise ostensibly nightmarish prospect—aside from the nanopore sensing paradigm discussed below—of a hole in a liquid cell window chosen for its impermeability, then one may gain even greater experimental access to the chemical processes occurring in liquids (Fig. 9c). ¹⁹⁷ A custom-designed vacuum compatible microfluidic electrochemical device was created with a thin-film SiN_x cover. By forming a ∼2 µm hole in that cover, the authors were able to then use in situ liquid secondary ion mass spectrometry (SIMS) to probe the electrochemistry, including the electrode-solution interface. ¹⁹⁸ A similar fluid cell—one with protected thin-film SiN_x window with a single micrometer-scale breach to allow unfettered access—had earlier allowed the characterization of a single cell by both super-resolution optical microscopy and time-of-flight secondary ion mass spectrometry (ToF-SIMS) and the combination of SEM and ToF-SIMS.^173,199 The compatibility of this platform of a micrometer aperture in a thin SiN_x membrane with a range of characterization tools has led to the coining of the term System for Analysis at the Liquid Vacuum Interface (SALVI).^198,199 The performance of this platform has been well-established and was recently anchored in a mini-review of in situ monitoring of catalytic processes in aqueous environments. ²⁰⁰ The available fabrication tools for Si-based cells, coupled with favorable material properties, allow for other functional architectures such as integrated microwells instead of a single sample volume. ²⁰¹

Beyond structural modifications of the basic nanofluidic platform, advanced detection schemes can be used to compensate for some material and fabrication limitations, and implementation of different experimental techniques can broaden the achievable insights (Fig. 1a). For example, an annular dark field detector can be used to yield atomic number contrast. Thus, when samples themselves offer low contrast—a prime example being biological material in a surrounding aqueous medium ¹⁰¹ —spatial resolution can be enhanced by using an annular dark field detector in conjunction with heavy-atom-containing labels such as gold nanoparticles.^34,202–204 This approach was used to enhance the spatial resolution in a fluidic cell for TEM where the liquid path length was micrometers thick, instead of the desired nanometer thickness.^34,202–204 Electron energy is also an important consideration when developing optimized sensing schemes—whether it be the incident electron energy or the transmitted electron energy. For example, energy-filtering of the electrons just before detection can improve the spatial resolution of the image, albeit it at the cost of detected electron intensity.^{11,40,205,206} Energy discrimination of the sample-transmitted electrons leads naturally to a dramatically broadened scope of sensing capabilities, including electron energy loss spectroscopy (EELS),^89,207,208 and to perhaps its ultimate extension, vibrational spectroscopy by EELS, with extremely high spatial resolution and different selection rules than optical approaches. ²⁰⁹ By opening up the interior of the TEM to liquid-compatible samples, other investigation tools become available, including X-ray energy-dispersive spectroscopy (XEDS) ²¹⁰ and time-resolved studies including and extending well beyond dynamic TEM (DTEM).^131,132,169

Nanopore Single-Molecule Sensing

Typical nanopore devices are constructed from the basic structure outlined in Fig. 2c, a nanofluidic channel through a thin, free-standing SiN_x membrane. Nanopore holders are constructed so that this nanopore provides the sole fluid path connecting electrolyte solutions on both sides of the membrane. The insulating, impermeable SiN_x thus acts as a physical barrier and as a support for the channel through which ions and molecules can pass under the influence of an electric field or other suitable driving force.^{21–29,211–216} The most common sensing configuration is the so-called resistive-pulse sensing mode (Fig. 10). In brief, ions pass through the nanopore in response to the applied cross-membrane electric field, giving a typically pico- to nanoampere current. The nanopore conductance magnitude is determined by the solution conductivity (and solution pH, through its effect on surface charges), nanopore size and shape, and also by the nanopore’s surface charge.^{24,113,217–221} Passage of an analyte through the nanopore can generate a perturbation in the measured current that arises from physical displacement of solution ions by the analyte and from more complex interactions between charges in solution and on the surface of the analyte and nanopore.^{24,28,52,221–223} It is convenient to generically refer to the current perturbations as “blockages,” although analyte properties and sensing conditions may lead even to current enhancements. ²²¹ With suitable nanopore and experimental design, the measured current perturbation generates a signal from a single molecule at a time that can be used for chemical identification or profiling. Nanopores can also be used in alternative sensing schemes, including for force spectroscopy and single molecule optical sensing, that will be outlined after a brief discussion of their fabrication. OPEN IN VIEWER

