Potential-Energy Gating: Mechanisms & Applications

• Potential-energy gating is a unifying principle where controlled energy landscapes enable, suppress, or redirect transport and state transitions in diverse systems.
• It finds applications in condensed matter, graphene nanostructures, quantum control, biochemical motors, and even power gating in electronic architectures.
• By acting as a state-dependent energetic filter, this mechanism enhances measurement fidelity, guides device optimization, and improves robustness in computational models.

Searching arXiv for the papers on arXiv and closely related uses of “potential-energy gating” to ground the article in current arXiv records. Potential-energy gating is used in the arXiv literature as a family of mechanisms in which a controlled potential landscape determines whether transport, state transitions, information transfer, solvent filling, observation trust, or hardware resources are effectively enabled, suppressed, or redirected. In condensed-matter systems, the relevant potential may be electrostatic or defect-mediated; in free-electron and quantum-control settings it may be an optical near-field or a time-dependent electric potential; in soft matter and biophysics it may be an electrochemical, hydrophobic, or mechanochemical free-energy landscape; in stochastic inference it is an explicit potential in the Bayesian update; and in computer architecture it appears as an opportunity for bank- or unit-level power gating inferred from workload dynamics (Xing et al., 2020, Hougland et al., 2021, Samin et al., 2016, Simeone, 12 Feb 2026, Klhufek et al., 8 Apr 2026).

1. Conceptual scope and governing idea

A common structure across these usages is that the gate is not merely a binary switch but a state-dependent energetic filter. In vanadium dioxide under ionic-liquid gating, the near-$\mu$ density of states is reshaped by oxygen-vacancy-related defect states rather than by a simple rigid-band field effect (Xing et al., 2020). In monolayer and bilayer graphene, electrostatic gate patterns define superlattice potentials, voltage-drop profiles, and valley-selective quantum point contacts (Hougland et al., 2021, Papior et al., 2015, Costa et al., 2016). In molecular motors and voltage-sensitive ion channels, gating is formulated through information flow, hydrophobic barriers, and electro-mechanical coupling of charged moieties (Takaki et al., 2021, Horng et al., 2017, Song et al., 2022). In bistable stochastic estimation, the observation noise covariance is modulated directly by the local potential energy, $R(x)=R_0[1+gV(x)]$, so that measurements are trusted near well minima and discounted near barriers (Simeone, 12 Feb 2026).

This suggests that “potential-energy gating” functions less as a single field-specific term than as a reusable organizing principle. The gate acts on whichever degree of freedom is operationally relevant: carrier density, thermopower, valley transmission, solvent composition, mechanical stepping, filter likelihood, or circuit leakage.

2. Correlated-electron and photoemission realizations

In vanadium dioxide, combined resistance and thermoelectric power measurements under ionic-liquid gating show a distinctive separation between charge transport and thermopower. In the ungated film, the metal-insulator transition occurs near $T_{\mathrm{MIT}}\approx 325$–$335$ K, the resistance changes by more than three orders of magnitude across the transition, and the Seebeck coefficient is strongly negative, evolving from metallic-like behavior above the transition to much larger semiconducting-like behavior below it. At finite gate voltage $V_g\le 1.0$ V, the transition is progressively suppressed to lower temperature and its resistance jump is reduced by more than an order of magnitude, yet the low-temperature resistance remains thermally activated,

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$

with $E_A$ decreasing as $V_g$ increases (Xing et al., 2020).

