Multilevel storage and linear optoelectronic response in mixed-dimensional photomemories

Abstract

The rapid evolution of artificial intelligence (AI) computing demands innovative memory technologies that integrate high-speed processing with energy-efficient data storage. Here, we report a mixed-dimensional photomemory device based on a CsPbBr₃/Al₂O₃/MoS₂ architecture, leveraging perovskite quantum dots (PQDs) as a photoactive floating-gate layer, a tunable Al₂O₃ dielectric, and a 2D MoS₂ channel. Optical and electrical characterization studies, including steady-state and time-resolved photoluminescence (PL), Kelvin probe force microscopy (KPFM), and current–voltage measurements, reveal the interplay of dielectric thickness and interfacial effects in governing charge transfer efficiency. By optimizing the Al₂O₃ thickness to 5.5 nm, we achieve precise control over charge transfer dynamics, enabling an optimal charge transfer rate with minimal optical energy (∼sub-pJ) to store a single positive charge in the PQDs. The device exhibits exceptional optoelectronic performance, including a nearly linear correlation between incident photon number and average photocurrent (I_ph(avg)) over two orders of magnitude, multilevel storage capability, and a memory window with a high on/off ratio. These findings establish a robust platform for next-generation perovskite-based photomemories, offering insights into energy-efficient, high-performance optoelectronic systems for advanced AI chip applications.

---

New concepts

A 5.5 nm Al₂O₃ dielectric layer optimizes charge transfer efficiency in the CsPbBr₃/Al₂O₃/MoS₂ photomemory, minimizing the optical energy to ∼0.228 pJ per stored hole. The photomemory achieves a near-linear photocurrent response to incident photons over two orders of magnitude, enhancing data precision. Combining illumination time and back-gate voltage enables 4-bit multilevel storage, increasing data density in perovskite-based photomemories. Interfacial dipole layers in ultrathin Al₂O₃ reduce charge transfer efficiency by altering band alignment, revealing a critical dielectric interface effect. Type-II band alignment in CsPbBr₃/MoS₂ facilitates rapid electron transfer and hole retention, enabling fast and efficient photomemory operation.

1. Introduction

With the exponential growth of artificial intelligence (AI) computing technologies, the unprecedented demand for high-speed telecommunications and energy-efficient data storage is increasingly challenging the limits of conventional electronic processors and memory architectures.¹ In particular, highly integrated photonic–electronic circuit chip systems leverage light as both a computational medium and communication signal, offering a promising path toward enhanced processing speed and reduced power consumption.^1,2 Notably, in conventional electronic chips, floating-gate memory devices—which employ an electrically isolated charge storage layer to retain charge over extended periods—have long been the benchmark for non-volatile data storage due to their excellent charge retention properties.³ Building on the intrinsic advantages of the floating-gate configuration, photomemory devices enable the transduction of optical communication signals into electrically stored non-volatile data, thereby establishing a direct interface between optical and electronic domains. This approach significantly enhances programming capabilities by replacing traditional electrical programming with photonic and optical communication techniques.^4,5 As a result, extensive research efforts have been dedicated to exploring photomemory functionalities in metal oxides, two-dimensional (2D) materials, and perovskite-based material systems.^4–7 Recent studies on MoS₂ have expanded its potential in both electronic and optoelectronic applications. Researchers have investigated the role of asymmetric Schottky contacts in MoS₂ field-effect transistors, proposing a back-to-back Schottky barrier model to elucidate the rectifying behavior and underscoring the device's high photoresponsivity.⁸ Studies have shown that memory performance, including hysteresis, can be modulated by controlling air pressure. Furthermore, leveraging the combination of a gate voltage and light illumination has been demonstrated to substantially enhance the memory window, paving the way for multi-level data storage.⁹

Among the various photomemory technologies reported in recent years,^10,11 perovskite-based materials exhibit significant optoelectronic properties, including a low exciton binding energy, excellent photo-absorbing ability, and feasible solution processing, making them ideal as photon absorption layers in photomemory applications.^12–14 In particular, PQDs exhibit a significant quantum confinement effect and a tunable band gap energy, making them highly suitable for the light absorption layer in optoelectronic memory devices.^15,16 Chang et al.¹⁷ precisely controlled the PQD/poly(3-hexylthiophene-2,5-diyl) interfacial area, enabling the construction of an efficient photoactive floating gate within a polystyrene-block-poly(ethylene oxide)(PS-b-PEO) composite film. This significantly enhanced the charge transfer rate and programming speed of the photomemory.¹⁷ Moreover, Chao et al.¹⁸ conducted in operando PL measurements, revealing a direct correlation between PQD crystal orientation of quasi-2D perovskites and photomemory performance. This approach yielded a device with 7-bit storage capacity and a rapid charge transfer rate.¹⁸ Among the various technologies, such hybrid nonvolatile photomemory has attracted considerable scholarly attention in recent years due to its exceptional photoelectric characteristics and ease of its use.^14,19–22

Nonetheless, several technical and fabrication challenges have arisen in the practical application of PQD-based photomemory, particularly regarding the aggregation of PQDs within the hybrid organic layer and its uneven vertical distribution near the channel, which can diminish the quantum confinement effect and create leakage pathways.²³ Such inefficient energy consumption during optical-to-electrical data conversion and undesirable data loss during both programming and reading operations in hybrid PQD-based photomemories limit their feasibility for next-generation photonic–electronic chip systems.²³ To advance optoelectronic performance, efforts to improve the energy efficiency of photomemory operation and expand programming functionalities in PQD-based photomemory devices are urgently needed for next-generation AI chip system applications. Moreover, mixed-dimensional devices have emerged as a critical hardware platform for integrating synaptic functions directly into electronic devices, thereby offering a path to mitigate the von Neumann bottleneck in artificial intelligence applications.^24–26 Within this field, optoelectronic synaptic devices are poised to confer a significant advantage over their electrically stimulated counterparts by providing higher operational speeds and reduced crosstalk.^27,28

