Superior performance of printed optoelectronic synapses based on defect-controlled monolayer MoS₂ with ultralow power consumption for neuromorphic computing

Abstract

Optoelectronic synapses (OES), which integrate photodetection and synaptic functions in a single platform, offer a promising approach to mimic the visual processing capabilities of the human brain. Two-dimensional (2D) materials are attractive for OES devices due to their excellent energy efficiency and high photoelectric conversion capability. However, most 2D material-based OES devices rely on conventional lithography and post-growth defect engineering, both of which require expensive cleanroom facilities and complex processing steps. Here, we present a hybrid fabrication strategy that integrates cost-effective printing technology, as an alternative to conventional lithography, with in situ defect engineering during CVD growth of monolayer MoS₂. The latter intrinsically tunes the defect density and eliminates the need for any post-growth treatments. This approach enables precise control over defect density while ensuring large-area uniformity and fabrication scalability. The resulting OES devices exhibit excellent photoresponsivity (10.3 A W⁻¹) and stable synaptic behaviors, including excitatory postsynaptic current, paired-pulse facilitation, short-term memory, long-term memory, and spike-timing-dependent plasticity. Remarkably, the device mimics human learning and forgetting with an ultralow energy consumption of 1.2 fJ per synaptic event, outperforming the energy efficiency of biological synapses (∼10 fJ). Furthermore, an artificial neural network trained using device-derived parameters achieves a recognition accuracy of 87.1% on the MNIST handwritten digit dataset. Density functional theory calculations elucidate the crucial role of in situ engineered sulfur vacancies in modulating carrier dynamics and defect-assisted charge trapping, providing a fundamental understanding of the light-induced synaptic behavior in the device.

---

Introduction

The von Neumann architecture has been the foundation of conventional computing systems to solve complex mathematical problems for decades. However, its inherent architecture, characterized by the physical separation of memory and processing units, leads to serious inefficiencies. The constant movement of data between the central processing unit and memory slows computation and leads to significant energy consumption.¹ The rapid growth of machine learning and artificial intelligence has significantly increased data processing demands, further amplifying these limitations and raising concerns about a potential energy crisis.^2–4 In-memory computing, inspired by the human brain, is gaining attention as a transformative approach to address the above challenge.^5,6 The human brain, comprising ∼10¹¹ neurons interconnected by 10¹⁵ synapses, operates as an exceptionally efficient computing system, consuming only ∼20 W of power.⁷ Mimicking these functionalities in electronic systems has the potential to revolutionize computation, achieving superior energy efficiency. Vision, which contributes approximately 80% of the sensory input processed by the human brain, is particularly critical in neuromorphic applications.^8,9 Inspired by the human visual system, artificial optoelectronic synapses (OES) have emerged as a transformative technology, integrating photodetection and synaptic functionalities into a single platform.^10–14 This system mimics the optical nerve system of the human retina, enabling simultaneous perception, processing, and memory storage of visual information without requiring external data transfer pathways. The increasing demand for intelligent image sensors with reconfigurable and self-learning capabilities emphasizes the importance of optoelectronic synapses in advancing artificial vision systems.^15–19 These devices enhance decision-making processes and improve the efficiency of human–machine interactions by enabling real-time visual recognition and interpretation.

Two-dimensional (2D) materials have emerged as fundamental building blocks for next-generation computing technologies, owing to their atomically thin structure, which enables exceptional energy efficiency.^20,21 Their strong light–matter interactions and high photoelectric conversion efficiency further make them ideal candidates for optoelectronic applications.²² Consequently, significant progress has been achieved in the fabrication of photodetectors based on 2D materials.^23–25 Notably, photodetectors can be engineered into optoelectronic synaptic devices by introducing trap states in the material, which induce persistent photoconductivity (PPC) and slow the carrier recombination dynamics.^14,26–32 Several approaches including ion beam irradiation,³³ chemical treatment,³⁴ and plasma treatment³⁵ have been employed to create defects in 2D materials. However, such post-treatment processes often cause structural damage to the material. To overcome this limitation, a single-step in situ growth of defect-rich 2D materials^36,37 has emerged as a promising strategy for fabricating high-performance synaptic devices. Among 2D materials, monolayer (1L) MoS₂ grown by chemical vapor deposition (CVD) exhibits intrinsic defects, particularly sulfur vacancies (V_s), which are energetically favorable.^38,39 Here, we present a one-step in situ V_s defect modulation in 1L MoS₂, offering a non-destructive and efficient approach to realizing optoelectronic synaptic devices.

Despite the significant potential of 2D materials and the ease of defect modulation through CVD, device fabrication has predominantly relied on intricate lithography techniques. This approach requires cleanroom environments and high-vacuum systems for metal deposition, rendering the fabrication process expensive and time-intensive.^28,40–42 Printing technology has recently emerged as a viable and cost-effective alternative for fabricating microdevices.^43–45 Printing is a non-vacuum, maskless process that operates at room temperature, significantly reducing fabrication costs. Thus, it has become a widely accepted method for fabricating various electronic devices including memristors,⁴⁶ capacitors,⁴⁷ field-effect transistors,⁴³ and photodetectors.^48–50 Printed devices often exhibit performance metrics comparable to devices fabricated using conventional methods. Recently, printing technology has also been used to fabricate synaptic transistors on various material platforms, such as carbon nanotubes, organic semiconductors etc.^51–53 However, printed opto-electronic synapses based on 2D materials have not been reported.

Here, we demonstrate CVD growth of defect-controlled monolayer MoS₂ as the photoactive material, combined with printed electrodes as source–drain contacts. This hybrid fabrication approach integrates the precision of the CVD-grown 2D material with the scalability and cost-effectiveness of printing technology. To the best of our knowledge, this is the first demonstration of a high-performance photodetector and OES device fabricated using this hybrid technique. The device exhibits excellent photodetection performance with an on–off ratio of ∼10⁴ and a peak responsivity of 10.3 A W⁻¹ at a 5 V external bias. The ultrahigh detectivity of 1.6 × 10¹² Jones makes the device efficient for detecting ultralow optical signals. Moreover, the OES device with V_s defects exhibits multiple synaptic behaviors, including an excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), short-term memory (STM), long-term memory (LTM), and spike-timing-dependent plasticity (STDP), all with excellent efficiency. The OES device achieves an extraordinary energy efficiency of only 1.2 fJ per synaptic event, lower than the typical ∼10 fJ by the human brain. Our hybrid approach offers a cost-effective pathway to fabricate high-performance optoelectronic devices, presenting a promising direction for the development of next-generation synaptic device technologies. Furthermore, DFT calculations provide theoretical insight into the underlying mechanism and the role of sulfur vacancies in OES operation.

Experimental section

Growth and device fabrication

Monolayer MoS₂ was grown on a SiO₂/Si substrate via CVD,⁵⁴ as described in the SI (Fig. S1). Briefly, molybdenum oxide (MoO₃) and sulfur (S) powders were employed as precursors with sodium chloride (NaCl) as a growth promoter. The growth was conducted at two distinct temperatures of 750 °C and 850 °C under a pressure of 1.0 mbar. Following the growth of the monolayer MoS₂ film, source and drain contacts were printed using a microcantilever-based printing technique. The Nano eNabler™ Molecular Printing System (BioForce Nanosciences, Inc.) operates similarly to inkjet printing but with higher precision. The device fabrication process is similar to those in our previous reports.^48,55,56 In brief, a new surface patterning tool (SPT) was cleaned with acetone, isopropanol, and deionized water, and subsequently dried using a nitrogen gas flow. The cleaned SPT underwent UV ozone treatment for 15 minutes to ensure surface activation. Silver nanoparticle (Ag NPs) ink was then loaded into the SPT for printing the contact electrodes. The source electrode was first printed onto the MoS₂ monolayer and dried at 100 °C for 30 minutes. After drying, the drain electrode was printed while maintaining a source–drain separation of ∼40 μm. The device was annealed at 100 °C to ensure stable contact formation. The successful integration of the printed electrodes onto the monolayer MoS₂ was confirmed by Raman spectroscopy.

For lithography-based device fabrication for comparison purposes, the 1L MoS₂ film was spin-coated with a photoresist, followed by soft baking and patterning using UV laser direct writing. After development, contact electrodes were deposited by thermal evaporation, and the lift-off process was used to complete the electrode patterning.⁵⁷

Characterization

Raman and photoluminescence spectroscopy measurements were performed using a Micro-Raman spectrometer (LabRam HR800, Jobin Yvon) with a 532 nm laser as the excitation source. The thickness of the films was measured using atomic force microscopy (AFM; Cypher, Oxford Instruments). The crystal structure of the monolayer MoS₂ was characterized using field emission transmission electron microscopy (FETEM, JEOL-2100F) performed at an accelerating voltage of 200 kV. To gain chemical insights into the material, X-ray photoelectron spectroscopy (XPS) measurements were carried out in a PHI 5000 VersaProbe III instrument (ULVAC-PHI, Inc.). All the electrical characterization processes were conducted using a Keithley 4200A-SCS parameter analyzer. The 405 nm and 532 nm laser sources modulated with a digital pulse generator were utilized to study the temporal response of the device.

