Nanostructured Interfaces for Single Molecule Sensing and Molecular Fingerprinting

Nanoscale interfaces can have a profound influence on sensor performance, arising from the increased surface-area-to-volume ratio on length scales <100 nm, and often on the emergence of new phenomena on this length scale and even enhancement of existing phenomena. These interfaces can be used to form sensing devices capable of molecular sensing and fingerprinting. Attaining rapid and reliable molecular information with low analyte concentrations and minimal instrument overhead is crucial for many fields including the pharmaceutical industry, food quality analysis, biomedicine, water quality analysis, etc., to meet the current demands of sample analysis. Nanoscale elements in these nanosensors, in amalgam with other physical and chemical driving forces are useful for attaining low limits of detection with the ultimate goal of observing one molecule-at-a-time. This proposal contains two approaches to develop nanostructured sensors—one optical and one non-optical—to reach this goal. The first study is designed to develop a non-optical sensor—a solid state nanopore—for carbohydrate biopolymers—a class of abundant biomolecules that nevertheless have not been extensively characterized like other biomolecules (DNA or proteins), due to inadequate sensing capabilities to easily tackle the molecular complexity by classical methods alone. Additionally, methods to enhance and control the pore surface chemistry are investigated. Second, a series of accessible and low-cost surface enhanced Raman substrates are fabricated on a range of supports using electroless gold plating, to create optical sensors with vibrational selectivity and multifunctional capabilities.

(black squares), and hyperbolic (magenta diamonds); and with an added access resistance term, by Equation xviii S1 (hollow markers)cylindrical with length (small circles) and cylindrical with a . effective length [43] (large circles). To not bias further analysis with an explicit choice of nanopore profile, the ,TEM expt were fit to Equation S1 with bulk and surface from the cylindrical model weighted by fit parameters: independent trials in ~8. 6, 9.8, 9.9, and 13.6 nm (left to right), are shown in    , carbohydrates are a vital class associated with mediating many   biological functions, including cell-cell interactions, cell proliferation, apoptosis   and microbial interaction with the body and thus are vital for disease detection and as drug targets. 4,5,6 Given that difficulties have severely hindered the use of conventional chemical analysis tools and methods, it is essential to develop new tools that may be better able to address key issues in glycomics-the comprehensive study of structure and function of carbohydrates. We proposed to use a nanopore-an effective and robust non-optical singlemolecule sensing device-operated on an ostensibly simple principle. Nanopore is a nanometer scale channel that connect two electrolyte compartments. A typical nanopore setup is shown in Figure 1 information about the molecule. This current blockage is characteristic of the molecular structure at a particular pH, salt concentration, temperature, pore shape, etc. 8 and by analyzing the current events using a given pore, indications of the molecular size, length, charge and concentration can be determined. Two of the most compelling advantages of nanopore sensors are the requirement of small sample volume and less complicated sample preparation that avoids molecular labelling, chemical modification or surface immobilization of the molecules so that its structure and properties are preserved. By providing a very small volume for the ions to pass through, nanopore sensors ensure that a single molecule interrupts a significant fraction of ions as it passes through the pore. This large signal eliminates the complication associated with enzyme amplification and attachment of identification groups, such as fluorophores. 7 This is vitally important because while DNA studies benefit from amplification techniques such as polymerase chain reaction (PCR), there is no equivalent for glycan analysis. Figure 1.2: Solid-state nanopore experimental setup. Nanopore is mounted in a PTFE cell and connects two electrolyte reservoirs. Ag/AgCl electrodes are immersed in each well to apply a voltage that results in a current corresponding to the pore size.

LIST OF TABLES
The first nanopore-based sensors used naturally-occurring nanopores-nanoscale devices mainly secreted by bacteria as exotoxins. They spontaneously insert into lipid bilayers and act as nano-gates for selected molecules to pass through. 7 While these biological nanopores have atomically highly reproducible composition and structure, the supporting lipid bilayer is not stable for use over an extended time period and the fixed pore size means they have a limited set of molecules they can sense in simple translocation-based schemes. To overcome these difficulties synthetic pores were introduced which are more durable, robust and size tunable. 8 Synthetic pores are stable over a vast range of pH and temperature conditions, compatible with methods to create pores of tunable sizes to match the analyte properties, amenable to surface chemical modification (with appropriate care) to manipulate the surface chemistry as required. 9 The need for minimal sample preparation and sample volume furthermore supports the prospect of a commercial sensor. 10 Silicon nitride is a material that had gained lot of attention due to its immanent robustness, nanofabrication compatibility, mechanical strength, chemical resistance, and dielectric strength. 11,12 With the intention of extending the sensing capabilities of these novel sensors for sugar structure and property determinations, we proposed to use silicon nitride nanopore sensors, that could enable the detection of as little as one molecule of sugar at a time. 7 While silicon nitride synthetic pores are size tunable by transmission electron, scanning electron, or He-ion microscopes we fabricated nanopores of desired size in-house by a recently discovered simple process called dielectric break down 13 and sizeanalyzed them by a conductance-based method based on a recently developed theoretical framework. 14 DNA translocation through synthetic nanopores has been widely explored and being a charged molecule, its transport via electrophoretic movement is widely studied. 1 In addition to electrophoretic movement , electroosmotic transport also plays a role in nanopore experiments if the nanopore wall is charged. 15 Kasianowicz et al. demonstrated polymer translocation through nanopores using polyethylene glycol (PEG) as the test molecule that indicated uncharged molecules could still be profiled by ion-adsorbed electrophoretic nanopore sensing. 16 Carbohydrates are polymers in biological systems and in theory should act similarly to DNA if charged and similar to PEG if uncharged. We intended establish the fundamentals of using nanopore sensors for single molecule sugar characterization. While no literature exists on using nanopore sensors for polysaccharide structure determination so far, there have been promising initial attempts to use oligosaccharides with nanopores, albeit with protein nanopores and not the solid-state nanopores of interest to us and the nanopore community. 18,19,20 We explored the possibility of extending the sensing ability of solid-state nanopores to polysaccharides as shown in chapter 2. Further, different strategies to enhance the nanopore sensing platform were studied and described in later chapters as well as adventitious nanostructured platforms discovered during these studies.
Enhancing nanopore sensing by surface coating the pore interiors with metal thin films are investigated in chapter 3. Chapter 4 describes a procedure developed for micro/nano patterning, inspired by work in chapter 3. To include a complete study of these nanochannels, chapter 5 and 6 provide a theoretical model to answer complications associated with nanopore size and shape determination.

OPTICAL SENSOR: SURFACE ENHANCED RAMAN SUBSTRATES
Raman spectroscopy is a vibrational spectroscopy technique for sample chemical analysis and vibrational fingerprinting. An energy level diagram for Raman scattering is shown in Figure 1.4. It provides a scope of information on functional groups of a molecule and allows vibrational fingerprinting. With no interference from water vapor, Raman spectroscopy can be used to analyze water-based samples in contrast to its counterpart-IR spectroscopy. Despite this advantage, the Raman scattering process produces generally weak signals, and this reality limited widespread adoption and routine use of the technique. Later, it was discovered that coinage metals (e.g.: gold, silver, copper) could enhance the molecular Raman signal. Based on this concept, surface enhanced Raman spectroscopic (SERS) substrates as shown in Figure 1.5-suitably coinage metal coated devices-were introduced that enhance the molecular Raman signal. 21 These consist of nanoscale metal structures that can provide higher surface area-tovolume ratio. Moreover, an incident beam of appropriate frequency could generate localized surface plasmon resonance of the metal that enhances the Raman signal of the analytes in the vicinity. Metal nanoparticles in the vicinity to each other could create hot spots that further enhance the Raman signal. Furthermore, nanoparticles of different shapes have been tested to create hot spots, e.g. pillars etc. and to provide higher surface area-to-volume ratio for the analyte binding as well as to provide sharp edges (as in nano-stars, pyramids and cubes) to enhance signal. 22,23,24,25 There is still some debate over the mechanism of enhancement, and out of the proposed mechanisms, electromagnetic and chemical enhancement mechanisms are widely accepted. 26  respectively. Obtained raw SERS spectra will be processed by custom-written Mathematica programs to baseline-correct the spectra. The relative performance of each substrate will be evaluated using calibration curves, and limit of detection (LOD) and enhancement factor (EF) calculations. For demonstration purposes, we would use 4-nitrobenzenethiol (NBT, also known as 4-nitrothiophenol) as the analyte to study the enhancing of its SERS signal by our series of SERS substrates.   in the United States, including ~100 deaths. [14][15][16][17][18][19] . Glycan samples can be challenged by heterogeneity and low abundance in addition to chemical and structural diversity, so while new analysis tools have been broadly called for, 12,13,20 singlemolecule-sensitive methods are a particularly compelling goal for glycomicsmore so given the absence of sample amplification techniques analogous to PCR for DNA sequencing 21 . Nanopore single-molecule methods have emerged as a powerful tool for characterizing DNA and proteins including aspects of sequence, structure, and interactions. [22][23][24][25][26][27][28] Monomer-resolved length determinations of more prosaic polyethylene glycol samples further buttress the potential of suitably configured nanopore assays for the analysis of polymers with biological utility. 29 The simplest implementation for nanopore measurements places the nanopore-a <100 nm-long nanofluidic channel through an insulating membrane-between two electrolyte solutions (Figure 2.1). Ion passage through the nanopore in response to a voltage applied across the pore gives the baseline "open pore" current, i 0 ; passage of a molecule into, across, or through the nanopore disrupts this ion flow to give a blocked-pore current, i b . A discernible current perturbation reveals the presence of an analyte, and the sign, magnitude, and temporal structure of i b depend strongly on size and shape of the analyte-and of the nanopore-and on the applied voltage and bulk and interfacial charge distributions. It thus provides insight into analyte presence, identity, and properties, including interactions between the analyte and pore interior or surface. [29][30][31][32] Analysis of the resistive-pulse characteristics of a sample offers the potential to glean molecular-level insights, but the i b characteristics can also be used more simply as benchmarks in quality assurance assays where atypical i b signal sample impurities.
Much groundwork must be laid, including proof-of-principle experiments, if nanopore methods are to emerge as a tool for glycan profiling-and by extension as a tool for -omics writ-large (spanning genomics, proteomics, and glycomics).
Protein nanopores, polymer, and glass-supported nanopores have been used to detect sugar-pore binding, polysaccharides, and enzyme-digested oligosaccharides. [33][34][35][36][37][38][39][40][41][42] While solid-state nanopores in thin (~10 nm) membranes have been often portrayed as the preeminent nanopore platform, their use to profile classes of molecules beyond DNA and proteins is in its infancy. These nanopores can be size-tuned 43  The potential for integration of additional instrumentation components, such as control and readout electrodes, around the thin-film nanopore core, is especially compelling. 28,44,45 Recent (nanopore-free) work on recognition electron tunneling measurements on polysaccharides, for example, has reaffirmed the importance of a nanopore development path that values augmented nanopore sensing capabilities. 46 A key question concerning the use of SiN x nanopores for polysaccharide sensing is whether this fabrication material is compatible with sensing glycans. The often challenging surface chemistry of SiN x (giving rise to a complex surface charge distribution) 44,45,47 may lead to analyte-pore interactions that hinder or prevent its use. Variability in polysaccharide electrokinetic mobility arising from differences in molecular structures may exacerbate the effect of these interactions. These issues become particularly important when analyte translocation through a constricted pore is required, such as in transverse electron tunneling measurements. 28,46 The aims of the present work were threefold: (1) to introduce and test the feasibility of SiN x nanopores for sensing polysaccharides; (2) to explore the preliminary performance of this class of nanopores in this implementation; and (3) to gauge the prospects of a clinically relevant assay to detect a toxic impurity in the anticoagulant heparin. The broader implications of the successful use of SiN x -a readily nanofabrication-compatible material-to form the nanopores would be to conceivably smooth the path to large-scale production and to provide a platform in a heparin sample [14][15][16][17] to make the analysis of heparin (~16 kDa) and OSCS by nanopore a compelling experimental test with clinical relevance. Analyte was added to the headstage side ("cis-" side, according to nanopore convention) unless otherwise noted, and applied voltages were referenced to the ground electrode ("trans-" side) on the other side. 2) that physically separated electrodes and nanopore, events were only detected when A1 was injected into the well proximal to the nanopore, thus supporting a signal generation mechanism involving interaction with the nanopore and not with the electrodes. This result did not, however, distinguish between passage-free collision with the nanopore opening ("bumping" or "blocking") or translocation through the pore. 32 Either mechanism (including extending the idea of "bumping" or "blocking" to allow for transient interactions of the analyte with the pore mouth), though, has the potential to deliver analytically useful sensing performance. Low analyte concentrations challenge the direct investigation of polysaccharide translocation through small, single nanopores. In one experiment to investigate this, a solution of A1 was added to the headstage side of a ~22 nm-diameter nanopore and was left overnight with a +200 mV applied voltage. The initially analyte-free contents of the ground-stage side were then transferred to the headstage side of a fresh ~17 nm-diameter pore,

