High-efficiency colloidal quantum dot infrared light-emitting diodes via engineering at the supra-nanocrystalline level

Colloidal quantum dot (CQD) light-emitting diodes (LEDs) deliver a compelling performance in the visible, yet infrared CQD LEDs underperform their visible-emitting counterparts, largely due to their low photoluminescence quantum efficiency. Here we employ a ternary blend of CQD thin film that comprises a binary host matrix that serves to electronically passivate as well as to cater for an efficient and balanced carrier supply to the emitting quantum dot species. In doing so, we report infrared PbS CQD LEDs with an external quantum efficiency of ~7.9% and a power conversion efficiency of ~9.3%, thanks to their very low density of trap states, on the order of 1014 cm−3, and very high photoluminescence quantum efficiency in electrically conductive quantum dot solids of more than 60%. When these blend devices operate as solar cells they deliver an open circuit voltage that approaches their radiative limit thanks to the synergistic effect of the reduced trap-state density and the density of state modification in the nanocomposite. PbS quantum dot ternary blends enable the realization of high-efficiency colloidal quantum dot infrared light-emitting diodes

CMOS-compatible and low cost NIR and SWIR LEDs 6,7,8 . In contrast to other highly performant solution processed materials such as polymers and dyes, whose bandgaps are mainly limited in the visible, CQDs offer a unique opportunity as they readily provide access to both the visible 9,10,11,12 and the infrared parts of spectrum 13 . In view of this, several efforts employing core-shell structures 14,15 , inter dot spacing engineering 16,17 , chemical passivation with perovskite 18 and organic-inorganic hybrid 19 approaches have been made to develop efficient CQD infrared-emitting LEDs. The use of core-shell CQD structures 14,15 to increase the PLQE, have reached EQE in excess of 4%. Alternatively, the use of appropriate host matrices has been considered as a means to suppress PLQE quenching in close-packed CQDs due to energy transfer. Initial reports have employed polymer host matrices 20,21,22,23 , yet with limited EQEs mainly due to the polymers' poor electron transport properties. Recently, an alternative matrix has been reported based on perovskite materials epitaxially connected to the CQD emitting species serving both as a chemical passivant of the QD surface and as an efficient carrier transport matrix, leading to EQE of 5.2% and power conversion efficiency (PCE) of 4.9% 18 . PCE in LEDs is defined as the ratio of optical output power over the electrical input power and it is of paramount importance when considering the power consumption of the device.
We posited that instead of relying solely on chemical passivation of the CQD emitting species, the use of a remote charge passivation mechanism induced from an appropriate matrix would be more robust and efficient in reducing trap state density in the CQDs 24,25 . Unlike prior approaches 18,22 , our implementation is based entirely on CQD materials. In doing so, we exploited the advances made in QD solids in terms of mobility and carrier diffusion length thanks to the progress in photovoltaic devices 26,27,28 , in which mobilities and carrier diffusion lengths in excess of ~10 -2 cm 2 V -1 s -1 and 230 nm respectively, have been reported , fulfilling the needs for efficient carrier transport in the typical thinner-than-solar-cells, LED devices.

