Size-controlled synthesis of individual halide perovskite nanocrystals at defined locations
In a typical synthesis, an ink solution composed of halide perovskite precursors (e.g., AX and PbX2 for the lead halide perovskite APbX3, where A is either an organic or inorganic cation and X is a halide anion) dissolved in a mixture of dimethyl sulfoxide (DMSO) and sulfolane was spin coated onto an array of ~1000 polydimethylsiloxane (PDMS) micro-pyramidal pens (pen array) (Fig. 1A, i). Unlike molecular or polymer inks used in conventional PPL, the liquid organic inks accumulate around the base of each pyramid due to the high surface tension and low viscosity of the ink and serve as a reservoir for continuous inking (Fig. 1, B and C). Once the arrays were inked, >100,000 droplet nanoreactors were deposited across a variety of substrates (Fig. 1A, ii and iii). Because of the high surface-to-volume ratio, these nanoreactors readily evaporate within seconds, which leads to the nucleation and growth of individual halide perovskite nanocrystals (iv). As a proof of concept, methylammonium lead bromide (MAPbBr3) nanocrystals, which exhibit strong PL (Fig. 1D and fig. S2), were synthesized on silicon substrates. In this case, each of the PDMS pyramidal pens created 121 crystals covering an ~0.024-mm2 area (yellow dashed box in Fig. 1D), and the entire substrate is covered with a highly ordered, periodic array as evidenced by the Fourier transform of the fluorescence micrograph (Fig. 1E). The morphology and chemical composition of the crystals were determined using scanning electron microscopy (SEM) and energy-dispersive x-ray spectroscopy (EDS) elemental mapping [note that Pb and Br are uniformly distributed throughout the individual nanocrystals (Fig. 1F)]. Atomic force microscopy (AFM) shows that these nanocrystals have a typical width-to-height ratio of ~3:1 (fig. S3). Transmission electron microscopy (TEM) imaging and selected-area electron diffraction (SAED) of the nanocrystals synthesized on a 15-nm-thick silicon nitride membrane confirm that the nanocrystals are single crystalline (Fig. 1, G and H). The SAED pattern along the  zone axis, along which a rectangular projection is observed for the nanocrystal, matches the simulated diffraction pattern for a cubic perovskite structure (Fig. 1I) (19). It is critical to note that the ink mixture used in these experiments has a low volatility and remains stable on the pen arrays for at least an hour of continuous patterning. The low volatility is important in generating high-quality single nanocrystals as higher-volatility mixtures did not yield comparable results (fig. S4).
In addition to controlling the location of individual nanocrystals, this method enables one to tune the crystal size by controlling both the initial precursor concentration in the ink and the extension length of the PDMS pens against the substrate (Fig. 2A and figs. S5 and S6). Using an initial ink concentration of 0.04 M and extending the PDMS pens only 1 μm result in the formation of ~50-nm MAPbBr3 nanocrystals, as evidenced by SEM (Fig. 2B). In principle, the synthesis of even smaller site-isolated nanocrystals should be possible; however, such structures are difficult to characterize and analyze using microscopy techniques. Because the PL intensity scales as a function of nanocrystal size, one can use this size tunability to create grayscale images at the microscale (Fig. 2, C to E). In addition, because this technique is substrate versatile, it can be used to generate comparable patterns on conductive indium tin oxide (ITO)–coated glass, glass slides, and silicon nitride membranes (Fig. 2, F to H, and figs. S7 and S8). However, the single nanocrystal per site yield and corresponding crystal quality depend on the roughness of the substrate. For example, for ITO with a root mean square (RMS) roughness (Rg) of 2.97 nm, multiple particles in each nanoreactor were typically observed, whereas ITO substrates with an Rg of 0.62 nm had a single-particle yield close to 100% (figs. S8 and S9). In addition, while large-area patterning with a controlled size gradient is possible on glass slides, as evidenced by dark-field (DF) scattering measurements (fig. S7E), the PL of these particles is nonuniform, suggesting poor crystal quality (Fig. 2G).