Introduction
and an appreciable number of A1-characteristic events (182 in 1 h) were detected again at +200 mV. Acid digestion was used as a signal generation and amplification technique (complete details in the Supplementary Information) to convert A1 polymers to many smaller fragment-derived species absorbing at ~270 nm. 51,52 This spectrophotometric assay (Supplementary Figure 2.3) was used to confirm translocation of polysaccharide through a ~9 nm SiN x nanopore.
The analyte-induced translocation blockage current, i b , is expected to be determined by the properties of the analyte and its size relative to the nanopore,  , so that the overall charge density of this molecule was expected to be higher than A1. Further analysis was consistent with alginate A1 having a ratio of guluoronic (G) to mannuronic (M) residues exceeding that of A2, with values from IR spectroscopy of ~63%G/37%M and ~57%G/43%M, respectively. 48 Nanopore profiling of A2 showed differences compared to A1. Using the same electrolyte for A2 as for A1, measurements generated a ~7-fold lower event frequency with longer durations for A2 compared to A1, in spite of at the 75-fold higher A2 concentrations required for reasonable measurement times. Enzymatic digestion of A2 produced events at a higher frequency than for undigested A2, but still at lower frequency than for A1. The events for the digested sample of A2 were ten-fold shorter-lived than for the A2 polymer, but not appreciably different in terms of blockage depth (Figure 2.3).   interfaces. 55,56 The effects of these and more complex interfacial phenomena emerged in one of the more startling observations in this work: that the voltage polarity for signal generation with both alginate samples was opposite to that expected for electrophoretic motion of an anionic polymer, whereas for heparin the voltage polarity was consistent with electrophoresis.
In addition, when comparing the two alginates, the more charge-rich A2 was detected at a lower event frequency than A1. Nanopore-based studies with The failure in 2008 to detect an OSCS contaminant in clinical heparin samples had previously led to patient morbidity and mortality, [14][15][16][17][18] so that our ability to use a simple nanopore-based assay to quantify heparin levels and detect OSCS at clinically meaningful contamination levels, is itself significant. In a broader sense, we expect that these initial results exploring polysaccharide structure can, by analogy with earlier nanopore DNA and protein sensing supporting genomics and proteomics, spotlight the potential of using nanopores as a tool for glycomics. The demonstration of polysaccharide translocation through nanofabrication-compatible SiN x nanopores portends the development of more sophisticated sensing schemes as seen in the use of nanopores for genomics.
Similarly, the successful use of chemical tuning-of electrolyte composition and by enzyme addition-to alter the nanopore signal generated by diverse polysaccharides suggests that nanopore glycomics might borrow from and extend upon similar approaches developed for nanopore genomics. There is an ongoing need in glycomics for new tools to cope with the analytical challenges caused by the structural and physicochemical complexity of polysaccharides, and by the often inherently heterogeneous nature of naturally derived carbohydrates. The demonstrations of nanopore sensing here provide a beachhead for ongoing efforts to develop solid-state nanopores as a promising platform technology for glycomics.

A full listing of the experimental details is available in the Supplementary
Information. Nanopores were formed via dielectric breakdown 43  Current blockages were extracted using a current-threshold analysis. All applied voltages are stated with the polarity of the electrode on the headstage side relative to ground on the ground side of the sample cell. 34 Code Availability. Labview source code to view the current event files can be supplied upon request.
Data Availability. The datasets generated during the current study are available from the corresponding author on reasonable request.

Author Contributions
All authors have contributed to, and approve, the manuscript. resistance, and dielectric strength [4][5] . Silicon nitride is thus ubiquitous as a structural and functional element in nanofabricated devices where it plays a variety of roles 2, 5-8 . Its surface chemistry, however, presents especial challenges given the complex mixture of silicon-, oxygen-, and nitrogen-bearing surface species 5 . The nominal surface modification of silicon nitride is frequently carried out in practice using silane-based modification of a silica layer that may itself not be welldefined 9 . Thus, there remains both a need and opportunity to expand the suite of approaches useful for surface functionalizing silicon nitride directly. Electroless deposition is a particularly compelling approach to film formation: deposition proceeds from solution allowing the coating of three-dimensional surfaces, including surfaces hidden from line-of-sight deposition methods; no electrochemical instrumentation is required; no electrical power must be supplied nor must the substrate be conductive; there is no need for expensive vacuum deposition equipment; and a variety of classical physicochemical parameters such as reagent composition, solution properties such as pH and viscosity, and temperature, are available to tune the film properties [10][11] . There is a wealth of familiar approaches for the electroless plating of substrates such as polymers, for example, but no established prior art for the direct metal-cation-mediated electroless plating of gold onto silicon nitride [12][13] . A particularly compelling sequence exists for the electroless gold plating of poly(vinylpyrrolidone)-coated polycarbonate substrates (Au/PVP) 13 : direct sensitization of the PVP surface with Sn 2+ , activation by immersion in ammoniacal silver nitrate to oxidize the surface Sn 2+ to Sn 4+ by reducing Ag + to elemental silver (producing, also, a small amount of silver oxide), and finally gold plating by galvanic displacement of the silver with reduction of Au(I) to Au(0) accompanied by the oxidation of formaldehyde.
Amine and carbonyl groups in the PVP layer were proposed to complex the tin cation during sensitization 13 . Extending this approach, Sn 2+ has been reported to complex effectively with oxygen-rich polymer surfaces 12 and with quartz and silica substrates 10, 14-16 . Tin(II) sensitization has also been reported on NaOH-roughened surfaces 17 , suggesting that a specific chemical interaction may not be essential 18 , and underscoring the utility of electroless plating for rough and high-surface-area surfaces where physical deposition is challenged 19 . In principle, though, a smooth silicon nitride substrate with a well-defined silica surface layer should be amenable to direct tin sensitization. Yet, electroless deposition of gold on planar silicon nitride has been limited to routes requiring the use of a silica layer with organic linkers and metal layers between the silicon nitride and gold overlayer 18 . In the first case, covalent attachment of an organic monolayer using silane chemistry can be beneficial for film adhesion, but adds operational complexity 18 and can constrain downstream processing conditions. In the second case, the intervening layers may lend beneficial properties, or may similarly introduce compositional constraints on applications, or morphological constraints on the final gold film nanostructure. Regardless of the ability to carry out a silica-based modification, it does not eliminate the benefits of a direct functionalization of silicon nitride. We present a dramatically simplified electroless gold deposition method in which we eliminate the initial covalent attachment of an organic monolayer to the substrate, and in which we do not need to initially mask the silicon nitride surface chemistry with a silica overlayer. Our method directly sensitizes the silicon nitride substrate with a Sn 2+ solution, followed by a series of metal ion treatments in which we exert control over the gold film thickness using process time and temperature. Film thicknesses ranged from 30 to 100nm for deposition times from 0.5-3h, and temperatures of 3 and 10°C. Elemental analysis of the gold film was carried out by energy-dispersive x-ray spectroscopy (EDS) and by x-ray photoelectron spectroscopy (XPS).

Full
Characterization details are provided in the Supporting Information.
Scheme 3.1. Electroless plating of silicon nitride. The silicon nitride-coated substrates are plasma-cleaned of organics and HF-etched before the surface is exposed to Sn 2+ ions which are oxidized during the redox-driven deposition of an elemental silver layer. Gold plating begins with galvanic displacement of the elemental silver.  Scheme 3.1 thus follows the plasma-based cleaning steps with an HF etch step that removes oxide and H-terminates the surface 22 , and ends with the gold plating treatments 13 . We note that in the absence of the HF-etching step, chips would sporadically be coated with patchy gold layers, but no uniform high-quality gold films were observed on these chips even after 3 hours in the gold plating solution.    Measured spectra from 1cm 2 silicon nitride substrates soaked in 0.01M NBT for 5 minutes: from a substrate electrolessly gold-plated at 3°C for 3 hours (red), from the same chip plasma cleaned, annealed at 280°C for 20 minutes, and plasma cleaned again before NBT exposure (blue), and from a sputtered (30s) gold film (black).

DISCUSSION
While the electroless gold plating was strongly sensitive to the surface preparation of the silicon nitride, we note, for completeness, that the exposed silicon at the edges of the chips was consistently gold-plated, regardless of whether the wafer was treated with HF, HNO 3  in the second, we prepared the tin sensitizing solution without adding tin. In none of the cases was the appreciable widening of the O1s peak observed. The broad, low-amplitude 102.5eV Si2p peak that appeared after Scheme 3.1 tin-sensitization of silicon also appeared after tin-free control processing, and it suggests submonolayer oxygen coverage that can arise from aqueous processing 23,26 . The analogous formation of silicon oxynitride 27-28 on the silicon nitride substrate would be more difficult to discern from the main Si2p peak due to spectral overlap. Tin oxidation states can be difficult to definitively identify by XPS measurement 16,29 , but the shifts of the best-fit ~487eV Sn3d 5/2 peak to lower binding energy after the addition of silver(I) ions to both substrates (by ~0.5eV for SiN x and ~0.15eV for Si), would be consistent in direction with the oxidation of tin(II). The Sn3d 5/2 peaks were affected by the substrate preparation, with ~0.2eV greater width on silicon and silicon nitride substrates that had not been treated with hydrofluoric acid, with an accompanying ~0.4eV shift to higher binding energy on the silicon substrate.
Overall, the XPS spectra suggest complex roles for oxygen and tin in the surface sensitization steps and, while the detailed mechanism of sensitization remains unresolved, adherence to Scheme 3.1 exposed the silicon-rich LPCVD silicon nitride surface for direct surface modification and yielded high-quality gold films.
In fact, in spite of complex and challenging surface chemistry, the choice of silicon nitride as a substrate opens a panoply of possible applications for consideration, and the use of a solution-based gold plating method allows us to coat surfaces that are difficult or impossible to reach by line-of-sight metal coating methods. We paid special attention in our development to be able to coat freestanding thin silicon nitride membranes. As a final demonstration of the capabilities of this method, we electrolessly gold plated micropore arrays fabricated in thin (200nm) silicon nitride membranes. and gold film thickness, either by fabricating pores with smaller initial sizes, or by increasing the plating time, this electroless plating process can also be used to 53 physically tune the pore dimensions. This method thus provides access to surfaces that may not be accessible to line-of-sight methods, and it moreover provides control over both surface physicochemical properties and physical dimensions of surface and internal pores 7 . In addition, the method is well-suited for tuning and enhancing the properties and performance of thin film and pore-based devices.

AUTHOR CONTRIBUTIONS
All authors have given approval to the final version of the manuscript. / ‡These authors contributed equally.

FUNDING SOURCES
NSF CAREER award CBET-1150085, in part by NSF EPSCoR Cooperative Agreement #IIA-1330406, and by the University of Rhode Island.