The thermopower, however, changes qualitatively. At $V_g=0$, $S(T)$ is typical of a semiconductor; at $R(x)=R_0[1+gV(x)]$0 V it becomes nearly temperature independent; and at $R(x)=R_0[1+gV(x)]$1 V it trends toward a metallic-like, roughly linear-in-$R(x)=R_0[1+gV(x)]$2 form. The paper interprets this contrast through the different observables sampled by conductivity and thermopower. The linear-response thermopower is written as

$R(x)=R_0[1+gV(x)]$3

with $R(x)=R_0[1+gV(x)]$4. In ungated VO$R(x)=R_0[1+gV(x)]$5, the electron-dominated semiconducting response is fit by a conduction-band form giving $R(x)=R_0[1+gV(x)]$6 about $R(x)=R_0[1+gV(x)]$7 eV below the conduction-band minimum and the donor level about $R(x)=R_0[1+gV(x)]$8 eV below $R(x)=R_0[1+gV(x)]$9. Under gating, a defect-band model composed of Gaussians,

$T_{\mathrm{MIT}}\approx 325$0

with representative parameters $T_{\mathrm{MIT}}\approx 325$1 for the defect state and $T_{\mathrm{MIT}}\approx 325$2 for the conduction band, reproduces thermally activated resistance together with apparently metallic thermopower. The central claim is that ionic-liquid gating primarily creates oxygen-vacancy defect states that reshape the density of states near $T_{\mathrm{MIT}}\approx 325$3, rather than acting as a purely electrostatic dopant (Xing et al., 2020).

A different condensed-matter realization appears in Cu-intercalated Bi$T_{\mathrm{MIT}}\approx 325$4Se$T_{\mathrm{MIT}}\approx 325$5, where the control parameter is photon energy in ARPES rather than an applied gate electrode. In Cu$T_{\mathrm{MIT}}\approx 325$6Bi$T_{\mathrm{MIT}}\approx 325$7Se$T_{\mathrm{MIT}}\approx 325$8, changing the photon energy from 20 to 110 eV produces a strong and reproducible change in surface-state occupancy, with a step-like feature around 30 eV and about a 150 meV shift in the Dirac-point binding energy. The surface-state Fermi-surface area roughly doubles, and the effect correlates with a photocurrent increase of more than an order of magnitude across the Bi $T_{\mathrm{MIT}}\approx 325$9 absorption edge; the relevant core-level energies are 23.8 eV and 26.9 eV, with work function $335$0 eV. The interpretation is a photoemission-induced photovoltage that changes near-surface band bending while leaving the global sample Fermi level aligned with the analyzer, thereby acting as a local gate on the topological surface state (Kordyuk et al., 2010).

3. Graphene, bilayer graphene, and electrostatic profile engineering

In monolayer graphene, doubly periodic electrostatic gating is proposed as a route to flat-band engineering without twist. The tight-binding Hamiltonian is

$335$1

and the continuum description near the Dirac points takes the form

$335$2

Several potential families are examined, including checkerboard $335$3, sinusoidal hexagonal $335$4, and Kagome $335$5. The paper reports that wallpaper group 17, and in particular the Kagome potential, works best for inducing a high degree of band flattening near zero energy. The most robust Kagome flattening appears around $335$6, the hexagonal sinusoidal case has a representative optimal amplitude around $335$7, and tight-binding and continuum calculations agree well for small to moderate amplitudes, especially below about $335$8. The required flattening amplitude scales approximately as $335$9; the examples given are 11.6 nm $V_g\le 1.0$0 about 5.5 eV and 50 nm $V_g\le 1.0$1 about 1.3 eV. The same work also emphasizes commensurability constraints and warns that some apparent flattening can be a band-folding artifact rather than the desired robust mechanism (Hougland et al., 2021).

In graphene nanojunctions at finite bias, electrostatic gating controls not only carrier density but also the internal voltage-drop profile. The paper states that for $V_g\le 1.0$2-type($V_g\le 1.0$3-type) carriers the voltage drop is located close to the electrode with positive(negative) polarity, i.e. the potential of the junction is pinned to the negative(positive) electrode. The mechanism is tied to graphene’s vanishing density of states near the Dirac point, modeled through electrode broadenings $V_g\le 1.0$4. Because left and right electrodes then expose different densities of states in the bias window, the injected nonequilibrium charges are unequal, the scattering region becomes net positive or negative, and the junction behaves as an extension of one electrode. The effect is reported to be independent of device length, survives electron-hole asymmetry, and can also be induced by dopant atoms such as Li and F. When the electrode density of states is artificially smeared, the asymmetry disappears, and the anti-symmetric profile is recovered when $V_g\le 1.0$5 (Papior et al., 2015).

