Intrinsic and Extrinsic Exciton Recombination Pathways in AgInS2 Colloidal Nanocrystals

Dipartimento di Scienza dei Materiali, Università degli studi di Milano-Bicocca, via Roberto Cozzi 55, 20125 Milano, Italy Dipartimento di Energia, Politecnico di Milano, Via Ponzio 34/3, 20133 Milano, Italy Istituto Italiano di Tecnologia, Via Morego 30, 16163 Genova, Italy Glass to Power SpA, Via Fortunato Zeni 8, 38068 Rovereto, Italy Beijing Key Laboratory of Construction-Tailorable Advanced Functional Materials and Green Applications, School of Materials Science & Engineering, Beijing Institute of Technology, Beijing 100081, China IFN-CNR, Piazza Leonardo da Vinci 32, 20133 Milano, Italy National High Magnetic Field Laboratory, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA


Introduction
Colloidal semiconductor nanocrystals (NCs) are a widely investigated class of solution-processable functional materials with growing potential in numerous optoelectronic and photonic technologies spanning from artificial lighting [1,2] and lasers [3][4][5][6] to photovoltaics [7,8], sensing [1,9], and bioimaging [10,11]. Among various NC compositions, ternary I-III-VI 2 NCs, such as CuInS 2 or AgInS 2 NCs, are rapidly emerging as nontoxic alternatives to conventional heavy-metal-based chalcogenides [12,13] and can be synthetized in large quantities via high-throughput, noninjection techniques using inexpensive green precursors [14]. In addition to size-tunable electronic properties [15] and high luminescence quantum efficiency [14,16,17], I-III-VI 2 NCs exhibit distinctive optical properties including broadband optical absorption from the UV to the near-IR spectral regions, spectrally separated from a long-lived photoluminescence by a large Stokes shift which makes them promising candidates as optical markers for time-gated fluorescence bioimaging [18,19] and as reabsorption-free emitters in photon management technologies such as large-area luminescent solar concentrators [2,[20][21][22]. Owing to their typically broad and featureless luminescence spectrum, CuInS 2 NCs have also been proposed as active materials in single-component white LEDs [23,24], enabling one to bypass the limitation of different turn-on voltages and degradation dynamics typically encountered in white-LEDs based on mixtures of NCs with complementary electroluminescence spectra [25,26]. Because of their relevance for applications, numerous studies have been devoted to investigating the fundamental physical processes responsible for the distinctive optical and electronic behaviors of CuInS 2 NCs [7,20,[27][28][29][30], as well as to identify and suppress competitive nonradiative processes affecting the luminescence yield [31]. Based on the strong similarities between the optical and magneto-optical behaviors of off-stoichiometry CuInS 2 NCs and Cu-doped II-VI chalcogenides [32][33][34], in such NCs, the large Stokes shift and long emission lifetime are typically ascribed to the recombination of a photogenerated conduction band (CB) electron with a photohole localized on an intragap defect state associated with Cu + defects [28,35]. Alternatively, a self-trapped exciton model has been proposed where the highest occupied molecular orbital is due to the 3d states of Cu + cations that localize the photohole also in the absence of lattice imperfections [32]. In stoichiometric, defect-free CuInS 2 NCs, on the other hand, the large Stokes shift and long-lived PL have been shown to originate from the detailed structure of the valence band (VB) featuring a high-energy hole sublevel with even-parity separated in energy from a lower-lying odd-parity sublevel [20,36]. As a result of parity selection rules, the absorption edge in stoichiometric CuInS 2 NCs is dominated by the transition coupling the even-parity hole sublevel and the 1S electron level (which features s-type, and even-parity symmetry), while the long-lived and largely Stokes-shifted emission occurs via the partially allowed transition between the CB electron and the odd-parity sublevel where the photohole thermalizes.
Despite their comparably promising optical properties, AgInS 2 NCs have been much less investigated than CuInS 2 NCs [37][38][39][40]. Recently, building upon the theoretical description of the VB structure of CuInS 2 NCs by Efros and coworkers [36], Baimuratov et al. [41] theoretically investigated the origin of the optical behavior of AgInS 2 NCs also by taking into account the contribution of electron-phonon coupling and the additional effect of point defects. To date, however, detailed experimental studies of the photophysical mechanisms in AgInS 2 NCs have not yet been reported, and thus, our understanding of this important class of NCs lags behind that of their copper-based analogues.