LED Architecture and Performance
We have considered two LED architectures, one comprising a binary blend of small PbS QDs with large bandgap serving as the carrier supplier for the large PbS QDs with smaller-bandgap that act as the carrier acceptor and emitting species (Fig. 1a). The second case comprises a ternary blend formed by the binary blend with the addition of ZnO nanocrystals (NCs), which serve as a high-bandgap electron-rich transporting medium and that is employed to further balance carrier injection in the active region as well as to further passivate remotely the traps of the PbS QDs 24 (Fig. 1b). The ligand exchange scheme that was employed for the active layers was based on a mixture of zinc iodide and 3-mercapto-propionic-acid (MPA) as it has delivered previously solar cells with long carrier diffusion lengths and high open circuit voltage 29 (see Methods for the details of device fabrication). Both structures also employ a ZnO front layer as an electron injecting, hole-blocking layer and 1, 2-Ethanedithiol (EDT) treated small PbS QD layer on top that facilitates hole injection and electron blocking at the back interface. The thickness of each of the layers used in high performance devices considered in this study are illustrated in the cross-sectional focused ion beam SEM images in Fig. 1c,d for the binary and ternary blends respectively. Typically, the optimal thickness of electron injecting, active and hole injecting layers are around 80 nm, 60 nm and 70 nm respectively.
The TEM images of Fig. 1 illustrate the effective blending of these QD species at the nanoscale and support the nature of the nanocomposite active layer. The corresponding band diagrams of the constituent materials used in these devices are shown in Fig. 1e, taken from UV photoelectron spectroscopy measurements 28,30 (UPS). According to this, both ZnO NCs and small PbS QDs serve as a type-I heterostructure with the emitting large PbS QDs. The small PbS QD matrix forms a marginal type-I heterostructure with the large PbS QD emitters in which the band offset confinement for both electrons and holes is between 0.1-0.2 eV. Based on the band diagram electron transport and injection takes place within the ZnO NC and the small PbS QD matrix given their matched conduction band levels, whereas hole transport is facilitated largely via the small PbS QD matrix. Figure 2a shows the radiance of the binary and ternary blend-based LED devices with applied bias voltage. The control device comprising only large PbS QD as the active layer is also plotted for comparison. All devices showed very high radiance of ~9 W sr -1 m -2 at 3.5 V, which is more than 50% higher compare to the previously reported PbS QD based IR LEDs 16,18 . It is noteworthy that the turn-on voltage for both binary and ternary blend devices is around 0.6 V, i.e. below the bandgap of emission, while the turn-on voltage of the control device was 0.87 V, i.e. matching closely the bandgap of emission (note: we have considered 1 nW radiance as the turn on power). The electroluminescence spectra with different values of voltage bias for the ternary device are shown in Fig. 2b with clear band-edge electroluminescence emission at subband gap voltages. This is the lowest turn-on voltage ever reported for a PbS QD based LED.
Although this below-bandgap turn-on value is not thermodynamically feasible in the absence of multi-carrier processes, it has been reported previously for polymer 31 and QD based LEDs 11 , attributed to Auger assisted charge injection processes. According to this, the low-energy barrier for electron injection results in electron accumulation at the active layer and hole transporting layer (HTL) interface which can fulfil the condition for Auger assisted charge injection 11,31,32,33 . In our case, we attribute this partially to the improvement of mobility and trap passivation with the mixed ligand treatment employed herein as well as the use of bulk heterojunctions. To examine the hypothesis about the role of the ligand passivation we fabricated a ternary blend active layer device using only MPA as the ligand (MPA ligand exchange yields carrier mobility lower than the ZnI2_MPA treatment). The device showed a much higher turn-on voltage (2.5 V) and lower radiance compared to the devices reported in this manuscript (Fig. S2). An additional plausible mechanism can be assigned to Auger assisted recombination in the small PbS QD matrix in which hole transport takes place via midgap delocalized states 34,35 and recombination with electrons transfers the energy to the remaining holes in those states enabling them to move to the valence band and subsequently inject into the emitting QDs. Yet, we consider that further studies are needed to shed light into this interesting effect.
Despite the similar radiance measured across those three LED devices, a large difference in their driving currents was recorded, with the control (single) device yielding very high leakage current that progressively decreases in the binary and ternary blend cases as shown in Fig. 2c.
This has a significant effect in the EQE of the LEDs as shown in nm wavelength. Previously reported best devices showed the peak EQE around 5% covering a similar spectral range 14,15,18 . We attribute this to the trap passivation and the reduction of leakage current in the blend-devices. The high EQE recorded in the binary and ternary blend devices, taken together with the low turn-on voltage results also in unprecedentedly high PCE. Figure 2e plots the PCE of the LED devices from which the ternary blend based device outperforms with a peak PCE as high as 9.3%, a nearly twofold improvement over the previous record 18 . The synergistic use of both PbS QD matrix dots and ZnO NCs has been instrumental in reaching high EQE. Binary control devices comprising PbS QD emitters with ZnO NCs have reached an optimum EQE of 1.8% (Fig. S4).
In course of device optimization and in order to demonstrate the versatility of this approach, we have fabricated and measured ternary devices with optimized blend ratio varying the bandgap of the PbS QD matrix as well as the bandgap of the PbS QD emitters, the results of which are summarized in Table 1. The electroluminescence spectra of various emitting PbS QD bandgaps are shown in Fig. S5a. The variation of the PbS QD matrix bandgap (Table 1 and