Computational details

The electronic band structure and density of states (DOS) of the pristine monolayer (1L) MoS₂ and its defect-rich configurations (sulfur vacancy) were investigated using DFT within a plane-wave basis set, as implemented in the Quantum ESPRESSO package.^58,59 A 5 × 5 × 1 monolayer MoS₂ supercell was constructed to model pristine and defect-rich systems. The exchange–correlation interaction was described using the generalized gradient approximation (GGA) with the Perdew–Burke–Ernzerhof (PBE) functional.⁶⁰ A plane-wave kinetic energy cutoff of 60 Ry was employed, and the Brillouin zone was sampled using a Monkhorst–Pack k-point⁶¹ mesh of 5 × 5 × 5. The electronic self-consistency criterion was set to an energy convergence threshold of 1 × 10⁻⁶ eV. A dense k-point grid (7 × 7 × 7) was used for non-self-consistent calculations.

Results and discussion

Material characterization

Defects play a pivotal role in governing the optoelectronic synaptic functionalities of two-dimensional materials. In CVD-grown monolayer MoS₂, such defects are intrinsically formed due to the high-temperature growth environment, and the defect density can be systematically tuned by controlling the growth parameters. Fig. 1(a and b) presents the optical microscopy images of CVD-grown multilayer (ML) and 1L MoS₂ films, respectively. Raman spectroscopy was employed to determine the layer number and assess the structural integrity of the films. Fig. 1(c) displays the Raman spectra of both samples, exhibiting two distinct characteristic peaks corresponding to E¹_2g and A_1g vibrational modes.^62,63 The E¹_2g mode corresponds to the in-plane vibrations of Mo and S atoms, while the A_1g mode represents the out-of-plane vibrations of S atoms. The peak separation (Δk) between these two modes was measured as 26.3 cm⁻¹ for the multilayer film and 19.1 cm⁻¹ for the monolayer film, confirming their respective layer thicknesses.^62,63 AFM measurements were performed to directly confirm the thickness of the films on a region where the film and substrate are distinguishable. The corresponding AFM image and height profile are provided in Fig. S2, confirming the monolayer and multilayer nature of the film. The thickness of the monolayer film is ∼0.8 nm whereas the multilayer film is ∼38 nm.

Fig. 1 (a) The optical microscopy image of a multilayer (ML) MoS₂ grown on a SiO₂/Si substrate. (b) The optical microscopy image of a monolayer (1L) MoS₂ grown on the SiO₂/Si substrate. (c) The corresponding Raman spectra of the film exhibit characteristic peaks, with Δk (peak separation) values of 26.3 cm⁻¹ and 19.1 cm⁻¹, confirming the multilayer and monolayer nature of the films, respectively. (d) The temporal response of the films under a 532 nm laser exhibits a sharp (ML film) and slow decay (1L film) nature upon removal of the optical stimulus. The inset shows the schematic of the device configuration.

The optoelectronic characteristics of the two films were investigated next under 532 nm laser illumination, as shown in Fig. 1(d). The inset illustrates the device configuration for both films. The monolayer MoS₂ film exhibits a photocurrent response nearly two orders of magnitude higher than that of the multilayer film, attributed to the direct bandgap nature of the monolayer. Furthermore, the monolayer film demonstrates a slow decay in the photocurrent upon removal of the optical signal, indicating the presence of a persistent photoconductivity effect, which is crucial for optoelectronic synaptic applications. This slow photocurrent decay in the monolayer is attributed to the higher defect density compared to that of the multilayer film.^64,65

Next, we tuned the growth temperature for the monolayer MoS₂ film to achieve a highly defective structure. We observed a distinct variation in defect densities in samples grown at 750 °C (low temperature, LT) and 850 °C (high temperature, HT). We employed room-temperature photoluminescence (PL) spectroscopy to qualitatively assess the defect densities, which are highly sensitive to defect states in two-dimensional semiconductors. Fig. 2(a) presents the PL spectrum of the LT-grown monolayer MoS₂. The spectrum has been deconvoluted into three Lorentzian components, revealing a sharp and intense neutral exciton (A⁰) peak with minimal contribution from the trion (A⁻) state. This spectral feature indicates a low concentration of charged defect states in the LT sample.⁶⁶ In contrast, the PL spectrum of the HT-grown monolayer (Fig. 2(b)) MoS₂ exhibits a broadened A⁰ peak along with a significantly enhanced A⁻ emission. The pronounced trion contribution is indicative of an increased density of negatively charged defect sites, primarily attributed to sulfur vacancies.^36,66,67 A comparative analysis of the spectral weight distribution between the LT and HT samples (Fig. 2(c)) further corroborates the higher defect density in the HT film. The inset of Fig. 2(c) presents the A⁻/A⁰ intensity ratio, which serves as a qualitative metric for defect density and highlights the enhanced trion formation in the HT sample. The PL spectra taken at various random points of the LT- and HT-samples (Fig. S3) further validate these characteristic features of the respective films. We performed temporal photocurrent measurements (Fig. 2(d)) to understand the impact of these defect states on the optoelectronic response. The HT-film demonstrates a significantly slower decay of the photocurrent compared to the LT-film. This prolonged decay is attributed to defect-mediated carrier trapping and delayed recombination dynamics.^68,69 The green square pulse represents the temporal profile of the laser excitation (on and off time). These observations highlight the strong dependence of the excitonic recombination dynamics and their temporal photoresponse on the defect landscape, which is modulated by the CVD growth temperature. Additionally, temporal photoresponse measurements under two consecutive optical pulses were conducted to evaluate the paired-pulse facilitation (PPF) index, an essential indicator of synaptic behavior. The HT sample exhibits a higher PPF index than the LT sample (Fig. S4), further indicating the higher density of defects and its role in governing the temporal photoresponse decay characteristics.

Fig. 2 Investigation of defect densities and optoelectronic behavior in 1L MoS₂ grown via CVD at two different temperatures. (a) Room-temperature PL spectrum of the low-temperature (LT, 750 °C) grown MoS₂ sample, exhibiting a sharp neutral exciton (A⁰) peak with minimal trion (A⁻) contributions, indicative of low defect density. (b) The room-temperature PL spectrum of the high-temperature (HT, 850 °C) grown sample reveals a broadened neutral exciton peak along with a significant trion contribution, suggesting an increased defect density. (c) Comparative analysis of the spectral weight between the LT and HT samples confirms a higher defect density in the HT-grown film. The inset presents the trion to neutral exciton (A⁻/A⁰) intensity ratio for both samples. (d) Comparative temporal photoresponses of LT and HT samples demonstrating a slower photocurrent decay in the HT sample, consistent with enhanced defect-mediated carrier trapping. The green square pulse symbol shows the laser on and off times.