In this study, we present a photomemory featuring a mixed-dimensional architecture that integrates CsPbBr₃ PQDs with 2D MoS₂, where an Al₂O₃ dielectric layer is strategically positioned in between the floating layer and the channel. By tuning the thickness of the Al₂O₃ layer, we precisely modulate the charge transfer dynamics of the photoexcited excitons in the CsPbBr₃/Al₂O₃/MoS₂ system, enhancing the optoelectronic performance of the photomemory. We proposed that charge transfer dynamics of photo-induced excitons is proposed to be predominantly governed by the interplay between the band structure modulation and the charge tunneling process, which is determined by the dielectric thickness. In the ultrathin regime, the positive dipole generates an intense electric field that forms considerable negative carriers within the MoS₂. This charge accumulation critically elevates the effective potential barrier at the CsPbBr₃/oxide junction, thereby modulating the device's charge transfer properties. As a result of this optimized charge transfer rate, a low optical energy of approximately sub-pJ is sufficient to generate a single positively effective charge stored within the PQDs Additionally, favorable electrical characteristics and nearly linear correlation between the total number of incident photons and the average photocurrent of PQD-based photomemories were discussed.

2. Results and discussion

2.1. Structural and material characterization of mixed-dimensional photomemories

In order to assess the performance of the mixed-dimensional phototransistor memory, back-gate devices with a floating layer of PQDs were meticulously manufactured, as demonstrated in Fig. 1(a) and (c). The design of devices incorporates a highly doped silicon (Si) wafer serving as the back-gate electrode, which is subsequently covered with a 100 nm silicon oxide (SiO₂) layer that functions as the gate dielectric. Double-layer MoS₂ materials were utilized as photomemory channels, whereas an Al₂O₃ dielectric layer acts as a tunneling oxide layer. Uniformly distributed CsPbBr₃ PQDs, synthesized through an efficient one-pot hot-injection methodology, serve as the photo-absorption layer as well as charge storage layer, constituting a vital component for the photomemory devices. The structural characteristics of the photomemory devices were systematically examined through a series of analytical material techniques, including transmission electron microscopy (TEM and high-resolution TEM (HRTEM)), energy-dispersive spectroscopy (EDS) elemental mapping, and X-ray photoelectron spectroscopy (XPS). The fabricated device is investigated using cross-sectional TEM as illustrated in Fig. 1(b), revealing the distinct CsPbBr₃ PQDs and the dielectric layer with a thickness of approximately 7 nm and 5.5 nm, respectively. Furthermore, the EDS mappings demonstrated the presence of cesium (Cs, lead (Pb), and bromine (Br)) elements arisen from the CsPbBr₃ PQD floating layer situated above the dielectric oxide, as presented in Fig. S1a–f.

Fig. 1 (a) Schematic illustration of the mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ device. (b) Cross-sectional transmission electron microscopy (TEM) image of the photomemory. (c) Top-view optical microscopy (OM) image of the photomemory. (d) TEM image of the CsPbBr₃ layer serving as the top-floating charge storage layer.

As depicted in Fig. 1(d), the PQDs synthesized through the hot-injection process exhibit exceptional crystallinity and homogeneous dispersion. The analysis of size distribution revealed an average diameter of 8 ± 2 nm. Furthermore, the inserted figure demonstrates that the HRTEM image of the PQDs reveals a lattice spacing of 0.32 nm, corresponding precisely to the (200) crystallographic plane of CsPbBr₃. In addition, three distinct electron diffraction rings in the SAED pattern of PQDs as displayed in Fig. S2 can be ascribed to the (100), (110), and (020) crystal planes, arranged in order from the innermost to the outermost rings. High-resolution XPS analyses are conducted to examine the chemical states of CsPbBr₃, as illustrated in Fig. S3a–c. The spectral analysis reveals the presence of well-defined distinct peaks corresponding to Cs, Pb, and Br within the PQDs, thereby affirming the integrity of their chemical composition. Importantly, we have realized that the oxide thickness of the dielectric layer (t_ox) is a critical parameter affecting the charge transfer dynamics of photo-excited excitons in PQDs, thereby determining the performance of photomemories. Accordingly, the precise thickness of the thin Al₂O₃ dielectric layer is quantified at the resolution of sub-nano level through atomic force microscopy (AFM). The height profile presented in Fig. S4c indicates an Al₂O₃ dielectric layer possessing the thickness of approximately 5.5 nm, which is consistent with the result of the cross-sectional TEM image (Fig. 1(c)). Furthermore, AFM measurements of the Al₂O₃ dielectric layers with different thicknesses are comprehensively documented in Fig. S4a–f, revealing that the Al₂O₃ films with a continuous and smooth surface can be achieved at any thickness. Interestingly, a significant elevation in the height profile can be observed at the peripheries of the Al₂O₃ layer, regardless of the thickness variations. We speculate that this may result from incomplete reactions occurring during the development process. To further investigate the characteristics of the channel material, Raman spectroscopy was employed to validate the successful synthesis of MoS₂, as shown in Fig. S5. The Raman spectrum, acquired using excitation at 532 nm in an ambient environment, displays two distinguishing vibrational modes: E¹_2g mode and A_1g mode at 383 and 406 cm⁻¹, respectively. These vibrational modes are indicative of the presence of the MoS₂ layer.²⁹ Additionally, high-resolution XPS spectra of MoS₂, depicted in Fig. S3f–g, exhibit two prominent peaks at binding energies of 229.2 eV and 232.3 eV, corresponding to Mo⁴⁺3d_5/2 and Mo⁴⁺3d_3/2, respectively. The peaks observed at 162.0 eV and 163.5 eV can be attributed to S²⁻2p_3/2 and S²⁻2p_1/2, respectively, further confirming the chemical composition and oxidation states of MoS₂. These comprehensive characterization studies provide insights into the structural and material properties of the mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ system, a crucial structure of the photomemory architecture.

2.2. Optical properties and charge transfer dynamics analysis

To further elucidate the optoelectronic dynamics of the mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ system, which predominantly influence the performance of the phototransistor memory, a series of optical characterization studies were conducted. First, in order to gain insight into the fundamental optical properties of our mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ photomemory system, the optical analysis using UV-visible absorption spectra and steady-state photoluminescence (PL) spectra was performed. As depicted in Fig. S6a, the CsPbBr₃ PQDs exhibited a broad absorption spectrum with a band edge approximately at 525 nm, highlighting its potential uses as an efficient photoactive material under visible blue and green light. Moreover, a pronounced emission peak with a narrow full width at half maximum (FWHM) indicates a bandgap of approximately 2.36 eV for the PQDs. MoS₂ also exhibits optical properties, with a PL peak at 666 nm, as shown in Fig. S6b.

