A photodetector that can change what wavelengths it sees, on command, is a different kind of instrument from one that merely counts photons. Conventional detectors measure intensity; a spectrometer measures a spectrum, and that has historically required something bulky — a grating, a filter wheel, a moving mirror — to separate light by color before it is detected. The alternative, now gaining ground, is a detector whose spectral response is itself programmable: sweep a voltage, record how the photocurrent changes, and reconstruct the spectrum computationally from a set of known response curves.
A paper published online on October 20, 2026, in ACS Nanoscience Au gives that idea a concrete and unusually fast device. Researchers at Sungkyunkwan University, VinUniversity, the Solar Energy Research Institute of Singapore, and Pusan National University built a vertical photodetector from a stack of monolayer graphene, multilayer tungsten diselenide (WSe2), and tin diselenide (SnSe2), and showed that an out-of-plane electric field does two things at once: it shifts the device’s spectral response through the Stark effect and a Franz–Keldysh-type modulation, and it rips bound excitons apart into free carriers efficiently enough to deliver a responsivity of 260 mA/W with response times around one microsecond at a drain-source bias of −1 V.
The result matters less as a record than as a demonstration that spectral tunability and respectable detector performance can coexist in the same atomically thin device. In this field, that combination is rare, and the reasons it is rare are worth understanding before the numbers.
The problem: excitons that will not let go
In a transition metal dichalcogenide (TMD) such as WSe2, an absorbed photon does not immediately produce a free electron and hole. It produces an exciton, a bound electron-hole pair held together by a Coulomb attraction that, in two-dimensional semiconductors, far exceeds room-temperature thermal energy — about 25 meV — and can reach roughly 30 percent of the bandgap. That strong binding is a blessing for light emission, where you want electrons and holes to meet and recombine radiatively. For a photodetector, it is an obstacle: a bound pair carries no current, and every exciton that recombines before dissociating is a photon wasted.
The standard ways around this are a built-in field from a junction, a type-II heterojunction offset that pulls the electron one way and the hole the other, or brute-force bias. The new device uses a vertical geometry to make the field do more work per volt. The channel length is not a lithographically defined gap between two side-by-side contacts; it is the thickness of the WSe2 flake itself, about 15 nm in the reported device. Carriers generated anywhere in that slab need only cross nanometers to reach an electrode, so collection is fast and the field across the slab is intense even at modest bias.
The electrode choice is the clever part. The bottom contact is monolayer graphene; the top contact is multilayer SnSe2, a degenerate, high-work-function material at about 5.1 eV. A metal bottom electrode would screen the gate field, shielding the WSe2 channel from the very voltage meant to tune it — the authors show exactly that failure in a control device with a metal bottom contact. Graphene, being a single atomic layer, screens weakly, so a gate voltage applied through the substrate actually reaches into the channel and tilts the bands. The device is thus a gate-tunable asymmetric Schottky photodiode whose diode character itself changes with bias: under strongly negative gate voltage the graphene and WSe2 both behave p-type and the junction is p–p; as the gate swings positive they turn n-type against the persistently p-type SnSe2 side, forming an n–p Schottky photodiode. Kelvin probe measurements put the surface potential difference between graphene and SnSe2 at about 0.23 eV, consistent with that band picture.
Two spectral effects from one field
The headline physics is the gate-tunable photocurrent spectrum. Sweeping the gate voltage from −50 V to +50 V, the authors record responsivity as a function of photon energy and watch two features move.
The first is the A-exciton peak near 1.63 eV, the familiar WSe2 absorption resonance. As the field increases, it redshifts by about 4.8 meV. That is the Stark effect in its quantum-confined form: the field pulls the electron and hole wavefunctions apart within the exciton, lowering its energy roughly in proportion to the square of the field, with an estimated exciton polarizability that the authors extract from the shift. A 4.8 meV shift is small in absolute terms — a few nanometers of wavelength — but it is a clean, reversible, electrically controlled movement of a spectral feature, and it grows monotonically with gate bias.