ACKNOWLEDGEMENTS
We thank Sarah Golden for custom software used in data analysis.  where the need for versatile metallizing approaches is clear. [3][4][5][6][7] The most common solid-state nanopores are <100 nm-diameter nanofluidic channels formed through <100 nm-thick, free-standing SiN x films, and nanopore-integrated metal films can enhance sensing capabilities by serving as optical elements such as light shields and plasmonic films, as electrodes for tunneling and other molecular control and sensing functions, and as a means to tune nanopore size and surface chemistry. [3][4][5][6][7][8] The nanoscale dimensions of the SiN x film and pore can be significant barriers to efforts to incorporate such functional metal films, particularly when the interior of the pore must be metallized. Solution-based metallization routes offer an appealing route with natural compatibility with nanofluidic devices. Surface capture of nanoparticles-by specific and nonspecific attachment mechanisms-is a possible solution-based route to surface metallization. [9][10][11][12] Electroless plating is a compelling alternative: a solution-based process useful for metallizing a wide variety of materials, including nonconductive and irregularly shaped materials. 7, 13-14 Solution access, rather than line-of-sight as in physical vapor deposition, dictates where surface plating will occur, so that electroless plating is an appealing choice for fashioning nanofluidic devices where even irregular and concealed surfaces may require metallization. To fully exploit solution-based metallization as a tool for micro-and nanofabrication, however, requires control not just over the plated film composition, thickness, and grain size, but also over its spatial disposition, which must be at least partly independent of underlying substrate patterning. 15 We wanted a patterning approach that did not need mechanical access to target surfaces, both to improve the generality of the approach, and to minimize the risk of damage that can accompany repeated handling of thin films-especially of free-standing thinfilms. We sought to develop a gentle, solution-based patterned metallization approach 16-17 capable of plating a range of even structured substrates, including inside existing (nano)fluidic channels. 3,7,[14][15]18 The horizons of single-molecule science have recently been dramatically expanded by the development of simple methods for fabricating nanopores: entirely solution-based processes requiring only uncomplicated instrumentation are removing barriers to the widespread use of nanopore methods. 19 To conserve the benefits of simple pore formation methods, our focus also included developing similarly widely-accessible, straightfoward solution-based approaches to patterned metallization. We therefore wanted to avoid the instrumentation and processing overhead associated with traditional photoresist-based approaches and more exotic analogues and alternatives. 11,[20][21][22][23] Instead, we chose to photo-pattern the covalent attachment of an organic monolayer to SiN x , 24 and to investigate its ability to then template the substrate metallization. By only attaching the protective layer where it was desired, rather than removing portions of a patterned photoresist film, for example, we sought to simplify the processing compared to conventional approaches. With the use of an initially liquid patterning precursor (here, 1-octene), we sought to gain greater tolerance to irregularities-including the presence of engineered structures such as nanofluidic channels-of the SiN x surface. For metallization, we initially adopted an electroless plating approach that had been specifically developed for goldplating SiN x . 7,25 The approach is outlined in Scheme 4.1, and full details of materials, instrumentation, and safety precautions are provided in the Supporting Information (SI). We had previously developed a gold electroless plating approach for SiN x that required a hydrofluoric acid (HF) etching step prior to surface metallization 7, 25 .
The HF etching step offered a natural point to incorporate patterned monolayer formation in an effort to guide the spatial extent of the substrate metallization. An alkane monolayer could be covalently linked to HF-etched SiN x through the photochemically-driven hydrosilylation of a 1-alkene. 24 Tremendous care must be exercised in the use of HF, and we detail the precautions-including additional protective equipment and monitored work-in the SI. The UV (254 nm) photoirradiation was through copper transmission electron microscopy (TEM) grid masks, with different bar sizes and spacings (see SI for specifications), that had been placed directly on the wafer (without securing them or preventing liquid access underneath), with both wafer and mask then immersed in the 1-alkene.
Plating selectivity depended on rigid adherence to the rinsing steps detailed in the SI, and, as in prior work, we ensured compatibility of the process with freestanding ultrathin SiN x membranes by avoiding ultrasonic cleaning steps. 20 Scheme 4.1: A SiN x substrate is (a) plasma treated and hydrofluoric-acid etched, then (b) immersed in 1-octene for photopatterning (254 nm) through a TEM grid. The patterned substrate is then (c) immersed in a series of metallizing solutions to yield (d) a patterned gold film. A detailed description of solution compositions and process flow is provided in the SI.
We proposed to spatially pattern LPCVD SiN x metallization by forming a physical barrier on the surface to control where the metal plating could take place.
The first step of patterned plating thus involved the formation of this patterned protective layer. In our prior work to develop an electroless gold plating procedure for SiN x , we found it was essential to first etch the SiN x surface with dilute HF. 7 This same initial etching step forms the starting point for the covalent attachment   We abandoned Sn (II)-sensitized electroless plating when efforts to improve the spatial selectivity by using different rinsing steps, for example, proved ineffective. We tested, instead, a palladium-based treatment 27    by supplanting the traditional use of charged-particle microscopes for fabrication, but nanopore profiling has customarily depended on microscopic examination. Our approach exploits the dependence of nanopore conductance in solution on nanopore size, shape, and surface chemistry in order to characterize nanopores.
Measurements of the changing nanopore conductance during formation by etching or deposition can be analyzed using our method to characterize the nascent nanopore size and shape-beyond the typical cylindrical approximation-in realtime. Our approach thus accords with ongoing efforts to broaden the accessibility of nanopore science from fabrication through use: it is compatible with conventional instrumentation and offers straightforward nanoscale characterization of the core tool of the field.

INTRODUCTION
A nanopore is a nanofluidic channel, with dimensions in all directions generally less than 100 nm, that can be used to deliver a host of capabilities for single-molecule sensing. 1-10 High-profile nanopore sensing efforts have targeted sequencing single strands of DNA and RNA; protein conformational analysis; and characterization of other biomolecules, molecular complexes, and nanoparticles. In the most straightforward implementation of nanopore sensing, the nanopore is the sole path connecting two reservoirs containing electrolyte solutions. Electrodes in each reservoir establish a potential difference across the nanopore that drives ions through the nanopore: passage of a target molecule, nanoparticle, or complex through the nanopore perturbs that ionic current and provides molecular-level information. That information naturally depends on the target's dimensions and physicochemical properties and the ionic solution composition, but it is also profoundly affected by the size, shape, and surface chemistry of the nanopore. In the case of a (cylinder-like) double-stranded DNA polymer that fills the entire length of a cylindrical nanopore as it transits through, a simple geometric treatment considering only the displacement of bulk ions by the polymer gives a straightforward expression for the macromolecule-induced conductance change 11 with 〈G〉 and 〈G b 〉 the time-averaged conductance through an unobstructed and DNA-containing nanopore, respectively, and r DNA and r 0 the cross-sectional radii of the molecule and nanopore. The expression does not capture the panoply of complex phenomena giving rise to conductance perturbations in nanopore sensing, 12-13 but does, in convenient closed form, appropriately underscore the importance of nanopore dimension. This geometric basis of the conductance change has been used to infer biopolymer conformation, for example: a foldedover polymer presents a larger effective cross-section than a linear one. 14 The more elusive dependence of current change on single-stranded DNA base sequence, for example, underpins efforts to sequence single strands of DNA using nanopores. 2,8 In a powerful implementation of nanopore force spectroscopy, details of interaction energetics can be revealed if, and only if, a nanopore size is properly engineered to sterically force the linearization of a folded moiety during passage, or rupture of an intermolecular complex by barring passage of one of the partners. [15][16][17] The ionic conductance (G), alone, of a nanopore with a charged surface can be expressed as the sum of a bulk and surface conductance term [18][19][20][21] when access resistance is negligible. 22 Overlapping Debye layers require a more sophisticated treatment, but need not be considered over a broad useful range of nanopore sizes and solution ionic strengths. [23][24] This simple formulation for G has been supported by experimental measurements in which nanopore conductance was measured for nanopores that had size and shape interrogated by combinations of transmission electron microscopy and electron energy loss spectroscopy. 13,18 The bulk conductance is determined by the solution conductivity, K, and a volume integral, A, over the unique nanopore shape: G bulk = K (∫ dz π(r(z)) mobility of counterions proximal to the pore surface, μ, the density of surface chargeable groups, σ, and an integral, B, over the surface of the nanopore: which may be only a subset of those needed to fully characterize a given nanopore profile, include the limiting radius (the minimum radius along the profile), r 0 , and total nanopore length, L, that can in some cases be equated with the supporting membrane thickness. The experimentally-supported 13, 18 treatment of the nanopore conductance here assumes axially and cylindrically symmetric nanopores in a size regime where access resistance is negligible, 22 and that any surface charge emerges from a singly ionizable surface species described by a characteristic pK a -A-H ⇌ -A -+ H + Native or engineered nanopore surface chemistry is an important element in nanopore performance, and contributor to nanopore conductance. The conductance can be naturally exploited for nanopore characterizations in conjunction with solution-based nanopore fabrication methods, and is especially useful when more complex methods present barriers to use. Charged-particle milling is an established, but challenging and burdensome, approach for formation of the smallest, <10 nm nanopores in thin membranes. [25][26][27][28] The use of (scanning) in the same holder where they will be used for experiments, and without the contamination and damage risks associated with charged particle techniques. A conductance-based characterization will not damage a molecular surface coating suitable for conductance-based sensing, and can harness the natural and direct connection to the nanopore surface chemistry that makes it a valuable method for characterizing chemically-tailored nanopores. 9,23,34,37 The conductance model is equally useful when a pore is formed and enlarged, and when an initially large pore is resized by solution-based deposition, including film growth. 9,19,35,38 Etching and deposition may be used in concert, with a pore being initially etched larger than desired to accommodate an electroless gold film, for example, that may ease nanopore surface chemical modification. In this work we wanted to understand how the measured conductance during nanopore fabrication-by deliberate expansion, closure, or both in consort-could be used to profile the nascent nanochannel. Simulations will focus, for expediency, on nanopores fabricated via deposition of surface coatings: the principles, however, are general.

THEORY
The algebraic structure of G = K • A + μ|σ| • B, and its underlying dependencies, means that a single-point conductance measurement can provide enough information to size a nanopore only when the shape is known and the fitting involves only a single geometric degree of freedom. Measurement of G versus K-by changing the electrolyte solution conductivity-for a given nanopore can provide greater insight into the nanopore size, shape, and surface chemistry. conductivity, nor should the nanopore surface chemistry change (except through deliberate action) throughout either type of fabrication process. We make the reasonable assumption that material transfer will be uniform across the surface, so that the nanopore shape will remain unchanged. Silicon nitride, the most common membrane material in which to form nanopores, is amorphous, and so will not inherently be prone to anisotropic etching. 39 Electroless plating, a surface deposition method that has been used with great success in resizing nanopores, 9 conformally coats even rough surfaces, 40 and film growth by polymer chain extension, for example, should be another effective route to reliably tune nanopore size. 41 We can then write conical-cylindrical, and hyperbolic ( Figure 5.1). 18, 21-22, 29, 32 For all profiles, we limited the {q j } to two free parameters per shape: (r 0 , L)-the limiting (minimum) radius and total nanopore length (see Tables S-1 and S-2 for notation and equations). Independent experimental studies of nanopore profiles 18,22 were used to guide the constraints and to make reasonable parameter value assignments to allow for numerical examples; the nanopore characterization method is general, however, and does not depend upon these particular numerical values. 21,23 We restricted the initial outer radius to be 10 nm greater than the initial limiting radius (not applicable to the cylindrical profile), [21][22] and fixed the initial cylinder length of the conical-cylindrical pore to be 0.6 times its initial total length. The deposited coating was piecewise curved to maintain a uniform coating thickness across the entire nanopore surface (

RESULTS AND DISCUSSION
The ability to characterize a nanopore in real-time, during its formation, using only its conductance, is an incredibly compelling goal. Its pursuit relies on the connection between the conductance of a nanopore and its size, shape, and surface chemistry, and its attainment hinges on properly exploiting the functional form of that connection. We will focus on nanopores fabricated by deposition of a coating onto the outer membrane surface and inner surface of an existing, larger pore, but similar arguments hold for a nanopore formed by etching of a smaller pore to create a larger pore.  The plotted lines denote the pairings of limiting nanopore radius, r 0 , and nanopore length, L, for each nanopore profile, that will produce a 200 nS conductance.
The most immediately striking consequence of a real-time measurement of the conductance is that, as shown in Figure 5.3, it reveals a clear distinction between different nanopore profiles. When different candidate profiles are used to fit experimental nanopore conductance data, the conductance versus time provides a means to determine nanopore shape and size. To produce the data plotted in Figure 5.3, we used the four representative nanopore profiles all with an initial 200 nS conductance and 10 nm total nanopore length. The initial nanopore limiting radii were ~6.4, 3.1, 5.5, and 4.0 nm, respectively, for the cylindrical, doubleconical, conical-cylindrical, and hyperbolic nanopore profiles. We calculated the conductance for each profile as the radii were reduced at the same rate, ν mt = 0.6 nm/h, during a simulated, deposition-based fabrication process. As shown below, the radius change after a given time must be known, but the method does not require a constant material transfer rate. We chose a constant rate, commonly observed in micromachining processing, 39 however, because it affords straightforward insights into the functional dependencies beyond what is revealed by the numerical results. Given the form of equation (5), it is perhaps unsurprising that even with constant ν mt (and therefore identical absolute rates of change of the radii across profile type), dG dt is not linear and depends on profile type (inset of Figure 5.3). The quantitative details of this behavior provide a means of extracting nanopore size and shape information from the measured conductance changes.   gives the corresponding initial nanopore lengths, L shape expt (t 0 ), for each nanopore profile. For each nanopore profile, we set the initial nanopore size, (r 0,shape expt (t 0 ), L shape expt (t 0 )), and used the progression of dimensions, (r 0,shape expt (t 0 )-Δr i (t 0 , t i ), L shape expt (t 0 ) + 2Δr i (t 0 , t i )), to simulate the post-deposition conductances G shape expt (t 1 ) and G shape expt (t 2 ). For a constant material transfer rate, ν mt , Δr i = (t i -t 0 )ν mt . While more generally Δr i = Δr i (t i , t 0 , ν mt (t)), the procedure implemented here relies on knowledge of this radius change only, not whether the material transfer rate is constant in time or not. We outline the conceptual framework for the characterization and provide a detailed step-by-step tutorial in the SI. The initial conductance, G shape expt (t 0 ), was used in conjunction with Figure   5.5.2 to establish the set of candidate {(r 0,shape (t 0 ), L shape (t 0 ))}, for each nanopore profile, whose members all have the initial conductance G shape (t 0 ) = G shape expt (t 0 ). The range of candidate sizes, for each candidate shape, is represented by the dotted lines in Figure 5.4a-d. Given G shape expt (t 0 ), alone, neither size nor shape can yet be determined. Each of these possible candidate geometries (size and shape) was then modified by the deposition of material to provide sets of nanopore dimensions given by {(r 0,shape (t 0 )-Δr i , L shape (t 0 ) + 2Δr i )} for times t 1 , t 2 , and t 3 , with corresponding sets of conductances {G shape (t 1 )}, {G shape (t 2 )}, and {G shape (t 1 )} (solid curves in Figure 5.4a-d). We then used the post-deposition G shape expt (t i ) to determine the nanopore size and shape. We found the initial limiting radius, r 0,shape (t 0 ), for each nanopore shape, that gave a conductance G shape (t 1 ) = G shape expt (t 1 ). That is, when the experimental nanopore was cylindrical, we found the r 0,shape (t 0 ) for cylindrical, double-conical, conical-cylindrical, and hyperbolic profiles that allowed the candidate pore conductance to match the experimental value, and plotted the radii in Figure 5.4e. Figure 5.4f-h are plots of the r 0,shape (t 0 ) when the conductances of double-conical, conical-cylindrical, and hyperbolic experimental nanopores were equated to the conductances of the same four candidate shapes. No matter the experimental profile, after two conductance values, all four candidate shapes-with different sizes-were equally viable conductance-based matches. By repeating this process by finding r 0,shape (t 0 ) to satisfy G shape (t 2 ) = G shape expt (t 2 ), the experimental nanopore size and shape both emerge. When the candidate nanopore profile matches the simulated experimental profile, all extracted r 0,shape (t 0 ) have the same value for all t i , which essentially delivers a simultaneous solution of G shape (t i , {q j (t i )}) = G shape expt (t i , {q j (t i )}) for all time-points. The curves in Figure 5.4e-h illustrate this successful characterization; the agreement is shown in terms of r 0,shape (t 0 ), but L shape (t 0 ) has the same behavior. Figure 5.4e plots the r 0,shape (t 0 ) when the simulated G cylindrical expt (t i ) values were fit using cylindrical, double-conical, conical-cylindrical, and hyperbolic profiles: only the cylindrical candidate nanopore returns the same r 0,shape (t 0 ) for different t i . Figures 5.4f-h show, by the constancy of the correct r 0,shape (t 0 ), the same successful capture of size and shape of double-conical, conical-cylindrical, and hyperbolic simulated experimental nanopores, respectively. Measurement of more conductance points does not provide more information, given the framework presented here, but can add numerical robustness to this approach. Alternatively, the formal need for only three conductance values allows one to piecewise repeat the shape-and size-profiling on independent sets of three conductance values throughout the duration of the fabrication, allowing for the possibility to extend this method to anisotropically-etching or -depositing materials. An extreme departure from the usual progression of conductance in time may signal the need for a more involved steady-state solution-based characterization of a pore after fabrication, 21 although even in this case the present time-dependent method should provide bounds on the evolving nanopore size. We note again, for generality, that while we used a constant ν mt , the plating rate must be known, but need not be constant. Fitting conductance values in time leverages the form of equation (2) to reveal the nanopore shape and extract dimensions from a solution-based nanopore fabrication method. generate characteristic conductance progressions that depend on the nanopore shape and initial size (conductance curves labelled with their particular Δr i ). Simulated experimental conductance data versus time for G shape expt (t 0 ) = 200 nS, r 0,shape (t 0 ) =3.5 nm pores of each shape were compared to the plots in (a-d) to reveal the (e) cylindrical (red), (f) double-conical (blue), (g) conical-cylindrical (black), and (h) hyperbolic (magenta) experimental nanopore size and shapes by the constancy of the fitting r 0,shape (t 0 ). The relevant experimental profiles for each column are inset in the top row.