In this work, we aim to contribute to this task by investigating the optical properties of AgInS 2 NCs using complementary spectroscopic techniques including continuous wave (cw) and time-resolved photoluminescence (PL), spectro-electrochemistry (SEC), ultrafast transient transmission (TT), and magnetic circular dichroism (MCD) experiments. Optical absorption and TT measurements enable us to resolve two spectral contributions responsible for the broadband absorption profile, while photoluminescence excitation (PLE) measurements reveal that the Stokesshifted PL can also be excited using resonant excitation as much as 550 meV below the dominant absorption edge. This suggests that, similar to CuInS 2 NCs, the PL derives from an optical transition coupling the CB minimum to an odd-parity VB sublevel where the photohole thermalizes. TT dynamics enable us to quantify the hole localization time (1.4 ps), and temperature-dependent MCD measurements corroborate the absence of paramagnetic intragap states involved in the Stokes-shifted PL of ternary I-III-VI 2 chalcogenide NCs. In order to identify the main nonradiative pathways affecting the PL quantum efficiency of our NCs-in view of their application as active materials in optoelectronic devices-we perform spectro-electrochemistry measurements and PL experiments as a function of temperature. The PL behavior under applied electrochemical potential reveals that thermally assisted surface electron trapping is the main nonradiative loss channel, whose suppression via electrochemical passivation of electron traps or by lowering the sample temperature to 3.5 K leads to PL quantum efficiency close to unity. Interestingly, consistent with recent results on defectfree CuInS 2 NCs [20], the PL dynamics below T ∼ 50 K show an almost 100-fold slowing down, at constant PL quantum yield (Φ PL ), thus pointing to a further fine structure of the emissive band-edge exciton state composed of the CB electron and the hole in the odd-parity VB sublevel. Crucially, side-by-side time-resolved PL measurements on Ag-doped CdSe NCs, where the photohole rapidly localizes on the Ag-dopant site, show exclusively the increase of the PL lifetime with decreasing temperature due to gradual suppression of thermally assisted nonradiative decay and no fine structure effects. This indirectly corroborates the picture that in ternary AgInS 2 NCs, the peculiar low-temperature PL kinetics are associated to the fine structure of the lowest excitonic level.
2.1.1. Synthesis of AgInS 2 Nanocrystals. For the synthesis of AgInS 2 NCs was followed a heat-up procedure. A mixture of Ag(OAc) (0.2 mmol), In(OAc) 3 (0.2 mmol), thiourea (0.2 mmol) 3 mL OLA, and 2 mL of DDT was loaded into a three-neck flask and was degassed under vacuum at 110°C for 1 hour. Then, under nitrogen flow, the temperature was raised to 230°C for 15 minutes to let the particles grow. Finally, the reaction was quenched by cooling the solution to room temperature. The NCs were cleaned by precipitation with methanol and redispersion in hexane.

Inductively Coupled Plasma Atomic Emission
Spectroscopy. ICP-AES analysis was performed by dissolving the NCs in HCl/HNO3 3 : 1 (v/v) and using an iCAP 6500 Thermo spectrometer.

Transmission Electron Microscopy.
High-resolution TEM (HR-TEM) and high-angle annular dark field-scanning TEM (HAADF-STEM) imaging was performed on an image-C S -corrected JEOL JEM-2200FS microscope (Schottky emitter), operated at an accelerating voltage of 200 kV.

Powder X-ray Diffraction.
Powder XRD patterns were acquired in Bragg-Brentano geometry with Cu Kα radiation (Panalytical X'Pert Pro powder diffractometer).