The HT sample was further characterized by various spectroscopic and microscopic characterization techniques due to its higher defect concentration, as mentioned above. Fig. 3(a) presents an optical image of the 1L MoS₂ film, highlighting a small uncovered region (marked by a circular spot) within the film. Raman mapping was conducted to examine the spatial uniformity of the sample, as shown in Fig. 3(b and c). The uniform intensity distribution of the E¹_2g mode (represented in red) and the A_1g mode (represented in green) confirms the high spatial uniformity of the monolayer film over a large area. A detailed Raman spectral analysis is provided in Fig. S5. The crystal structure and atomic arrangement of the 1L MoS₂ film were analyzed using TEM. The low-magnification TEM image of the film is shown in Fig. S6a, while Fig. S6b presents the corresponding high-resolution image. The processed image in Fig. 3(d) depicts the hexagonal crystal symmetry with an interplanar spacing of 0.27 nm, corresponding to the (100) planes of monolayer MoS₂. Additionally, the selected area electron diffraction (SAED) pattern in Fig. 3(e) exhibits a hexagonal symmetry, confirming the hexagonal lattice structure of the MoS₂ film. Energy-dispersive X-ray spectroscopy (EDS) analysis conducted at multiple random regions across the film (Fig. S7) confirms the presence of sulfur vacancies (V_s) in the sample. Low-temperature PL measurements were conducted to further characterize the material and evaluate the defect landscape.³⁶ The PL spectrum taken at 80 K under 514.5 nm laser excitation is shown in Fig. 3(f), which exhibits two prominent peaks centered at ∼1.85 eV and ∼2.03 eV, corresponding to the A and B excitonic transitions of 1L MoS₂, respectively.^70,71 These peaks arise from direct band-to-band transitions in 1L MoS₂ and are attributed to the splitting of the valence band at the K point of the Brillouin zone, due to the strong spin–orbit coupling as demonstrated in Fig. 3(g).⁷⁰ The PL spectrum was deconvoluted into four Lorentzian peaks to gain a deeper insight into these defects. The A exciton peak includes contributions from neutral excitons and negatively charged trions located at 1.84 eV and 1.80 eV, respectively. The trions result from interactions between excitons and electrons in the system, which are primarily introduced by V_s defects, acting as n-type dopants.⁶⁶ These V_s defects generate a surplus of free electrons, facilitating trion formation in the material. The substantial trion contribution in the PL spectrum provides direct evidence of the significant presence of V_s defects in the sample.^36,66,67 Additionally, a defect-related peak was observed at 1.76 eV, attributed to transitions involving in-gap states within the forbidden energy gap.⁷² These states are introduced by structural defects present in 1L MoS₂. Thus, the PL spectrum indicates the presence of significant structural defects in 1L MoS₂, which contribute to a slower decay of photogenerated carriers compared to the faster decay observed in defect-free or low-defect samples.³⁶ XPS measurements were performed to further investigate the chemical composition and defect states. Fig. S8 presents the full XPS spectrum of the MoS₂ sample, showing peaks corresponding to Mo, S, O, C, and Si (coming from the substrate), with no additional impurity peaks detected, thereby confirming the purity of the sample. Fig. 3(h) displays the high-resolution XPS spectrum for the Mo 3d states, featuring peaks at 228.3 eV and 231.4 eV, which correspond to the Mo 3d_5/2 and Mo 3d_3/2 doublets, respectively, along with the S 2s peak at 225.5 eV. Similarly, Fig. 3(i) shows the high-resolution XPS spectrum of the S 2p states, with peak positions at 161.1 eV and 162.3 eV. All observed peak positions are in agreement with previously reported values of monolayer MoS₂.⁷³ Furthermore, the stoichiometry of the S to Mo ratio (1.85) indicates the presence of sulfur vacancies (V_s) in the MoS₂ lattice. This observation is consistent with the defect-related features observed in the PL spectrum and EDS data.

Fig. 3 Characterization of the defect-rich 1L MoS₂ film (HT). (a) The optical microscopy image of the film; the region highlighted by the circle indicates the area without the film. (b and c) The red and green color maps correspond to the E¹_2g and A_1g Raman modes of 1L MoS₂, indicating the uniformity of the film. The black region corresponds to the regions without a film. (d) The HRTEM image displays hexagonal crystal symmetry with an interplanar spacing of 0.27 nm, corresponding to the (1 0 0) planes. (e) The SAED pattern of MoS₂ shows the hexagonal crystal structure. (f) The low-temperature PL under 514.5 nm laser excitation demonstrates the prominent defect-induced peak (X_D) along with a significant trion contribution. (g) Schematic of the exciton-related radiative transition of neutral excitons (A⁰), charged excitons (trions, A⁻), B excitons, and a defect-induced peak (X_D). (h and i) The high-resolution XPS spectra of Mo 3d and S 2p show the characteristic peak associated with MoS₂.

Photodetector performance

The high surface-to-volume ratio and direct bandgap nature of the monolayer MoS₂ enable efficient optical absorption, making it highly suitable for photodetection applications. We first compared the optoelectronic response of the printed device with that of a reference device of similar dimensions, fabricated using conventional photolithography. Both devices exhibit comparable photoresponse characteristics (Fig. S9) with the printed device demonstrating a slightly higher on/off ratio. This confirms that the contact quality achieved through our low-temperature, resist-free printing method is comparable or superior to that obtained via lithography. The absence of photoresist residues and lift-off imperfections in the printed devices may contribute to the improved interfacial quality and reduced contact resistance, which is a desirable attribute in device fabrication.

We carried out a detailed optoelectronic investigation of the printed device next. A schematic of the cross-sectional view of the device configuration is depicted in Fig. 4(a). Fig. 4(b) presents the optical image of the printed device, where the CVD-grown monolayer MoS₂ serves as the photosensitive material and the printed silver nanoparticle-based electrodes are the source–drain contacts. The integrity of the monolayer MoS₂ film in the electrode channel is confirmed through Raman spectroscopy mapping analysis. The red and green maps correspond to the E¹_2g and A_1g Raman modes of monolayer MoS₂, respectively. The Raman intensity map confirms the uniform integration of the printed electrodes onto the monolayer MoS₂ film without disrupting its structural continuity. Fig. S10 shows a snapshot of the Raman spectroscopy screen, captured following the Raman mapping of the device. Furthermore, we performed low-temperature PL spectroscopy on the monolayer film before and after electrode deposition to assess any possible defects/imperfections generated during the electrode printing process. The PL spectra in Fig. S11 do not exhibit any additional defect-related emission features or noticeable peak broadening after printing. Moreover, the ratio of trion to neutral exciton intensities remains nearly unchanged. These observations indicate that the printing process does not introduce new defects into the monolayer MoS₂, thereby preserving its structural and optical integrity, all of which are critical for reliable and reproducible device performance.

Fig. 4 The photoresponse study of the printed device under 532 nm laser illumination. (a) A schematic of the device cross-sectional view; 1L MoS₂ acts as the active channel with Ag printed electrodes as source–drain contacts. (b) Optical image and Raman mapping of a specific region of the printed device. The red and green maps correspond to the E¹_2g and A_1g Raman modes of the monolayer MoS₂, demonstrating the integrity of MoS₂ in the channel. (c) I–V characteristics of the PD under dark and light conditions. (d) Variation of responsivity and detectivity of the device with the intensity. (e) Variation of the photocurrent with light intensities. The obtained θ value of 0.89 demonstrates the presence of trap states in MoS₂. (f) Variation of responsivity and detectivity of the device with external bias voltages.

The photoresponse of the device was investigated across the full visible spectrum to determine the optimal excitation wavelength. Fig. S12 shows the spectral response, revealing a broad photoresponse with a maximum sensitivity around 520 nm. Based on this observation, detailed photoresponse measurements were subsequently carried out under 532 nm excitation. The current–voltage (I–V) characteristics of the printed device, under dark conditions and 532 nm illumination at varying intensities, are shown in Fig. 4(c). A substantial rise in current is observed under illumination, which is attributed to the absorption of photons, generating electron–hole pairs within the MoS₂ layer. As the intensity of the incident light increases, more photons interact with the MoS₂, leading to a higher generation of charge carriers and, consequently, an increased photocurrent. The dark current of the device at 0.5 V was 1.2 × 10⁻⁹ A, and the photocurrent increased to 1.6 × 10⁻⁵ A under 532 nm illumination at an intensity of 123.7 mW cm⁻². This four-order of magnitude increase in current highlights the excellent photoelectric conversion efficiency of the printed device.

Some key figures of merit such as responsivity (R), specific detectivity (D*), and external quantum efficiency (EQE) were evaluated to quantitatively study the photodetection performance of the device. Responsivity, defined as the photocurrent generated per unit of incident power, is given by:²³

where I_Ph (I_light–I_dark) refers to the photocurrent, P is the intensity of the light, and A is the effective device area. Specific detectivity (D*), which indicates the ability of a device to detect low-power optical signals, is calculated using:²³Here, q is the electronic charge, and J_d stands for the dark current density. The EQE, representing the ratio of photo-generated carriers to incident photons, is determined as:²³where h is the Planck constant, λ is the wavelength of the light and c is the speed of light. The peak responsivity and detectivity of the device were found to be 10.3 A W⁻¹ and 1.6 × 10¹² Jones, respectively, under 532 nm illumination at a 5 V bias and 0.1 mW cm⁻² intensity. Both R and D* decreased with increasing light intensity, as depicted in Fig. 4(d), with values dropping to 4.3 A W⁻¹ and 6.7 × 10¹¹ Jones at 123.7 mW cm⁻². This decrease is attributed to enhanced carrier recombination and saturation effects at higher intensities. Increasing carrier density at higher incident powers increases the recombination probability, thereby reducing the photocurrent. Additionally, saturation effects limit carrier generation and collection efficiency, and localized heating at high intensities may further degrade performance. The relationship between photocurrent and light intensity was studied, as shown in Fig. 4(e). The exponent θ was extracted by fitting the experimental data with the equation:⁵⁰

I_Ph ∝ P^θ (4)

where θ represents the response quality with intensity. The θ value was found to be 0.89 under 532 nm excitation. The deviation of θ from the ideal value of unity is attributed to complex processes involving carrier generation, trapping, and recombination, with carrier trapping likely resulting from intrinsic defects in the as-grown monolayer MoS₂.