CONCLUSIONS
The charged-particle, complex instrumentation approaches that dominated early nanopore fabrication methods allowed, in principle, for high-resolution nanopore characterizations, although such capability was rarely employed beyond determining a limiting radius. These instrumental approaches face limitations such as high likelihood of surface contamination and inability to probe soft (e.g. organic) nanopore coatings, and they add workflow steps that could be costly in time and instrumentation. Even so, since the nanopores were formed in these instruments, it was expedient to follow fabrication with the chosen degree of characterization in the same instrument. The ongoing development of completely solution-based methods-including the advent of new techniques-to fabricate nanopores has ushered in an exciting new area for nanofluidics, generally, and nanopore science in particular. Nanopores can now be formed in their native liquid environment, and without the instrument and workflow cost of charged-particle methods. We have modelled the nanopore conductance with a simple framework that nevertheless includes an explicit surface chemistry term and has demonstrated concordance with independent experimental characterizations of nanopore sizes and shapes of most importance for routine use in single molecule science. 13 and (2) the formation of two small pores instead of one larger one. Our simulations demonstrated that the time-dependent conductance formalism supports the detection and characterization of defects, as well as the determination of pore number, but with implementation performance depending on the measurement context and results. In some cases, the ability to discriminate numerically between the correct and incorrect nanopore profiles was slight, but with accompanying differences in candidate nanopore dimensions that could yield to post-fabrication conductance profiling, or be used as convenient uncertainty bounds. Timedependent nanopore conductance thus offers insight into nanopore structure and function, even in the presence of fabrication defects.

INTRODUCTION
Nanopores are a rising tool for single-molecule science, featuring prominently in DNA sequencing efforts, but with broader reach into biophysics, and bioanalytical and materials chemistry. [1][2][3][4][5][6][7][8][9][10][11][12] The nanopore heart of these techniques is a nanofluidic channel generally less than 100 nm in all dimensions, formed through a membrane or support, with the particular dimensions dictated by the analyte and method. The essential determinants of nanopore performance include the elements of three general nanopore-specific parameter groupings: nanopore size, shape, and surface chemistry. [13][14][15][16][17][18][19] Even the most basic nanopore operating configuration illustrates the importance of these parameters, and also provides a means for assaying them. A nanopore is positioned as the sole fluid path where σ is the nanopore surface charge density that attract counterions of mobility, μ. The pore has a radius, r(z) , that can vary along length, L, of the pore (aligned with the z-axis as shown in Figure S6 Inaccurate nanopore models will affect the quality of conductance characterizations, and other work has shown (and taken advantage of) the influence of internal nanopore structural irregularities on analyte current blockages. [32] While it is essential to control the size of isolated nanopores for single-molecule characterization and sensing applications; the use of arrays of nanopores as filters for physical and chemical separations multiplies the challenges and underscores the need to detail the formation of even single nanochannels. [11] The extreme, ~10 nm feature size has historically been challenging to nanopore fabrication (and characterization) efforts. Methods have tended to be instrumentation-intensive, using charged-particle microscopes such as scanning and (scanning) transmission electron microscopes (SEM and (S)TEM), and helium ion microscopes, or ion accelerator facilities to prepare membranes for subsequent chemical etching steps. [33][34][35][36][37] More recently, ~20 V potentials applied across thin membranes immersed in electrolytes conventionally used for nanopore experiments resulted in (controlled) dielectric breakdown of the films, and could produce size-tuned nanopores following voltage-assisted etching. [38] This truly low-overhead approach can yield <10 nm diameter nanopores, and produces them reliably wetted for use, without the risks of drying and surface contamination from steps such as TEM-based fabrication (or examination). A similarly all-solutionbased approach uses deposition of largely conformal films to shrink suitable pores to the desired final dimension. [9,39] By deliberately and beneficially removing high-magnification charged-particle microscopes from the fabrication workflow, however, the opportunity to immediately image the fabricated pores is lost. We therefore explored existing nanopore conductance formalisms [13,18] and developed a framework to use conductance to characterize nanopore size, shape, and surface chemistry. [14][15][16] We most recently showed that the method could yield real-time insight into these nanopore properties during solution-phase fabrication processes such as those outlined above. [14] In all instances, however, the simulations assumed perfectly formed single nanopores. Here we (1) deliberately introduce defects into the pore models, and we moreover (2) allow for the possibility that a measured conductance arises from two separate nanopores forming in the same membrane (denoted a double pore). The latter allowance arises from TEM observations, post-pore fabrication, showing that dielectric breakdown formation of nanopores using unoptimized multilevel pulse-voltage injection could yield more than one pore. [40] Conductance-based measurements should allow for these realities, at least through the setting of reasonable uncertainty levels. We focus here on nanopores formed in thin, free-standing silicon nitride membranes, so that our numerical simulations use parameter values from the most commonly used nanopore material platform. The films are amorphous and thus not inherently prone to anisotropic etching, [41] and silicon nitride is notably resistant to structural and chemical modification absent deliberate action.

METHODS
The form of Equation 1 means that a single measured conductance does not yield a single unique solution for the nanopore size and shape. [14][15][16] One can gain more degrees of freedom by measuring the conductances at two different solution conductivities, K, [15,16]  ( This particular implementation can determine geometries with two free parameters, and we chose the limiting (minimum) radius, r 0 (z, t), and the total nanopore length, L(t). [14] The presence of a defect disrupts the usual cylindrical symmetry. For a membrane with more than one nanopore, the nanopores are conductors in parallel (with identical surface chemistries and electrolyte contents) so that their conductances would be added directly, G = ∑ G n n . Using a single measurement of the conductance at a single time t i , it is not possible to distinguish between a single large pore and two smaller pores, or between a pore with or without a defect, when G(t i , {q j (t i )})= G(t i , {q j ' (t i )}). [14] The size-and geometrydependence of the conductance change in time, however, experimentally supported nanopore profiles, and then fit the data using candidate nanopore profiles. [16,18] The focus was whether including either defects or double pores would negatively affect the feasibility of the approach augured by the formalism. To allow this emphasis, the effect of measurement noise on the conductance was neglected. The change in nanopore radius in time, dr dt = v mt , occupies a privileged role as the material transfer rate (with opposite signs for etching and deposition). We used a constant |ν mt | = 0.6 nm/h to highlight the nonlinear dependence of conductance on geometry in Equations 1, 3, and 4, and in keeping with the linear etch rates common to micromachining, but the method does not depend on that particular magnitude or time-dependence. [14,41] We chose four nanopore profiles finding widespread use: cylindrical, double-conical, conical-cylindrical, and hyperbolic ( Figure S6.1), but the method does not hinge on these particular choices. [13,16,18,37,42] The label r 0 is used here to denote the radius of the cylindrical pores, and the minimum radius (at any given time) of the pores with radii varying with z; "pinch" and "outline" labels will be introduced for the r 0 of cylindrical nanopores with defects. All profiles were conventionally restricted to two free parameters, each, (r 0 and L) with the outer radius of the three tapered profiles fixed to be 10 nm greater than their corresponding r 0 , and the initial length of the inner cylinder of the conical-cylindrical pore restricted to 0.6 times its overall length, L(t 0 ), where t 0 is the starting time. To model the double pore case, the two pores were set to be identical. Parameter values and calculations were consistent with previous work: [14][15][16]22] 1 M potassium chloride electrolyte solution in water, K=14.95 S·m -1 , pH 7.0, and silicon nitride surface pK a =7.9, with σ calculated in the usual way. [16,22] The influence of solution pH is outlined in Figure S6.3 and the discussion immediately preceding it. For the defect-free pores, surface-deposited films were treated in a piecewise curved manner to maintain a uniform surface coating thickness ( Figure S6.1) across the entire nanopore surface. [14] For the case of the pores with defects ( Figure 6.1a) the half-cylinder protrusions running along the full length of the pore interior were centered on the pore outline, opposite each other. Simulations of G(t i ) were performed using 0.01 nm step sizes in the nanopore radius (or 1 minute increments given v mt ), and fits to r 0 (t 0 ) versus t were plotted using 0.05 nm increments.