2.1.6. Spectroscopic Studies. Absorption spectra of NCs in solution were measured with a Cary 50 UV-Vis spectrophotometer. Steady-state PL measurements were performed by exciting samples at 3.06 eV with ps-pulsed diode lasers (Picoquant LDH-P series, ∼70 ps pulses). The emitted light was dispersed with a spectrometer and detected with a charge-coupled device. Photoluminescence quantum yield measurements were performed in an integrating sphere on NCs deposited onto silica substrates using the same excitation source and detection setup. The same films were mounted in the variable temperature insert of a closed cycle He cryostat for monitoring the dependence of the emission yield with temperature. Transient PL measurements at RT were carried out using the same excitation source at 3.06 eV from a pulsed diode laser. The emitted light was collected with a photomultiplier tube coupled to time-correlated single-photon counting unit (time resolution ∼400 ps). Temperature-dependent PL and trPL measurements were carried out on NC thin films drop-casted on quartz substrates, excited using a frequency tripled pulsed Nd:YAG laser at 3.49 eV with a 150 Hz repetition rate (pulse duration, 5 ns), and mounted inside a cryostat with optical access.
2.1.7. Transient Transmission Spectroscopy. For the ultrafast transient transmission measurements, the laser source was a Ti:sapphire laser with chirped pulse amplification (Coherent LIBRA-HE), which provided 95 fs pulses at 800 nm at a repetition rate of 2 kHz. The excitation pulses at 1.97 and 2.3 eV were obtained by noncollinear optical parametric amplification in a beta-barium borate crystal. Pump pulse duration about 100 fs. The probe beam was a white light supercontinuum generated by focusing a small fraction of the fundamental beam onto a 2 mm thick sapphire plate. After chopping the pump beam at 1 kHz, pump and probe were focused on the sample by means of a lens and a spherical mirror. Pump fluence on the sample position was~11 μJ cm -2 . A computer-controlled optical multichannel analyzer working at the repetition rate of the laser source acquires the map of the differential transmission ΔT/ T = ðT on − T off Þ/T off , as a function of the pump-probe time delay; T on and T off are the probe spectra transmitted by the excited and unperturbed samples, respectively.
The ITO-coated surface was first covered with ZnO NPs (Nanograde, ∼50 nm diameter) to avoid quenching of NC emission by fast charge/energy transfer to the ITO. The ZnO NP layer (∼60 nm thick, measured using a Dektak profilometer) was deposited by dip-coating the glass/ITO substrate into an ethanol suspension of ZnO NPs (2 mg·ml −1 , one dip for 10 s) and annealed at 150°C for 10 min in a nitrogen glovebox. To test the stability of the glass/ITO/ZnO NP substrates during potential scans, we performed control experiments in which we monitored changes in optical absorption spectra for prolonged exposures to negative and positive potentials. The results of these measurements indicate that the substrates are unaffected by either positive or negative EC potentials for exposure times of tens of minutes, which are much longer than the measurement time used in our SEC experiments (∼10 min). The NCs were deposited onto the ZnO NP layer as a few-monolayer-thick film by dip-coating from a dilute toluene solution. The ZnO NP layer used in this study was not treated with crosslinkers, and therefore, it represented a dielectric tunneling barrier of ∼1 V. The introduction of the additional ZnO spacer as well as the presence of insulating surface ligands can also lead to an appreciable attenuation of the actual shift of the FL compared to the nominal applied EC potential. The ITO was connected as a working electrode to the potentiostat (Bio Logic SP-200 Research grade potentiostat/galvanostat), and the film was placed into a quartz cuvette filled with the electrolyte (0.1 M tetrabutylammonium perchlorate (TBAClO 4 ) in propylene carbonate). Ag and Pt wires were used as pseudoreference and counter electrodes, respectively. All potentials reported in this work were measured relative to the quasireference Ag electrode during staircase voltammetry scans (10 s per scan). The film was excited at 3.06 eV with a pspulsed-wave diode laser, and the emitted light was collected with a focusing lens and detected with a charge-coupled device spectrometer.
2.1.9. MCD Measurements. MCD measures the normalized difference in transmission between left-and right-circularly polarized light through films of nanocrystals in the Faraday geometry [43]. Nanocrystal films were mounted in the variable-temperature insert (1.5-300 K) of a 7 T superconducting magnet with direct optical access. Probe light of 3 Energy Material Advances tunable wavelength was derived from a Xenon lamp directed through a spectrometer. The probe light was mechanically chopped at 137 Hz and was modulated between right and left circular polarizations at 50 kHz using a photoelastic modulator. The transmitted light was detected with a silicon avalanche photodiode.