Photoluminescence Quantum Efficiency Studies
The origin of this high EQE lies on very high PLQE of the QD films employed and we therefore explored in more depth the role of the host matrix on the PLQE of the emitting QD species. In  Fig. S14, we plot the internal quantum efficiency (IQE) of the LED devices defined as the ratio of photon generated inside the device to the injected charge carriers. The ternary blend devices yield a peak IQE value of 33.3% whereas the binary blend shows IQE of 17.5%. The improvement in EQE and thereby IQE is nearly twofold, i.e. much higher than the improvement in the PLQE. In Table 2 we summarize the different efficiency factors that determine the EQE of an LED according to the following equation 18  transporting layers were around 40 nm, 220 nm and 30 nm respectively and the PV structure followed a typical CQD solar cell 43 . The dark J-V characteristics, plotted in Fig. 4a, provide initial features of suppressed recombination upon blending, as evidenced by the significantly lower reverse current of the blended-layer diodes compared to the single QD-layer diode case.
The J-V curves of those cells under simulated AM1.5 solar illumination is shown in Fig. 4b.
The VOC of the binary device increased to 0.59 V from 0.39 V of the single QD based device, which is a typical VOC value for PbS QD solar cells of the same bandgap 28 . The large VOC deficit (defined as the deficit of VOC from the bandgap voltage) for PbS QD based solar cells has been believed to be a result of significant presence of non-radiative, in-gap traps 44 . The increased VOC (and subsequent decrease in VOC deficit) can therefore stem from reduced trap state density in accordance to the drastically improved PLQE values recorded for the binary blend films (Fig. 3b). The addition of ZnO further improves VOC to a value of 0.69 V for the ternary blend device. This is a notably large value of VOC that approaches the radiative limit 40 considering the fact that the solar cell harnesses photons with energies down to 0.92 eV. We have calculated the radiative VOC limit of the PV devices following standard analysis (Supplementary information S13) 40,41 . Figure  In order to thoroughly examine whether the trap state reduction is the sole responsible for this notable increase in VOC, we performed SCAPS simulations of a single layer device in which we have varied the trap state density across the experimentally obtained range of values.
According to SCAPS the expected VOC increase from single to binary to ternary devices is 40 mV and 100 mV respectively, whereas the observed increase in VOC from single to binary is around 200 mV. SCAPS simulations confirm that upon trap state reduction from single to binary VOC increases but to a lesser extent than what we have experimentally measured, whereas the further increase in VOC from binary to ternary is accounted for trap state reduction.
We therefore sought additional mechanisms at play in the binary blends that may contribute to this high value of Voc. The binary blend devices comprise ~7.5% lower bandgap emitter QDs

Additional information
Supplementary information is available in the online version of the paper. Reprints and permissions information is available online at www.nature.com/reprints.
Correspondence and request for materials should be addressed to G.K.

Competing financial interests
G.K and S.P have filed a patent application related to this work.

Synthesis of PbS QDs:
Schlenk technique was used to synthesize PbS QDs. 830 nm and 940 nm excitonic peak based PbS QDs were synthesized following the standard recipe.

ZnO nanocrystals preparation:
ZnO nanocrystals were prepared following a previously reported method 24  completion of the reaction, the heating source was removed and the solution was allowed to cool down slowly to room temperature. The solution was then centrifuged at 3500 rpm for 5 min. The supernatant was discarded and an equal amount of methanol was added and centrifugation repeated. After three rounds of purification, the NCs were dispersed in a solution of 2% butylamine in chloroform for base layer formation and in 5% butylamine in toluene for ternary blend formation.