The device performance was also studied under varying bias voltages at a fixed light intensity (0.1 mW cm⁻²), as shown in Fig. 4(f). Responsivity and detectivity values increased from 0.2 A W⁻¹ and 3.7 × 10¹¹ Jones to 1.9 A W⁻¹ and 9.4 × 10¹¹ Jones as the applied bias was raised from 0.1 V to 1.0 V. This trend is attributed to the enhanced transport of photogenerated carriers under higher electric fields, which facilitates more efficient extraction and movement toward the electrodes, thereby improving both R and D*. Furthermore, the photoresponse of the device under 405 nm excitation was evaluated, the details of which are given in Fig. S13. The variation of the EQE with the intensity of light and the applied bias is shown in Fig. S14, under the illumination of 532 and 405 nm excitation. The EQE increases with the applied bias voltage, and it decreases at higher incident power.

Optoelectronic synaptic characteristics

After achieving exceptional photoelectric conversion efficiency, the photonic synaptic plasticity of the printed device was thoroughly studied for the HT monolayer device. The conductivity increases under optical stimulation in a conventional photodetector, but rapidly returns to the baseline once the light source is removed. A synaptic device, however, retains this optical information even after the optical stimulus is removed. This synaptic behavior relies on persistent photoconductivity as mentioned before,⁷⁴ where the increase in conductivity does not immediately decay after the light is switched off. The current decays rather very slowly, retaining the high current value for a prolonged period. The basis for PPC lies in the intrinsic lattice defects within the active material, which generate multiple localized energy states in the forbidden band. The electronic states trap photogenerated electrons, thereby slowing down recombination and resulting in PPC, and this phenomenon can be modulated to emulate synaptic characteristics.

Fig. 5(a) shows the I–V characteristics of the device, where the green curve corresponds to 532 nm illumination, and subsequent curves depict dark conditions measured at various time intervals after the light was switched off. Notably, the dark current does not immediately revert to its initial value after the removal of the light, but rather decays gradually over time. The dark current does not fully stabilize even after six hours of light-off condition, indicating significant PPC within the device. The dark current variation at 0.5 V with time is shown in the inset of Fig. 5(a). Fig. 5(b) illustrates the temporal response of the device to a single 532 nm optical pulse, showing a robust photoresponse with PPC. The time-dependent decay of the current, shown in Fig. 5(c), is fitted using a double-exponential decay model:

where I₀, C₁, and C₂ are the constants, and τ₁ and τ₂ are the relaxation times. The values of the relaxation time were found to be 0.49 and 11.17 minutes, respectively. The PPC enables the device to function as an OES, capable of mimicking visual perception and memory retention, analogous to human brain functionality. Here, the optical stimulus acts as the presynaptic signal, while the device current serves as the synaptic weight.

Fig. 5 Optoelectronic synaptic characteristics of the printed device. (a) I–V characteristics of the device under light (green curve) and all the other curves represent dark conditions taken continuously (after switching off the light) at different time intervals, indicating a slow decay of the photocurrent. The inset shows the variation of the dark current at 0.5 V with time. (b) Temporal response of the device under 532 nm illumination, demonstrating PPC behavior. (c) Decay of photocurrent with time at 0.5 V, where double-exponential fitting yields time constants of 0.49 and 11.17 minutes. (d) Temporal response under two identical optical pulses of 1 s durations each showing an enhanced excitatory post-synaptic current (EPSC) for the second pulse, mimicking biological synaptic behavior. (e) Paired-pulse facilitation (PPF) index as a function of the pulse interval, showing a decrease in PPF with increasing interval. The double-exponential fitting provides characteristic time constants. (f) Illustration of human learning, where the sensory input is stored as short-term memory (STM) with attention and is converted to long-term memory (LTM) through repeated training. (g) Analogous learning behavior in the device: single optical pulses induce STM, while repeated pulse exposure converts STM to LTM, with EPSC incrementally increasing with each pulse (indicated by an arrow). (h) Memory retention time comparison between single-pulse and 150-pulse stimulation, showing retention times enhanced from 23.2 s to 84.4 s.

Paired-pulse facilitation (PPF) is a key feature of biological synapses, where closely spaced presynaptic spikes cause the second postsynaptic response to exceed the first. Two consecutive optical pulses were applied to the device to replicate this behavior, and the resulting excitatory post-synaptic current (EPSC) was recorded, as shown in Fig. 5(d). The second pulse current is notably higher than the first, emulating biological synaptic behavior. This effect arises because the photo-generated carriers from the first pulse do not fully recombine before the arrival of the next pulse, leading to an accumulation of carriers and enhanced photoconductivity with successive pulses. The PPF index, a measure of the learning rate, is defined as the ratio of the second to the first ΔEPSC (the difference in EPSC and dark current); it is measured as 158% for a 1 s pulse width. This index decreases as the interval between pulses increases, mirroring biological synapses, as shown in Fig. S15. The variation of the PPF index with the pulse interval is shown in Fig. 5e and fitted using the following double-exponential model:

where I₀, C₃, and C₄ are the constants, and τ₃ and τ₄ are the relaxation times for the rapid and slow decay components. The relaxation times are found to be 0.39 s and 2.65 s, respectively. The slow decay component is approximately seven times longer than the rapid one, closely similar to the biological synapse.

Memory processing occurs in biological systems through three key units: a sensory unit that perceives external information, short-term memory (STM) that retains this information briefly, and long-term memory (LTM) where information is stored for extended periods following repeated training or rehearsal. The learning process is illustrated in Fig. 5(f). STM to LTM transition was achieved in the device by applying repeated optical pulses (∼150 pulses), as shown in Fig. 5(g). The EPSC rises incrementally with each pulse, illustrating the ability of the device to mimic synaptic behavior. The inset shows a linear plot of this relationship, revealing how the pulse count modulates the EPSC and memory retention duration. Fig. 5(h) presents the memory retention (M(t)) curve for the single and multiple pulse irradiation. Here M(t) is defined²⁸ as

M_t = (I_t − I_off)/(I_max − I_off) (7)

where I_t is the current value at any time t, I_off is the dark current and I_max is the maximum current value. The relaxation of photoinduced memory follows a Kohlrausch stretched-exponential function:²⁸

M_t = C₅ exp[−(t/τ₅)^β] (8)

where C₅ is a constant and τ₅ is the memory retention time. For single and multiple pulses, memory retention times are found to be 23.2 s and 84.4 s, respectively, showing nearly four-fold enhancement with multiple pulses. Thus, memory retention can be significantly improved by increasing the pulse count, demonstrating the device's potential as a tunable optical synaptic system. All synaptic measurements here were carried out under 532 nm optical excitation, as this wavelength yielded the best synaptic behavior among the tested wavelengths (405, 450, 532, and 640 nm), as shown in Fig. S16. Additionally, the PPC behavior at different bias voltages is also presented in Fig. S17, demonstrating the robustness of the synaptic response across various operating conditions.

STM and LTM are two distinct types of memory processes within the human brain, both considered to arise from mechanisms of synaptic plasticity. In biological systems, STM reflects transient memory states that decay over time, while LTM represents a more durable retention of information through strengthened synaptic connections. Mimicking these memory processes in artificial synaptic systems is central to developing OES devices, and achieving the STM to LTM transition depends on modulating the synaptic weights by adjusting key parameters such as the pulse frequency, pulse duration, pulse number, and light intensity of the optical stimuli. STM can be effectively converted into LTM through repeated training via optical pulses. Fig. 6(a) illustrates this transformation achieved by systematically increasing the optical pulse frequency. When the pulse frequency is high, the photogenerated carriers have limited time to recombine before the next light pulse arrives, leading to an accumulation of charges and enhancing the conductivity in the device. When a pulse frequency of 0.1 Hz is applied, the EPSC of the device reaches approximately 3.37 × 10⁻⁷ A. However, this EPSC increases nearly threefold to 1.01 × 10⁻⁶ A when the frequency is raised to 5 Hz. Similarly, the decay current (measured after 100 seconds of removing the optical stimuli) increases from 2.90 × 10⁻⁸ A to 9.70 × 10⁻⁸ A, demonstrating that enhanced pulse frequency extends memory retention. The transformation of STM to LTM through pulse frequency modulation is further evidenced in Fig. 6(b), where both EPSC and decay current exhibit a frequency-dependent response. This increase in retention time with higher pulse frequency, as seen in Fig. S18(b and e), demonstrates an efficient STM to LTM transition, mirroring the frequency-based memory reinforcement seen in biological synapses.

Fig. 6 Short-term memory (STM) to long-term memory (LTM) conversion through repeated training by various means of optical stimuli. The right y-axis of (b, d and f) represents the decay current, measured after 100 s of removal of the optical stimuli. (a) Synaptic weight modulation as a function of different optical pulse frequencies. (b) Frequency-dependent modulation of EPSC and decay current. (c) Synaptic weight modulation with varying optical pulse widths. (d) Dependence of EPSC and decay current on pulse width. (e) Synaptic weight modulation as a function of pulse number. (f) Variation of EPSC and decay current as a function of pulse number. (g) Simulated photocurrent mapping: the upper images show that the EPSC increases with the number of pulses, demonstrating learning characteristics. The lower images show the STM to LTM transformation as the pulse count increases (current values are magnified by a factor of 3 for enhanced visibility). The mapping for the memory is done based on the decay current. (h) Synaptic weight modulation with increased light intensities.