Size dependence of single-nanocrystal PL emission
To understand the PL properties of individual MAPbBr3 nanocrystals of different sizes, we prepared size-gradient nanocrystal arrays with an interparticle spacing of ~5 μm on silicon wafers. Single-nanocrystal emission spectra were collected by focusing the excitation laser onto an ~2-μm spot around each nanocrystal (Fig. 3A) and then correlated with nanocrystal size, as determined by SEM. Unexpectedly, high-resolution PL (HRPL) spectroscopy reveals an emission peak that contains multiple shoulders, as evidenced by the peaks in the second derivative of the spectra (Fig. 3B), which do not originate from the spectrometer (fig. S10). This observation indicates that multiple emission modes may be present. Because no prior knowledge of these modes is available, a direct fit of the HRPL spectra using multiple peak functions is unreliable. As such, we assumed that each mode could be described by an arbitrary peak broadened by a Gaussian point spread function, whereby each spectrum was iteratively deconvolved using the Richardson-Lucy algorithm (Fig. 3B, dashed curve; see also text S1) (20, 21). The shoulders in the HRPL spectrum can be decomposed into multiple emission modes, and their relative peak intensities (when normalized) still match with the envelope shape of the overall HRPL spectrum when the intensity is adequately higher than the noise level. The presence of multiple modes suggests that the band edges in these materials are defined by various lattice imperfections, such as emissive defects (22, 23) and lattice distortion (24, 25). Such a highly defective halide perovskite lattice may no longer be considered continuous; it may not have the well-defined band edges typically observed with large single crystals.
Using this high-throughput approach, we studied the emission from nanocrystals as a function of size from ~460 to ~110 nm. Smaller nanocrystals show an HRPL spectrum blue-shifted to higher energies (Fig. 3C), even though the dimensions of all these nanocrystals are well above the Bohr radius in MAPbBr3 that is required for quantum confinement (26). Similar blue shifts have been observed in polycrystalline thin films and microstructures of halide perovskites; however, their origin is under debate with several proposed explanations. These pertain to surface depletion (27), surface emission (28), substrate-induced strain (29–32), free carrier formation (33), and photon reabsorption (34). To gain further insight into this size-dependent emission phenomenon, we deconvolved spectra from nanocrystals of different sizes to study the peak energies and relative intensities of all modes (Fig. 3D). The peak positions for all modes are almost the same; however, their relative intensities vary, resulting in the apparent blue shift of the overall emission. Quantitatively, four major modes labeled as a, b, c, and d were selected for comparison. When all deconvolved spectra were normalized to the range [0, 1], the relative intensity of modes a and b decreased as crystal size decreased, while mode d increased (Fig. 3E). The peak energies of all four modes exhibit a slight blue shift on the order of a few millielectron volts (Fig. 3F), far below the observed overall PL blue shift for the single nanocrystals of different sizes measured here or for the polycrystalline structures reported in the literature (27, 35). Note that all these modes are closely correlated to the modes present in the bulk crystal, suggesting that they are intrinsic to the crystal and share the same physical origin. These results provide strong evidence that two different types of potentially size-dependent effects exist in halide perovskite nanocrystals: surface depletion–constraint quantum confinement and substrate-induced strain [internal pressure on the megapascal scale (36); note that heteroepitaxy is not relevant in this study, as evidenced by the SEM images] are potentially responsible for the slight blue shift of each emission mode. On the other hand, the overall PL shift as a function of crystal size is a result of the systematic intensity modulation of these modes, which has a different physical origin (vide infra).
