A new era in optoelectronic applications has emerged with the recent development of metal halide perovskites with the chemical formula ABX3 (where A is an organic cation, B is Pb, and X is a halide anion) [1–3]. All-inorganic cesium lead halide perovskites (CsPbX3, where X is a halide) exhibit promising characteristics, such as a long carrier diffusion length, high and broad absorption, tunable bandgap, and high carrier mobility, and have low-temperature fabrication processes [4–9]. These novel properties, which can potentially enhance the quantum efficiency in a broad spectral range from ultraviolet to near-infrared at a low cost, have gained increasing interest in the fields of perovskite light-emitting diodes (LEDs) and solar cells [2,10–12].

In conjunction with their high quantum yield characteristics, a broad range of memory devices can be realized by combining the advantages of the intrinsic hysteresis of perovskites (caused by ferroelectricity, ion/defect migration, and charge trapping), low-voltage operation, long retention time, and high ON/OFF ratio [13–15]. Numerous methodologies of quantum dot (QD) synthesis have been developed in the past decades, and these techniques are categorized based on the phase of the particle growth environment, either a liquid or vapor [16,17].

Despite the successful synthesis and size tuning of QDs using vaporphase methods, extensive research has been focused on liquid phase synthesis. This is due to the requirement of expensive devices for vapor-phase methods and the difficulty in sample retraction after epitaxial growth [18,19]. Numerous synthesis techniques have been developed based on liquid-phase methods, such as hot injection, noninjection, and aqueous synthesis. Among them, hot injection has been widely applied due to its high production yield, short fabrication time, high crystalline quality, and high tunability of the QD size, which depends on the synthesis temperature. Recent studies on metal halide perovskite QDs have adopted the hot-injection method, in which the QDs are capped by long, hydrophobic alkyl ligands, such as oleic acid (OA) and oleylamine (OAm), leading to colloidal stabilization in nonpolar solvents and higher stability over long-term operation.

However, even with the unique photovoltaic and optoelectronic characteristics, the practical application of perovskite QDs has been impeded by the deterioration of charge transport characteristics of the QDs owing to the insulating behavior of the long alkyl ligands and reduced carrier injection from the adjacent hole- and electrontransporting layers, decreasing the quantum efficiency. Moreover, the remaining long ligands and precursors are purified by adding polar solvents after centrifuging the QD mixture, resulting in a decrease in perovskite stability, thereby necessitating the minimization of conventional purification steps. To overcome these limitations, various studies have suggested the use of a halide ion pair ligand [didodecyl dimenthylammonium bromide (DDAB)], instead of OA or OAm, with a relatively short bromide ion pair ligand to enhance the poor charge injection observed when using OA and OAm. These attempts have successfully achieved a high quantum yield with unprecedented stability. Rather than simply decreasing the length of the ligands, in another study, a mixed solvent [hexane/ethyl acetate (EA)] was employed to optimize the surface ligand density in the QD solution to promote dynamic ionic interactions [20].

Other research studies on perovskite LEDs adopted the centrifugation of QDs in polar solvents (e.g., dimethyl sulfoxide, ethanol, and toluene) to purify QD solutions and selectively filter the unwanted constituents remaining in the solution. Despite these developments in the purification processes, the aforementioned methods have the following drawbacks. First, the DDAB-treated QDs should be washed with anti-solvents to remove by-products, including superfluous ligands, which could significantly deteriorate the charge transport of QDs. This degradation and increased quenching may adversely affect the merits of this method. The excessive use of polar solvents may deteriorate the photoluminescence (PL), colloidal stability, and structural integrity of the QD films. Furthermore, these methods do not target scalable applications, such as memory devices.

Therefore, we introduced a new type of surface ligand engineering: a washing process involving the drop casting of a methanol (MeOH) solvent on a spin-coated perovskite QD film. The use of a high-polarity solvent at the spin-coating site enabled the simultaneous removal of excess ligands and residual reaction solvents while avoiding the selfaggregation of QDs. Therefore, the fabrication of memory devices with perovskite QDs in conjunction with the drop-casting purification method may provide a new way of fabricating low-voltage memory devices.