Another way for modulating synaptic weights is by varying the pulse width, which directly affects the generation and accumulation of photocarriers. Fig. 6(c) illustrates the synaptic weight enhancement through the increase of pulse width. More carriers contribute to increasing the EPSC by extending the duration of light exposure. The EPSC measures 2.54 × 10⁻⁷ A with a 5 s pulse width, whereas a 200 s pulse width increases it to 6.24 × 10⁻⁷ A. Similarly, the decay current, assessed 100 seconds after light deactivation, rises from 1.20 × 10⁻⁸ A to 1.19 × 10⁻⁷ A, as the pulse duration extends. This significant increase in current retention correlates with the length of the applied pulse, as depicted in Fig. 6(d), indicating that the longer light exposure reinforces the synaptic weight and enhances memory retention. Fig. S18(a and d) displays the positive correlation between memory retention time and the pulse duration, showing that longer pulses can progressively convert STM into LTM. This finding is consistent with memory formation in biological synapses, where increased stimulation time strengthens synaptic connections and facilitates memory retention.

Synaptic modulation can also be achieved by increasing the pulse number at a fixed frequency. Fig. 6(e) shows how repeated pulses at a frequency of 0.1 Hz enhance the EPSC and memory retention. Initially, the EPSC measures approximately 1.47 × 10⁻⁷ A with one optical pulse; however, it doubles to 2.90 × 10⁻⁷ A after 25 pulses. Similarly, the decay current, recorded 100 seconds after stimulation, increases from 1.26 × 10⁻⁸ A to 5.69 × 10⁻⁸ A as the pulse number increases (Fig. 6(f)). The EPSC and decay current grow steadily with an increase in the pulse count, mimicking the biological process where repetitive learning consolidates STM into LTM. The memory retention time also extends as the pulse count rises, indicating the gradual enhancement of memory retention with repeated stimuli (Fig. S18c and f). All the photoresponse characteristics discussed so far were measured under a positive external bias. However, the device also exhibits stable photoresponse behavior with clear synaptic characteristics under a negative external bias, as shown in Fig. S19.

We have also simulated the synaptic behavior of the device in MATLAB to visually represent the learning and memory processes through photocurrent mapping under patterned optical stimuli, as depicted in Fig. 6(g). The simulation demonstrates synaptic weight modulation in response to optical pulses using an ‘S’ shaped illumination pattern. The photocurrent intensity increases relative to the background in the illuminated regions corresponding to the ‘S’ pattern, demonstrating selective pattern recognition. As the number of optical pulses increases, the photocurrent within the patterned region rises, as shown in the upper part of Fig. 6(g), illustrating the learning behavior of the device. Notably, the photocurrent in the ‘S’ region does not immediately return to the baseline even after the optical stimuli are removed, confirming its ability to retain memory. An increase in optical pulse numbers further enhances memory retention, facilitating the transition from STM to LTM, as shown in the lower part of Fig. 6(g). This behavior highlights the ability of the device to detect optical stimuli, learn through reinforced activation, and retain memory over time, mimicking biological synaptic processes. Additional simulations (Fig. S20 and S21) further illustrate the influence of pulse width and frequency on synaptic behavior. Fig. 6(h) demonstrates that higher light intensities generate a greater number of photogenerated carriers, which subsequently enhances the EPSC. When the light intensity is increased from 1.32 mW cm⁻² to 11.47 mW cm⁻², the EPSC rises from 2.53 × 10⁻⁷ A to 1.72 × 10⁻⁶ A, indicating a substantial increase in synaptic weight. Correspondingly, the decay current also follows a similar trend, facilitating the conversion of STM to LTM through higher-intensity light stimuli. Fig. S22 further illustrates that memory retention time is positively correlated with light intensity, suggesting that stronger stimuli contribute to more durable memory formation, analogous to high-intensity training sessions in biological learning. Thus, the device demonstrates enhanced memory retention and STM to LTM conversion through finely tuned external stimuli, indicating its potential use in developing intelligent memory systems for future neuromorphic computing applications.

The human-like learning behavior of our device is demonstrated here by the application of repetitive optical pulse stimulation to the device. The corresponding temporal response under repetitive optical stimuli is shown in Fig. 7(a). When exposed to a continuous pulse train, the EPSC of the device exhibits a cumulative increase with each successive pulse, demonstrating its inherent learning capability. The device exhibits a gradual increase in current with each pulse, reaching from 1.3 µA to 3.1 µA over approximately 130 pulses which demonstrates the learning capability of the device. Thus, approximately 130 pulses are necessary for the device to learn a specific task for the first time. When the optical stimulus is removed, the current begins to decay, analogous to the process of forgetting observed in human memory. During subsequent learning phases, the device exhibits an accelerated response, taking only 41 pulses to reach the same current range of 1.3 µA to 3.1 µA. This reduction in required stimuli reflects the human tendency to relearn forgotten information more rapidly than during initial learning sessions. In the third learning phase, the device requires just 27 pulses to relearn the same task, showing a progressive decrease in learning time with repeated exposure to the same stimuli. This iterative learning process suggests that the device requires less time and fewer stimuli to recall previously learned information, which is analogous to the human learning process. The STM gradually transitions into LTM with continued stimulation and repeated learning, reflecting the mechanism of human memory consolidation through multiple training. The device also replicates human-like forgetting behavior. In humans, repeated learning leads to more prolonged retention and slower forgetting. Similarly, the decay time or forgetting time in our device progressively increases with each successive learning session. After the first learning cycle, the forgetting time is approximately 11.05 s, extending to 12.18 s and 15.02 s during the second and third cycles, respectively, as shown in Fig. 7(b). These findings indicate that as the device continues to learn through repeated optical pulses, it retains information for longer durations, mimicking the human process of memory reinforcement through repeated exposure.

Fig. 7 Human-like learning and forgetting behavior of the device. (a) Temporal response of the device under continuous pulsed 532 nm laser illumination. The first learning cycle requires ∼130 pulses, which decreases to 41 and 27 pulses for the second and third learning cycles, respectively. (b) The forgetting curves (decay current) corresponding to the first, second, and third learning cycles demonstrate an increase in data retention time, from 11.05 s after the first cycle to 12.18 s and 15.02 s after the second and third cycles, respectively. (c) Temporal response of the device at a 1 mV bias, showing an energy consumption per single event of 1.2 fJ.

Energy efficiency is a critical parameter in the development of neuromorphic devices, which aim to emulate the speed and low power consumption of biological synapses. Energy-efficient operation is particularly essential in neuromorphic applications to ensure scalability and practicality. The electrical energy consumption per synaptic event can be calculated using the formula:^10,52

E_electrical = I_peak × V_d × t (9)

where I_peak is the peak current, V_d is the applied drain to source voltage, and t is the pulse width. In this study, I_peak = 3.0 pA, V_d = 1 mV, t = 400 ms. Therefore, the total electrical energy consumption is 1.2 fJ (Fig. 7c), which is remarkable considering that the energy consumption by the human brain is ∼10 fJ. The optical energy consumption can be estimated using the relationship:

E_optical = P × A × t (10)

where P is the power density of the optical spike, A is the effective device area, and t is the optical spike duration.¹⁰ Here, P = 25 mW cm⁻², A = 3 × 10⁻⁴ cm², and t = 400 ms. The total optical energy consumption was calculated to be 3.0 μJ.

A comparison of the metrics of our device against reports in the literature for both printed and some representative non-printed 2D material-based OES devices, including MoS₂, is presented in Table S1, highlighting the superior performance achieved here. Finally, four similar types of devices were fabricated by the same process to address the repeatability of the device performance, and the corresponding I–V characteristics are shown in Fig. S23. All four devices demonstrated persistent photoconductivity with similar characteristics under similar optical pulses. The figures of merit, including the on/off ratio, responsivity, and decay time, were determined for all devices. The comparable values obtained across the four samples confirm the robustness and reliability of the printed device fabrication process. To further confirm the operational repeatability and stability of an individual device, repeated optical pulse measurements were carried out using a 532 nm laser with alternating ON (30 s) and OFF (30 s) cycles for a total duration of approximately 15 minutes. The measurements were performed at four different bias voltages (0.1, 0.2, 0.3, and 0.4 V). The device exhibited a cumulative increase in current with successive pulses in all cases, indicating stable synaptic-like behavior without any observable degradation, as shown in Fig. S24.

In this study, all optical synaptic characteristics were demonstrated for a device with a 40 µm printed channel length. To systematically examine the influence of channel length, additional devices with varying channel lengths (100, 80, 60, and 40 µm) were fabricated, and their temporal photo-responses were measured. As shown in Fig. S25, all devices exhibit stable and reproducible photoresponses with distinct synaptic-like behavior. The photocurrent increases slightly with decreasing channel length, which can be attributed to the lower series resistance and shorter carrier transport paths in shorter channels. Although further reduction in channel length could potentially yield improved performance, this lies beyond our current fabrication capabilities. A more detailed channel-length-dependent study with shorter channels is, therefore, a promising direction for future investigation.