Thin-film SiN_x is a standard material for nanofabrication, but small-diameter (< 10 nm), solid-state nanopores formed in thin-film membranes historically represented extreme fabrication challenges. Laborious efforts requiring significant operator expertise saw charged-particle milling—in transmission and scanning electron microscopes, by focused ion beams, or in helium ion microscopes, for example—deliver the desired structures, albeit with fabrication yields that dramatically diminished with sizes < 50 nm.^{21,23,25,27,144,224–231} Chemical etching of damage tracks left in thin Si-based thin films by relativistic ion bombardment can produce single nanopores and (irregularly spaced) nanopore arrays, but the approach does require the use of large-scale facilities. ²³² More specialized approaches exist alongside efforts to draw on conventional techniques such as atomic-layer deposition (ALD).^{21,23,25,27,144,224–227,230,233,234} Dielectric breakdown has emerged as a simple, yet extremely flexible and powerful, technique to fabricate nanopores: it requires little more than the usual apparatus used for nanopore sensing and instantly wets the nanopore (contrary to TEM-based methods, for example, where surface contamination poses wetting and other challenges).^235–237 In overview, nanopores as small as 1 nm in diameter and ∼10 nm long can be fabricated with some flexibility in fabrication method. Characterization of nanopores can be done using charged-particle microscopes and related techniques such as electron tomography and EELS,^{219,220,238–241} or by much less instrumentally intensive (ionic) conductance-based approaches.^{113,217–220} A conductance-based characterization is gentler than charged-particle-based approaches and is also more naturally connected to the nanopore surface chemistry—especially to any organic monolayers that may be damaged in the TEM, for example—and to the usual measurable of nanopore single molecule measurements (that is, to the ionic current as a function of time). The advantages of using a conventional nanofabrication material are seen naturally in applications, some of which are outlined below, in which additional functional elements are added to the base nanopore nanofluidic platform. Nevertheless, efforts to continue to develop nanopores using other thin-film materials such as molybdenum disulfide, but with graphene seemingly foremost among them, continue; approaches often see the use of an underlying thin-film SiN_x support.^242–248 While lacking the electronic properties of graphene, carbon nanomembranes (see discussion above) present an appealing potential ultrathin carbon-based alternative. ²⁴⁹ Nanopores, fabricated using a suitable selection of means and materials of construction, and employed in a variety of configurations and approaches, are an exciting and promising motif for single molecule sensing, confinement, and manipulation.^{20,21,23,27,28,59,222,250–270}

Conclusion

Free-standing thin-film structures feature both prominently and unobtrusively in a wide range of tools and methods for molecular-level research. Suitable thin films can provide transparency through composition and path length, and that transparency can extend across a wide range of spectroscopically and crystallographically important wavelengths to include charged-particle transparency. Highly localized micro- and nanoscale apertures can be opened in sufficiently robust films when transparency cannot be otherwise achieved. Such transparent, or locally transparent, thin-film window constructs thus enable studies ranging from linear and ultrafast spectroscopies across the visible, IR, and X-ray to X-ray photoelectron spectroscopy, mass spectrometry, and transmission electron microscopy, and many others. A less inconspicuous role uses the mechanical properties of particular thin-film windows to not merely contain samples within well-defined environments, but to provide protection from severe challenges to sample integrity such as vacuum conditions necessary for some experimental methods. The use of thin-walled nanofluidic cells to allow transmission electron microscopy on liquid samples is but the most dramatic of such a role. Thin-film elements can be more forcefully engaged, as seen in nanopore single molecule sensing applications. Nanochannels formed through thin films and having tailored size, shape, and surface chemistry, can be used as a powerful class of tool for single molecule sensing and manipulation. The capabilities of this basic element, alone, have dramatically expanded the toolkit of single molecule science, but it can be even further expanded when additional functional elements and techniques can be integrated with the core nanopore structure. Thin-film SiN_x membranes, by virtue of their composition, properties, structure, and fabrication compatibility and infrastructure, figure prominently in all of the structures, devices, and approaches featured in this review article. We therefore used this canonical structural and functional element to demonstrate key performance criteria and principles that will govern the development and application of new materials and structures. Free-standing thin films are truly an enabling element for molecular science.