The second feature is more interesting, and the authors are appropriately careful about it. At higher gate bias, a new responsivity peak emerges near 1.86 eV, a photon energy where the independently measured WSe2 absorption spectrum shows no corresponding peak. It redshifts by about 17 meV as the gate bias increases. The authors assign this to Franz–Keldysh-type modulation: in a strong field, the band edges tilt in real space, electron and hole wavefunctions develop Airy-function tails that overlap below the bandgap, and optical transitions become possible at energies that would otherwise be forbidden. The feature is therefore described as consistent with Franz–Keldysh-type behavior rather than proven to be the textbook effect — a distinction worth preserving, since room-temperature broadening smears spectral features and the internal field cannot be extracted from the applied bias alone, as the authors themselves note. The drain voltage is distributed across the WSe2 layer and both interfaces, so the field inside the absorber is not simply the bias divided by the thickness.
Under drain-source bias the same story plays out with a performance payoff. At Vds = −1 V, reverse current flows, the built-in potential is high, and the photocurrent spectrum rises across the board, with responsivity at about 2.4 eV substantially exceeding the A-exciton peak. External quantum efficiency reaches up to 65 percent at a photon energy of 2.48 eV under that bias. The Franz–Keldysh-type feature, visible between Vds = −0.5 V and 0 V, disappears outside that range, which the authors attribute to changes in field distribution and competing photocurrent mechanisms rather than to any single mechanism switching off. That kind of candor about what a bias sweep does not uniquely determine is rarer in device papers than it should be.
The numbers, with their conditions attached
The device’s headline metrics, all reported at room temperature, are: responsivity of 260 mA/W; rise and decay times of approximately 1 microsecond and 1.1 microseconds under 642 nm illumination at Vds = −1 V and zero gate bias; detectivity of 6.1 × 10^9 Jones; and a linear dynamic range of about 67 dB, measured over incident powers from 0.74 nW to 1,704 nW with a near-unity power-law exponent, indicating clean linear behavior over that range. At zero bias under 520 nm illumination, the rise and decay times are 1.08 and 1.2 microseconds, and repeated measurements across gate biases stay below 1.5 microseconds. The active area is estimated at roughly 100 square micrometers, defined by the overlap of the three flakes.
Those numbers deserve the standard reader’s checklist from any photodetector claim, and the paper supplies most of it: wavelength, bias, power range, and area are all specified, and the response is time-resolved rather than inferred from a frequency rolloff. But one number in the paper cuts against a purely photovoltaic reading of the responsivity. The authors calculate a photoconductive gain of approximately 2.45 × 10^3 and conclude that field-driven carrier separation coexists with a substantial trap-assisted photogating contribution: carriers trapped in localized states act as a local gate, modulating channel resistance so that one absorbed photon can drive many carriers through the circuit. That is gain, not quantum efficiency, and it helps explain how responsivity reaches 260 mA/W in a device whose absorber is only 15 nm thick. It also means the headline responsivity should not be compared naively with gain-free photodiodes, and it is a reason the authors’ own measurement choice matters: the spectral photocurrent was recorded in AC mode with chopped illumination and lock-in detection, which preferentially captures the component synchronized with the optical modulation and suppresses slowly varying trap signals — though, as they note, faster trapping and detrapping cannot be fully excluded.
None of this diminishes the result. It locates it correctly: a fast, gate-tunable, gain-assisted vertical detector, not a record-breaking photodiode. The microsecond response is the genuinely notable part. Phototransistors with large gain are typically slow, because the trapped carrier must eventually be released; here the short vertical channel keeps extraction fast even while gain is present, and repeated on/off cycling at gate biases of −50, 0, and +50 V shows stable, reversible switching.
Why a tunable spectrum is worth more than a fast pixel
The application the authors point to is the miniaturized computational spectrometer, and the logic is worth spelling out. A conventional spectrometer disperses light and measures each wavelength channel separately, which sets a minimum size dictated by optics. A computational spectrometer replaces the dispersive element with a detector — or a small set of detectors — whose response varies in a known way with some control parameter, then recovers the spectrum by solving an inverse problem. The detector’s tunability is the dispersive element.