Simulation of the MNIST digit recognition

Inspired by the neuromorphic learning behavior of our OES device, a single-layer perceptron model based on an artificial neural network (ANN) was developed to emulate the neuromorphic visual processing capability of the device. The image recognition task was performed using the modified National Institute of Standards and Technology (MNIST) dataset.^75,76 The ANN architecture consists of an input layer, a fully connected synaptic layer and an output layer as seen in Fig. 8(a). Each pixel of the MNIST image corresponds to an input neuron, which processes the input vector. The output values are computed by multiplying the input vector by the synaptic weight matrix, followed by the application of a softmax activation function to obtain class probabilities. Synaptic weight updates in the neural network are governed by a conductance-based rule designed to emulate synaptic plasticity. The weight update dynamics are described by the equation:⁷⁷where G_n represents the synaptic conductance when the n^th pulse is applied to the device. The parameters α and β are the changing step sizes of the conductance and nonlinearity, respectively. The G_min and G_max denote the minimum and maximum conductance values, respectively. Long-term potentiation (LTP) is induced through the application of continuous optical pulse stimuli, while long-term depression (LTD) is achieved by applying a series of negative electrical voltage pulses, as illustrated in Fig. 8(b). The α and β parameters are extracted by fitting the conductance data for LTP and LTD curves, as shown in Fig. S26. The confusion matrix of the ANN, shown in Fig. 8(c), illustrates the classification performance, where the diagonal elements represent correctly predicted digits. The corresponding recognition accuracy after 100 training epochs was found to be 87.1% (Fig. 8(d)).

Fig. 8 MNIST handwritten digit recognition using a conductance-mediated artificial neural network (ANN). (a) Schematic of the single-layer perceptron ANN architecture implementing a conductance-based weight update mechanism for digit recognition. (b) Experimentally obtained optical long-term potentiation (LTP) and electrical long-term depression (LTD) characteristics were used to extract parameters for the ANN weight update equation. (c) Confusion matrix showing the performance of the trained ANN; correct predictions lie along the diagonal. (d) Recognition accuracy reaches 87.1% after 100 training epochs.

Mechanistic insights into photodetection and defect-mediated synaptic behavior

The photodetection and synaptic functionalities in the monolayer MoS₂ device originate from the interplay between photoexcitation-driven carrier dynamics and defect-mediated charge trapping processes. When the device is illuminated with light energy greater than the bandgap of the material, monolayer MoS₂ efficiently absorbs the photons, resulting in the generation of electron–hole pairs. These photogenerated carriers are subsequently separated and transported across the channel under the influence of the externally applied bias and the built-in potential at the metal–semiconductor interface. The carrier drift and extraction contribute to an instantaneous rise in photocurrent, corresponding to the conventional photoconductive response that governs the photodetection behavior of the device.

To elucidate the origin of the long-lived and history-dependent photoresponse, we carried out density functional theory (DFT) simulations. These calculations specifically examine the influence of V_s defects that are intrinsically introduced during high-temperature CVD growth. The V_s defects critically influence the synaptic response by modifying the carrier dynamics under optical excitation. The presence of sulfur vacancies was confirmed using a range of spectroscopic characterization techniques, including PL, XPS, and EDS, as seen in the previous sections. The electronic band structure of pristine monolayer MoS₂, as shown in Fig. 9(a), reveals a direct bandgap at the K point, where the valence band maximum (VBM) and conduction band minimum (CBM) coincide with a band gap value of 1.68 eV.⁷⁸ However, the introduction of V_s defects (8%) alters this band structure, leading to the formation of localized in-gap states within the forbidden energy region, as illustrated in Fig. 9(b). The density of states (DOS) analysis further validates this observation, showing defect-induced in-gap states in the defect-rich MoS₂ sample (Fig. 9(c)). Additionally, a slight reduction in the bandgap is observed, which aligns with previous reports on defect-engineered MoS₂.⁷⁹ Thus, the DFT simulations reveal that the V_s defects in monolayer MoS₂ introduce localized electronic states within the bandgap. Additional calculations for lower vacancy concentrations (2%) yield similar results (Fig. S27), confirming the intrinsic robustness of these defect-induced electronic states. These defect-associated in-gap states act as effective carrier-trapping centers that profoundly modulate carrier relaxation dynamics under optical excitation (schematically illustrated in Fig. 9d). Upon photoexcitation, a fraction of the photogenerated electrons becomes localized at V_s-related trap sites instead of recombining directly with holes in the valence band. The trapped carriers modify the local electrostatic potential and inhibit immediate recombination, thereby sustaining an elevated channel conductance even after the optical excitation is removed. The gradual thermal or field-assisted release of these carriers gives rise to the PPC, a characteristic signature of defect-mediated carrier retention in 2D semiconductors. This defect-mediated trapping and de-trapping process forms the physical basis for the synaptic behavior observed in the MoS₂ device. Each optical pulse acts as an excitatory stimulus, analogous to a presynaptic spike in a biological synapse, where the immediate conductance increase corresponds to an excitatory postsynaptic current (EPSC). Repeated or closely spaced optical pulses result in partial accumulation of trapped charge carriers, thereby increasing the overall channel conductance and mimicking the process of short-term memory or paired-pulse enhancement. As the pulse frequency or duration increases, a larger fraction of defect states becomes filled, and the conductance change persists for longer durations, reproducing long-term memory behavior. The gradual recovery of conductance upon removal of the optical input is governed by the de-trapping time constants, which determine the transition between STM and LTM states.

Fig. 9 DFT modeling of pristine and defect-rich 1L MoS₂. (a) Calculated electronic band structure of monolayer MoS₂, exhibiting a direct bandgap nature. (b) Electronic band structure of monolayer MoS₂ with sulfur vacancy (V_s) defects (8%), highlighting the defect-induced modifications in the band structure. (c) Density of states (DOS) comparison between pristine and defect-rich MoS₂, the emergence of in-gap states within the forbidden energy gap in the defect-rich sample. (d) Schematic representation of the energy band diagram, depicting carrier dynamics under illumination and dark conditions.

It is interesting to note that despite the presence of defects, the photodetection performance is considerable in the 1L MoS₂ sample. The observation of a high photocurrent implies that charge separation is more efficient than the trapping and subsequent release of carriers, which mainly determines the photocurrent decay timescale. The latter determines the synaptic behaviour. Note that the photocurrent measurements are performed under a 5 V bias, while the synaptic measurements are performed at a lower voltage (typically 0.5 V). Due to the high bias voltage and associated high electric field, carrier separation is more efficient across the device, resulting in a high photocurrent (ON state) in the device. Thus, the device seamlessly integrates photodetection and optical synaptic operations within a single monolayer MoS₂ channel, wherein the balance between immediate carrier extraction and delayed defect-assisted recombination dictates the evolution of conductance states under optical stimulation.

Conclusions

We have demonstrated an optoelectronic synapse with a defect-engineered monolayer MoS₂ as the active sensing material, integrated with printed Ag electrodes. This innovative hybrid approach combines the superior photoelectric conversion efficiency of monolayer MoS₂ with a cost-effective, scalable fabrication technique. The device shows exceptional photodetection performance with a responsivity of 10.3 A W⁻¹ and a high detectivity of 1.6 × 10¹² Jones. The intrinsic defects in the monolayer MoS₂ introduce localized trap states within the forbidden energy gap, giving rise to persistent photoconductivity, and facilitate synaptic functionalities. The modulation of synaptic characteristics is systematically achieved by varying key optical parameters including the pulse frequency, pulse number, pulse duration, and intensity of light, enabling the demonstration of human learning and forgetting behavior. Notably, the OES device exhibits exceptional energy efficiency with an operational electrical energy requirement as low as 1.2 fJ per synaptic event. We implemented a single-layer ANN framework using our experimental data to demonstrate its potential for neuromorphic computing, achieving 87.1% accuracy in recognizing MNIST handwritten digits. This demonstrates the capability of our OES device to function as a high-performance synaptic element for next-generation neuromorphic computing. Our work highlights a new paradigm that integrates the unique properties of two-dimensional materials with a simple, low-cost, and environmentally sustainable printing process. The demonstrated high-performance synaptic behavior, combined with the flexibility and scalability of the fabrication technique, paves the way for the development of advanced optoelectronic devices. This approach can readily be extended to other 2D material systems, enabling broad applicability in neuromorphic computing and beyond.

Author contributions

SD: methodology, investigation, data curation, formal analysis, and writing – original draft; AKM: methodology; MM: formal analysis and writing – review & editing; PKG: project administration, funding acquisition, resources, supervision, and writing – review & editing.