To shed light on the role of the dielectric layer thickness in governing the charge transfer dynamics in the mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ structures, a series of photomemory devices with varying Al₂O₃ dielectric thickness t_ox were fabricated accordingly. The steady-state PL spectrum measurements were performed and the degree of PL quenching served as a quantitative measurement of charge transfer efficiency.^18,29 As shown in Fig. 2(a), a 5.5 nm Al₂O₃ dielectric layer demonstrates a remarkable 95% PL intensity quenching compared to pure CsPbBr₃, indicating optimal charge transfer efficiency. The charge storage capacity within the floating PQD layer is significantly modulated by both the charge transfer efficiency of photo-induced excitons and the thickness of the dielectric layer. Thus, precisely controlling the thickness of the Al₂O₃ dielectric layer is crucial for regulating the charge transfer process in the CsPbBr₃/Al₂O₃/MoS₂ system. For instance, when the Al₂O₃ dielectric thickness is reduced to 2.5 nm, the storage of photo-induced charge becomes more challenging. In other words, a thinner dielectric layer hampers the ability to isolate stored charges in the PQDs, thereby resulting in an inevitable decrease in PL quenching to 37%. Conversely, increasing the dielectric thickness beyond 5.5 nm impeded charge transfer through the tunneling mechanism, resulting in enhanced recombination within the PQDs. Our findings indicate that a 5.5 nm Al₂O₃ dielectric layer yields the lowest PL peak intensity, reflecting an optimal balance between charge recombination and charge transfer.

Fig. 2 (a) Steady-state photoluminescence (PL) spectra of CsPbBr₃ distributed on the surface of the Al₂O₃/MoS₂ structure with varying Al₂O₃ thicknesses under 365 nm excitation. (b) Exciton lifetimes of the photomemory device with different thicknesses of the Al₂O₃ dielectric layer. (c) Schematic illustration of the energy band alignment governed by the competing effects of barrier height and dielectric thickness modulation.

To further conduct a more in-depth analysis of the lifetime of photo-induced excitons and the charge transfer efficiency (CTE) between the CsPbBr₃ and MoS₂ materials in the mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ structure system, time-resolved PL spectroscopy was employed. Fig. S7 presents the fluorescence decay characteristics at a wavelength of 405 nm for CsPbBr₃ PQDs, comparing them with those of the CsPbBr₃/Al₂O₃/MoS₂ samples. The observation of a faster fluorescence decay for CsPbBr₃/Al₂O₃/MoS₂ structures than that for CsPbBr₃ PQDs indicates a more pronounced charge transfer process of the photo-induced excitons within the photomemory. The decay trend in the time-resolved PL spectra was modeled using a biexponential equation shown in eqn (1),

where τ₁ and τ₂ represent the long-lived and short-lived exciton lifetimes, respectively, while A₁ and A₂ denote the intensity ratios.^17,30 Therefore, the average lifetime, denoted as τ_average, can be calculated by using eqn (2), as described below

Subsequently, the charge transfer efficiency (CTE) was evaluated employing eqn (3).³¹

where τ_CsPbBr3 represents the τ_average of CsPbBr₃ and τ_{CsPbBr3/tox/Al2O3/MoS2} represents the τ_average of the CsPbBr₃/Al₂O₃/MoS₂ structure. All parameters are systematically summarized in Table S1. It is widely recognized that τ_ave serves as a reliable indicator for evaluating the efficiency of photo-excited charge transfer within the heterojunctions, where a shorter τ_ave represents a faster charge transfer process. As shown in Fig. 2(b), the average fluorescence decay time of 0.38 ns for the 5.5 nm Al₂O₃ case exhibited a significant reduction compared to the 1.49 ns for pure CsPbBr₃, corresponding to a CTE of 73%. The experimental results of TRPL and steady-state PL strongly highlighted that the CsPbBr₃/Al₂O₃/MoS₂ structure with a 5.5 nm Al₂O₃ dielectric layer achieves an optimal charge transfer rate in our mixed-dimensional material system.

It is well established that photo-induced charges can migrate more readily through thinner Al₂O₃ dielectric layers due to the increased tunneling probability, assuming only the effect of tunneling barrier thickness is considered. However, our observations reveal a paradoxical decrease in charge transfer efficiency when the Al₂O₃ dielectric thickness is reduced below 5.5 nm. We attribute this phenomenon primarily to the presence of an interfacial dipole layer generated by positively charged defects within the Al₂O₃ dielectric surface.³² This interfacial dipole layer, which has been observed in other oxide dielectric systems, effectively modulates the band edge of MoS₂.³³ Although such interfacial dipole layers are formed only near the dielectric interfaces, their electric field penetrates the entire dielectric under ultrathin conditions, significantly modulating the band alignment of n-type MoS₂ and the band diagram of the dielectric layer.^34,35 Accordingly, we propose that the charge transfer dynamics of photo-induced excitons is predominantly governed by the interplay between the band structure modulation, induced by interfacial dipole layers, and the charge tunneling process, which is determined by the dielectric thickness, as depicted in Fig. 2(c). When the Al₂O₃ dielectric thickness falls below a critical threshold thickness, a significant accumulation layer of electrons forms on the n-type MoS₂ side due to electrostatic induction from interfacial positive charges on both sides of ultrathin Al₂O₃ dielectric layers. Although the reduced thickness enhances tunneling probability, the resulting interfacial dipole layer strongly modulates the band structure, effectively increasing the effective barrier height (ΔE_B) for charge transfer, as shown in the left-hand side of Fig. 2(c). As the dielectric thickness increases, the charge transfer mechanism becomes predominantly governed by the tunneling process, as depicted in the right-hand side of Fig. 2(c).