To reveal the nature of these multiple emission modes that show size-dependent intensity modulation behavior, we systematically analyzed single-nanocrystal emission from crystals with different structures and in different environments. By placing the nanocrystal arrays in high vacuum (<~10−4 Pa), deep traps on the surface that are usually blocked by oxygen and water molecules were exposed (37), which resulted in a substantial reduction in PL intensity (by a factor of >300; Fig. 3G). The emission peak energy is almost unchanged in the presence or absence of the blocker molecules (in air versus in vacuo), suggesting that the emission modes originate from the interior of the crystals and that surface defects are not involved significantly. Moreover, we partially excited the nanocrystals using an ~2.33-eV laser and compared the HRPL spectra with the fully excited ones to reveal the relationship among different emission modes. The HRPL spectra in the <2.32-eV (filter cutoff) region are unchanged from the fully excited spectra, suggesting that the energy states associated with these emission modes have relatively fixed densities and are independent from one another. We further confirmed that this multimode, size-dependent emission behavior is not a result of organic cation rotation (38) or DMSO insertion (39) in the MAPbBr3 crystals, because all-inorganic CsPbBr3 nanocrystals and DMF (N,N′-dimethylformamide)–derived MAPbBr3 polycrystals both show similar effects (figs. S13 and S15). Together, the data are consistent with two plausible pathways: (i) a change in the relative density of emissive states or (ii) a redistribution of the emission intensities at different energies induced by crystal size variation.
To identify the most probable explanation for the size-dependent emission behavior, we studied the excitonic properties of single MAPbBr3 nanocrystals in detail. When cooled to 10 K in vacuo, individual MAPbBr3 nanocrystals exhibit well-defined emission depending on the power of the focused excitation laser (Fig. 4A). As the laser power increases from 0.1 to 20 μW, the lower-energy tail in the emission peak indicative of the formation of bound excitons gradually diminishes. The intensity of the main emission peak (I) follows a power law against the excitation power (P)
(1)with exponent values 1 < k < 2 (Fig. 4B), which confirms that the emission is predominantly excitonic (40) and rules out nonlinear optical generation in this system. When heated between 100 and 150 K, a gradual transition in emission energy was observed (Fig. 4, C and D), consistent with a phase transition (41, 42). Multi-peak features are present at all temperatures and can be deconvolved below ~200 K when the signal-to-noise ratio is sufficient. These results suggest that the multiple modes are associated with defects intrinsic to the crystal. We further studied the emission from single nanocrystals at room temperature in the atmosphere, excited by cyclically varying excitation power, and a hysteresis loop was observed in the emission intensity (Fig. 4, E and F). Specifically, when the laser power is higher than ~10 μW, PL intensity from the nanocrystal decreases due to photoinduced damage. The peak energy shows almost no change before the laser power reaches the damage threshold when the process is dominated by free exciton-like emission with a power-law slope k = 1.14 (fig. S17). We did not observe evidence of an additional bound exciton peak at low excitation power (Fig. 4E). These results are consistent with the interpretation in micrometer-sized and bulk crystals that various intrinsic defect states exist in proximity to the band edges (22, 23) and that deep traps are protected due to screening (24, 25). In addition, the damage-induced peak shift is also less than 10 meV (Fig. 4G), much smaller than the size-induced blue shift that spans tens of millielectron volts. As the emission modes that constitute the PL peak are independent of each other (Fig. 3H), these results suggest that the hot carrier recombination (band filling) effect (43) or defect density variation are not the main contributors to the size-dependent emission in halide perovskite nanocrystals. As a result, a redistribution of emission intensity due to the interaction between emitted photons and the crystal, i.e., photon reabsorption and possible photon recycling (44), is most likely responsible for this size-dependent behavior. The overall PL energy shift of ~26 meV in the nanocrystals (crystal size from ~460 to ~110 nm, estimated thickness from ~150 to ~40 nm; Fig. 3C) is consistent with the depth-dependent cathodoluminescence energy shift due to reabsorption, as reported in the literature (45).
