High-bandwidth low-current measurement system for automated and scalable probing of tunnel junctions in liquids

Tunnel junctions are an emerging class of electronic single molecule sensors for freely diffusing molecules in liquid media.¹ A tunnel junction is typically a <3 nm wide gap between two conductive faces of gold, at least one of which is atomically sharp. Tunnel junctions are frequently formed as a mechanically controlled break junction (MCBJ) or in a scanning tunneling microscope break junction (STM-BJ), where the gap can be mechanically reconfigured. STM-BJ experiments have already shown the label-free identification of individual DNA nucleotides in solution and in seven of the twenty standard (canonical) amino acids.^2,3 Covalently bonding molecular tags to the tunneling electrodes can significantly enhance the sensitivity of tunnel junction sensors and can work with even wider nanogaps (∼2–6 nm). This approach is called recognition tunneling and can even distinguish between molecular isomers or detect minor structural differences in amino acids and peptides.⁴ Using engineered binding sites, the electrical activity of single proteins, including phi29 DNA polymerase, bound to nanogap electrodes have been recorded.⁵ All these developments are working toward the goal of creating electronic single molecule sensors and sequencers for DNA, RNA, and eventually proteins using scalably manufactured nanogap electrodes as the foundation.^1,5–7

While the reconfigurable nanogap techniques (i.e., MCBJ and STM-BJ) have demonstrated the single molecule sensing capability of nanogap tunnel junctions, the vision for high-throughput sensing applications typically involves on-chip integrated arrays of fixed-gap devices.^8–10 The repeatable and precise manufacturing of sub-10 nm nanogaps remains extremely challenging using conventional semiconductor manufacturing techniques.^11,12 Various non-commercial techniques have been used to fabricate arrays of fixed-gap devices in a format relevant for electronic single molecule sensing research.^8,13–19 Regardless of the method used to fabricate devices, molecule sensing typically takes the form of current measurement at a fixed voltage bias (<1 V) at high temporal resolution (>100 kS/s) and >10 kHz bandwidth after filtering.^1,20,21 Typical current levels are in the range of 1 pA–1 µA. Commercially available measurement instruments that incorporate a voltage source and high-bandwidth low-current amplifiers with adjustable gain are typically used to perform these measurements. 1 and 10 MHz bandwidth instruments are also available, whereas 100 kHz has been the most commonly used.^22,23 Past studies on nanogap, as well as nanopore sensors, have outlined the following general requirements to achieve low-current high-bandwidth measurements using an individual sensor: (1) parasitic capacitances between the amplifier input stage and the sensor terminals have to be low, and this can be accomplished by making the signal path short; ^24,25 (2) the junction metal area exposed to liquids should be minimal to decrease faradaic currents and broadband electronic noise due to double layer capacitances;²⁶ and (3) electromagnetic shielding to prevent coupling of extraneous noise.^27,28 Nanogap and nanopore measurements have typically been performed on single sensors in a shielded measurement setup. Scaling up measurements from one sensor to an array of sensors in a mature and highly integrated sensor system can be achieved using custom-designed multichannel ASICs, on top of which the sensor array can be directly integrated.⁶ In contrast, in an experimental setup with a single measurement channel, this must be done using a motorized three-axis stage, such as in an automated semiconductor probe station optimized for DC and pulsed measurements. However, such a system would introduce additional noise into high-bandwidth low-current measurements. Furthermore, while wire-bonding is one way to achieve miniaturized electrical connections, this can make packaged devices susceptible to electrostatic damage,²⁹ and it is also a laborious process while prototyping devices. Therefore, reversible electrical connections between the devices on the chip and the measurement instrument are a preferable approach for experimental sensor systems, provided the signal path can be kept short. Due to the need to perform measurements under liquid immersion and permit easy exchange of media, the setup must also accommodate microfluidic components, again without having an adverse impact on the noise floor.

Here, we present a measurement system to perform scalable high-bandwidth low-current measurements of on-chip integrated arrays of tunnel junctions under liquid immersion, which is a typical use case in electronic single molecule sensing. By automating data acquisition with a motorized stage and minimizing significant sources and paths introducing electronic noise, we can probe hundreds of tunnel junctions while collecting high-bandwidth current traces that record sub-millisecond long current spikes seen when molecules transiently interact with tunnel junctions. We use electromigrated tunnel junctions as the prototype sensor in this study and describe the fabrication of the microfluidic system and its integration with the tunnel junctions, which enabled our experiments.