Post-fabrication comparisons of electron microscopic and steady-state
conductance measurements support the independent use of Equation 1 for nanopore characterization. [13,16,18,20,21,24] Conductance measurements recorded during a fabrication process such as dielectric breakdown, however, occur in a different context than post-fabrication measurements. [38,43] In Figure 6.2, we used experimental multilevel pulse-voltage injection (MPVI) nanopore formation measurements-both steady-state and time-dependent-by Yanagi et al. [43] to test whether a formalism such as Equation 1 would yield reasonable real-time size determinations using the time-dependent conductance of a forming nanopore.
Yanagi et al. [43] measured the steady-state conductances, G, of post-fabrication pores and then used TEM imaging to determine their mean r 0 . With appropriate consideration of the usual caveats of EM nanopore characterization [14,16], along with possible consequences of nanopore dewetting and handling, post-fabrication electron microscopy provides a valuable, albeit instrumentation-and expertiseintensive, measure of nanopore size. Unsurprisingly, we obtained good fits to postfabrication data using Equation 1 (Figure 6.2a)-in particular with a conicalcylindrical profile with conventional constraints (see above)-and using Equation S1 (Equation 1 with an access resistance term-see discussion below) with cylindrical models with effective or adjustable fitting parameters. To correlate Yanagi et al.'s [43] measured G and mean r 0 without biasing the fit with an explicit choice of nanopore shape, we modified the cylindrical model of Equation S1 by replacing G bulk with αG bulk , and G surface with βG surface . We optimized the parameters α and β using the fit to the experimental data (with known r 0 , L, and G) in Figure   6.2a to correlate experimental post-fabrication nanopore conductances and mean nanopore radii by TEM, r 0,TEM α,β (G). We then used r 0,TEM α,β (G) to convert Yanagi et al.'s [43] time-dependent measurements of the conductance into nanopore size as a function of time, r 0,TEM α,β (t i ) (Figure 6.2b). In this context, the function r 0,TEM α,β (G) is thus better thought of as simply a fit function relating nanopore conductance and TEM-based size, rather than representing a particular model choice for the nanopore conductance. Finally, for each G(t i ) data point of Figure 6.2b, we calculated r 0,candidate (t i ), with all other parameters fixed, for each of the candidate nanopore profiles, and compared the results with r 0,TEM α,β (G) (Figure 6.2c). The experimental G(t i ) of Yanagi et al. [43] was fit best, using Equation 6.1, by a conical-cylindrical model with overall length equal to the nominal membrane thickness. The cylindrical model using Equation S1 and with an effective length equal to a fraction of the nominal membrane thickness [43] did not fit as well as the conical-cylindrical model, but outperformed the remaining candidates. Overall, Equations 6.1 and S6.1 produce reasonable nanopore sizes when applied to conductance data recorded during nanopore fabrication. As discussed in earlier work [14], a time-dependent material-transfer rate, ν mt (t), is no impediment to the time-dependent conductance profiling framework. [14] As the first application of Equation 6.1 to more complex nanopore configurations, we investigated the effect of defects on our ability to extract reasonable geometric descriptions of nanopore sizes. conductance. With larger initial defect size, the initial radius of the cylindrical outline of the nanopore (the "outline radius", r 0 outline (t 0 )) must also be larger to compensate for the internal volume lost for ionic transport. Defects distort the circular symmetry of the nanopore and introduce "pinch points" (as illustrated in by the conductance and radii). We attempted to fit these data by using the (known) material transfer rate and varying the dimensions of three candidate nanopore profiles: a defect-free cylindrical nanopore, and profiles with 0.1 and 1.0 nm-radii defects. The question was whether fitting to the G case sim (t i ) would reveal the existence and size of defects. A step-by-step tutorial for this process is provided in earlier work, [14] which we abbreviate here to allow a suitable focus on fabrication irregularities. The initial conductance, G case sim (t 0 ), was used to determine the (infinite) set of {(r 0,candidate (t 0 ), L candidate (t 0 ))} for which G candidate (t 0 ) = G case sim (t 0 ).
After the dimension changes from depositing material at the known rate (outline and pinch radii diminish at ν mt , whereas the cylinder length increases at 2ν mt ), only one pairing (r 0,candidate (t 0 ), L candidate (t 0 )) for each candidate also satisfied G candidate (t 1 ) = G case sim (t 1 ). This answer gave the unique initial nanopore size for each candidate with its specified defect size, but could not be used to identify the simulated defect size. That is, all three candidate profiles could exactly reproduce the two simulated conductances. After propagating the deposition one more time from the three different (r 0,candidate (t 0 ), L candidate (t 0 )), only one pair of initial nanopore dimensions gave G candidate (t 3 ) = G case sim (t 3 ). Figure 6.3 summarizes this behavior: the ordinate is the initial nanopore radius, r 0,candidate (t 0 ), that, after deposition until time t i , would give G candidate (t i ) = G case sim (t i ) (the dimensions at time t i are readily calculated from the initial dimensions and the known material transfer rate). When the candidate profile (here, defect size) matches the simulated profile, then all the r 0,candidate (t 0 ) from each t i are equal to each other, and equal to r 0,case sim (t 0 ), and the line connecting the data is horizontal. When the candidate profile is incorrect, then the plotted data is no longer horizontal. Thus, in Figure   6.3a, when the simulated data is generated using a cylindrical pore with a 0.1 nmradius defect, only the fit data using the 0.1 nm-defect candidate pore is perfectly horizontal. The defect-free nanopore fit data is close to horizontal and overlaps substantially with the outline radius of the simulated pore, but the 1 nm-defect fit data has a larger nonzero slope and is therefore the incorrect candidate. While r 0 outline (t 0 ) of the 1 nm-defect candidate was not substantially larger than the true r 0 outline (t 0 ), its small r 0 pinch (t i ) would suggest an incorrect threshold for analyte size-exclusion. Figure 6.3b shows that a 1 nm-defect simulated pore is successfully fit only with a 1 nm-defect candidate pore, and that radii for the remaining two formation, between an ideal pore of a given shape, and a pore of a different shape, but with defects. It may be necessary to then resort to more involved postfabrication approaches. [15,16,44,45] Indeed, one may be forced to adopt a strategy of repeated cycles of incomplete fabrication-with real-time profilingfollowed by more in-depth characterization. In such a case it is important to understand the inherent uncertainties-such as the error in r 0 -of these real-time characterization procedures to ensure that the fabrication cycles do not pass by the desired final size.
A second complication for nanopore formation is the formation of more than one pore when only one is intended. Microscopy can be used to directly enumerate the pore number, but at the cost of instrumentation and user burdens, and possible nanopore surface contamination, among other drawbacks. We wanted to determine if conductance could provide any insight into this possible problem of multipore formation. We explored the case of double pores of matching size and shape. Figure S6.4 illustrates that the conductance change in time provides the prospect of differentiating between single and double pore systems, just as it did for single pores of different shapes. [14] To explore whether the conductance time trace could reliably determine the size and number of the pores during their fabrication, we simulated conductances for single and double pore configurations of the four profiles in Figure S6 the relative sizes of analyte and pore. [15,16,44,45]

CONCLUDING REMARKS
The performance of a nanopore used for applications such as singlemolecule sensing, separations, and manipulations is dictated in large part by its size, shape, and surface chemistry. These three parameter groupings underpin the nanopore conductance and allow a suitable analysis framework to use straightforward measurements of the conductance as a means to gain insight into these nanopore properties. Nanopore conductance is routinely used to coarsely gauge nanopore size during use, typically with at least the assumption of a cylindrical shape, and then often with deliberately incorrect parameter constraints to ensure that reasonable numerical estimates of the radius are nevertheless     [43] were plotted versus several models: Equation 1 (solid markers)cylindrical (red circles), double-conical (blue triangles), conical-cylindrical with an inner cylinder length of 0.6L (black squares), and hyperbolic (magenta diamonds); and with an added access resistance term, by Equation S1 (hollow markers)cylindrical with length L (small circles) and cylindrical with a 0.37L effective length [43] (large circles). To not bias further analysis with an explicit choice of nanopore profile, the r 0,TEM expt were fit to Equation S1 with G bulk and G surface from the cylindrical model weighted by fit parameters: αG bulk and βG surface (orange triangles-r 0,TEM α,β (G)). (b) Time-dependent conductance measurements were taken from the experimental work of Yanagi et al. [43] and were used with r 0,TEM α,β (G) to determine r 0,TEM α,β (t i ). (c) Candidate profiles matching those in (a) were used at each discrete value of G(t i ) to calculate an r 0,candidate (t i ). The figure compares the fit and experimentallyderived radii where the correct candidate size should result in a straight line at a ratio of 1. Selected data markers are shown for clarity.   conical-cylindrical (black squares), and hyperbolic (magenta diamonds) profiles were used to simulate nanopore conductance values versus time. Eight candidate profiles (4 shapes, single and double) were used to fit (a-d) single pore simulated data and (e-h) double pore data from the 4 shapes. All experimental pores were initially 200 nS conductance. The correct nanopore shape was indicated by the constancy of the fit to r 0 (t 0 ) in time, and is labelled with the corresponding shape and number of pores. Selected data markers are shown for clarity.
The necessary gold film structure can arise from both the metallization process and the underlying support structure, and the structure of the support can deliver additional functions including analytical capabilities such as physical filtering. We used electroless plating as a general approach to create a library of SERS substrates: SERS-active gold films on a range of supports made from a variety of materials, made with a mixture of simple and complex fabrication histories, and offering a selection of structurally-derived functions. The result was that supports with existing functions had their capabilities enhanced by the addition of SERS sensing. Electroless plating thus offers a host of beneficial characteristics for nanofabricating multifunctional SERS substrates, including: tolerance to substrate composition and form factor; low equipment overhead requirements; process chemistry flexibility-including compatibility with conventional top-down nanofabrication; and a lengthy history of commercial application as a simple metallization technique. We gold-plated standard nanofabrication-compatible silicon nitride support surfaces with planar and porous architectures, and with native and polymer-grafted surface chemistries. We used the same plating chemistry to form SERS-active gold films on cellulose fibers arrayed in commercial filter paper and formed into nanocellulose paper. In a functional sense, we used electroless plating to augment nanoporous filters, chromatography platforms, and nanofabrication building blocks with SERS capability.

INTRODUCTION
Surface-enhanced Raman spectroscopy (SERS) is a tool at the forefront of chemical analysis for analytes ranging from single molecules to bacterial cells. [1][2][3][4][5] Raman enhancement is engineered by tuning SERS substrate design parameters such as elemental composition; the size and shape of nanoscale elements; closerange interparticle spacing responsible for hot spots; and patterning of solid substrates that can include ordered and random hierarchies across short, long, and multiple length scales. 1, 3, 6-10 Physical structure of the SERS-active metal layereither its inherent structure or the structure imposed upon it by an underlying support layer-is a critical and performance-determining factor. Considerable effort has been devoted to crafting a host of solid-supported SERS substrates, with results that inspire further efforts to improve and expand fabrication options, sensing capabilities, and sensing performance. 1,3,[7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26] Top-down nanofabrication using conventional and unorthodox approaches can produce exquisitely structured substrates, but can require substantial practitioner expertise along with expensive, specialized, and complicated instrumentation, and can moreover substantially limit the palette of fabrication materials. SERS substrates developed outside the material and processing constraints of conventional micro-and nanofabrication have been compelling. Both approaches and material sets hold promise. We sought, therefore, to develop a general route for nanofabricating SERS substrates that would bridge both paradigms-to draw on the strengths of each, and to be useful for both.
Conventional micro-and nanofabrication approaches offer well-established, highly optimized, large-scale manufacturing capabilities for reproducibly fabricating nanoscale structures. A less conventional fabrication material such as paper offers a myriad of advantages that have driven its adoption as a material of choice for low-cost diagnostics for use in resource-limited settings. 23,[27][28] The genesis for the present work was the discovery that gold films we had electrolessly plated onto silicon nitride as part of a nanofabrication effort were also capable, easily and without optimization, of generating reproducible SER spectra. 29 We wanted to take a variety of interesting and functional support materials and structures, and determine if a simple electroless plating process could make them SERS-activethereby augmenting their core functions by creating multifunctional SERS substrates. This goal of multifunction does not exclude the conventional quest for maximum signal enhancement, but does require that SERS substrate evaluation be application-context dependent. Paper, for example, can support a SERS-active metal component, offers obvious advantages such as low-cost and ubiquity, and has a pore structure that could improve sensing selectivity through separations by chromatography or by physical filtering. 18-21, 23-26, 28, 30-42 Electroless plating is a robust technique for surface metallization, wellestablished in commercial manufacturing applications for forming decorative, electrical, and optical elements, and with excellent substrate tolerance. 17,24,29,33,41,[43][44][45][46][47][48][49][50][51][52] Objects are immersed in liquid baths, with solution access and homogeneity dictating the uniformity of the plating: rough and large-area surfaces can be coated without the geometric-including line-of-sight-constraints of physical vapor deposition. Equipment overhead is minimal, the surface being plated need not be conductive-allowing for support material tolerance-and the plating occurs without the need for external electrical power. Electroless plating is inherently different than the capture, by nonspecific or specific attachment protocols, of preformed, frequently ligand-coated solution-phase nanoparticles onto a surface: 11-12, 15-16, 18, 30-32, 36-38, 40 the electrolessly plated metal film structure, properties, and composition can be controlled through surface pretreatment, plating bath formulation, and process conditions, and can occur on a timescale that can be measured in minutes. Vitally important for our pursuit of a library of multifunctional SERS substrates, electroless plating is, in principle, compatible with coating sophisticated top-down nanofabricated, and low-cost bottom-up assembled structures and surfaces.
The term "electroless deposition" is used to describe a number of different plating mechanisms, including autocatalytic, substrate-catalyzed, and galvanicdisplacement processes. 50 We adopted a single electroless plating process that had been optimized for coating nonconductive porous plastic membranes. 49 In brief, a Sn (II) solution is used to sensitize the surface which, when treated with an ammoniacal silver nitrate solution, undergoes a redox reaction to produce a nanoscopic metallic silver layer. Gold plating is then accomplished by immersing this surface in a Au (I)-containing plating bath: the aurous ions galvanically displace silver, giving gold particles that catalyze the reduction of aurous ions by formaldehyde also present in the bath. Tin-based sensitizers provide fairly indiscriminate surface sensitization, which is beneficial since tolerance to surface composition is a desired goal of our SERS substrate fabrication explorations. There is also much flexibility in plating chemistry after sensitization, allowing full access to the metals typically used for SERS. While silver coatings can be produced through electroless plating, the chemical stability of gold motivates our testing of gold-coated substrates for SERS activity. The use of a conventional electroless plating protocol, with only minor material-specific modifications in washing steps, allowed us to focus on support material composition and physical structure-and thereby, function-in our exploration of whether electroless plating could be a general tool for incorporating SERS sensing capabilities into already functional and structured materials and platforms.
We selected a range of support structures and material compositions to explore the generality of using electroless plating to form a library of SERS substrates. Silicon-rich LPCVD silicon nitride (SiNx) films on silicon were chosen for their ability to support a variety of nanofabricated structures and roles. [53][54][55] Polished SiNx films ensured the nanoscale gold grain structure would be the dominant substrate structural feature. Silicon nitride films with nanoscale throughchannels introduced key structural features (the individual nanochannels and the nanochannel array) underpinning designer filters and multifunctional chemical analysis platforms using plasmonic nanopores. [56][57] Surface-grafting of an acrylatebased polymer generated a more subtle structural modification of the planar SiNx thin film, and was intended to increase the number of possible sensitizer interaction sites on the film. Our next selection was standard filter paper, a frequent actor in paper-based low-cost diagnostics. 23,27 We explored the effect of fiber dimensions and spacing, by electrolessly plating and attempting to record SER spectra from standard filter paper and nanocellulose fiber paper-the fourth and fifth choices of material and structure. We characterized a commercial substrate (Silmeco) based on a gold-coated nanopillar array architecture 9 and etched away its gold coating to expose the sixth surface for examining electroless plating for SERS: a nanopillar array. Given the vastly different SERS substrate configurations, and the often severe approximations necessary to calculate enhancement factors, 46 we used a comparison framework designed to compare SERS performance across disparate substrates. The method yields a SERS enhancement value (SEV), which is defined as the ratio of the analyte concentrations that produce the same instrument response by normal Raman and SER measurements. 58 While spectral acquisition was formalized to allow comparisons between substrates, it nevertheless cannot account for the performance benefits of matching substrate function to a particular application.