In bilayer graphene, electrostatic gates define a quantum point contact by imposing an interlayer bias that opens a gap. For aligned bias on both sides of the constriction, the transmission displays the usual QPC steps and is valley independent. For anti-aligned bias, one side has the opposite bias polarity, the low-energy dispersion becomes directionally asymmetric, and only one Dirac valley is efficiently transmitted. The valley polarization is defined as

$V_g\le 1.0$6

and the paper reports that $V_g\le 1.0$7 can approach 1, with larger $V_g\le 1.0$8, smaller $V_g\le 1.0$9, and larger $R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$0 improving polarization. The proposed device is entirely electrostatic: it does not rely on edge termination, strain-induced pseudomagnetic fields, or defect engineering (Costa et al., 2016).

4. Optical and quantum-coherent implementations

A free-electron realization appears in sequential phase-locked optical gating. There, two spatially separated, phase-locked plasmonic near-fields generated by dipolar plasmons in gold nanorods act as a localized, time-dependent optical potential for a slow electron wavepacket. The simulations use 600 eV electrons, so the interaction with each near-field extends over several optical cycles. The initial optical phase of the localized dipolar plasmon at the first interaction zone, the relative phase between the two zones, and the polarization state of light act as control parameters for the longitudinal and transverse recoil of the electron wavefunction. The paper presents the standard PINEM ladder expansion

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$1

and evaluates the final kinetic-energy expectation value from the momentum distribution after interaction. The reported effect is selective acceleration or deceleration of electron energy along specific diffraction angles, with diffraction orders as large as $R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$2. The sequential scheme is explicitly described as Ramsey-type phase control, in which in-phase fields enhance the PINEM pattern and opposite phases can cancel energy exchange (Chahshouri et al., 2023).

A formally different but related quantum-control usage is the synthesis of unitary gates by weak perturbation of a charged particle in a one-dimensional box. The controlled Hamiltonian is

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$3

with

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$4

The evolution is treated in the interaction picture with a Dyson expansion truncated at low order, and the target gate mismatch is minimized in Frobenius norm,

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$5

The main example is the quantum Fourier transform gate. Performance is monitored through the noise-to-signal energy ratio,

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$6

which the paper states decreases with time and approaches a steady nonzero value because of finite-dimensional truncation and finite Dyson order. In this setting, the gate is literally an engineered electric potential that reshapes the accessible unitary evolution (Gautam, 2023).

5. Soft matter, molecular motors, and ion-channel gating

In dimeric molecular motors, gating is recast as an information-theoretic and thermodynamic coordination cost between the two heads. The energetic decomposition

$R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$7

splits ATP free-energy transduction into apparent entropy production and information flow between the heads. Positive chemical-state information flow, $R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$8, is identified with positive cooperativity and efficient head-head communication. For kinesin-1, the paper places the critical force at $R=R_0\exp\!\left(\frac{E_A}{k_BT}\right),$9 pN; for myosin V, $E_A$0 pN. Below $E_A$1, $E_A$2; above it, effective information flow ceases and backward stepping becomes more likely. The work also states that $E_A$3 is independent of the input ATP free energy. For kinesin-1, the input chemical energy is $E_A$4, only about 45% of that input is used for mechanical stepping at $E_A$5 pN, the ideal stall-force estimate is $E_A$6 pN, and the measured stall force is $E_A$7–$E_A$8 pN because backward stepping becomes significant before the ideal stall point (Takaki et al., 2021).