Our measurement system is essentially an electrical probe station in a shielded enclosure with custom-designed miniature probes and electrical and microfluidic feedthroughs, enabling automated data acquisition of high-bandwidth current traces typical for electronic single molecule sensing experiments using arrays of tunnel junctions (Fig. 1). The principal components of the setup are (1) a three-axis motorized stage with a mounting plate to seat the chip containing arrays of tunnel junctions; (2) a custom-made two-pin probe card with fine-pitch spring-loaded connectors, on which (3) a high-bandwidth low-current amplifier instrument or head stage can be directly mounted; (4) electrical feedthroughs to pass data, control, and power signals, and to connect additional instruments, such as a DC SMU in our case; (5) microfluidic feedthroughs to pass gases and liquids across the Faraday cage; and (6) a battery-powered light source and a USB-powered CMOS camera mounted on a parfocal zoom lens for navigation and sample monitoring. The stage, camera, and electrical instruments are all controlled using a single PC using USB interfaces. The design and operational considerations for these components, their individual performance, and the total system are described in the following sections. The details of the key components acquired to build the experimental measurement system are summarized in the supplementary material, Table 1.

For electronic single molecule sensing in liquids, using tunnel junctions or solid-state nanopores, current signals are collected using a low-current high-bandwidth amplifier by sampling at >100 kHz and filtering the signal to >10 kHz while measuring current levels between 10 pA and 100 nA. A DC voltage bias of 50 mV–1 V is typical. While instruments capable of even higher bandwidths in the 1–10 MHz range are emerging, capacitance noise dominates at high frequencies, and only a handful of highly optimized nanopore sensors have yielded sensible data at such high bandwidths.^31,32 We developed our setup using an off-the-shelf 100 kHz bandwidth low-current amplifier instrument (e1b, Elements srl, Italy), which can perform all the typical measurements required for single molecule sensing using tunnel junctions. Furthermore, the compact form factor (30 × 15 × 74 mm³, weighing 40 g) enabled us to develop a miniature custom electrical probe on which the amplifier can be mounted to keep the measurement signal path from the chip to the amplifier short (∼25 mm).

Miniature electrical connections to sensitive on-chip electrical and electronic devices are typically made by (1) wire bonding after fixing the chip in a ceramic package or on a PCB (printed circuit board) or (2) using needle probes in a probe station. Initially, we wire bonded our devices but found that after wire bonding, they were extremely susceptible to catastrophic failure by electrostatic discharge, particularly when plugged into a PCB socket or when connections to instruments were made. Moreover, since wire bonds are permanent connections that are time-consuming to produce, it is preferable to use probe contacts in an experimental measurement setup. We, therefore, abandoned any further use of wire bonding in our study and focused on developing a probe-based electrical contacting procedure. While needle probes on micromanipulators with shielded cables are suitable for low-noise DC measurements, they are not optimal for high-bandwidth (>10 kHz) low-current (<10 nA) measurements due to the finite capacitance of shielded cables (∼100 pF/m) and long signal path (>10 cm).²⁴ To minimize the parasitic capacitance and signal path length between the tunnel junction and the amplifier while making reversible electrical connections, we developed a custom probe card using fine-pitch spring-loaded gold-plated connectors (KHW-040-001R2; FIXTEST Prüfmittelbau GmbH, Germany), so-called “pogo probes,” on which the low-current amplifier was directly mounted (supplementary material Fig. S1).³³ The probe card consists of two pogo probes soldered at its tip at a separation of 800 µm from each other (supplementary material Fig. S2), with which two-probe contacts were made to 100 µm wide gold contact pads on-chip. The direct mounting of the low-current amplifier minimized the stray capacitance of the contact path between the electronics and the on-chip tunnel junction devices to <1 pF, essential for minimizing the baseline noise, as shown by the measurements. This was vital to minimizing baseline noise in the signal, as shown by the measurements in Fig. 2, where our custom probe with sub-pF capacitance shows a noise floor that is ∼20 times lower (comparing rms current values at 10 kHz bandwidth) than while using a 27 pF lead capacitance, which is typical of standard needle probe connections with ∼10 cm long shielded cables.