EXPERIMENTAL
A detailed listing of materials and exposition of methods is provided in the Supporting Information. All substrates were electrolessly gold-plated by sequential immersion in the same series of tin (II) chloride-, ammoniacal silver nitrate-, and sodium gold sulfite-containing solutions (Scheme S4.1), with appropriate rinsing steps in between immersions. The solutions were prepared as previously reported. 29,59 Immediately prior to direct plating of bare silicon or silicon nitride surfaces, they were oxygen-plasma-treated and then etched with dilute hydrofluoric acid. The severe chemical hazards presented by hydrofluoric acid require special precautions such as those detailed in the Supporting Information. A subset of cleaned and etched planar silicon nitride supports was polymer-coated by formation of a covalently-linked sodium polyacrylate film before electroless plating, and once polymer-coated, was treated neither with plasma nor hydrofluoric acid. Silmeco gold-coated nanopillar SERS substrates were used, as-supplied, for comparison measurements. These silicon nanopillar substrates were also immersed in iodide-based gold etchant and then, after plasma treatment and HF etching, electrolessly gold-plated. Whatman 1 filter paper was plated without modification.
Nanocellulose fibers were formed between two glass slides into a crude paper-like mat ~1 mm thick (referred to as "nanocellulose paper") before plating. Surface characterization of the plated metal films was performed by field emission scanning electron microscopy (FE-SEM), x-ray photoelectron spectroscopy (XPS), and surface enhanced Raman spectroscopy (SERS). SER spectra were acquired at an excitation wavelength of 785 nm, with a ~100 µm diameter (full-width-half-maximum) beam, and at an excitation power of ∽57 mW for cellulose and as-provided Silmeco, and ∽250 mW for all other substrates. Standard solutions of 4-nitrobenzenethiol (NBT) in ethanol were prepared, covering a concentration range from 5×10 -9 to 1×10 -4 M. All measurements (save for replated Silmeco) were performed with the substrates immersed in the standard solutions. Substrates were immersed in standard NBT solutions and SERS spectra were recorded every 2 minutes until saturation of the signal level. Following piecewise linear background subtraction (details provided in the SI), the data was analyzed according to a framework using receiver operating characteristic (ROC) curves and kinetic analysis to calculate the SEV. 58 Figure 7.1a shows photographs of the complete set of materials before and after electroless gold plating: we use the term "support" to denote a material prior to gold plating, and the term "substrate" to denote a gold-plated support. All supports were successfully gold-plated by the series of baths of Scheme S7.1, as confirmed by visual inspection and XPS analysis ( Figure S7.1). All plated substrates could be used to record SER spectra of 4-nitrobenzenethiol (NBT). The support composition, however, placed restrictions on the experimental parameters.
Lower excitation power was required to avoid signal saturation using the assupplied Silmeco substrates, and substrate damage using the cellulose-based substrates. The higher excitation power left a through-hole in the paper substrate, as shown in Figure 7.1b, and a hollow in the thicker nanocellulose substrate after 10 exposures (~60 s each) when both were irradiated when dry; fume evolution was observed when immersed in ethanol. No damage was apparent when unplated paper that had been soaked in NBT was irradiated, so that the damage mechanism is reasonably ascribed to photothermal transduction by the gold film. This susceptibility of paper to burning is a noted benefit of using paper diagnostics in resource-limited settings where safe disposal options for biocontaminated devices may be limited. 23,27 Figure 7.1: a) Representative substrates before (supports, top row) and after (bottom row) electroless gold plating. Left to right: Silicon nitride, polymer-grafted silicon nitride, paper, nanocellulose paper, nanopillar silicon (Silmeco etched of its as-supplied gold coating), silicon nanoporous substrates. b) Laser-induced damage at 250 mW sets an excitation power limit for paper (top, showing a through-hole) and nanocellulose paper (bottom, showing a hollow in the thicker substrate).
None of the (gold-free) supports produced detectable Raman spectra of NBT at a drop-cast ~10 -4 M test dose, and the (gold-plated) substrate analyte-free background spectra were, excepting a small ~1340 cm -1 peak in paper, flat and featureless in the key spectral regions used to benchmark the substrate performance ( Figure S7.2). Figure 7.2 shows a representative background-subtracted SER spectrum from each substrate type using a 10 -5 M NBT solution. The principal spectral features are consistent across substrate type, including the most intense signal from the NO 2 symmetric stretch, centered at ~1330 cm -1 in all spectra. The intensity ratio of this peak to the 880 cm -1 ethanol peak, R NBT EtOH ⁄ , was used to construct the response versus concentration curve for each substrate type in Figure   S7.3 in the Supporting Information. These response curves had profiles typical for this class of experiment. 58,60 The Raman spectral intensity at a given analyte concentration was strongly dependent upon the support material and preparation, with a substantial penalty in signal strength imposed by the excitation power limitations required by the cellulose substrates. The use of polymer-grafted silicon nitride substrates resulted in the highest signal at all concentrations compared to all other electrolessly plated substrates, most notably when compared at low analyte concentrations. To quantify the SERS performance, representative ROC curves were constructed to calculate the SEV for each substrate: 0.646×10 3 (paper), 0.694×10 4 (porous silicon nitride), 2.34×10 5 (nanocellulose), and 5.91×10 5 (silicon nitride), and at least 9.33×10 5 for both polymer and Silmeco substrates. Following low signal intensities in the test measurement for replated Silmeco substrates in Figure 7.2, we pursued structural characterization (vide infra)-instead of further spectral characterization-in an effort to understand this lower response compared to as-supplied Silmeco substrates. For the Silmeco and polymer substrates, even the measurement at the lowest concentration demonstrated a better than 90% probability of detection for a 10% probability of false alarm and due to this, we can report only a minimum SEV. 58 These results emerged from proof-of-principle experiments of the general utility of electroless plating for SERS substrate creation rather than from longerterm substrate-specific optimizations. They are thus useful, when paired with the demands of a particular application, for indicating where efforts to gain additional enhancement might be warranted. The polymer-grafted silicon nitride is of note not simply for providing the largest SEV of our electrolessly plated substrates, but as an example of the benefits of nanoscale tailoring of SERS substrates, and for serving as a bridge between substrates based on traditional, silicon-containing nanofabrication materials, and those based on larger organic polymer fibers. More broadly, the design of a SERS substrate type should balance, in an applicationspecific way, the SEV and any special capabilities, such as filtering, offered by a given substrate. For example, gold films electrolessly plated onto and into these membrane filters can be used to physically optimize filter performance by tuning pore dimensions; to chemically optimize filter performance by serving as a first step in surface functionalization; and to augment filter performance by adding SERS-sensing capabilities in addition to separation. 29,61 Ultrathin, nanofabricated membrane filters, such as nanoporous silicon and silicon nitride, offer significant advantages over conventional polymer ultrafiltration membranes. 54,[62][63][64][65][66][67][68][69][70] Mechanically robust, unsupported ultrathin filters allow for high hydraulic and diffusive permeabilities. The material properties and ultrathin dimensions allow for the straightforward fabrication of smooth pores in controllable, well-defined sizes with narrow size distributions, and with high areal densities. The short, smooth walls do not suffer the drawbacks of flow resistance and sample losses due to the tortuosity and large surface area of conventional, thicker (polycarbonate) tracketched membranes. Such high-throughput, low-loss nanoporous membranes can be custom-fabricated with pore dimensions and characteristics optimized to filter micrometer-scale organisms such as bacteria, or even to separate macromolecules. Sensitivity might be enhanced by optimizing pore dimensions and distributions to form a nanoplasmonic array, 56 but at the cost of filtration performance (and selectivity). 57 A different example of the need to balance SEV and other application demands is illustrated in Figure S7.4: electrolessly gold-coated paper was used for the SERS readout of a crude paper-based assay that performed physical filtration and chromatographic separation. This multifunction capability augments the spectral selectivity of SERS for greater ease of analysis of multicomponent samples, but by no means circumscribes the utility of SERSactive paper. Indeed, the development of paper-based diagnostics has been characterized by the incorporation-by a variety of approaches, sophisticated and simple-of ever-greater function into paper-based supports. 23,[27][28]42 One means to create useful multifunctional SERS substrates-or even highly optimized SERS-only substrates-is through the deliberate incorporation of    [56][57] and the nanoporous membrane was moreover freestanding between support bars (not shown) so that it was electrolessly gold-plated within the pores and on both sides of the membrane. We avoided any ultrasonic cleaning steps that might cause rupture of this thin porous membrane, and we were consistent in this purposeful omission across all substrates. The three substrates were composed of nanostructured gold films with low-and high-aspect ratio grains, but the preponderance and character of the high-aspect ratio structures differed dramatically between the substrate types. The polymer-grafted silicon nitride gold film bore the greatest number of integral high-aspect ratio features, and with a unique grain structure characterized by the prevalence of larger, sharper, and more finely substructured gold flakes that projected from the surface. These flakes provide an increase in surface area for chemisorption of the NBT, and more significantly, are nanostructured on a length scale favorable for the existence of hot spots, and with an aspect ratio amenable to signal enhancement by the lightning rod effect. 4 The nanoporous substrate imposed gaps between gold grains, although on length scales optimized, in this substrate, for filtering rather than hot spot formation. 57 The loss of planar substrate area might be compensated for by plating sufficiently long pores, but the nanochannel surface is normal to the conventional substrate surface, and longer pores would affect through-pore flow rates. Overall, detrimental decreases in sensitivity from surface area losses to pores may be quickly outpaced by beneficial gains to analytical performance through the selectivity and throughput that emerges from careful tuning of the pore geometry to support rapid and tuned sample filtering.  in underlying layers. The pore, or void space, size distribution in paper can be controlled during its manufacture, and is an important metric when selecting commercial filter paper, for example. The hand-fabricated nanocellulose substrate was highly textured and convoluted, without the fiber bundling, alignment, and low packing density that produced obvious microscale voids in the paper substrate.
The ability of electroless plating to coat rough, nonplanar surfaces-beyond what was seen in the plating of the curved pore walls orthogonal to the planar upper surface of the porous silicon nitride film-is dramatically illustrated by the impressive surface coverage. Thick, porous supports such as the nanocellulose paper have a large surface area for plating-distributed throughout their interiorand require a greater minimum plating solution volume than a planar support.
Similarly, most of the plated gold surfaces will be able to bind analyte but will be optically inaccessible, and must be considered when aliquoting samples. Even after addressing these issues, the available signal strength using the cellulose-supported substrates was limited by the lower allowable excitation intensity. The fiber-based construction of the cellulose substrates, however, is an intriguing structural design feature that can provide additional analytical capabilities such as swab sampling and chromatographic separation. 35,44,71 The cellulose substrates are evocative of other fiber-mat platforms used for SERS, [11][12][14][15][16][17][18][19][20][21][22] with paper supports being available at scale and at low cost using well-established manufacturing methods.
When the ability to filter or chromatographically separate a sample using a SERSactive porous substrate is desired in addition to SERS sensing, one must consider the effect of the pore size on each capability-and on the interplay between each capability. Pore size is tunable through support fabrication or through the plating time-dependent thickness-within the limits of cost and available gold in the plating bath-of the plated gold layer. The flexibility, simplicity, and ease-ofhandling of these nanofiber-based substrates stand in stark contrast to the more delicately engineered Silmeco nanopillar arrays, particularly for applications in resource-challenged settings. The superb Raman enhancement that the nanopillar substrates provided when used as-supplied, without modification, reinforces the utility of rationally patterning traditional micro-and nanofabrication materials to create SERS substrates. One must, however, be careful during handling and solution processing to prevent unwanted damage or modification of such high-aspect ratio features: 9 the gold-etched surface shows some broken nanopillars. SEM images in Figure 7

CONCLUSIONS
Electroless plating is a robust method for fashioning a variety of materials, exhibiting a range of structural features and capabilities, into SERS-active substrates. The general electroless plating procedure we employed was able to successfully plate gold onto planar, porous, nanopillar, and fibrous surfaces; into well-defined nanochannels and variably-sized void volumes; onto traditional nanofabrication-compatible materials; and onto less conventional device platform materials such as paper that are important in the domain of low-cost diagnostics.
All resulting substrates in our library were capable of generating SER spectra. This electroless plating approach produced nanostructured films where the size, shape, and position of the gold grains could be tuned by the particular material and form factor of the support material being plated, and this tuneability was evident from both microscopic imaging and SERS intensities. The underlying support structure for the gold plating did more than imprint structure on the gold film, though.
Electroless plating of already functional structured supports created multifunctional SERS substrates. The force of the work presented here is thus both foundational and prospective: there is much promise in exploring electroless plating-including extensions such as patterned electroless plating 51, 55 -as a straightforward, robust, and low-overhead method to create custom SERS-active substrates that augment the compelling material properties, structures, and capabilities of their supports. Multifunctional SERS substrates require a rich, and application-specific, context and framework for design and performance evaluation. The substrate must, of course, generate a useful Raman spectrum, but the particular implementation-from design and fabrication to end-use-dictates the balance between Raman enhancement and other capabilities such as integral sample processing. This balance dictates how to tune the electroless plating process chemistry, and the support structure, to optimize the SERS substrate. We believe that electroless plating has great potential in the creation of multifunctional SERS substrates useful for answering a host of design and sensing challenges.