To further explore the charge transfer dynamics of the photo-induced excitons in PQDs, we employed Kelvin probe force microscopy (KPFM), a widely used technique for probing surface potential, to investigate the surface potential characteristics of the CsPbBr₃ PQD layer under various wavelengths. As shown in Fig. S8a, the initial spatial potential distribution of the CsPbBr₃ PQD layer revealed an average surface potential of 370 mV. After exposure to excitation light with a wavelength of 980 nm for 60 seconds, the average surface potential increased slightly to 387 mV (Fig. S8b). This indicates that 980 nm light, which is not absorbed by CsPbBr₃ PQDs, results in a surface potential comparable to the initial state. Furthermore, as displayed in Fig. S8d, the exposure of the CsPbBr₃ PQD layer to 405 nm excitation light for 60 seconds led to an average surface potential of 504 mV, suggesting the accumulation of photo-induced charge within the floating PQDs. In addition, an average surface potential of 409 mV was measured after exposure to the 632 nm excitation light as illustrated in Fig. S8c. Compared to the initial state, the increase in surface potential at 632 nm may be attributed to charge leakage from MoS₂. Histogram distributions of the surface potential extracted from Fig. S8a–d are summarized in Fig. 3, highlighting a pronounced positive shift in surface potential under 405 nm excitation compared to the other cases. This result underscores the strong wavelength dependence of charge carrier behavior in our study.

Fig. 3 Histogram of KPFM surface potential distributions under different wavelength excitations.

2.3. Electrical performance of CsPbBr₃/Al₂O₃/MoS₂ photomemories

In addition to the optical measurement results, the fundamental memory characteristics were further evaluated through the current–voltage (I–V) measurements. These were performed by sweeping the back-gate voltage (V_GS) from −60 to 20 V while maintaining a constant drain voltage (V_DS) of 5 V, as shown in Fig. 4(a). Simultaneously, the electrical transfer characteristics of different 5.5 nm devices were measured, as illustrated in Fig. S9, confirming the reproducibility of the study. The threshold voltage shift (ΔV_th), a key parameter for assessing a memory device's data storage capacity, is typically defined as the voltage shift between different memory states. We extracted the V_th of the photomemories in their initial state and after photo-programming for various dielectric layer thickness t_ox using the linear extrapolation method.³⁶ After the photo-programing process with the blue light illumination (405 nm, 7.29 μW cm⁻², 180 s), a pronounced negative voltage shift in the electrical transfer characteristics of the photomemory was observed, resulting from additional holes stored in the floating PQDs. As depicted in Fig. 4(b), the photomemory device incorporating a 5.5 nm-thick Al₂O₃ dielectric layer exhibited optimal electrical performance, featuring a wider memory window and a higher on/off current ratio, compared to other dielectric layer thicknesses. Serving as a direct measure of stored charge, the memory window provides verification that a larger window under a consistent external energy density directly reflects a more efficient charge transfer rate.³⁷ Notably, the mixed-dimensional photomemory with a 5.5-nm Al₂O₃ dielectric layer demonstrated not only optimal memory characteristics in terms of the memory window and on/off ratio but also fast charge transfer rate of the photo-induced excitons. Furthermore, the effective hole carrier density (n_h) per unit area within the floating PQD layer can be roughly estimated using eqn (4).^4,38

n_h = C_G|ΔV_th|/q (4)

where the gate capacitance (C_G) is defined as the capacitance per unit area of the SiO₂ dielectric layer. The ΔV_th and q refer to the threshold voltage shift and the elementary charge, respectively. For devices utilizing a 100 nm-thick SiO₂ back-gate dielectric, the charge storage density can be calculated roughly to be as high as 2.57 × 10¹⁴ cm⁻². The energy density of excitation light for the photo-programming process can be calculated to be 1.31 mJ cm⁻² in the present study. Based on this measurement, the energy cost per stored hole was then calculated by dividing the energy density of excitation light by the charge storage density (2.57 × 10¹⁴ # cm⁻²), resulting in a value of 0.228 pJ per hole. Moreover, the performance of the device under a low power density of 7.29 μW cm⁻² is exceptionally low compared to those in many other recent studies on optoelectronic memory devices. These experimental observations highlight the low energy consumption of the photo-programming process, enabling energy-efficient data storage in our mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ photomemory.

Fig. 4 Measured optoelectronic characteristics of the mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ photomemory. (a) Electrical transfer characteristics of the photomemory programmed by blue light illumination (light: 405 nm, 7.29 μW cm⁻², 180 s) and subsequently erased by a back-gate voltage pulse at V_GS = 60 V for 2 s. (b) On/off ratio and threshold voltage shift (ΔV_th) measured in the dark and after light illumination (light: 405 nm, 7.29 μW cm⁻², 180 s) for devices with different Al₂O₃ dielectric thicknesses. (c) Real-time I_DS curves of the photomemory measured under different light illumination durations (1 s to 180 s) with an applied gate voltage of V_GS = −50 V (light: 405 nm, 7.29 μW cm⁻²). (d) Logarithmic plot of the I_ph(avg) as a function of the incident photon number under V_g = −50 V (light: 405 nm, 7.29 μW cm⁻²), obtained by averaging the measured photocurrent (I_ph) after illumination termination. (e) Multilevel behavior of the photomemory under varying illumination durations with V_g varying from −50 V to 10 V (laser: 405 nm, 7.29 μW cm⁻²). (f) Heatmap of the photoprogramming-to-initial-current ratio (I_ph/I_ini), illustrating the correlation between illumination time and V_GS values (light: 405 nm, 7.29 uW cm⁻², 60 s).

Previous experiments (Fig. 2 and 4(b)) indicate that the CsPbBr₃/Al₂O₃/MoS₂ photomemory with a 5.5 nm-thick Al₂O₃ dielectric layer exhibits optimal optical properties and superior memory characteristics compared to other dielectric layers with different thicknesses. Therefore, we concentrate on subsequent experimental investigations on the photomemories using this specific dielectric layer thickness. After the erasing process, in which a back-gate voltage V_GS of 60 V is aplied for 2 s, the electrical transfer curve nearly returns to its initial state. This recovery is likely due to the neutralization of trapped holes by electrons injected from the MoS₂ side.