Since electrical connections to the tunnel junction devices are made one device at a time by placing a pair of pogo probes on the gold contact pads of a device, we used a three-axis motorized stage [8-MTF-102LS05 (XY Scanning stage); 8MVT100-25-1 (Vertical stage); Standa UAB, Lithuania] in our measurement system to automatically and sequentially measure large arrays with hundreds of tunneling junction devices. As the motorized stage had to be operated inside the Faraday cage, and the introduction of any powered device inside the Faraday cage was an additional source of electrical noise, we characterized the impact of the motorized stage on noise in the tunneling measurements. Various available operational modes of the stage were evaluated, and the resulting noise level was compared to the baseline noise level (power cord disconnected from the socket). The ability to program the stage motors to power down after a stage movement to contact a new device was essential to decrease the noise back to the baseline level (Fig. 3). When the stage has completed a movement, the default setting is to put it into a state of “reduced power,” where it still introduces overwhelming noise into the signal path, as shown in Fig. 3. Setting the stage to an idle state using “soft stop” or “hard stop” modes, as possible in this system, lowered the noise floor to a level indistinguishable from the baseline at frequencies >10 Hz. In a measurement scenario, the slight difference at frequencies <10 Hz becomes inconsequential upon the application of a non-zero voltage bias, as the 1/f noise in a measured device dominates the device behavior. Therefore, by powering down the motorized stage using the “soft stop” setting after reaching a target position, we had the ability to scale measurements in a high-bandwidth, low-current probe station with enough room to fit mm–cm scale microfluidic components. Note that the microcontrollers for the stage (8SMC5-USB-B8-B9; Standa UAB, Lithuania) and the power supply were located outside the Faraday cage.

To ensure the controlled wetting of tunnel junction sensors and to create a system requiring only small liquid volumes for liquid exchange while limiting evaporation, we fabricated and bonded flow cells to the microfabricated chips. We used a thermosetting polymer, OSTEMER (OSTE 322, Mercene Labs AB), as the material for our flow cell. OSTE was chosen because of its chemical resistance to most common solvents and its impermeability to air when fully cured, leading to insignificant evaporation loss of liquid during long measurement periods.³⁴ Furthermore, the dual cure process enabled by the OSTE polymer, which we will describe next, permitted the ability to fabricate and bond a very thin flow cell (<200 µm thick) with a combination of through-holes for electrical contacts and microfluidic inlets and outlets as well as enclosed microchannels in the same layer.

A typical sensor chip in our study was 15 × 45 mm² in size, consisting of 12 columns, each column with up to 100 tunneling junction sensors in an array, as shown in Fig. 4. Further details of the fabrication of nanogap tunnel junction sensors starting from microfabricated gold nanoconstrictions using controlled electromigration are described in Sec. III. We made 150 µm thick OSTE flow cells with through-openings for electrical connections and twelve 200 µm wide and 50 µm tall embedded fluidic channels with independent inlets and outlets (Fig. 4). The process sequence for the flow cell fabrication proceeds as follows (Fig. 5): (1) fabrication of a silicon master mold by photolithography and dry etching using a combination of silicon oxide and photoresist masks; (2) PDMS mold casting using the silicon master mold; (3) reaction injection molding of OSTE in a spring-loaded aluminum and glass assembly by UV exposure (365 nm) through the PDMS mold to partially cure the liquid OSTE into a sticky semi-solid; (4) aligned bonding of the OSTE flow cell to a microfabricated chip containing arrays of gold nanoconstrictions [Fig. 4(d)]; and (5) mechanical clamping and thermal curing to fully cross-link OSTE and permanently bond the flow cell to the chip. Figure 4(d) shows the chip and the flow cell, along with a picture of the bonded chip. More information on the specific processes and materials is provided in the supplementary material, Note 1.

A thin flow cell is necessary to use the sub-pF custom probes reported in this work, with each probe touching the device ∼400 µm from the tunnel junction. The thin flow cell also permits the use of wire bonding using standard wedge or ball bonding heads if required. We developed a spring-loaded assembly for reaction injection molding of OSTE to tune the applied force while pressing the soft PDMS mold onto the hard surface of the assembly to the right extent [Fig. 5(b), supplementary material Fig. S3]: too little and through holes were not fully open in the final OSTE flow cell due to wicking of liquid OSTE; too much and there was warping of the inlet and outlet hole areas due to excessive residual stress, which resulted in improper bonding of the OSTE flow cell to the chip. Photographs of the spring-loaded assembly and its component parts are shown in the supplementary material, Fig. S3. Furthermore, we developed a micromanipulator-enabled aligned bonding process to bond the flow cell across the 12-channel chip with an alignment precision better than 50 µm using a standard stereo microscope (supplementary material, Fig. S4).