The inorganic perovskite CsPbBr₃ QDs used in our experiment were synthesized using a typical hot-injection method with Cs-oleate and PbBr2 precursors [21], as illustrated in Fig. 1(a). In a 100 ml threeneck round-bottom flask, 20 ml of octadecene (ODE) (90 %, Thermo-Fisher), and 276 mg of lead (II) bromide (99.999 %, Thermo-Fisher) were added, and the moisture was removed at 120 °C. Following 1 h of drying, 2 ml of OA (90 %, Sigma-Aldrich) and 2 ml of OAm (70 %, Sigma-Aldrich) were added, and the mixture was heated. When it reached 150 °C, the synthesis of CsPbBr₃ was continued by injecting 1.6 ml of Cs-oleate, which was prepared by mixing 250 mg of cesium carbonate (99.9 %, Sigma-Aldrich), 1 ml of OA, and 10 ml of ODE. After 30 s, the synthesized QD solution was rapidly cooled in an ice bath, followed by centrifugation at 20 °C and 15,000 rpm for 30 min to separate the by-products. Following the removal of the supernatant, the remaining pellets were dispersed in toluene and centrifuged at 4 °C and 8,000 rpm for 20 min. Finally, the CsPbBr₃ QD solution was filtered using a 200 nm membrane filter.

Figure 1. Schematics of (a) the synthesis of CsPbBr₃ QDs in toluene and (b) purification of QDs using the drop casting of polar solvents.

The solution was analyzed using nuclear magnetic resonance (NMR) (Agilent 600 MHz Premium COMPACT NMR spectrometer) to investigate the existence of long-ligand residuals (i.e., OA, OAm, and ODE) in the QD solution, and the results are presented in the right side panel of Fig. 1(a). Comparing the spectra of the residuals, the spectrum of the centrifuged QDs exhibited terminal alkene resonance peaks of OA and OAm at ∼ 5.3 ppm, although the ODE resonances disappeared. These results prove that the two thorough centrifugation/filtering processes completely removed the excessively long ligands from the CsPbBr₃ QDs. Subsequently, we used polar solvents, such as dichloromethane, 1, 2-dichloroethane, EA, and MeOH in an additional purification step, as illustrated in Fig. 1(b). Specifically, during 60 s of spinning on the spin coater at 1,000 rpm, 1 ml of the solvent was directly dropped onto the sample surface. Such drop-cast polar solvents contributed to the washing away of the long ligands residing on the surface of the spin-coated CsPbBr₃ film through ion trapping.

Figure 2 illustrates the degree and quality of purification achieved by washing solvents of different polarities. As illustrated in Fig. 2(a), the pre-washed QDs exhibit a strong chemical shift at 5.3–5.4 ppm [22], indicating the existence of OA and OAm impurities. Notably, as we applied the spin coating of polar solvents, a gradual decay of the vibration mode corresponding to the impurities was observed. Significantly, for the strongest polarity used among the chosen solvents (i.e., MeOH), the chemical shift indicating the ligand impurities nearly disappeared. Such a dependence on solvent polarity is observed in the results of Fourier transform infrared (FT-IR) spectroscopy measurements, as illustrated in Fig. 2(b). The vibrational mode of C=O stretching at 1,710 cm⁻¹, indicating the presence of OA, gradually narrowed, and finally disappeared when a washing solvent with stronger polarity was used.

Figure 2. Identification of impurities and crystal structures of CsPbBr₃ QDs depending on the washing solvents (spectra shown from the top down are in the increasing order of polarity indices of washing solvents, marked in parentheses): (a) NMR, (b) FT-IR spectroscopy, and (c) XRD.

X-ray diffraction (XRD) analysis was performed on the perovskite QDs coated on the silicon wafer, and the results are shown in Fig. 2(c) (the silicon wafer peak at ∼ 33° was removed during data processing). Notably, all the patterns exhibit distinct peaks at 2θ = 15.1 and 30.7°, which are assigned to the diffractions of (100) and (200) planes, respectively, indicating that the CsPbBr₃ QDs can maintain their crystal structure through our purification step. Notably, in the MeOH-washed case, even more peaks of the standard XRD patterns of cubic CsPbBr₃ appeared at 21.5, 37.8, and 43.9°, corresponding to diffractions of (110), (211), and (220), respectively, signifying the enhanced crystallization of QDs. Additionally, the sharp PL peak observed after MeOH purification (Fig. 3) supports the well-maintained quantum confinement capability of the QDs. Compared to that of the pre-washed sample, the slight red shift of the PL spectrum after purification may also verify the removal of long ligands. This shift could be attributed to the increased coalescence of QDs and their increased size [23].

Figure 3. Photoluminescence spectra of pre-washed and MeOH washed CsPbBr₃ QDs.

