Active Dual-Nanopore System

• Active dual-nanopore systems are two-pore platforms that use controlled voltages, forces, and feedback to regulate single-molecule translocation.
• They implement architectures such as tug-of-war, tandem cells, and vertical stacks to achieve bidirectional scanning and improved analyte detection.
• Real-time FPGA control and GPU-based data sieving enable high-bandwidth measurements, addressing challenges in polymer dynamics and electrical noise.

An active dual-nanopore system is a two-pore single-molecule platform in which voltages, forces, or feedback logic are used to control transport through and between two nanopores rather than merely observe uncontrolled translocation. In the solid-state DNA implementations, a single dsDNA molecule can be co-captured by two closely spaced pores, subjected to opposing electrophoretic forces in a tug-of-war configuration, and then driven through repeated bidirectional scans (“flossing”) or resensing cycles while two ionic-current channels and inter-pore time-of-flight measurements report on the same molecule. Across the literature, the term also covers serial two-pore architectures such as tandem cells with an upstream processing pore and a downstream sensing pore, as well as vertically stacked dual pores that provide complementary readout of small analytes (Liu et al., 2018, Seth et al., 2021, Dong et al., 17 Oct 2025).

1. Device architectures and physical layouts

The canonical solid-state active dual-nanopore device uses two nanopores in a common membrane, each connected to its own fluidic channel and source electrode, with a shared chamber on the opposite side. In the experimentally realized tug-of-war device, two pores in a common 36 nm-thick SiN membrane were positioned with inter-pore distances below $0.8\,\mu\text{m}$, with examples of $0.73$, $0.68$, and $0.62\,\mu\text{m}$; a dual-channel voltage-clamp amplifier independently applied $V_1$ and $V_2$ and measured independent ionic currents $I_1$ and $I_2$ (Liu et al., 2018). In the dual solid-state barcode platform, the same underlying geometry is described as a DNA molecule simultaneously captured by two pores drilled through the same membrane, separated by a center-to-center distance $d_{LR}$, with pore thickness $t_{\text{pore}}$; only monomers inside cylindrical pore regions experience pore forces $0.73$0 or $0.73$1 (Seth et al., 2021).

A second architecture is the tandem or serial two-pore cell. In the exonuclease-based tandem cell, the structure is explicitly $0.73$2, with an upstream biological nanopore used for strand threading and an exonuclease bonded to its trans side, and a downstream solid-state nanopore used for base detection. The intermediate trans1/cis2 compartment and a profiled electric field across the downstream nanopore enforce ordered, one-at-a-time transport of cleaved mononucleotides (Sampath, 2014).

A third architecture is vertically stacked rather than laterally separated. In the amino-acid dual nanopore platform, a top 2D MoS$0.73$3 pore and a 3 to 5 nm thick SiN pore are arranged in series with a vertical separation of $0.73$4, with diameters of $0.73$5 and $0.73$6, respectively. Because the two pores are distinct constrictions with different thickness and diameter, they can in principle provide two blockade amplitudes, two dwell times, and a transit time for a single analyte (Lin et al., 23 Dec 2025).

Related fabrication work has also established a route toward electronically active nanopores by embedding a single-walled carbon nanotube across the diameter of a solid-state nanopore and insulating it everywhere except at the pore. The exposed nanotube segment can function as a local FET channel or, after ablation, as tunneling electrodes, and the method is described as directly extensible to more complex dual-nanopore systems (Sadki et al., 2013).

2. Feedback control, operating modes, and measured observables

Active operation is defined by event-triggered voltage control. In the dual-pore DNA tug-of-war device, the control sequence was implemented on an FPGA with $0.73$7 response time. DNA was initially captured at pore 1 under $0.73$8 mV while $0.73$9 mV; pore-1 capture was detected when $0.68$0 pA for at least $0.68$1. Approximately $0.68$2 later, the FPGA set $0.68$3 mV to promote co-capture at pore 2; once pore 2 was detected by the same blockade criterion, $0.68$4 was switched to a positive reverse value while $0.68$5 remained fixed, thereby initiating tug-of-war (Liu et al., 2018).