Voltage-gated ion channels provide another explicit potential-energy picture. In the continuum gating-current model, changes in membrane potential alter the electric potential energy of the positively charged arginines in the S4 voltage sensor, which move through a hydrophobic plug and drag the S4 segment. The model combines Poisson–Nernst–Planck-steric transport for Na$E_A$9, Cl$V_g$0, and four arginines with a mechanical model for S4: $V_g$1 The hydrophobic plug is represented by a barrier $V_g$2, arginines are attached to S4 by spring potentials, and the system is derived by energy variational methods. A major claim is that exact conservation of current is essential; the paper states that chemical-reaction-type models based on ordinary differential equations cannot capture such interactions with one set of parameters and may inadvertently violate conservation of current. The model reproduces equality of on and off charge, saturating voltage dependence in the $V_g$3 curve, and many details of the shape of gating current as a function of voltage (Horng et al., 2017).

A related but distinct channel model treats gating as the motion and collapse of a hydrophobic, ion-free bubble in the pore/filter region of a Kv channel. Outside the bubble, ions satisfy a PNP system; inside, ions are excluded and the bubble carries a uniform permanent negative charge density. The bubble interface evolves according to

$V_g$4

so the applied voltage changes the electrostatic force on the bubble and drives collapse. The opening delay is interpreted as the time required to cross this energetic landscape,

$V_g$5

Randomness in bubble and channel sizes is then used to reproduce stochastic single-channel behavior, ensemble currents, and the Cole-Moore delay under different holding potentials (Song et al., 2022).

In porous membranes filled with aqueous mixtures, an external electric potential can reversibly gate large pores between an “off” co-solvent-rich state and an “on” water-rich state. The model couples mixture free energy, preferential ion solvation, electrostatics, and surface affinity. For hydrophilic ions, positive $V_g$6 makes water-rich regions favorable, so voltage-driven ion accumulation drags water into the pore. The paper reports that for a hydrophobic pore the water fraction exceeds 0.5 at roughly $V_g$7 V for $V_g$8 M and $V_g$9 V for $V_g=0$0 M, and that the transition can be discontinuous at $V_g=0$1 but gradual at $V_g=0$2 and $V_g=0$3. This is an explicitly thermodynamic gate: the field alters local composition and phase state rather than merely driving ionic flux (Samin et al., 2016).

6. Potential-energy gating in stochastic state estimation

In bistable stochastic systems, potential-energy gating is formulated as a physics-informed modification of Bayesian filtering. The core gating law is

$V_g=0$4

where the observation noise covariance increases with potential energy. A second term adds direct potential regularization to the update objective,

$V_g=0$5

The intended behavior is that observations are trusted near potential minima and progressively discounted near the barrier separating metastable wells. The method is implemented in Extended, Unscented, Ensemble, and Adaptive Kalman filters and in particle filters, and introduces exactly two additional hyperparameters, $V_g=0$6 and $V_g=0$7 (Simeone, 12 Feb 2026).

The benchmark system is an overdamped Langevin process in a Ginzburg–Landau double-well potential,

$V_g=0$8

with minima at $V_g=0$9 and barrier height $S(T)$0. In synthetic tests with 10% outlier contamination and 100 Monte Carlo replications, the paper reports RMSE improvements over the standard EKF of $S(T)$1 for PG-PF, $S(T)$2 for PG-EKF, $S(T)$3 for PG-AKF, $S(T)$4 for PG-UKF, $S(T)$5 for PG-EnKF, and $S(T)$6 for the naive topological baseline; all improvements are statistically significant at $S(T)$7-level significance under paired Wilcoxon signed-rank tests. The naive baseline,

$S(T)$8

uses only distance to the nearest well, and its comparison is used to argue that the continuous energy landscape adds an additional $S(T)$9 percentage points beyond topology alone. The paper further reports that even when assumed potential parameters deviate by 50% from their true values, improvement never falls below 47%; under spontaneous Kramers-type transitions PG-EKF retains 67.7% improvement whereas NT-EKF drops to 30.3%; and in an illustration using NGRIP $R(x)=R_0[1+gV(x)]$00O the fitted asymmetry parameter is $R(x)=R_0[1+gV(x)]$01 with bootstrap 95% confidence interval $R(x)=R_0[1+gV(x)]$02, while outlier fraction explains 91.1% of the variance in filter improvement (Simeone, 12 Feb 2026).