To investigate the feasibility of the as-obtained purified QDs for memory applications, we first tested the capacitive characteristics by fabricating a metal-insulator-semiconductor-insulator-metal capacitor, where the combination of a gold top-contact electrode, CsPbBr₃ QDs:PS, CsPbBr₃ QDs, Al₂O₃, and a heavily doped Si layer served as the cascading configuration. The capacitance of the MeOH-washed sample gradually decreased in the high-frequency range, whereas that of the prewashed sample exponentially decreased [Fig. 4(a)]. The capacitive transition at low frequencies indicates the long-term relaxation of the charge carriers in the organic capacitor. Charges localized in relatively deep states do not respond to high frequencies because of their slow release from trap states. Therefore, it is suggested that deeper states exist for trapping mobile charges in the pre-washed sample, owing to organic residuals. The structure was adopted as a gate dielectric in our study of the memory device, as shown in Fig. 5, to achieve a high level of capacitance at a low gate voltage (V_g). The dislocated charges derived from space charge polarization accumulated at the interface of the CsPbBr₃ QDs and Al₂O₃, enhancing the internal electric field within the depletion region. In general, the organic residuals in semiconductors decrease their capacitance at high frequencies, owing to the lag of dipole polarization change under the applied electric field, which is known as anomalous dielectric dispersion. As illustrated in Fig. 4(a), both the pre-washed and MeOHwashed samples follow this trend. Notably, the MeOH-washed sample maintained a reasonably high level of capacitance at high frequencies. By contrast, the pre-washed sample was unable to efficiently passivate the high-frequency charge oscillation at > 1 MHz. These results prove the effectiveness of our purification step using MeOH, which removed the residual long ligands and recovered the intrinsic capacitive performance of the QDs, even at high frequencies. In addition, the V_g dependence test in a range from −2 to +2 V suggests its feasibility in low-voltage memory operation, as shown in Fig. 4(b).

Figure 4. Capacitive characteristics of pre-/MeOH-washed CsPbBr₃ QDs. (a) Frequency dependence and (b) V_g sweep from −2 to +2 V.

Figure 5. Memory characteristics of pre-washed and MeOH-washed CsPbBr₃ QD memories: I–V characteristics of (a) pre-washed and (b) MeOH-washed CsPbBr₃ memories (V_g sweep range of ±1, 2, 3, 4, and 5 V and V_d = −5 V). Endurance properties of (c) pre-washed and (d) MeOH-washed CsPbBr₃ memories. Normalized photo-response curves (excitation at 450 nm and V_d = −1 V) of (e) pre-washed and (f) MeOH-washed CsPbBr₃ memories.

A heavily n-doped bare silicon (100) wafer was diced into the dimensions of 1.5 × 1.5 cm² and sonicated in acetone and isopropyl alcohol for 10 min. As a dielectric layer, the Al₂O₃ film was deposited with 200 cycles of atomic layer deposition on the prepared substrate at 200 °C, resulting in a thickness of ∼ 25 nm. Trimethylaluminum and water were used as Al and O precursors, respectively. Thereafter, the following steps were repeated four times: CsPbBr₃ QD solution was spin-coated on the Al₂O₃ layer at 1,000 rpm for 1 min and then baked at 180 °C for 30 s in a nitrogen atmosphere. The baked sample was purified by MeOH drop casting, and a mixture of 2 mg of polystyrene (PS, Mw ≈ 192,000, Sigma-Aldrich) and 1 ml of CsPbBr₃ QD solution was agitated on the sample for 2 h. The CsPbBr₃/PS solution was spin-coated at 4,000 rpm and baked in a glove box at 180 °C for 30 s. Subsequently, a 50 nm-thick pentacene layer was deposited on the CsPbBr₃/PS composite film as an active layer using thermal evaporator at a rate of 0.2−0.3 Å/s. Finally, ∼ 50 nm gold electrodes were deposited through a shadow mask through thermal evaporation at a pressure of 10⁻⁶ Torr and used as the source−drain contacts.

To compare the memory characteristics of the pre-/MeOH-washed devices, whose optical microscope images are illustrated in the inset of Fig. 5(a), we measured the current−voltage (I−V) characteristics under dark conditions, as shown in Figs. 5(a) and 5(b). Our memory device exhibited a counter-clockwise hysteresis window and considerable threshold voltage shifts with dual sweeps of V_g values ranging from ±1 to ±5 V (V_d = −5 V), indicating low-voltage ferroelectric transistor memory characteristics. For both V_g sweeps, the memory window (△VTh) is negligibly narrow at a gate sweep range of ±1 V, and it gradually increases as a wider sweep range is utilized, indicating the tunneling of charge carriers in the channel through the CsPbBr₃ QDs−PS composite layer and charge trapping in the CsPbBr₃ QDs. For the MeOH-washed sample, △VTh reached 1.45 V at the V_g sweep range of ±5 V, showing an ON/OFF ratio of ∼ 103. Significantly, the memory window is proportional to the number of trapped charges. The asymmetric hysteresis of the transfer curve arises because most of the charge carriers in pentacene are more easily trapped than the electrons in the CsPbBr₃ QDs. Although at a gate sweep range of ±5 V, the △VTh of MeOH-washed sample appears to be narrower, and its ON/OFF ratio is improved by approximately 22.5 times (10 to 225).