In the flossing formulation, active control is expressed through a reversible differential bias. For motion from right to left, the effective differential bias is written as

$0.68$6

and for motion from left to right,

$0.68$7

Flipping the sign of $0.68$8 produces repeated $0.68$9 and $0.62\,\mu\text{m}$0 scans while the DNA remains captured. The primary experimental observables are dwell times, defined as the duration of current blockage while a tag resides in a pore, and time of flight (TOF), defined as the time for a tag to travel from one pore to the other (Seth et al., 2021).

A resensing implementation uses sequential rather than simultaneous two-pore control. In the DNA nanostructure platform, a “resensing cycle” starts with capture at pore 1, then voltage reversal drives the same molecule back through pore 1 into the common chamber, after which the field drives it toward pore 2, where it is sensed again. Each detection is followed by an 11 ms push-in period and a 10 ms zero-voltage wait before reversal. A triple-scan event yields eight features: $0.62\,\mu\text{m}$1 Here $0.62\,\mu\text{m}$2 is conductance blockade amplitude, $0.62\,\mu\text{m}$3 is event duration, $0.62\,\mu\text{m}$4 is pore-1 resensing time, and $0.62\,\mu\text{m}$5 is inter-pore time of flight (Dong et al., 17 Oct 2025).

Real-time operation at scale requires a separate control and acquisition layer. The Data Sieving framework performs rolling-average filtering and windowed min-max triggering directly on the GPU, selectively stores event-centered snapshots rather than continuous raw streams, maintains downsampled baseline metadata, and drives autonomous closed-loop actuation such as declogging pulses. The framework was demonstrated at up to 27 MHz per channel and explicitly presented as hardware-agnostic for solid-state nanopore arrays and adaptable to coupled dual-pore control logic (Cartiglia et al., 2 Apr 2026).

3. Tug-of-war physics, tension propagation, and escape dynamics

The central physical distinction between an active dual-nanopore system and a single passive pore is that the molecule is a flexible polymer subjected to two localized drives. In the strict tug-of-war model, equal and opposite forces satisfy

$0.62\,\mu\text{m}$6

yet the mean first passage time $0.62\,\mu\text{m}$7 still depends on the force magnitude because the inter-pore segment is stiffened by the opposing pulls. Under a moderate net bias $0.62\,\mu\text{m}$8, the mean first passage time follows the same scaling ansatz as the single-pore case,

$0.62\,\mu\text{m}$9

with pore friction contributing additively in the dual-pore geometry (Bhattacharya et al., 2020).

Brownian-dynamics studies of semi-flexible chains in three-dimensional double-nanopore escape further show that the segment between pores becomes locally stiffer than the bulk chain, and that escape time factors into a mechanism-dependent $V_1$0 term and a stiffness-dependent prefactor,

$V_1$1

For the 3D case this gives the $V_1$2 dependence explicitly verified in simulation (Seth et al., 2020).

For barcode-bearing DNA, the decisive nonequilibrium effect is tension propagation. When a tag passes a pore, monomers just behind it are suddenly pulled more strongly; rapid uncoiling continues until the tension front reaches the next, more massive and more highly damped tag. The result is an oscillatory velocity profile in which monomers between tags can move faster than the tags themselves, while tag velocities form lower-envelope points of the velocity profile. This is why the average chain velocity $V_1$3 is relatively uniform even though tag-specific dwell and TOF velocities fluctuate and are systematically lower for TOF (Seth et al., 2021). An allied simulation strategy for 48.5 kbp λ-DNA with structural protein tags reaches the same conclusion: nonequilibrium tension propagation, together with the effective charge, mass, and geometry of the tag, explains the disparate velocity variation from one tag to another during active dual-pore flossing (Seth et al., 2022).

The tug-of-war state itself can be described as one-dimensional sliding of contour between two absorbing boundaries. In the 2018 experimental study, the coarse-grained coordinate $V_1$4 obeys

$V_1$5

with $V_1$6. The observed slowdown as a function of voltage tuning was described by a first-passage analysis for a one-dimensional sub-diffusive process, with fitted parameters $V_1$7, $V_1$8, $V_1$9, $V_2$0, and $V_2$1 mV; in one device the mean tug-of-war duration reached about $V_2$2 ms, approximately $V_2$3 the single-pore time (Liu et al., 2018).