To verify the memory characteristics, a systematic investigation was conducted to explore the photo-programming conditions by examining the photomemory device's response to multiple wavelengths and varying power densities. Photo-programming was performed utilizing lasers with wavelengths of 405, 520, 632, and 980 nm, as shown in Fig. S11a. Under 632 nm laser illumination, a photocurrent was observed, as its photon energy was sufficient to generate photo-induced excitons in MoS₂. However, no memory behavior was detected after the illumination was terminated. Nonvolatile characteristics were observed only when the laser energy exceeded the bandgap of CsPbBr₃, allowing charge storage even after light was removed. This finding is consistent with the KPFM results, further confirming the role of wavelength in memory behavior. Fig. S11b illustrates the effect of varying light intensities from 2.16 to 7.29 μW cm⁻² at λ = 405 nm. At insufficient intensities, data stability was compromised; however, a distinct memory characteristic emerges at an intensity of 7.29 μW cm⁻².

To further demonstrate the photo-programming behavior, the real-time response of the drain current (I_DS) was investigated under 405 nm light excitation (7.29 μW cm⁻²) at fixed voltages of V_DS = 5 V and V_GS = 0 V. As shown in Fig. S12, drain current I_DS was initially maintained at a relatively high current level (∼10⁻⁷ A), indicating the turn-on state of the device at V_GS = 0 V. Upon light exposure, the I_DS exhibited a sharp increase, reaching ∼10⁻⁵ A, before gradually decaying to ∼10⁻⁶ A after the illumination ceased. Notably, a photocurrent of ∼10⁻⁵ A was observed within just 1 second of light excitation, demonstrating the device's rapid photoresponse capability. Despite its strong data retention and reliable performance under various operating conditions, the device faces challenges in accurately distinguishing multilevel data states over different illumination periods. Over time, all data states tend to converge, likely due to the interfacial electric field generated by trapped holes in CsPbBr₃ and accumulated electrons in MoS₂.^39,40 This phenomenon reduces the data density of top-floating-gate memories, limiting their commercial viability. To address this issue, a back-gate voltage V_GS was applied at the bottom gate terminal. As shown in Fig. 4(c), under 405 nm illumination (7.29 μW cm⁻²), real-time response measurements of I_DS confirm the effective implementation of multilevel storage, enabled by electrical depression. Notably, applying a negative V_GS induces minimal transport current in n-type materials in accumulation mode, forming a depletion layer within the channel. This, in turn, generates a positively charged layer in MoS₂, which attracts photo-generated electrons in CsPbBr₃. Simultaneously, photo-induced holes become localized within the floating PQD layer, further facilitating long-term data retention and enabling multilevel storage through electrical depression. To investigate the relationship between data storage density and the amplitude of V_GS = 0 V, the temporal I_DS was measured at V_DS = 5 V, as studied in Fig. S12. The findings further confirm the robust retention capabilities of all photomemories under various electrical depression conditions and demonstrate that distinguishable multilevel behavior can be achieved with a larger negative V_GS.

Conventional memory devices are generally restricted to binary data storage, allowing only two discrete levels per device. In contrast, the photomemory investigated in this study exhibits a multilevel data storage function, modulated by varying the illumination period (405 nm, 7.29 μW cm⁻²), as depicted in Fig. 4(d). To evaluate the precision of storage data, which is crucial for optimal memory performance, we calculated the I_ph(avg) by averaging the I_DS after the light was turned off and extracted the corresponding standard deviation. The result clearly demonstrates that the photomemory achieves distinguishable multilevel states. Moreover, the magnitude of the error bars decreases with increasing illumination time, indicating improved stability of data. In addition, the I_ph(avg) shows a strong correlation with the number of phonons (N_ph), following a power-law dependence of 0.85. This suggests that the photomemory possesses well illumination-time-dependent controllability. The relationship between I_ph(avg) and N_ph under varying V_GS conditions is systemically analyzed. Notably, under V_GS = −50 V, our mix-dimensional devices achieve an almost linear correspondence, providing critical insight into mitigating nonlinearity challenges in optoelectronic devices. Furthermore, multilevel storage capabilities can also be attained by varying the illumination power (405 nm, 5 s and 10 s), as evidenced by the observations presented in Fig. S13a and b. Due to limitations in our laboratory equipment, shorter photo-programming times could not be explored. However, we propose that our device exhibits multilevel behavior under laser pulsing due to its favorable band alignment and efficient charge transfer. As shown in Fig. 4(e), a significant enhancement in the I_ph/I_ini is observed under negative bias conditions, which is in agreement with the previously discussed phenomenon. Furthermore, the CsPbBr₃/Al₂O₃/MoS₂ photomemory devices can achieve 4-bit data storage by integrating two input parameters: photo-driven (illumination time) and voltage-driven (amplitude of V_GS). The output is characterized by the power (n) of the I_ph/I_ini, as depicted in Fig. 4(f). In particular, the highest current level, indicated by the red region, is observed at a photo-programming duration of 180 s and V_GS = −50 V, indicating that these specific parameters create optimal conditions for attaining the highest I_ph/I_ini ratio. Subsequently, the heat map not only illustrates the correlation between the I_ph/I_ini ratio and the two input parameters but also enables the fine-tuning of the input parameters to achieve well-defined and reliable states. This provides valuable insights for precisely controlling photomemory levels, ensuring accurate data.

Based on the experimental results from KPFM analysis and the corresponding electrical properties, a possible operation mechanism for our mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ photomemory is proposed as illustrated in Fig. 5(a)–(c). Upon exposure to 405 nm excitation light, photons are initially absorbed by the floating PQD layer, leading to the generation of photo-induced excitons in CsPbBr₃ PQDs. Notably, our mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ system exhibits a type-II band alignment, which facilitates electron transfer.^41,42 As shown in Fig. 2(c), the conduction band energy (E_c) of MoS₂ and CsPbBr₃ is −4.2 eV and −3.3 eV, respectively; while their valence band energies (E_v) are −6.0 eV and −5.7 eV, respectively.⁴² This band alignment enables efficient charge transfer, allowing photo-induced electrons to move from the floating PQD layer to the E_c of the MoS₂ channel under illumination, leaving holes in the E_v of CsPbBr₃ PQDs. Furthermore, applying a negative back-gate voltage during the photo-programming significantly enhances the possibility of electron transfer likely due to the attractive electrostatic force arising from positively charged impurities within the depletion layer of the MoS₂ channel. During the erasing process, applying a positive gate bias allows the trapped holes in PQDs to be electrically neutralized by injected electrons from the n-type MoS₂ channel under strong accumulation conditions. Herein, the successful demonstration of this novel mixed-dimensional CsPbBr₃/Al₂O₃/MoS₂ phototransistor memory exhibits linear correlation and multilevel storage capabilities.