GENERAL NANOPORE SENSING PROCEDURE
Nanopores in the ~10 nm-thick silicon nitride membranes were fabricated by controlled dielectric breakdown using 11-15.5 V DC applied potentials. 3 The nanopore formation was carried out in 1 M KCl electrolyte, HEPES-buffered to pH ~7, and the membranes and pores were secured in custom-machined PTFE holders with ~500 µL sample wells. Nanopore conductances, G, were the slope of the linear fit to the experimental Ohmic current-voltage data, measured in 1 M KCl electrolyte buffered with HEPES at pH ~7. The corresponding nominal nanopore diameters were calculated using a conductance model (including bulk, surface, and access resistance terms) and cylindrical nanopore shape

ENZYMATIC DIGESTION FOR SPECTROSCOPIC MEASUREMENTS
A 2250 µL aliquot of 0.2% (w/v) A1 was added to a 150 µL aliquot of 1 unit/mL alginate lyase and heated in a water bath at 37˚C for 30 minutes. The procedure was repeated for sample A2, but the sample was diluted with 10 mL H 2 O before spectral acquisition.

ENZYMATIC SAMPLE PREPARATION FOR NANOPORE SENSING
For enzymatic digestion, samples of 3% (w/v) A2 were mixed with alginate lyase where the parameters of the unmodified Gaussian function are as conventional -A i , μ i , and σ i are the magnitude scaling, expected value, and standard deviation.
The step function, (1 + θ), was set to 1 forf b < f b cutoff + W bin , and 0 otherwise, so that the fit function covers only the accessible experimental data (f b cutoff was the threshold for event extraction  Comparison of the intensity of the guluronic (G) unit absorption at ~1025 cm -1 to the mannuronic (M) unit absorption at ~1100 cm -1 allows calculation of the M/G ratio that varies with particular alginate source. 14 Using this approach, alginate A1 was determined to be ~63%G/37%M, and alginate A2 was ~57%G/43%M. These relative proportions were supported by additional analysis: in Supplementary Figure 3b, the particular alginate lyase was a mannuronic lyase, so that the greater absorption from the digestion of A2 than A1 was consistent with a greater proportion of M in A2.
Supplementary filter was applied to these histograms, which then had their counts normalized. (7) The event duration threshold was taken to be the nearest bin at the distance of three standard deviations (after the 0.5% filter) from the bin with the maximum number of counts. (8) When events had been detected at log 10 τ above this threshold, the recognition flag was set to red to signal the presence of heparin; it was otherwise left white.
Supplementary Even dilute hydrofluoric acid presents significant chemical hazards upon operator exposure, requiring special working precautions. All beakers for HF containment were polypropylene, instead of glass which can be etched and rendered permeable. Dilute (5%) stock solutions were purchased to avoid handling concentrated solutions and Calgonate (Port St. Lucie, FL) 2.5% calcium gluconate gel was kept at hand in case of accidental skin exposure. To minimize exposure risk, all personnel wore a full faceshield, a disposable polypropylene apron and thick neoprene long-sleeved gloves over standard chemical safety glasses, laboratory coat and long-sleeved nitrile gloves, respectively. Finally, we employed a "buddy system" so that one researcher monitored the other's work with HF. All labware and gloves were thoroughly rinsed with water after use.

PREPARATION OF AMMONIACAL SILVER NITRATE 3
This solution was prepared by adding 4 drops of 1M sodium hydroxide solution to 0.010g of silver nitrate. Ammonium hydroxide was slowly added, dropwise, until all traces of dark precipitate had dissolved. The solution was then diluted to a final volume of 10mL using ultrapure water.
Ammoniacal silver nitrate solution can form explosives if allowed to dry. This solution should be prepared on only a scale sufficient for immediate use, and should preferably be deactivated by precipitation by the addition of dilute hydrochloric acid or sodium chloride prior to disposal 6 .

PREPARATION OF SODIUM GOLD SULFITE 4, 7
The synthesis of the gold plating solution was in accordance with the Abys et al. patent 4 modified by the addition of a drying step 7 , as described here. 0.275g sodium tetrachloroaurate dihydrate was added to approximately 15 mL ultrapure water at 80°C with stirring. To this solution were added 1.500g barium hydroxide octahydrate and 54μL of 50% w/w sodium hydroxide to yield an orange-yellow precipitate. The solution was boiled until all visible water had evaporated, and then allowed to cool to room temperature. The precipitate was slurried with approximately 10mL of ultrapure water and filtered through a medium porosity Buchner funnel. The precipitate was slurried with approximately 10mL of ultrapure water, heated to 60-65°C with stirring, cooled, and then filtered (bis). The precipitate was then slurried with approximately 20mL of ultrapure water, and 0.500g sodium sulfite was added to the solution. The solution was heated to 60-65°C with stirring until the precipitate turned blue-purple. This solution was filtered while still warm, and the resulting filtrate was diluted to a final volume of 25mL. The pH was adjusted with 1M sodium hydroxide to a final pH above 10.

CHARACTERIZATION
Gold film depositions were carried out in triplicate at each temperature and time point, and the 3°C trial was repeated so that each film thickness was based on deposition and measurements from between 3-6 different silicon nitride chips (allowing for occasional chip breakage). A step edge from gold film to exposed silicon nitride substrate was created by selectively removing gold film with adhesive tape (Scotch® 810 Magic™ tape) or, when film adhesion to the substrate was stronger, with a gentle pass of plastic tweezers across the substrate. AFM measurements of gold film thickness were performed in tapping mode at 0.1Hz across 10μm × 10μm segments of the step edge with an AFM Workshop (Signal Hill, CA) TT-AFM (equipped with SensaProbesTM190-A-15, 190kHz, aluminumcoated probes with tip radius <10 nm). Line profiles at several points across the step edge were analyzed, using the planar silicon nitride surface as a reference for quadratic background subtractions. For each background-subtracted profile, the means of the coated and uncoated sides were calculated (omitting large particle outliers from the statistics), and averaged for each chip over several profiles. These mean step heights were then averaged over each deposition time and temperature point, propagating the standard deviation as an uncertainty to yield the final reported step heights (Figure 3.1).
Gold film morphology was examined using a Zeiss Sigma VP FE-SEM at an electron energy of 8keV (Oberkochen, Germany), and elemental analysis by EDS was performed on the same instrument equipped with an Oxford Instruments X-MaxN 50mm 2 silicon drift detector (Concord, MA). Custom code was written in Mathematica 9 (Wolfram Research, Champaign, IL) to yield gold film grain size estimates via watershed analysis. X-ray photoelectron spectroscopy was used for the majority of the elemental analysis. XPS spectra were acquired using a PHI 5500 system (Physical Electronics, Inc., Chanhassen, MN) using unmonochromatized Al Kα radiation (1486.6 eV) and an aperture size of 600 × 600μm 2 . Survey scans were performed with 0.8eV step sizes and 20ms per step, with a pass energy of 187.85eV and 10 scans per spectrum. High resolution spectra were recorded with 50 scans per spectrum, 0.1eV step sizes, 40ms per step and a pass energy of 23.50eV. Spectra were analyzed initially with Multipak 6.1 (Physical Electronics). All curve fitting was performed using XPSPeak 4.1 8 using linear baselines and the minimum meaningful number of fixed 90% Gaussian-10% Lorentzian peak profiles per peak, with all other peak parameters free. To compensate for substrate charging, we aligned the N1s peak from silicon nitride substrates to 398.00eV, and the lower binding energy Si2p peak from silicon substrates to 99.25eV 9 , shifting spectra by up to 0.49eV. The particular choice of reference precludes analysis based on the binding energy, alone, of that component of the XPS spectrum. We chose these peaks, rather than the commonly used C1s peak 10 , because they had better signal-to-noise ratios; the peak fitting reliability would be less frequently compromised by the presence of multiple contributing features; and the C1s binding energy, itself, has been shown to be variable, notably in response to the particular surface treatment of silicon 9,11 . To gain a measure of the binding energy uncertainties useful for guiding the interpretation of binding energy shifts, and of the consistency of the reference alignment, we fit the main,

MATERIALS AND EQUIPMENT
To photoprotect the LPCVD SiN x films, we purchased 1-octene (O4806, 98%) and 11-bromo-1-undecene (467642, 95%) from Sigma-Aldrich ( The chips were rinsed with dichloromethane, allowed to dry, rinsed by isopropanol, and then processed in the metal-ion-containing solutions. The chips were then thrice-rinsed in alternating methanol and water, and dried in an argon stream.

PD (II) / AG (I) / AU (I)
Similar to the previous procedure, but with the Sn (II) step replaced with a Pd (II)-based treatment. The patterned chips were immersed in 1 M hydrochloric acid for 5 minutes, washed with isopropanol, and then immersed for 1 hour in 2 mL of the palladium surface treatment solution, followed by 3 rinses, each, of 1 M hydrochloric and water, a 5 minute soak in 2 mL of ammoniacal silver nitrate solution, one rinse with methanol and three rinses with water. The chips were then submerged in the Au (I) bath as described in the previous section.

AG (I) / AU (I)
The patterned SiN x chips were immersed in 1 M hydrochloric acid for 5 minutes, washed with isopropanol, and then immersed for 5 minutes in 2 mL of ammoniacal silver nitrate solution followed by one rinse with methanol and three rinses with water. The chips were then submerged in the Au (I) bath as described in the two previous sections. Champaign, IL) to analyze gold film properties.

GRID RECOGNITION
To distinguish between grid and grid-free zones of an FE-SEM or DHM contour image, each image was first filtered using a median filter with an appropriate pixel value threshold (usually 5), followed by image binarization (with automatic thresholding) and color-negation.

THICKNESS OF DEPOSITED GOLD
ImageJ 8 was used to extract raw gold film thickness data from a DHM image at 5× magnification, provided by Lyncée Tec, of a gold replica of a 100 mesh grid. The grid recognition algorithm was used to distinguish between grid and grid-free zones of a given contour plot. The mean film thickness with standard deviation (~23±1. 5 nm) was calculated by averaging across 10 such grid images each with metal-plated grid lines containing at least 35,000 pixels. where A min and A max are the initial and final values, and x 0 and dx are the center and slope (spatial resolution) of the edge transition. These were set as free parameters for fitting the EDS line profiles using the "Automatic" setting of the nonlinear-model-fit in Mathematica. The mean spatial resolution (as the mean dx, with standard deviation) from the EDS line profiles was 0.9 2 ±0.2 4 µm.