All perfusion of liquid media into the integrated sensor chip was carried out from outside the Faraday cage using a syringe pump, and microfluidic tubing crossed the cage through custom machined ports. Microfluidic connections from the chip to the “world” were made by interfacing PTFE tubing (1.6 mm OD, 1 mm ID) to the 12 independent channels of the flow cell using a custom-designed PDMS adapter held in place by mechanical clamping (supplementary material, Fig. S5). The tubing was then connected to the syringe pumps, or the gas lines as required using an “L-valve,” enabling easy switching of media during an experiment. An outlet reservoir within the Faraday cage was used to collect the liquid flowing out of the chip. The choice of liquid medium was limited by the susceptibility of the flow cell to polar aprotic solvents, such as DMSO and acetonitrile, which led to the bulging and delamination of the flow cell over time; other polar and non-polar solvents (acetone, ethanol, isopropanol, mesitylene, and deionized water) did not show any such effect and were successfully contained by the flow cell over long periods of time (>1 week).

As liquid perfusion into the integrated sensor chip was carried out from outside the Faraday cage, microfluidic tubing crossed the cage through custom machined ports. To prevent the introduction of additional electronic noise from outside the cage into the electronic signal path when using even a moderately electrically conductive liquid, such as deionized water (DIW), it was essential not to have a continuous fluid volume from outside the cage to the devices on the chip.²⁸ Breaking this “liquid bridge” across the Faraday cage by perfusing air in the tubing until it crossed into the Faraday cage was enough to eliminate this source of noise (Fig. 6). The rms current (I_rms) at 10 kHz bandwidth of a tunnel junction immersed in DIW was 30 pA with a liquid bridge, compared to 24 pA when the bridge is broken, and 23 pA in the air (i.e., not immersed in DIW). The liquid bridge couples in noise predominantly at low frequencies (<100 Hz) and the magnitude of noise coupled in varies unpredictably. Interestingly, the electrical noise that coupled in due to the liquid bridge was also dependent on the conductivity of the liquid. A liquid bridge with DIW introduced a higher noise (I_rms = 42 pA, 10 kHz) compared to mesitylene (I_rms = 28 pA), which is comparable to the baseline in the air (I_rms = 25 pA). This effect was also observed when the liquid bridge was broken with the baseline in DIW decreasing (I_rms = 30 pA) but still higher than that of mesitylene, which remained unchanged (supplementary material Fig. S6). This indicates that a Teflon tube filled with mesitylene, which is a non-polar, non-conductive liquid, does not act as an unwanted, unshielded electrical path across the Faraday cage at the bandwidth of our measurements. Nevertheless, we broke the liquid bridge before performing any measurements under liquid immersion, regardless of the liquid used. Note that the AC–DC adapter, i.e., the power source for the syringe pump, was switched off during measurement, as this was a major noise source despite being outside the Faraday cage and could not be eliminated regardless of the state of the liquid bridge.

To demonstrate the functioning of our measurement system for a typical use case, we present measurements of a current trace recorded at a fixed bias (100 mV) of a gold tunnel junction immersed in DIW. The gold tunnel junction used for this measurement was made using controlled electromigration and self-breaking of microfabricated gold nanoconstrictions, which were passivated with silicon oxide everywhere except at a 2 µm diameter circular area around the tunnel junction [Fig. 4(d)]. The device fabrication process and electromigration procedure are described in detail elsewhere.³⁰ Briefly, the gold nanoconstrictions were fabricated by photolithography and dry etching of a 27 nm thick gold layer with a 2 nm chromium adhesion layer evaporated on a 300 nm thick silicon oxide on silicon substrate. Macroscale electrical contacts were patterned by lift-off and sputtering. Next, the devices were passivated by depositing a layer of silicon oxide, which was then patterned and wet etched to open a 2 µm diameter circular area over the nanoconstrictions that later form the tunnel junctions after controlled electromigration and self-breaking.