Results
AgInS 2 NCs were synthesized following the heat-up method proposed by Torimoto and coworkers [16] using a metal acetate precursor ratio of 1 : 1 In : Cu and thiourea in a mixture of oleylamine and 1-dodecanthiol solution. Using such a precursor ratio, we obtained stoichiometric AgInS 2 NCs with In : Ag : S ratios of to 1.06 : 1 : 2.04, as confirmed by elemental analysis by inductively coupled plasma atomic emission spectroscopy (ICP-AES). The obtained NCs were of tetragonal crystal structure with radius distribution of 2:3 ± 0:4 nm (Figures 1(a)-1(c)). The optical absorption spectrum of the NCs in toluene (Figure 1(d)) shows the characteristic broadband profile of I-III-VI 2 NCs with an absorption shoulder at 2.25 eV and a low-energy tail extending into the PL peak. The broad (FWHM~0.3 eV) PL spectrum is centered at 1.64 eV corresponding to a Stokes shift as large as 650 meV from the absorption shoulder. The Φ PL at room temperature is 40 ± 5%. Consistent with previous reports [38], the corresponding PL dynamics (Figure 1(e)) are nearly single-exponential-except for a slightly faster initial drop possibly due to charge trapping-with lifetime τ = 820 ns (as extracted from single exponential fitting of the dominant PL decay excluding the initial fast drop). Interestingly, the PLE spectrum collected at the PL maximum resembles nearly perfectly the low-energy absorption tail, and the respective PL spectra excited at progressively lower energy (Figure 1(f)) show no significant difference in shape or energy position with respect to each other or in comparison with the PL spectrum excited well above the absorption shoulder at 3.1 eV. The constant PL linewidth suggests that the exciton thermalizes to the same state below the dominant absorption shoulder, independently from the excitation energy.
More detailed insights into the structure of the absorption spectrum and the carrier dynamics are obtained by ultrafast TT spectroscopy using a spectrally selective pump operating in the low-fluence regime (when the number of photons absorbed per NC per pulse is much less than unity). Upon pumping above the absorption shoulder at 2.3 eV (Figure 2(a)), the differential transmittance (ΔT/T) spectrum exhibits two intense bleaching bands at~2.25 eV and 2.05 eV (indicated as "A" and "B") in good correspondence with the absorption shoulder and with the low-energy tail of the steady-state absorption profile, respectively. The inspection of TT dynamics in Figure 2(b) shows an almost instantaneous photobleaching of the A transition-comparable with the temporal resolution of the experimental setup (~100 fs)-while the B signal grows more slowly with a rise time of 1.4 ps. These results support the VB structure model with the A band corresponding to the parity-allowed optical transition coupling the even-parity VB sublevel and the CB minimum, which also features an even-parity symmetry. On the other hand, the B band is associated with a lowerenergy transition coupling the odd-parity VB sublevel and  Energy Material Advances the CB. This transition exhibits small oscillator strength due to the parity selection rules. The slow photobleaching risetime observed for the B signal (Figure 2(b)) is related to hole dynamics; in particular, pumping above the absorption shoulder, it describes thermalization of the photohole from the higher-lying even-parity hole sublevel to the odd-parity sublevel. Following this picture, pumping directly within the B bleaching band should not change the TT spectrum since promoting electrons from the odd states into the CB inhibits also the A band transition. Furthermore, direct excitation of odd VB sublevel would result in an instantaneous bleaching also for the B signal. To explore this scenario, we selectively pump at 1.97 eV, far below the absorption shoulder. The resulting ΔT/T spectrum is nearly unchanged showing both A and B bands (Figure 2(c)), while bleaching dynamics are now also instantaneous for the B signal (Figure 2(d)). As a result, the overall TT analysis corroborates the VB structure model showing two transitions separated in energy by~0.2 eV that we ascribe to odd-and even-parity VB sublevels, and reveals an efficient photohole thermalization process occurring at the picosecond timescale. We point out that similar TT dynamics have been observed [34] for Ag:CdSe NCs where silver atoms undergo a transient photooxidation from the nonmagnetic Ag + ([Kr]4d 10 ) configuration to Ag 2+ ([Kr]4d 9 ) which behaves as a paramagnetic radiative acceptor centre for a CB electron. Crucially, the unpaired spin of the lone electron in the d-shell of Ag 2+ couples strongly with the