Longer dsDNA introduces an additional asymmetry. In the 166 kbp T$V_2$4-DNA study, tug-of-war initiation could place about $V_2$5 of the contour in one channel, yielding asymmetric exit probabilities, two-peak dwell-time distributions at some $V_2$6, and free-end escape dynamics whose velocity increased linearly with the exiting-pore voltage. Free-end motion was much faster than tug-of-war sliding and was measurably slowed by the presence of an additional strand between the pores, consistent with increased hydrodynamic friction in folded configurations (Liu et al., 28 Aug 2025).

4. Time-domain inference, barcode decoding, and common errors

Dual-pore sensing is fundamentally a time-to-distance conversion problem. For a tag $V_2$7, the TOF velocity is defined as

$V_2$8

and dwell velocities are constructed from pore thickness divided by pore residence time. A common experimental approximation is then to infer the distance between tags $V_2$9 and $I_1$0 from the tag-time-delay $I_1$1 and an average of tag velocities. In the double-nanopore barcode study, Brownian dynamics showed that this approximation is uncontrolled because the monomers between tags do not move at the tag velocities (Seth et al., 2021).

The resulting error is systematic rather than random. For TOF, all $I_1$2 were below $I_1$3, so inferred barcode distances were consistently underestimated. For dwell-based velocities, $I_1$4 lay above or below $I_1$5, giving mixed overestimation and underestimation. The physical origin is the same tension-propagation picture described above: the segment between tags includes many monomers moving closer to the chain-average velocity than to the tag velocities (Seth et al., 2021).

Two explicit correction schemes were proposed. Method I uses a known end-to-end tag distance,

$I_1$6

and then reconstructs any segment by

$I_1$7

If $I_1$8 is not known, the average chain velocity can instead be estimated from scan length and scan time,

$I_1$9

Method II refines short-range segments with a weighted segment velocity

$I_2$0

which assigns tag velocities only to a small number of monomers adjacent to the tags and the scan-average velocity to the remainder. A normalized tag-time-delay matrix,

$I_2$1

obeys an approximate sum rule,

$I_2$2

which can be used to detect missing tags or inconsistencies (Seth et al., 2021).

Protein-tag discrimination introduces a related, but distinct, inference problem. In the dual-pore simulations for streptavidin-like markers on λ-DNA, dwell-time asymmetry between entrance and exit pores was shown to be strongly dependent on tag charge, length, and morphology. For charged tags, the entrance-pore dwell time becomes broader and larger because the tag moves against the local field at one pore and with the field at the other; generic power-law dependences were reported for scaled dwell asymmetry versus tag charge and tag length. The same study explicitly matched experiment and simulation through the Peclet number, obtaining $I_2$3 and $I_2$4, thereby validating the coarse-grained mapping (Seth et al., 2022).

5. Applications and representative analyte classes

The most mature application is controlled DNA translocation and mapping. In the 2018 dual-pore tug-of-war experiments on λ-DNA, co-capture was achieved for $I_2$5 of molecules initially captured at pore 1, and the tug-of-war state produced a maximum two-order of magnitude increase in average pore translocation time relative to the average time for single-pore translocation. Under optimal conditions, more than $I_2$6 of events were unfolded, and mono-streptavidin tags that were generally not resolvable in single-pore measurements became resolvable in both $I_2$7 and $I_2$8 traces (Liu et al., 2018).

Protein-decorated dsDNA constructs constitute a second application class. The dual-pore coarse-grained simulations of 48.5 kbp λ-DNA with seven streptavidin-ssDNA tags identified key parameters controlling experimentally measurable dwell times and TOF velocities, showed that heavy spherical tags create pronounced local minima in velocity profiles, and showed that extended sidechain tags interact more strongly with the nonuniform field beyond the pores. These results were presented as a route to discriminate different neutral and charged tags of different origins and to refine genomic-length inference in experimental dual-pore setups (Seth et al., 2022).