Fig. 5 Operation mechanism of the photomemory device. (a) Photomemory programming mechanism under blue light. (b) Photomemory photo-programming mechanism. (c) Photomemory electrical erasing mechanism.

2.4. Operational insights into CsPbBr₃-based photomemories

It is well known that a large V_DS leads to significant power consumption and reading disturbances.⁴³ From the perspective of flash array technology, maintaining V_DS below the device's threshold voltage V_th is crucial to prevent unintended data writing. The photomemory exhibits remarkable nonvolatility even at a reduced V_DS = 5 V after photo-programming with 405 nm light (7.29 μW cm⁻², 30 s). This enhanced performance can be attributed to the appropriate carrier mobility of the MoS₂ channel and significant photo-absorbing properties of PQDs, contributing to improved energy efficiency. Additionally, the n-type flash memory offers advantages such as high electron mobility, increased charge capacity, and reduced read voltage. The elevated electron mobility, in particular, enables faster programming and erasing operations.^44–46 Nevertheless, research specifically focusing on the n-type photomemory remains limited. In our study, MoS₂ was employed as an n-type channel, and an optimized Al₂O₃ dielectric thickness effectively mitigated data leakage, one of the key challenges associated with n-type channels.

To elucidate the dynamics of the photo-induced charge transfer process in CsPbBr₃/Al₂O₃/MoS₂ photomemories, I_DS was monitored over time under blue-laser illumination (405 nm, 7.29 μW cm⁻², 60 s) at V_DS = 5 V as described using eqn (5),⁴⁷

where t denotes the time, t₀ denotes the time constant at the onset of illumination, I_ini represents the initial current, A is a scaling constant, and τ_on characterizes the photo-response time associated with the onset of illumination. The floating-gate memory is known by its reliable data retention capabilities; however, it typically exhibits longer programming durations compared to SRAM and DRAM technologies.⁴⁸ As shown in Fig. S14, photomemory demonstrates a τ_on of 4.37 s, where shorter τ_on values indicate a faster charge transfer of photo-induced excitons from the absorbing layer to the conducting channel layer. Such fast τ_on stands out prominently among perovskite-based photomemory devices, as summarized in Table S2. This behavior may be attributed to the favorable band alignment between MoS₂ and CsPbBr₃, along with the optimized Al₂O₃ thickness, which enhances charge transfer efficiency. This phenomenon establishes a link between high-speed programming and stable data retention in perovskite-based systems; however, there is still plenty of room for further research on photo-programming speed in photomemories. In addition, the stability of photomemory switching operation is also evaluated through programming/reading/erasing/reading (PRER cycles, as presented in Fig. S15).

3. Conclusion

This report presents a high-performance mixed-dimensional photomemory device that integrates the superior photo-absorption of top-floating CsPbBr₃ PQDs and a bottom-channel MoS₂ with an optimally thickened Al₂O₃ dielectric layer to achieve exceptional optoelectronic performance. By fine-tuning the Al₂O₃ dielectric layer to an optimal thickness of 5.5 nm, we realize a highly efficient charge transfer rate, enabling low-energy photo-programming (∼0.228 pJ per stored hole) and robust multilevel data storage with a nearly linear photocurrent response across two orders of magnitude. Comprehensive structural, optical, and electrical analyses elucidate the critical role of dielectric thickness in balancing tunneling probability and interfacial dipole effects, with the 5.5 nm Al₂O₃ layer striking an ideal equilibrium for charge transfer and retention. The fast photoresponse, wide memory window, and reliable nonvolatile characteristics of the devices demonstrate the potential as a building block for energy-efficient photonic–electronic systems. These results not only advance the understanding of charge dynamics in perovskite-based heterostructures but also pave the way for high-performance photomemories tailored to the demands of next-generation AI computing technologies.

4. Experimental section

4.1. Materials

Sulfide (S, 99.5%, powder), caesium carbonate (Cs₂CO₃, 99.998%, powder), 1-octadecene (ODE, 90%, solvent), lead(II) bromide (PbBr₂, 99.998%, powder), oleylamine (OL, 80–90%), and ethyl acetate (99.5%) were purchased from Thermo Fisher Scientific. Molybdenum trioxide (MoO₃, 99.998%, powder) and oleic acid (OA, 90%) were purchased from Alfa Aesar. Sodium chloride (NaCl, 99.5%, powder) and 950 PMMA-A4 were purchased from Honeywell and Kayaku, respectively.

Synthesis of CsPbBr₃. The synthesis of CsPbBr₃ PQDs was conducted utilizing the hot injection method. The preparation of the Cs-oleate solution commenced with the dissolution of 100 mg of Cs₂CO₃ in 5.0 ml of ODS and 0.3 ml of OA as a solvent. The contents were subjected to a temperature of 150 °C for a duration of 10 minutes until complete solubilization was achieved, resulting in the formation of a clear solution. In the second step of the experimental procedure, 69 mg of PbBr₂, along with 5 mL of ODE, 0.5 mL of OA, and 0. 5 mL of OAm, were introduced into a reaction flask. The mixture was subsequently heated to 120 °C until all components were fully dissolved. Following this dissolution phase, the temperature was increased to 155 °C to facilitate the subsequent reaction. Subsequently, Cs-oleate was injected into the lead oleate solution for 5 s and rapidly cooled in the ice bath. Subsequently, the reaction product was subjected to centrifugation at 10 000 rpm for a duration of 4 minutes. The resulting precipitate was then dispersed in a mixture of hexane and ethyl acetate, after which a second round of centrifugation was conducted at 9000 rpm for 10 minutes.