band-edge exciton, resulting in the characteristic nonlinear temperature and magnetic field MCD dependence due to sp-d exchange interactions [44]. Therefore, magneto-optical experiments can be used to corroborate the absence (or reveal the presence) of intragap metastable Ag 2+ -like acceptor states also in our AgInS 2 NCs. In Figure 2(f), we report the MCD spectra at increasing temperature from 2.5 K to 20 K at constant magnetic field (H = 6 T). Importantly, opposite to what observed in Ag:CdSe NCs, the magnetic response of our stoichiometric AgInS 2 NCs is temperature independent, same as what found in stoichiometric CuInS 2 NCs [20]. Consistently, the MCD intensity at 2.5 K follows a linear trend with H, thus confirming the absence of sp-d exchange interaction between the semiconductor bands and Ag 2+ lone spins. Therefore, our experimental results support a scenario in which the broad absorption profile is dominated by parity selection rules, while the long-lived PL arises from the parity-forbidden recombination of the 1S e electron into the odd-parity VB state activated by photohole localization. Within this framework, and in accordance with theoretical predictions [41], strong electron-phonon coupling could be responsible for the homogenously broadened PL spectrum shape in AgInS 2 NCs [45], as previously reported in single-particle PL measurements [46][47][48] (FWHM~0.2-0.35 eV). Consistent with [41], electron-phonon coupling also accounts for the addi-tional~450 meV spectral separation between absorption

Energy Material Advances
shoulder and PL peak, leading, together to the~200 meV separation due to the odd-even-parity VB effect to the observed 650 meV Stokes shift. With a view to optimize these NCs for optoelectronic applications, we conducted SEC experiments to obtain a picture of the principal loss channels. Electrochemical measurements are performed as cyclic staircase voltammetry for negative and positive electrochemical potentials (V EC ) against a silver-wire pseudoreference electrode. These measurements can be interpreted in terms of a deliberate modulation of the Fermi level (FL) where negative (positive) V EC is equivalent to raising (lowering) the FL, resulting in a reversible passivation (activation) of surface electron traps associated with undercoordinated surface atoms or dangling bonds [31]. In Figure 3(a), we show the complete set of PL spectra for AgInS 2 NCs under application of negative V EC cyclically scanned between 0 and -2.4 V. Notably, the PL maximum exhibits a 0.18 eV red-shift after drop-casting on indium tin oxide (ITO)/ZnO substrate, ascribable to energy transfer processes in aggregated NC films as previously reported in core-shell CuInS 2 /ZnS NCs [49]. To quantify the effect of raising the FL on the PL intensity, we plot the integrated PL intensity as a function of the applied potential (Figure 3(b)), showing almost no effects at the initial stage of the potential ramp (up to V EC = −1:2 V). This effect is common in SEC experiments using ITO/ZnO substrates, where the ZnO layer represents a dielectric tunnelling barrier resulting in a voltage drop across it [9]. By lowering the working electrode potential, the PL intensity undergoes a sudden growth by~120% at -2.25 V. At even more negative potentials, the PL exhibits a reversible intensity drop likely associated with direct electron injection in the CB resulting in the activation of nonradiative Auger recombination [31,43].
The PL evolution has been monitored during several cycles showing good reproducibility and symmetry by lowering and raising the applied potential, thus indicating that the voltage sweep does not damages the NC surfaces and that the passivation and activation of surface traps occur with similar time kinetics. On the other hand, by lowering the FL (Figure 3(c)), we observe a slight PL dimming (~15%) at 1.8 V that almost completely recovers when returning to V EC = 0 V. The overall SEC behaviour-illustrated in Figure 3(d)-suggests that surface electron trapping (ET) is the dominant loss channel with respect to competitive surface hole trapping, and that its suppression would ultimately lead to Φ PL close to unity. Similar suppression of electron trapping can be achieved growing a large bandgap shell; however, the use of zinc chalcogenides is typically accompanied by Zn diffusion into the NC core which modifies the electronic structure of the material. Hence, a defect-mediated radiative recombination is commonly invoked in alloyed AgInS 2 /ZnS NCs to model their optical behaviour, where the defect's nature and position within the NC volume determine the PL energy and decay rate [50].