DNA nanostructure characterization uses a different active mode: recapture and resensing of the same object. In the dual-pore resensing platform, Random Forest classification applied to the eight triple-scan features achieved accuracy $I_2$9 for nunchuck versus compact seeds and $d_{LR}$0 for three-class discrimination among compact, looped, and fringed-looped nanostructures, substantially outperforming single-scan single-pore classification. A finite-element drift-diffusion model of the TOF process yielded diffusion coefficients and rod-length estimates such as $d_{LR}$1 nm for nunchucks and $d_{LR}$2 nm for compact seeds (Dong et al., 17 Oct 2025).

The analyte scope now extends below polymers. In the stacked MoS$d_{LR}$3/SiN dual nanopore, single O-Phospho-L-tyrosine translocations were measured at 400 mV with RMS current noise of $d_{LR}$4 and $d_{LR}$5. Simulations predicted distinct fractional blockades for the two pores, and the architecture was explicitly framed as a route to five-parameter analyte characterization: two blockade amplitudes, two dwell times, and a transit time (Lin et al., 23 Dec 2025).

A more serial, chemistry-coupled application is the tandem cell for exonuclease-based sequencing. There, two nanopores and three compartments enforce ordered delivery of cleaved mononucleotides into a downstream sensing pore; according to the Fokker-Planck model, bases enter the downstream pore in their natural order with probability approaching 1, are detected without loss, and do not regress after entering trans2, so sequencing efficiency is determined solely by the discrimination level among base types inside the downstream pore (Sampath, 2014).

6. Data infrastructure, limitations, and future directions

Because active dual-pore devices generate long, high-bandwidth, event-sparse recordings, data handling has become part of the system definition. Data Sieving was introduced as a GPU-accelerated, multichannel acquisition framework that performs real-time event detection directly in the measurement pipeline. It reduced stored data volume by up to $d_{LR}$6, handled four channels at 27 MHz with aggregate $d_{LR}$7 at less than $d_{LR}$8 GPU utilization on an RTX 4080 Super, and used continuous baseline monitoring to trigger autonomous declogging pulses; in high-concentration DNA experiments, automatic declogging reduced time spent in a non-productive clogged state to near-zero without interrupting parallel measurements (Cartiglia et al., 2 Apr 2026). This suggests that scaling active dual-pore logic to many pore pairs is as much a digital-architecture problem as a membrane-fabrication problem.

Several limitations recur across the literature. The coarse-grained barcode models represent dsDNA as 1024 beads, each bead corresponding to about 46 or 47 bp, and model tags by increased mass and friction rather than explicit protein structure. Idealized pores are typically treated as cylindrical regions or as simplified wall-and-hole geometries, with overdamped Brownian dynamics and no explicit hydrodynamic interactions or full electrostatics. Even in experimentally anchored models, field profiles are usually imposed or computed in simplified geometries, and asymmetries such as pore-to-pore diameter mismatch or capacitive transients during voltage reversal remain significant practical issues (Seth et al., 2021, Seth et al., 2022, Dong et al., 17 Oct 2025).

The next directions are correspondingly clear in the cited work. Proposed extensions include longer DNA and more irregular barcode architectures, explicit coupling of Brownian dynamics to electrostatics and hydrodynamics, real-time feedback control embedded directly in simulations, smaller dual pores via tip-controlled local breakdown, guiding fields to reduce escape probability in resensing devices, and FPGA or ASIC implementations of low-latency event detectors for large arrays. Graphene or multilayer series-pore systems are also explicitly proposed as a route to slow dsDNA through a series of nanopores, while vertically stacked heterogeneous pores are being pushed toward richer multi-parameter readout for amino acids and, ultimately, protein sequencing (Seth et al., 2020, Lin et al., 23 Dec 2025, Cartiglia et al., 2 Apr 2026).

Taken together, these studies define the active dual-nanopore system not as a single hardware format but as a control paradigm: two nanopores are used to shape polymer or analyte motion, to produce redundant or complementary sensing channels, and to convert current blockades, dwell times, and inter-pore delays into structural information under explicitly modeled nonequilibrium dynamics.