Fabrication of the photomemory device. First, a substrate composed of highly doped silicon with a 100 nm layer of SiO₂ was utilized. The substrate underwent a sequential cleaning process, involving sonication in ethanol, DI water, acetone, and IPA. Double-layer MoS₂ was subsequently transferred onto a silicon wafer utilizing a PMMA-assisted wet transfer technique. The source and drain contacts were subsequently delineated utilizing digital light processing (DLP) lithography. This procedure was followed by the deposition of 10 nm of Bi and 40 nm of Au electrodes through the method of thermal evaporation. Subsequently, the tunneling oxide layer composed of Al₂O₃ was meticulously patterned and deposited upon the active area, featuring a channel length (L) of 20 μm and a width (W) of 40 μm. The device was annealed at a temperature of 180 °C for a duration of 2 hours in an Ar-protected atmosphere, with the objective of enhancing the quality of the film and the properties of the interface. Finally, as-fabricated CsPbBr₃ was deposited onto the device utilizing a spin-coating technique at 3000 rpm for 30 s. Subsequently, a solvent annealing process was conducted under ambient conditions at a temperature of 80 °C for a period of 20 minutes.

4.2. Characterization

Raman spectroscopy data (MRI-1532A) were acquired utilizing an excitation wavelength of 532 nm. The optical absorbance spectra of the photomemory were obtained utilizing a UV-visible absorption spectrometer (U-3010, Hitachi), while PL (PerkinElmer LS55) spectroscopy was conducted at an excitation wavelength of 350 nm. The TRPL spectra were obtained by exciting the samples with a wavelength of 405 nm and the data were collected using a fiber-optic configuration in conjunction with a PicoQuant Picoharp 300 and IDQ SPAD detector. AFM measurements conducted in tapping mode were employed to investigate the alterations in the thickness of Al₂O₃ using a Dimension Icon instrument from Bruker. Furthermore, the potential difference was assessed using KPFM prior to laser application, with a lift height of 100 nm, a spring constant of 2.8 N m⁻¹, and a frequency range of 60–80 kHz for the platinum probe. XPS measurements were performed utilizing a multifunctional analysis platform (ESCA VPIII, ULVAC-PHI), which incorporated an electron source for the purpose of charge compensation.

TEM images were acquired using a JEM-ARM200FTH microscope operating at an accelerating voltage of 200 keV to examine the distribution of perovskite nanocrystals and the structural characteristics of photomemories. The power density of the laser illumination was quantified utilizing an optical power meter (Model 1830-R, Newport). Multilevel behavior was achieved by using a Thorlabs LDC220C laser controller to accurately modulate the laser power. The distinct memory states were programmed by switching the laser on and off at consistent time intervals, demonstrating the device's precise multilevel capabilities. To assess the electrical properties of the photomemory, all measurements were performed utilizing a Keithley B1500A semiconductor parameter analyzer under ambient conditions at room temperature.

Conflicts of interest

The authors declare no conflicts of interest.

Data availability

Data for this article are available from Ming-Yen Lu.

Supplementary information: The supplementary information contains the EDS mappings, XPS, AFM, KPFM, and the IV curves of the samples. See DOI: https://doi.org/10.1039/d5nh00397k.

Acknowledgements

This study was supported financially by the National Science and Technology Council (NSTC 112-2221-E-007-043, NSTC 113-2124-M-007-002, NSTC 114-2221-E-492-019, and NSTC 113-2221-E-007-MY3). We thank Mr Y. S. Chen of the Instrumentation Center at National Tsing Hua University for technical support. Mr. C. Y. Tsai acknowledged the support by UMC Fellowship.