In the specific experiment to produce the tunnel junction used for the present study, we electromigrated 50 gold nanoconstrictions in DIW one at a time and then performed current–voltage sweeps (I–V) on all 50 devices and found that 24 produced tunnel junctions. Further details, along with the estimated conductance values, are presented in the supplementary material, Fig. S7, Note 2 and Table 2. While the yield of tunnel junctions was 48%, we believe that the yield can be improved by better controlling the electromigration process, for example, by lowering the voltage ramp rate or increasing the heat sinking of the nanoconstrictions to further minimize thermal runaways.^30,35 Furthermore, using a more uniform and defect-free metal layer for the gold nanoconstrictions might improve the electromigration and self-breaking behavior, for instance, by employing alternative metal deposition techniques, such as atomic layer deposition.^36–38 After this, we recorded current traces at 100 mV bias on all 50 devices and reviewed them to choose one example device to highlight the current sensing resolution that we achieved using our measurement system. The current level was stable at around 0.21–0.25 nA in this chosen device over a measurement period of 10 s for all but a short ∼97 ms interval where rapid, repetitive switching between this baseline current level and a higher current level were observed [Figs. 7(a) and 7(b)]. Event analysis to identify continuous stretches of well-separated discrete current levels implemented using a custom MATLAB script showed that 305 events occurred over 97 ms. The events at both high (0.53 ± 0.02 nA) and low (0.25 ± 0.02 nA) current levels were narrowly distributed in amplitude and widely distributed in duration [Figs. 7(c) and 7(d)]. The duration of high- and low-level events (i.e., dwell times) were similarly distributed, with the shortest identified events ∼0.04 ms and the longest events ∼2 ms long [Fig. 7(d)]. The logarithm of dwell time values of events at high and low levels grouped around 0.26 and 0.19 ms respectively, with the junction dwelling at the high level for ∼61% of the interval during which rapid switching was observed. The standard deviation of the current signal (low-pass filtered to 20 kHz), calculated from the steady baseline level, where no switching was observed, was 0.03 nA for this tunnel junction device, which enabled clear separation between the two current levels during event analysis. Rapid switching between two current levels has been observed in both bare tunnel junctions, as is the case for our example shown here, as well as in molecular junctions forming in nanogaps.^15,39,40 The amplitudes and dwell times we have reported here also cover the range of values of relevance for dynamic molecule sensing using tunnel junctions.^16,20,21

This study highlights the development of a measurement system to perform automated high-bandwidth low-current measurements on arrays of on-chip integrated tunnel junctions for use in single molecule sensing at bandwidths up to 100 kHz and current levels in the pA range. We developed critical electrical and microfluidic components, processes, and experimental configurations that enable scalable low-noise high-bandwidth current recording in tunnel junctions under liquid immersion and described their effect on lowering noise in current signals. Using an electromigrated gold tunnel junction immersed in DIW, we detected rapid switching events separated by ∼280 pA, with dwell times as low as 40 µs, demonstrating the capability of our setup for single molecule sensing applications. We expect that our setup and findings will provide a significant boost to the continued development of fabrication and sensing applications for next-generation on-chip integrated tunnel junction and nanogap-based single molecule sensors, as well as to other types of electronic single molecule sensing platforms, such as solid-state nanopores and nanoscale field-effect transistors, which can be manufactured as arrays and also require high-bandwidth low-current measurement capabilities.

The supplementary material contains the details of key components procured to build the setup, a description of the microfluidic integration procedure, photographs of the setup components, and detailed information about the junction devices measured.

S.N.R. and S.J. thank M. Bergqvist for technical assistance with the construction of the setup. S.N.R. thanks T. Winkler for helpful discussions on microfluidic fabrication. S.N.R. thanks J. Campion and J. Oberhammer for the initial loan of motorized stages for testing. The authors acknowledge financial support from the Swedish Research Council (VR Research Environment Grant No. 2018-06169) and the Swedish Foundation for Strategic Research (SSF Grant Nos. ITM17-0049 and STP19-0065). S.N.R. acknowledges funding from the Ragnar Holm Foundation at KTH.

F.N. and G.S are co-founders of Zedna AB, a startup working towards the commercialization of crack-defined nanogap and nanopore manufacturing technologies.

S.N.R. and S.J. contributed equally to this work.

Shyamprasad N. Raja: Conceptualization (lead); Data curation (equal); Formal analysis (lead); Funding acquisition (supporting); Investigation (equal); Methodology (lead); Project administration (supporting); Software (lead); Supervision (lead); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Saumey Jain: Data curation (equal); Formal analysis (supporting); Investigation (equal); Methodology (supporting); Software (supporting); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Javier Kipen: Software (supporting); Writing – review & editing (supporting). Joakim Jaldén: Funding acquisition (supporting); Writing – review & editing (supporting). Göran Stemme: Funding acquisition (supporting); Writing – review & editing (supporting). Anna Herland: Funding acquisition (supporting); Supervision (supporting); Writing – review & editing (supporting). Frank Niklaus: Funding acquisition (lead); Project administration (lead); Supervision (equal); Writing – review & editing (equal).

The data supporting the findings of this study are available from the corresponding authors upon reasonable request.