SELECTIVITY
Pixel values corresponding to grid and grid-free regions of grid-recognized FE-SEM images were used to build histograms for each region. A single Gaussian fit was made to each of the histograms using the following equation, where A 2 , μ, σ, and x are the amplitude coefficient, mean, standard deviation, and pixel intensity, respectively. All parameters were left free during the fit to the histogram, using Mathematica's nonlinear-model-fit method with "Automatic" setting. The selectivity was then defined, in a classical signal-to-noise sense, as selectivity = μ grid region -μ grid-free region σ grid-free region so that 0 is the lower bound and larger values represent superior selectivity. Figure   S

METHOD OF CALCULATING VOLUME (A) AND SURFACE (B) INTEGRALS
Integrals were calculated using Mathematica 10.3.1 (Wolfram Research, Champaign, IL) in the following manner, where r int (z) is a 3 rd -order polynomial interpolation of r(z) sampled with a step height, Δz = 0.0001 nm, along the z-axis from z initial to z final . Here, z initial and z final are 0 and L for all profiles except the hyperbolic profile for which they are set to -L 2 ⁄ and L 2 ⁄ , respectively. r(z) = (R-Δr i ⋅ sin β)-y ⋅ tan β y → 0 to (L-l) 2 Step 1: First conductance value, G cylindrical expt (t 0 ) = 200 nS This conductance could be generated equally well by any appropriate combination of nanopore shape and geometric parameters, (r 0,shape (t 0 ), L shape (t 0 )), plotted in Step 1 in construction of Figure 5.4: Plots of r 0 (t 0 ) versus conductance for (a) cylindrical, (b) double-conical, (c) conical-cylindrical, and (d) hyperbolic nanopore shapes for an initial conductance of 200 nS.
Step 2: Second conductance value, G cylindrical expt (t 1 ) = ∽ 114.5 nS Knowing the change in radius, Δr 1 = 0.5 nm, we take each possible (r 0,shape (t 0 ), L shape (t 0 )) from Step 1 and calculate the conductance for each profile given (r 0,shape (t 0 )-Δr 1 , L shape (t 0 ) + 2Δr 1 ). The ordinate of the G shape (t 1 ) point shows that the initially (but now smaller) 200 nS conductance pore must have had an initial limiting radius, Step 2 in construction of Figure 5.4: Plots of r 0 (t 0 ) with conductance for (a) cylindrical, (b) double-conical, (c) conical-cylindrical and (d) hyperbolic nanopore profiles with Δr 1 = 0.5 nm, and (e) the corresponding r 0 (t 0 ) for each candidate profile.
Step 3: Third conductance value, G cylindrical expt (t 2 ) = ∽ 67.3 nS Knowing the change in radius, Δr 2 = 1.0 nm, we take each possible (r 0,shape (t 0 ), L shape (t 0 )) from Step 1 and calculate the conductance for each profile given (r 0,shape (t 0 )-Δr 2 , L shape (t 0 ) + 2Δr 2 ). The ordinate of the G shape (t 2 ) point shows that the pore must have had an initial limiting radius, r 0 (t 0 ), The consistent value of r 0 (t 0 ) in panel e (and of the L(t 0 ) that we don't show) for the cylindrical trial profile tells us that the simulated pore was cylindrical, and that its initial size was (r 0,cylindrical expt (t 0 ), L cylindrical expt (t 0 )) = (3.5 nm, 3.8 nm).
Step 4: Additional conductance values, G cylindrical expt (t i ) Additional conductance values can be collected and used to, for example, improve the robustness of the r 0 (t 0 ) determinations.
Step 4 in construction of Figure 5. (S1) where the second fraction arises from a common formulation of the nanopore access resistance, 2 G access ⁄ (where there is a 1 G access ⁄ contribution from each open side of the nanopore). [2][3][4][5][6] More complex treatments exist that also include a surface term in the access resistance, and others have noted the difficulty of treating the access resistance of other nanopore shapes. [2,3] To investigate the effect of including the access resistance into the conductance modelling, we used equation (S1) to calculate the conductances of nanopores with selected aspect ratios, L(t 0 )/r 0 (t 0 ), and then fit the results to the cylindrical conductance model of equations (1) and (S1), where access resistance is neglected in equation (1).
Simulation results are shown in Figure S6.2.
If one rewrites equation (S1) more generally, G = ( , beyond that shown in equation (S1) for a cylindrical nanopore, remains challenging, especially if nanopore surface contributions are to be included. [2,6] Scaling arguments and earlier work, [2] however, offer a possible approach in which setting G access = αKr 0 is followed by numerical calculations of α, a parameter dependent on nanopore shape. . In (b) and (d), we fit candidate pore models with and without access resistance using the conductance data in (a) and (c) that included the access resistance. There are three correct fits in (b) and (d)-one for each L(t 0 )/r 0 (t 0 )-that are indicated by the horizontal slope of the fit r 0 (t 0 ) versus t data. Neglecting the access resistance when fitting the conductance-versus-time simulations results in a ~2 nm overestimate of the nanopore dimensions and a nonzero slope that indicates the incorrect fit. The simulations used step sizes in the nanopore radius of 0.01 nm to calculate G versus t, and 0.05 nm to determine r 0 (t 0 ).
The dependence of nanopore conductance show in Equation (1) is explicitly on solution conductivity, K, and implicitly on solution pH through its effect on the surface charge density, σ (and, where a surface can carry a solution-pH-dependent charge of either polarity, through the mobility of the counterion, μ).
Here we take the reasonable step of treating the case where the solution conductivity is not itself dependent on pH. Thus, without change of either nanopore dimension or solution conductivity, a change of solution pH can change the nanopore conductance-especially at lower solution conductivities. [7,8] This behavior is shown in Figure S6 where the parameter χ(pH) is used to explicitly carry the pH-dependence of the nanopore conductance (calculated relative to a particular chosen reference pH). In this form, with μ|σ(pH ref )| and χ(pH) constant in time for a given fixed solution composition as for Equation (1), the consequence of solution pH is simply a reweighting of the surface contribution to the conductance, relative to the behavior at the reference pH. Figure S6.3 shows the time-dependence of the conductance of the nanopore conductance at several pH values, and their successful use to correctly recover the nanopore size.

ELECTROLESS PLATING
Electroless plating baths were prepared as previously reported 1 (note: a mass of 0.1500 g of barium hydroxide octahydrate was incorrectly reported previously 2 as 1.500 g). Material-specific preliminary processing steps preceding the electroless plating method are detailed below, before a more general discussion of the electroless plating steps outlined in Scheme S1.

MATERIAL-SPECIFIC SURFACE PREPARATION
Hydrofluoric acid presents significant chemical hazards, so that we adopted special operating procedures when working with it. glasses, a disposable polypropylene apron over a standard laboratory coat, and thick neoprene long gloves over extended-cuff nitrile gloves. We also used a "buddy system" so that one researcher supervised the other's work with HF. All labware, gloves, and working areas were thoroughly rinsed with water after use.

POLYMER-GRAFTED SILICON NITRIDE
A subset of purchased planar silicon nitride films (with films on silicon supports cut to 1 cm×1 cm) was polymer-grafted, as described briefly here, before electroless plating. The as-supplied silicon nitride-coated substrates were exposed first to 10 minutes of a nitrogen plasma, and then to 5 minutes of an oxygen

SILICON NANOPILLAR ARRAY (GOLD-ETCHED SILMECO)
A number of the commercial gold-coated silicon nanopillar SERS substrates were immersed in gold etchant under vacuum (to remove any initial air layer and any generated bubbles preventing full etching solution access between the pillars) for 30 minutes and then washed with copious amounts of water. A gold coating was no longer visible, and while x-ray photoelectron spectroscopy (XPS) analysis showed low residual amounts of gold, there was no measurable SERS response from the gold-etched Silmeco substrates before they were electrolessly plated according to Scheme S1.

CELLULOSE
Whatman 1 filter paper substrates were used without modification.
Nanocellulose fibers were formed into a crude paper-like mat by filtering the assupplied slurry of nanocellulose in water with a polyethersulfone membrane with 0.1 μm pores. When most of the water had filtered through, the resulting paper-like mat (hereafter referred to as "nanocellulose paper") was compressed to ~1 mm thickness (thickness chosen for fabrication convenience) between two glass slides in a custom-designed, 3D printed holder and left to dry under vacuum in a desiccator for two days before plating.

SILICON-AND SILICON NITRIDE SURFACES
Prior to plating, the planar and nanoporous silicon nitride chips, and the gold-etched silicon nanopillar array, were subjected to cleaning and etch steps.
Nitrogen and oxygen plasma treatment were used to remove organic contaminants and hydrofluoric acid etching was used to remove surface oxide layers, as described above and also in reference 1. Plasma-based surface pretreatments were not performed for the surfaces bearing organic moieties.

ELECTROLESS PLATING SCHEME
Scheme S1 illustrates the general electroless plating process which followed the previous material-specific surface preparation steps, and consisted of sequential plating bath immersions interleaved with rinsing steps. Electroless plating of planar and porous silicon nitride, polymer-grafted silicon nitride, and gold-etched Silmeco was carried out for 2 hours at ~3°C with gentle rocking of the plating baths. Whatman 1 filter paper substrates and nanocellulose paper were electrolessly plated at room temperature for 2 hours with gentle rocking using a BenchRocker 3D (Benchmark Scientific, Edison, NJ, USA), and then vacuum dried (~15 minutes) as the final step. Plating bath volumes were 2 mL, 2 mL, and 1.5 mL for tin-, silver-, and gold-containing solutions for all substrates except for nanocellulose paper for which the volumes were tripled. Solvent washes between metal ion baths were identical for all plated materials: after tin, rinsing and 5 minutes of soaking in methanol followed by drying; after silver, soaking in methanol for 5 minutes and in water for 5 minutes; and after gold, alternate rinses with methanol and then water at least three times. 211 Scheme S1. Process flow for the electroless plating steps common to the plating of each support type.   As-acquired spectra of support materials, substrates, and analyte. Spectra are displayed at full vertical range at left, and scaled at right to more clearly reveal the details of the baseline. (a) 1.67×10 -4 M NBT in acetonitrile was added to each element (drop-casting followed by 5 minutes of air-drying: 20 µL aliquots for silicon-and silicon-nitride-containing elements; 5 µL aliquots for commercial silicon nanopillar and nanoporous silicon nitride; and by soaking for 5 minutes followed by 5 minutes of vacuum drying: 1 mL for paper and 10 mL for nanocellulose paper), with the solvent allowed to dry before spectral acquisition. (b) Elements were immersed in 10 -4 M solutions of NBT in ethanol and spectra were recorded after signal level saturation in time.
Solutions were covered in aluminum foil to minimize any photodamage and stored around 3°C in the refrigerator when not in use. Solutions were allowed to reach room temperature before use. An R3000QE Raman Systems spectrometer was used for all SERS measurements, with an excitation laser wavelength of 785 nm set to a power of 57 mW on cellulose and as-provided Silmeco substrates, and 250 mW power on all other substrates. The full-width-half-maximum excitation spot size was ~100 µm, measured at the substrate surface with the reader head placed at a slight stand-off of ~2.0 mm from the substrate. Each substrate was placed in a glass beaker and a spectrum was acquired at this point to ensure that the substrate was not contaminated. The substrate was then immersed in ethanol and spectra were collected every 2 minutes for about 20 minutes. Once this ethanolonly blank experiment was done, the substrate was removed from solution and dried under nitrogen before being immersed in the standard NBT solution. A spectrum was recorded every 2 minutes until equilibrium was reached, and then the rinsing, drying, immersion, and signal acquisition were repeated for all NBT standard solution from lowest to highest concentration. To provide (unenhanced) Raman spectra for the SEV analysis, 5 the same procedure was repeated using a gold-free silicon nitride substrate, using NBT concentrations in the range of 2 × 10 -4 M to 2.5 × 10 -3 M.

SPECTRAL ACQUISITION FOR DRIED SAMPLES
A 1.67×10 -5 M solution of NBT in acetonitrile was prepared and a 5 µL aliquot was pipetted onto the Silmeco substrate. The substrate was allowed to airdry for about 5 minutes before spectral acquisition, and the Raman spectrometer read head was aligned with the center where the pipette tip had been for dropcasting. There was a slight ~1.2 mm stand-off between the SERS substrate and the pipette tip and read head to prevent mechanical damage to the SERS substrate (the nanopillar substrates were especially susceptible to scratches). Excitation power was 250 mW. This alignment of pipette tip and read head was repeated for the other drop-cast spectra in Figure S7.2a, and additional details specific to each substrate are provided in the figure caption.

SPECTRAL ANALYSIS
All spectra were analyzed by custom programs written in Mathematica 11.2 (Wolfram Research, Champaign, IL). Acquired spectra were backgroundsubtracted using piecewise linear fitting between local minima that were selected using a relative thresholding approach to bracket known spectral peaks. To obtain the SEV for all substrates, the remainder of the analysis was performed according to Guicheteau et al. 5 For each spectrum we calculated the ratio of the area of the  normal Raman measurements, with solid lines to aid the eye. Spectra were acquired using 250 mW excitation, except as noted: for cellulose substrates and commercial substrate, excitation was limited to 57 mW. Limits of detection (LOD = 3s blank sensitivity ⁄ ) were estimated by fitting the first 3-4 data points of each response curve to a straight line. The sensitivity was equated to the linear slope and the standard deviation of the blank, s blank , was calculated from experimental measurements. The LOD, in matching order to the substrates, were 2.58×10 -10 , 2.7×10 -10 , 2.13×10 -10 , 1.08×10 -9 , 1.16×10 -8 and 3.62×10 -11 M, but these should be understood, along with the data below, as providing a benchmark for optimizing the application-specific substrate preparation. Figure S7.4: We constructed a crude paper-based assembly to demonstrate the prospects of using electrolessly gold-plated supports as multifunction SERS substrates. This assembly incorporated physical filtration of a heterogeneous sample, chromatographic separation of a multicomponent mixture, and SERS readout. The sample was constructed from NBT in acetonitrile and 4aminothiophenol (ATP) in ethanol, with dirt added to the mixture. The mixture was spotted onto chromatography paper (7.5 cm×2.5 cm), which physically filtered the dirt (a view of the back shows the dirt did not fully penetrate through the paper). A separation was run in 4% (v/v) ethyl acetate in hexane. Iodine staining allowed visual determination of the ATP retention time (photograph shown as an inset), but SERS was needed to localize the NBT spot. After sampling then separation, squares of electrolessly gold-coated paper were placed on a glass slide underneath the two individual analyte spots. Transfer of the separated analytes was achieved using 10-40 µL drops of ethanol and SER spectra were then recorded from each piece of electrolessly gold-plated readout paper.