A recurring misconception addressed by this framework is that robustness can be made state agnostic. The paper explicitly distinguishes potential-energy gating from purely statistical robust filters, which treat all regions of state space identically, and from constrained filters, which impose hard bounds rather than modulating observation trust continuously (Simeone, 12 Feb 2026).

7. Architectural power gating and time-resolved memory demand

In embedded Transformer inference, the relevant “potential” is not a physical free-energy function but a time-resolved opportunity for switching off hardware that is not needed at a given instant. TRAPTI is a two-stage methodology in which Stage I uses TransInferSim to produce a cycle-level on-chip SRAM occupancy trace $R(x)=R_0[1+gV(x)]$03, memory access statistics, and an execution timeline; Stage II then reuses that trace offline to explore banked SRAM organizations and power-gating configurations (Klhufek et al., 8 Apr 2026). The active-bank demand is defined by

$R(x)=R_0[1+gV(x)]$04

where $R(x)=R_0[1+gV(x)]$05 is total SRAM capacity, $R(x)=R_0[1+gV(x)]$06 is the number of banks, and $R(x)=R_0[1+gV(x)]$07 is a headroom factor. Total SRAM energy is decomposed as

$R(x)=R_0[1+gV(x)]$08

The study compares GPT-2 XL and DeepSeek-R1-Distill-Qwen-1.5B under the same accelerator configuration, sequence length 2048, 8-bit quantization, and baseline 128 MiB SRAM. GPT-2 XL reaches a peak SRAM requirement of 107.3 MiB, or 84% of 128 MiB, whereas DeepSeek-R1-Distill-Qwen-1.5B reaches 39.1 MiB, or 31% of 128 MiB, a 2.72× reduction in peak on-chip memory utilization. With bank counts $R(x)=R_0[1+gV(x)]$09 and conservative $R(x)=R_0[1+gV(x)]$10, TRAPTI reports up to 78% SRAM energy reduction; the best gains often occur around $R(x)=R_0[1+gV(x)]$11 or $R(x)=R_0[1+gV(x)]$12, and switching overhead is stated to be negligible overall (Klhufek et al., 8 Apr 2026).

At the scale of the full neural processing unit, ReGate extends this logic from SRAM banks to multiple hardware blocks. The paper reports that 30%–72% of energy while busy is static power dissipation and that 17%–32% of total energy is wasted when chips are idle within the datacenter duty cycle. ReGate therefore assigns different control planes to different components: hardware-managed power gating for systolic arrays, inter-chip interconnect, and HBM controllers, and software-managed power gating for vector units and SRAM. A new ISA instruction, setpm, exposes on, auto, and off modes, with sleep additionally available for SRAM. For systolic arrays, the design supports cycle-level power gating at processing-element granularity by propagating control along the deterministic dataflow; for vector units, software-managed gating yields 3.3× higher VU energy savings on average than hardware-managed approaches; and for SRAM, 4 KB segments can be placed in ON, SLEEP, or OFF states. Across workloads, ReGate reduces NPU energy consumption by 8.5%–32.8%, 15.5% on average, with less than 0.5% performance degradation in the full design and total hardware area overhead under 3.3% of a TPUv4i-class chip (Xue et al., 4 Aug 2025).

These architectural papers use “gating” in a control-theoretic rather than microscopic sense, but they retain the central logic of the broader term: a state-dependent landscape determines when a resource should remain active and when it should be energetically suppressed. In this case, the relevant state variables are occupancy traces, idle intervals, and break-even times rather than carrier densities or free-energy minima (Klhufek et al., 8 Apr 2026, Xue et al., 4 Aug 2025).