Conflicts of interest

The authors declare no competing financial interest.

Data availability

The data supporting this article have been included as part of the supplementary information (SI). Supplementary information is available. See DOI: https://doi.org/10.1039/d5nr03748d.

Acknowledgements

The authors acknowledge the Central Instruments Facility (CIF) and the Centre for Nanotechnology, IIT Guwahati, for providing various characterization tools. Financial support from MEITY (Grant No. 5(1)/2022-NANO) is acknowledged for carrying out a part of this work.

References

1. X. Zou, S. Xu, X. Chen, L. Yan and Y. Han, Sci. China: Inf. Sci., 2021, 64, 160404.
2. A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser and I. Polosukhin, Advances in neural information processing systems 30, ed. I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan and R. Garnett, 2017.
3. Y. LeCun, Y. Bengio and G. Hinton, Nature, 2015, 521, 436–444.
4. M. M. Najafabadi, F. Villanustre, T. M. Khoshgoftaar, N. Seliya, R. Wald and E. Muharemagic, J. Big Data, 2015, 2, 1.
5. A. Sebastian, M. Le Gallo, R. Khaddam-Aljameh and E. Eleftheriou, Nat. Nanotechnol., 2020, 15, 529–544.
6. J. D. Kendall and S. Kumar, Appl. Phys. Rev., 2020, 7, 011305.
7. D. S. Jeong, K. M. Kim, S. Kim, B. J. Choi and C. S. Hwang, Adv. Electron. Mater., 2016, 2, 1600090.
8. G. Wang, R. Wang, W. Kong and J. Zhang, Cogn. Neurodyn., 2018, 12, 615–624.
9. S. Gao, G. Liu, H. Yang, C. Hu, Q. Chen, G. Gong, W. Xue, X. Yi, J. Shang and R.-W. Li, ACS Nano, 2019, 13, 2634–2642.
10. Y. Wang, L. Yin, W. Huang, Y. Li, S. Huang, Y. Zhu, D. Yang and X. Pi, Adv. Intell. Syst., 2021, 3, 2000099.
11. N. Ilyas, J. Wang, C. Li, D. Li, H. Fu, D. Gu, X. Jiang, F. Liu, Y. Jiang and W. Li, Adv. Funct. Mater., 2022, 32, 2110976.
12. X. Chen, B. Chen, B. Jiang, T. Gao, G. Shang, S.-T. Han, C.-C. Kuo, V. A. L. Roy and Y. Zhou, Adv. Funct. Mater., 2023, 33, 2208807.
13. D. Hao, Z. Yang, J. Huang and F. Shan, Adv. Funct. Mater., 2023, 33, 2211467.
14. N. Li, S. Zhang, Y. Peng, X. Li, Y. Zhang, C. He and G. Zhang, Adv. Funct. Mater., 2023, 33, 2305589.
15. L. Gu, S. Poddar, Y. Lin, Z. Long, D. Zhang, Q. Zhang, L. Shu, X. Qiu, M. Kam, A. Javey and Z. Fan, Nature, 2020, 581, 278–282.
16. L. Mennel, J. Symonowicz, S. Wachter, D. K. Polyushkin, A. J. Molina-Mendoza and T. Mueller, Nature, 2020, 579, 62–66.
17. J. Chen, Z. Zhou, B. J. Kim, Y. Zhou, Z. Wang, T. Wan, J. Yan, J. Kang, J.-H. Ahn and Y. Chai, Nat. Nanotechnol., 2023, 18, 882–888.
18. L. Chen, M. Ren, J. Zhou, X. Zhou, F. Liu, J. Di, P. Xue, C. Li, Q. Li, Y. Li, L. Wei and Q. Zhang, Proc. Natl. Acad. Sci. U. S. A., 2024, 121, e2407971121.
19. K. Li, X. Wang, Y. Wu, W. Deng, J. Li, J. Li, Y. Zhao, Z. Chen, D. Yang, S. Yu and Y. Zhang, InfoMat, 2025, 7, e70007.
20. C. Liu, H. Chen, S. Wang, Q. Liu, Y.-G. Jiang, D. W. Zhang, M. Liu and P. Zhou, Nat. Nanotechnol., 2020, 15, 545–557.
21. S. Das, A. Sebastian, E. Pop, C. J. McClellan, A. D. Franklin, T. Grasser, T. Knobloch, Y. Illarionov, A. V. Penumatcha, J. Appenzeller, Z. Chen, W. Zhu, I. Asselberghs, L.-J. Li, U. E. Avci, N. Bhat, T. D. Anthopoulos and R. Singh, Nat. Electron., 2021, 4, 786–799.
22. J. Cheng, C. Wang, X. Zou and L. Liao, Adv. Opt. Mater., 2019, 7, 1800441.
23. M. Long, P. Wang, H. Fang and W. Hu, Adv. Funct. Mater., 2019, 29, 1803807.
24. J. Zha, M. Luo, M. Ye, T. Ahmed, X. Yu, D.-H. Lien, Q. He, D. Lei, J. C. Ho, J. Bullock, K. B. Crozier and C. Tan, Adv. Funct. Mater., 2022, 32, 2111970.
25. M. Abdullah, M. Younis, M. T. Sohail, S. Wu, X. Zhang, K. Khan, M. Asif and P. Yan, Small, 2024, 20, 2402668.
26. T. Ahmed, S. Kuriakose, E. L. H. Mayes, R. Ramanathan, V. Bansal, M. Bhaskaran, S. Sriram and S. Walia, Small, 2019, 15, 1900966.
27. M. M. Islam, D. Dev, A. Krishnaprasad, L. Tetard and T. Roy, Sci. Rep., 2020, 10, 21870.
28. Z.-D. Luo, X. Xia, M.-M. Yang, N. R. Wilson, A. Gruverman and M. Alexe, ACS Nano, 2020, 14, 746–754.
29. S. Seo, B.-S. Kang, J.-J. Lee, H.-J. Ryu, S. Kim, H. Kim, S. Oh, J. Shim, K. Heo, S. Oh and J.-H. Park, Nat. Commun., 2020, 11, 3936.
30. M. M. Islam, A. Krishnaprasad, D. Dev, R. Martinez-Martinez, V. Okonkwo, B. Wu, S. S. Han, T.-S. Bae, H.-S. Chung, J. Touma, Y. Jung and T. Roy, ACS Nano, 2022, 16, 10188–10198.
31. L. Dong, S. Yuan, G. Wei, P. Zhu, S. Ma, B. Xu and Y. Yang, Small, 2024, 20, 2306998.
32. H. Ma, H. Fang, X. Xie, Y. Liu, H. Tian and Y. Chai, Nano-Micro Lett., 2024, 16, 104.
33. M. Telkhozhayeva and O. Girshevitz, Adv. Funct. Mater., 2024, 34, 2404615.
34. L. Gao, Q. Liao, X. Zhang, X. Liu, L. Gu, B. Liu, J. Du, Y. Ou, J. Xiao, Z. Kang, Z. Zhang and Y. Zhang, Adv. Mater., 2020, 32, 1906646.
35. S. Sovizi, S. Angizi, S. A. Ahmad Alem, R. Goodarzi, M. R. R. Taji Boyuk, H. Ghanbari, R. Szoszkiewicz, A. Simchi and P. Kruse, Chem. Rev., 2023, 123, 13869–13951.
36. I. H. Abidi, S. P. Giridhar, J. O. Tollerud, J. Limb, M. Waqar, A. Mazumder, E. L. Mayes, B. J. Murdoch, C. Xu, A. Bhoriya, A. Ranjan, T. Ahmed, Y. Li, J. A. Davis, C. L. Bentley, S. P. Russo, E. D. Gaspera and S. Walia, Adv. Funct. Mater., 2024, 34, 2402402.
37. S. Nandi, S. Ghosal, M. Meyyappan and P. K. Giri, Mater. Horiz., 2025, 12, 4274–4288.
38. W. Zhou, X. Zou, S. Najmaei, Z. Liu, Y. Shi, J. Kong, J. Lou, P. M. Ajayan, B. I. Yakobson and J.-C. Idrobo, Nano Lett., 2013, 13, 2615–2622.
39. K. Wu, Z. Li, J. Tang, X. Lv, H. Wang, R. Luo, P. Liu, L. Qian, S. Zhang and S. Yuan, Nano Res., 2018, 11, 4123–4132.
40. F. Liao, Z. Zhou, B. J. Kim, J. Chen, J. Wang, T. Wan, Y. Zhou, A. T. Hoang, C. Wang, J. Kang, J.-H. Ahn and Y. Chai, Nat. Electron., 2022, 5, 84–91.
41. K. Liu, T. Zhang, B. Dang, L. Bao, L. Xu, C. Cheng, Z. Yang, R. Huang and Y. Yang, Nat. Electron., 2022, 5, 761–773.