Complementary indications about loss channels in AgInS 2 NCs come from temperature-dependent steady state and time-resolved PL experiments. In Figure 4(a), we report a set of representative PL spectra at different temperatures in the range from T = 300 K to T = 3:5 K for AgInS 2 NCs showing monotonic growth of the PL intensity with decreasing T, saturating at Φ PL~1 00% below 50 K (Figure 4(b)) and followed by a slight drop possibly due to nonradiative competitive channels which become competitive with the darkexciton emission at low T. Nearly 100% PL efficiency indicates essentially complete suppression of thermally activated nonemissive channels and that the system has reached its  Energy Material Advances radiative regime. The relative Φ PL increase at cryogenic temperatures is consistent with SEC experiments at negative potentials, suggesting that electron trapping at the NC surface is phonon-assisted. By fitting the integrated PL intensity evolution with T to the function [45,51] I PL ðTÞ = c/ð1 + q * exp ð−E a /k B TÞÞ where c is a scaling factor, k B is the Boltzmann constant, and q is a fitting parameter acting as a weight for the probability of this process, we obtain the trapping activation energy E a~2 9 meV, similarly to previous reports [20]. To further investigate temperature-assisted losses, we perform timeresolved PL measurements, as a function of T, showing gradual lengthening of the PL dynamics with decreasing T (Figure 4(c)). We notice that for T > 50 K, the zero-delay PL intensifies with no significant variation of τ by cooling down the system before reaching Φ PL maximum at 50 K: this behaviour is largely due to suppression of ultrafast charge trapping channels depleting the NC bands faster than our temporal resolution. Strikingly, by lowering the temperature below 50 K, PL kinetics slows down dramatically and becomes double-exponential with an initial fast portion followed by a tens-of-microseconds long component, nearly two orders of magnitude slower than kinetics at 300 K (Figure 4(d)). This behaviour occurring at constant Φ PL is in very good agreement with previous observations for stoichiometric CuInS 2 NCs [20] and suggests a further fine structure of the band edge exciton featuring a higher lying emissive state and a nonemissive state in thermal equilibrium with the environment and responsible for the long PL decay at low T. In analogy to stoichiometric CuInS 2 NCs, both such states belong to the weakly allowed exciton consisting of a CB electron and a hole in the odd-parity VB sublevel. Therefore, we label the long-lived weakly emissive exciton as "dark" (D) and the higher-lying emissive exciton as "grey" (G), to distinguish the current case from the case of CdSe quantum dots, where the emissive state is optically allowed and therefore referred to as "bright." By assuming a Boltzmann distribution of excitons between these two dark and grey states, we model the transition between the slow and fast decay regimes with a three-level scheme (inset of Figure 4(c)), by expressing the radiative decay rate as where k D and k G are the radiative decay rates of dark and grey excitonic states, respectively, k B is the Boltzmann constant, and Δ S being the respective splitting energy. The model yields a Δ S value of 3.2 meV consistent with what found in CuInS 2 NCs of similar size [20]. Importantly, analogous temperature-controlled time-resolved PL measurements on Ag:CdSe NCs (Figures 4(e) and 4(f)) show exclusively the lengthening of PL decays with decreasing T due to suppression of thermally activated nonradiative channels and no fine structure effects, thus corroborating the picture that the observed optical behaviour is direct consequence of the band edge exciton structure of AgInS 2 NCs and not to the involvement of localized defect states.

Discussion
In summary, we combined ultrafast TT and MCD spectroscopies as well as temperature-dependent PL and SEC experiments to investigate the recombination mechanisms in ternary AgInS 2 NCs. Optical absorption and TT measurements reveal two spectral contributions responsible for the broadband absorption profile, in agreement with recent theoretical models. Temperature-dependent MCD experiments indicate the absence of Ag-related intragap acceptor states acting as radiative recombination centers. Time-resolved PL measurements as a function of temperature show a further fine structure of the band-edge exciton that is not present in Ag:CdSe NCs, which we associate with the intrinsic (not trap-mediated) PL nature in AgInS 2 NCs. Finally, temperature-dependent PL and SEC experiments reveal that surface electron trapping is the main nonradiative loss channel that can be suppressed via electrochemical passivation of electron traps or by lowering the temperature to 3.5 K, leading to PL efficiency close to unity.

Data Availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Conflicts of Interest
The authors declare no conflict of interest.