To test the cycling performance, 400 writing−reading−erasing−reading cycles on each memory were performed with the following gating parameters: Write: V_GS = −5 V; Read: V_GS = −2 V; Erase: V_GS = +5 V; and Read: V_GS = −2 V, and the results are shown in Figs. 5(c) and 5(d). The reading of the pre-washed memory failed after approximately 270 cycles, and its ON/OFF ratio was unstable during operation. By contrast, the MeOH-washed sample could reasonably maintain its normal operation throughout the cycling test. The retention times of the pre- and MeOH-washed CsPbBr₃ devices were determined to be ∼2 × 10³ and 2 × 10⁴ s, respectively. The poor retention characteristics of the prewashed device indicate that residual organic impurities induce the rapid charge relaxation of the trapped charges. The MeOH-washed sample maintained ON- and OFF-current states over a relatively long time after the writing and erasing processes, indicating excellent endurance. To extend the aforementioned application to optoelectronics (e.g., phototransistor memory), the pre-/MeOHwashed CsPbBr₃ devices were utilized as photodetectors and tested without an additional bias voltage. Notably, photosensors consisting of CsPbBr₃ QDs are known to exhibit slow responses, within a few seconds, and low ON/OFF ratios of approximately >1 [24]. However, to overcome these limitations, previous studies have suggested different fabrication techniques (e.g., the addition of ZnO NPs to the CsPbBr₃ precursor). This study aimed to observe the effectiveness of the purification method; therefore, our investigations were performed at the boundary of comparative analysis with the original properties of CsPbBr₃. To present the photoswitching characteristics, the photocurrent was measured at V_g = −1 V under the alternating dark condition and 450-nm monochromatic irradiation, which is reversibly switched using an optical shutter, and the result is shown in Figs. 5(e) and 5(f). The ON/OFF switching could maintain a nearly identical output for multiple cycles, indicating the reproducibility of our device. The highest ON/OFF ratios for pre- and MeOH-washed samples were 2.10 and 1.95, respectively. A representative cycle is shown in the inset for a detailed representation of the rise/decay time, and the time constants were estimated based on the curve fits of the transients. In terms of the rise time (t_rise, from 10 to 90 % of the saturated value), the MeOH-washed sample exhibits a highly enhanced temporal response of t_rise = 0.887 s compared to the relatively slower response (t_rise = 2.154 s) of the pre-washed sample. By contrast, the decay time (t_decay, from 90 to 10 % of the peak value) is similar for the pre-washed sample as t_decay ≈ 6.4 s and the MeOH-washed sample as t_decay ≈ 7.5 s. The asymmetric rise and decay suggested a parabolic optical conductivity detector. Therefore, the above photosensing test successfully demonstrated the effectiveness of our purification method, which removed long-chain ligands and improved the rise time by approximately 2.4 times without considerably deteriorating the ON/OFF ratio.

In this study, we reported a new purification method for inorganic perovskite CsPbBr₃ QDs by drop casting a strong polar solvent, MeOH, to remove the residual long ligands that impede the charge transport characteristics. This approach can utilize the advantages of the conventional application of long ligands in QD solutions to enhance the dispersion of QDs. Meanwhile, the issue of poor charge transport by long ligands can be addressed by spin coating the QDs in advance and subsequently removing the excessively long ligands in the quasi-solid phase. The removal of the long ligands was confirmed by NMR, FT-IR spectroscopy, and XRD. In terms of memory device operation, the enhanced semiconducting characteristics of the purified QDs were first proven by observing the highly maintained capacitance at a high frequency of approximately 1 MHz. In addition, an increased ON/OFF ratio during memory operation was observed, and the device exhibited a robust operation throughout the cycling performance test. Enhanced optoelectronic operation was confirmed by achieving a fast photoresponse.

This work was financially supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (Grant No. 2022R1F1A1072339 and RS-2023-00220077).

The authors declare no conflicts of interest.