Together, we conclude that two factors primarily cause the size-dependent energy shift of the PL peaks for halide perovskite nanocrystals: (i) defects at the noncontinuous electronic band edges result in excitonic emissions with varying energies, and (ii) the reabsorption of higher-energy photons changes the intensity distribution of these emission modes (Fig. 4H). For larger crystals, photons emitted by higher-energy modes are more efficiently reabsorbed, creating electron-hole pairs that typically relax to lower-energy states nonemissively and causing a decrease in the intensity of higher-energy modes (IHE). In addition, the relative intensity of the lower-energy modes (ILE) in larger crystals might be exaggerated further due to re-emission (44). It is important to note that the wavelength of the emitted photons (typically 520 to 550 nm) is larger than the nanocrystal dimensions involved in this study (typically 100 to 550 nm). Therefore, the photon energy transfer in nanocrystals is highly localized unlike what occurs in the bulk and microcrystals, which is typically described by a semiclassical light propagation model (35). Critically, this dominant pathway is independent of the surface or strain effects, which cause a minimal energy shift (on the order of 100 meV) as a function of crystal size.
Nanocrystal libraries and devices of halide perovskites
By changing the chemical precursors, AX and PbX2 (different cations, A, and halides, X), a library of halide perovskite nanocrystal arrays can be synthesized and studied. Specifically, nanocrystals of solution-processable halide perovskites MAPbI3, MAPbBr3, MAPbCl3, CsPbI3, and CsPbBr3 were synthesized using PPL and the appropriate precursor inks (Fig. 5, A to E). In addition to these “3D” (three-dimensional) halide perovskites, a layered Ruddlesden-Popper halide perovskite was synthesized in nanocrystal array format with butylammonium bromide (BABr) and PbBr2 as the precursors in the inks (Fig. 5F and fig. S22). SEM reveals the presence of thin steps on the surface of the rectangular nanocrystals, indicative of the targeted 2D layered structure. Critically, AFM identifies single- and double-layer step heights of ~1.3 and ~2.6 nm, consistent with the reported layer thickness in bulk crystals (1.4 nm) and single-layer sheets (1.6 nm) (46). With the exception of nonemissive δ-CsPbI3 that is formed due to thermodynamic limitations at room temperature, all halide perovskite nanocrystals exhibit well-defined PL emission (Fig. 5I). By sequentially patterning MAPbI3 (red), MAPbBr3 (green), and MAPb(Br0.4Cl0.6)3 (blue) nanocrystals on the same substrate, light-emissive RGB pixel arrays with all three colors were synthesized (Fig. 5J and fig. S23). This capability for synthesizing position-defined halide perovskite nanocrystals with controlled emission wavelengths points toward a new way of creating multicolor micropixels, potentially suitable for high-density display technologies.
Last, halide perovskite nanocrystals synthesized by PPL can also be used to prepare photovoltaic devices, including miniaturized solar cells. As a proof of concept, a hole-transporter-free (47) solar cell was constructed by first patterning a single MAPbBr3 nanocrystal on an ITO-coated glass substrate and then connecting its top surface using a Pt/Ir-coated conductive AFM probe (Fig. 6A). An in situ AFM stage was used to illuminate the nanocrystals (455- or 530-nm LED light source) during the experiment (fig. S24). In the dark, no appreciable photocurrent was observed, while illumination under 455-nm LED light (~3.6 mW/cm2) immediately triggered measurable photocurrents across the nanocrystal (Fig. 6B), although significant hysteresis was observed between the forward and backward scans, presumably due to ion migration or conditioning of the probe-crystal contact (48). This photovoltaic response was observed in all four site-isolated nanocrystals studied with an open-circuit voltage (VOC) between 1.06 and 1.21 V (fig. S25 and table S1). Under prolonged light illumination, the nanocrystals showed varying responses due to the unstable contact between the AFM probe and nanocrystal, and significant material degradation was observed as the light intensity was increased (Fig. 6C). Specifically, VOC drops to <0.8 V when the light intensity exceeds 120 mW/cm2, indicating an increase in defect density and highlighting the importance of stabilizing halide perovskite nanocrystals for photovoltaic applications. Similar behavior but with less activity was observed with a 530-nm LED source (fig. S26).