42. C.-Y. Wang, S.-J. Liang, S. Wang, P. Wang, Z. Li, Z. Wang, A. Gao, C. Pan, C. Liu, J. Liu, H. Yang, X. Liu, W. Song, C. Wang, B. Cheng, X. Wang, K. Chen, Z. Wang, K. Watanabe, T. Taniguchi, J. J. Yang and F. Miao, Sci. Adv., 2020, 6, eaba6173.
43. S. Conti, G. Calabrese, K. Parvez, L. Pimpolari, F. Pieri, G. Iannaccone, C. Casiraghi and G. Fiori, Nat. Rev. Mater., 2023, 8, 651–667.
44. X. Chen, X. Wang, Y. Pang, G. Bao, J. Jiang, P. Yang, Y. Chen, T. Rao and W. Liao, Small Methods, 2023, 7, 2201156.
45. Y. Dai, Q. He, Y. Huang, X. Duan and Z. Lin, Chem. Rev., 2024, 124, 5795–5845.
46. Z. Peng, A. Grillo, A. Pelella, X. Liu, M. Boyes, X. Xiao, M. Zhao, J. Wang, Z. Hu, A. D. Bartolomeo and C. Casiraghi, Mater. Horiz., 2024, 11, 1344–1353.
47. R. Worsley, L. Pimpolari, D. McManus, N. Ge, R. Ionescu, J. A. Wittkopf, A. Alieva, G. Basso, M. Macucci, G. Iannaccone, K. S. Novoselov, H. Holder, G. Fiori and C. Casiraghi, ACS Nano, 2019, 13, 54–60.
48. S. Debnath, K. Ghosh, M. Meyyappan and P. K. Giri, Nanoscale, 2023, 15, 13809–13821.
49. J.-W. T. Seo, J. Zhu, V. K. Sangwan, E. B. Secor, S. G. Wallace and M. C. Hersam, ACS Appl. Mater. Interfaces, 2019, 11, 5675–5681.
50. S. Debnath, M. Meyyappan and P. K. Giri, ACS Appl. Mater. Interfaces, 2024, 16, 9039–9050.
51. L. Shao, H. Wang, Y. Yang, Y. He, Y. Tang, H. Fang, J. Zhao, H. Xiao, K. Liang, M. Wei, W. Xu, M. Luo, Q. Wan, W. Hu, T. Gao and Z. Cui, ACS Appl. Mater. Interfaces, 2019, 11, 12161–12169.
52. K. Liang, R. Wang, B. Huo, H. Ren, D. Li, Y. Wang, Y. Tang, Y. Chen, C. Song, F. Li, B. Ji, H. Wang and B. Zhu, ACS Nano, 2022, 16, 8651–8661.
53. J. Shi, J. Jie, W. Deng, G. Luo, X. Fang, Y. Xiao, Y. Zhang, X. Zhang and X. Zhang, Adv. Mater., 2022, 34, 2200380.
54. A. Kaium Mia, S. Aswal, S. Debnath, V. Vretenar and P. K. Giri, J. Mater. Chem. C, 2025, 13, 22821–22831.
55. S. Debnath, S. Ghosal, M. Meyyappan and P. K. Giri, Flexible Printed Electron., 2024, 9, 035005.
56. D. Sahu, S. Debnath, S. Ghosal and P. K. Giri, ACS Appl. Mater. Interfaces, 2024, 16, 49544–49555.
57. S. Debnath, S. Dey and P. K. Giri, ACS Appl. Mater. Interfaces, 2025, 17, 8219–8230.
58. P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, A. Dal Corso, S. de Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A. P. Seitsonen, A. Smogunov, P. Umari and R. M. Wentzcovitch, J. Phys.: Condens. Matter, 2009, 21, 395502.
59. P. Giannozzi, O. Andreussi, T. Brumme, O. Bunau, M. Buongiorno Nardelli, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, M. Cococcioni, N. Colonna, I. Carnimeo, A. Dal Corso, S. de Gironcoli, P. Delugas, R. A. DiStasio, A. Ferretti, A. Floris, G. Fratesi, G. Fugallo, R. Gebauer, U. Gerstmann, F. Giustino, T. Gorni, J. Jia, M. Kawamura, H.-Y. Ko, A. Kokalj, E. Küçükbenli, M. Lazzeri, M. Marsili, N. Marzari, F. Mauri, N. L. Nguyen, H.-V. Nguyen, A. Otero-de-la-Roza, L. Paulatto, S. Poncé, D. Rocca, R. Sabatini, B. Santra, M. Schlipf, A. P. Seitsonen, A. Smogunov, I. Timrov, T. Thonhauser, P. Umari, N. Vast, X. Wu and S. Baroni, J. Phys.: Condens. Matter, 2017, 29, 465901.
60. J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett., 1996, 77, 3865–3868.
61. H. J. Monkhorst and J. D. Pack, Phys. Rev. B, 1976, 13, 5188–5192.
62. H. Li, Q. Zhang, C. C. R. Yap, B. K. Tay, T. H. T. Edwin, A. Olivier and D. Baillargeat, Adv. Funct. Mater., 2012, 22, 1385–1390.
63. C. Lee, H. Yan, L. E. Brus, T. F. Heinz, J. Hone and S. Ryu, ACS Nano, 2010, 4, 2695–2700.
64. R. Dagan, Y. Vaknin, A. Henning, J. Y. Shang, L. J. Lauhon and Y. Rosenwaks, Appl. Phys. Lett., 2019, 114, 101602.
65. M. D. Siao, W. C. Shen, R. S. Chen, Z. W. Chang, M. C. Shih, Y. P. Chiu and C.-M. Cheng, Nat. Commun., 2018, 9, 1442.
66. P.-C. Shen, Y. Lin, C. Su, C. McGahan, A.-Y. Lu, X. Ji, X. Wang, H. Wang, N. Mao, Y. Guo, J.-H. Park, Y. Wang, W. Tisdale, J. Li, X. Ling, K. E. Aidala, T. Palacios and J. Kong, Nat. Electron., 2022, 5, 28–36.
67. Y. Zhu, J. Lim, Z. Zhang, Y. Wang, S. Sarkar, H. Ramsden, Y. Li, H. Yan, D. Phuyal, N. Gauriot, A. Rao, R. L. Z. Hoye, G. Eda and M. Chhowalla, ACS Nano, 2023, 17, 13545–13553.
68. A. George, M. V. Fistul, M. Gruenewald, D. Kaiser, T. Lehnert, R. Mupparapu, C. Neumann, U. Hübner, M. Schaal, N. Masurkar, L. M. R. Arava, I. Staude, U. Kaiser, T. Fritz and A. Turchanin, npj 2D Mater. Appl., 2021, 5, 15.
69. M. Tarik Hossain, T. Jena, S. Debnath and P. K. Giri, J. Mater. Chem. C, 2023, 11, 6670–6684.
70. A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C.-Y. Chim, G. Galli and F. Wang, Nano Lett., 2010, 10, 1271–1275.
71. S. Mouri, Y. Miyauchi and K. Matsuda, Nano Lett., 2013, 13, 5944–5948.
72. Y. Zhu, J. Lim, Z. Zhang, Y. Wang, S. Sarkar, H. Ramsden, Y. Li, H. Yan, D. Phuyal, N. Gauriot, A. Rao, R. L. Z. Hoye, G. Eda and M. Chhowalla, ACS Nano, 2023, 17, 13545–13553.
73. X. Chen, Y. J. Park, M. Kang, S.-K. Kang, J. Koo, S. M. Shinde, J. Shin, S. Jeon, G. Park, Y. Yan, M. R. MacEwan, W. Z. Ray, K.-M. Lee, J. A. Rogers and J.-H. Ahn, Nat. Commun., 2018, 9, 1690.
74. A. George, M. V. Fistul, M. Gruenewald, D. Kaiser, T. Lehnert, R. Mupparapu, C. Neumann, U. Hübner, M. Schaal, N. Masurkar, L. M. R. Arava, I. Staude, U. Kaiser, T. Fritz and A. Turchanin, npj 2D Mater. Appl., 2021, 5, 1–8.
75. Y. Lecun, L. Bottou, Y. Bengio and P. Haffner, Proc. IEEE, 1998, 86, 2278–2324.
76. L. Deng, IEEE Signal Process. Mag., 2012, 29, 141–142.
77. J. Sun, S. Oh, Y. Choi, S. Seo, M. J. Oh, M. Lee, W. B. Lee, P. J. Yoo, J. H. Cho and J.-H. Park, Adv. Funct. Mater., 2018, 28, 1804397.
78. H.-P. Komsa and A. V. Krasheninnikov, Phys. Rev. B:Condens. Matter Mater. Phys., 2015, 91, 125304.
79. K. S. Kumar, A. K. Dash, H. Hasna Sabreen, M. Verma, V. Kumar, K. Watanabe, T. Taniguchi, G. S. Gautam and A. Singh, 2D Mater., 2024, 11, 045003.

---