References

1. B. J. Shastri, A. N. Tait, T. Ferreira de Lima, W. H. Pernice, H. Bhaskaran, C. D. Wright and P. R. Prucnal, Nat. Photonics, 2021, 15, 102–114.
2. R. Li, Y. Gong, H. Huang, Y. Zhou, S. Mao, Z. Wei and Z. Zhang, Adv. Mater., 2025, 37, 2312825.
3. D. Kahng and S. M. Sze, Bell Syst. Tech. J., 1967, 46, 1288–1295.
4. M. P. Lu, M. Y. Lu and L. J. Chen, Adv. Funct. Mater., 2014, 24, 2967–2974.
5. M.-P. Lu, C.-W. Chen and M.-Y. Lu, Phys. Rev. Appl., 2016, 6, 054018.
6. A. C. Gadelha, A. R. Cadore, K. Watanabe, T. Taniguchi, A. M. de Paula, L. M. Malard, R. G. Lacerda and L. C. Campos, 2D Mater., 2019, 6, 025036.
7. Y. H. Chao, J. C. Chen, D. L. Yang, Y. J. Tseng, C. H. Hsu and J. Y. Chen, Adv. Funct. Mater., 2022, 32, 2112521.
8. A. Di Bartolomeo, A. Grillo, F. Urban, L. Iemmo, F. Giubileo, G. Luongo, G. Amato, L. Croin, L. Sun and S. J. Liang, Adv. Funct. Mater., 2018, 28, 1800657.
9. A. Kumar, E. Faella, O. Durante, F. Giubileo, A. Pelella, L. Viscardi, K. Intonti, S. Sleziona, M. Schleberger and A. Di Bartolomeo, J. Phys. Chem. Solids, 2023, 179, 111406.
10. M. Zubair, Y. Dong, B. Cai, X. Fu, H. Wang, T. Li, J. Wang, S. Liu, M. Xia and Q. Zhao, Appl. Phys. Lett., 2023, 123, 051102.
11. F. Cao, E. Hong, Z. Hu, Y. Liu, B. Sun, J.-H. He and X. Fang, Nano Energy, 2024, 120, 109135.
12. I. H. Chao, Y. T. Yang, M. H. Yu, C. H. Chen, C. H. Liao, B. H. Lin, I. C. Ni, W. C. Chen, A. W. Ho-Baillie and C. C. Chueh, Small, 2023, 19, 2207734.
13. B. J. Moon, Y.-S. Song, D. Son, H. Y. Yang, S. Bae, S.-K. Lee, S. H. Lee and T.-W. Kim, Small Sci., 2023, 3, 2300068.
14. H.-T. Hsu, D.-L. Yang, L. D. Wiyanto and J.-Y. Chen, Adv. Photonics Res., 2021, 2, 2000185.
15. A. Stephen, A. Biju and J. Peediyekkal, J. Lumin., 2024, 120462.
16. L. Protesescu, S. Yakunin, M. I. Bodnarchuk, F. Krieg, R. Caputo, C. H. Hendon, R. X. Yang, A. Walsh and M. V. Kovalenko, Nano Lett., 2015, 15, 3692–3696.
17. Y. H. Chang, C. W. Ku, Y. H. Zhang, H. C. Wang and J. Y. Chen, Adv. Funct. Mater., 2020, 30, 2000764.
18. S. P. Prakoso, M.-N. Chen and Y.-C. Chiu, J. Mater. Chem. C, 2022, 10, 13462–13482.
19. E. Ercan, Y. C. Lin, W. C. Yang and W. C. Chen, Adv. Funct. Mater., 2022, 32, 2107925.
20. E. Ercan, Y. C. Lin, L. C. Hsu, C. K. Chen and W. C. Chen, Adv. Mater. Technol., 2021, 6, 2100080.
21. Z. Shen, Z. Yang, Y. Zhou, Y. Ye, B. Ye, Q. Huang, W. Wu, H. Hong, Z. Hong and Z. Meng, J. Mater. Chem. C, 2023, 11, 3213–3226.
22. H. Yang, Y. Yan, X. Wu, Y. Liu, Q. Chen, G. Zhang, S. Chen, H. Chen and T. Guo, J. Mater. Chem. C, 2020, 8, 2861–2869.
23. E. Ercan, J.-Y. Chen, C.-C. Shih, C.-C. Chueh and W.-C. Chen, Nanoscale, 2018, 10, 18869–18877.
24. Y. Sun, Y. Ding and D. Xie, Adv. Funct. Mater., 2021, 31, 2105625.
25. Y. Zhang, J. Guo, Y. Xu, W. Huang, C. Li, L. Gao, L. Wu, Z. Shi, C. Ma and Y. Ge, Nanoscale Horiz., 2020, 5, 847–856.
26. M. Y. Lu, Y. T. Chang and H. J. Chen, Small, 2018, 14, 1802302.
27. Q. Liu, L. Yin, C. Zhao, J. Wang, Z. Wu, H. Lei, Y. Liu, B. Tian, Z. Zhang and Z. Zhao, Nano Energy, 2022, 102, 107686.
28. Z. Zhang, P. Lin, Q. Liao, Z. Kang, H. Si and Y. Zhang, Adv. Mater., 2019, 31, 1806411.
29. M. A. Pimenta, E. Del Corro, B. R. Carvalho, C. Fantini and L. M. Malard, Acc. Chem. Res., 2015, 48, 41–47.
30. S. T. Chang, W. C. Chen, C. F. Lin, P. Z. Yu, C. H. Tsai, C. J. Cho, Y. C. Lin, W. C. Chen and C. C. Kuo, Adv. Opt. Mater., 2024, 12, 2400470.
31. J. Y. Chen, D. L. Yang, F. C. Jhuang, Y. H. Fang, J. S. Benas, F. C. Liang and C. C. Kuo, Adv. Funct. Mater., 2021, 31, 2105911.
32. T. Li, B. Wan, G. Du, B. Zhang and Z. Zeng, AIP Adv., 2015, 5, 2400470.
33. I. Shlyakhov, J. Chai, M. Yang, S. Wang, V. Afanas’ev, M. Houssa and A. Stesmans, Phys. Status Solidi A, 2019, 216, 1800616.
34. J. Church, C. Weiland and R. Opila, Appl. Phys. Lett., 2015, 106, 171601.
35. Q. Chen, C. Wang, Y. Li and L. Chen, J. Am. Chem. Soc., 2020, 142, 18281–18292.
36. A. Ortiz-Conde, F. J. García-Sánchez, J. Muci, A. T. Barrios, J. J. Liou and C.-S. Ho, Microelectron. Reliab., 2013, 53, 90–104.
37. Y. Chen, H. Li, Q. Wang, Z. Shang, M. Shi, X. Zhou, X. Xu, X. An, R. Huang and M. Li, 2025.
38. H. Lai, Z. Lu, Y. Lu, X. Yao, X. Xu, J. Chen, Y. Zhou, P. Liu, T. Shi and X. Wang, Adv. Mater., 2023, 35, 2208664.
39. C. C. Shih, Y. C. Chiu, W. Y. Lee, J. Y. Chen and W. C. Chen, Adv. Funct. Mater., 2015, 25, 1511–1519.
40. W. C. Yang, Y. C. Chiang, J. Y. Lam, T. H. Chuang, E. Ercan, C. C. Chueh and W. C. Chen, Adv. Electron. Mater., 2020, 6, 2000458.
41. L. Guo, Y. Gu, Z. Yang, S. Tian, X. San, J. Liu, L. Gao, S. Qiao and S. Wang, Adv. Mater. Interfaces, 2021, 8, 2002231.
42. X. Song, X. Liu, D. Yu, C. Huo, J. Ji, X. Li, S. Zhang, Y. Zou, G. Zhu and Y. Wang, ACS Appl. Mater. Interfaces, 2018, 10, 2801–2809.
43. Y. Cai, Y. Luo, S. Ghose and O. Mutlu, 2015.
44. H.-S. Shih, S.-W. Fang, A.-C. Kang, Y.-C. King and C.-J. Lin, Appl. Phys. Lett., 2008, 93, 213503.
45. B. Hafsi, A. Boubaker, D. Guerin, S. Lenfant, A. Kalboussi and K. Lmimouni, Org. Electron., 2017, 45, 81–88.
46. Y. Li, R. Huang, Y. Cai, F. Zhou, X. Shan, X. Zhang and Y. Wang, IEEE Electron Device Lett., 2007, 28, 622–624.
47. J. C. Chen, Y. D. Lu and J. Y. Chen, Adv. Sci., 2023, 10, 2301028.
48. C. Liu, X. Yan, X. Song, S. Ding, D. W. Zhang and P. Zhou, Nat. Nanotechnol., 2018, 13, 404–410.

---