That architecture only works if the tunable responses are sufficiently distinct, stable, and repeatable. The paper checks two of those boxes: repeated spectral scans under identical bias show stable peak positions, and the device switches reversibly over long on/off cycling. It does not yet demonstrate a reconstructed spectrum — no spectrum is recovered from the device’s own response curves in this paper — so the spectrometer claim remains a direction, not a demonstration. That gap is the honest boundary of the work, and it is where the next paper in this line should be judged.
The broader context helps calibrate expectations. Electrically programmable spectral responses have already been shown in other forms: deformable MoS2 homojunctions for reconstructive microspectrometry, devices exploiting electrically tunable band alignment for multispectral response, reconfigurable graphene plasmonic detector arrays for multichannel infrared imaging, and ferroelectric-controlled plasmonic surfaces for neuromorphic vision preprocessing. What distinguishes the Gr/WSe2/SnSe2 device in that landscape is the combination of microsecond speed, gate tunability through two distinct field effects, and a vertical geometry whose channel is defined by flake thickness rather than lithography. The comparison chart in the paper, benchmarking responsivity against response time for previously reported van der Waals photodetectors, places it in a favorable corner — with the usual caveat that compared devices were measured under different areas, wavelengths, and biases, so the chart is an overview, not a ranking.
What stands between this device and a product
Three gaps matter most, and the paper is forthright about the first. The study characterizes a representative device in detail but establishes no device-to-device statistics. For a computational spectrometer, that is not a footnote: reconstruction algorithms depend on knowing each pixel’s response curve, and if every exfoliated flake behaves differently, calibration becomes per-device work rather than a factory step. The fabrication route — mechanically exfoliated flakes, dry transfer with optical alignment, electron-beam lithography — is a laboratory route, not a manufacturing one. Scaling to arrays, let alone wafer scale, requires growth and transfer processes that preserve the weak-screening graphene bottom electrode and the clean interfaces the physics depends on.
The second gap is the operating voltage. Gate sweeps from −50 V to +50 V are routine in a probe station and awkward in a battery-powered sensor. The physics does not obviously require those voltages — the field matters, not the voltage, and a thinner gate dielectric or a shorter channel would lower them — but no low-voltage variant is shown here, and any claim about portable spectrometers has to reckon with the driver electronics.
The third gap is the one the authors flag themselves: the spectral features are broad at room temperature, and the Franz–Keldysh-type assignment is consistency, not proof. A computational spectrometer lives or dies on how distinguishable its response curves are, and broad, overlapping features make the inverse problem harder. Narrower lines, larger shifts, or more independent tuning knobs — a second gate, a different TMD, a resonant cavity — would all sharpen the instrument.
None of these gaps is a reason to dismiss the device. They are the difference between a physics result and a product, and the paper is careful not to blur it. What it establishes is a mechanism with unusual clarity: a graphene bottom electrode that lets a gate field into a vertical WSe2 channel, a Stark shift and a field-induced below-bandgap feature that together reshape the photocurrent spectrum, and a short channel that converts that reshaped absorption into a fast, high-responsivity signal. For a field that has long treated exciton binding energy as a tax on detector performance, a device that uses an electric field to both tune the spectrum and pay that tax is a useful new data point — and a plausible building block for spectrometers that trade gratings for voltages.
Sources
Vu Khac Dat, Minh Dao Tran, Minh Chien Nguyen, Van Dam Do, Ngoc Thanh Duong, Yu Jin Jang, and Ji-Hee Kim, “Electric-Field-Controlled Optical Response in Vertical Gr/WSe2/SnSe2 Heterojunction Photodetectors,” ACS Nanoscience Au (2026), published online October 20, 2026, DOI: 10.1021/acsnanoscienceau.6c00136. Received July 27, 2026; accepted October 9, 2026. All device figures in this article — 260 mA/W responsivity, ~1 μs response at Vds = −1 V, 65% external quantum efficiency at 2.48 eV, 6.1 × 10^9 Jones detectivity, 67 dB linear dynamic range, ~4.8 meV Stark shift, ~17 meV Franz–Keldysh-type shift, ~100 μm² active area, and 15 nm / 10 nm WSe2 / SnSe2 thicknesses — are taken from this paper. Context on prior computational-spectrometer and programmable-detector approaches follows the paper’s own cited literature; no additional performance claims are introduced.