LED device preparation:
LEDs were prepared on cleaned ITO coated glass. The electron transporting layer (ZnO) was prepared by spin coating ZnO nanocrystals in chloroform (40 mg mL -1 ) with a spin speed of 4000 rpm. The procedure was repeated once more to have a thicker film of approximately 80 nm. The active emitting layer was grown on top of the ZnO layer. Before mixing, all QDs and ZnO NC solutions were prepared in separate vials with the same concentration (30 mg ml -1 ).
For binary blends, emitter PbS QDs were mixed to donor PbS QDs with different volume ratios.
Ternary blends were formed by mixing ZnO NC solution to the binary blend with different volume ratios. During the film formation, QDs were treated with ZnI2 and MPA mixed ligand as described in our previous report 29S . The mix ligand was prepared by mixing 25 mM ZnI2 in methanol and 0.015% MPA in methanol solutions. The ZnO substrates were covered with 50 μL of QD solutions and spun immediately with 2500 rpm for 15 sec. Then, the spin coater was stopped to add few drops of mixed ligand to treat for 5 sec. After that, the spincoater was started to dry the film which was then washed with few drops of methanol. The procedure was repeated thrice to get an average thickness of 60 nm. The hole transporting layer was formed by using small diameter PbS QDs treated with 0.02% EDT in acetonitrile solution. The back electrode was formed with Au deposition through a pre-patterned shadow mask in thermal evaporator (Nano 36 Kurt J. Lesker) at a base pressure of 10 -6 mbar. The active area for each device is 3.14 mm 2 .

LED performance characterization:
All the devices were fabricated and characterized in ambient air conditions. Current densityvoltage (J-V) characteristics were recorded using a computer-controlled Keithley 2400 source measurement unit. To calculate the EQEs, electroluminescence from the front face of the device was detected using a calibrated Newport 918D-IR-OD3 germanium photodetector connected to Newport 1918-C power meter in parallel to the J-V measurements. A Shadow mask of 3 mm in diameter was placed in front of the device to minimize the waveguide effect from ITO coated glass. Lambertian emission was assumed. The thickness of the glass substrate was considered during the solid angle measurement. The radiance was further verified with a NIST certified 818-IG InGaAs photodetector with calibrated DB15 module by Newport.

PL, PLQE & PL decay measurements:
PL measurements were performed using a Horiba Jobin Yvon

EL measurements:
Spectral EL measurements were also performed using a Horiba Jobin Yvon iHR550 Fluorolog system coupled with a Hamamatsu RS5509-73 liquid-nitrogen cooled photomultiplier tube.
The voltage bias to the device was applied with a Keithley 2400 source measurement unit. The acquired spectra were corrected using the system response factor provided by the manufacturer.

Photovoltaic device preparation and characterizations:
The PV device preparation follows a similar procedure as the LED described above other than the thickness of the respective layers. The ZnO base layer was deposited thinner compared to LED device (~40 nm). The active layer was prepared much thicker in order to absorb sufficient photons (~200-220 nm Thermal admittance spectroscopy (TAS) measurements: The measurements were performed with the PV devices in a Lakeshore four probe cryogenic chamber controlled by a Lakeshore-360 temperature controller. The frequency dependent capacitance was measured with an Agilent B1500 connected with an external capacitance measurement unit. The temperature was varied from 220 K to 320 K to acquire the frequency dependent capacitance variation. The voltage dependent capacitance was measured with the same instrument in order to obtain the value of depletion width and built-in voltage. The detailed data analysis procedure is described in the Supplementary information S11.

TEM measurements:
The bright field (BF) TEM images of the films have been obtained with JEOL JEM-2100 (LaB6 electron gun) transmission electron microscope, operating at 200 kV. The samples were prepared by spin coating the QD solutions onto 300-mesh carbon-coated copper grid at 2500 rpm. Then, the ligand exchange with ZnI2_MPA was performed in line with the aforementioned device fabrication procedure.

Data Availability
The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request    Decrease of reverse saturation current is observed in accordance with the VOC evolution from single to ternary blend devices (a).The VOC of the devices increases from single to binary blend and further to ternary blend (b). c. Radiative VOC limit estimation by calculating luminescence EQE of PV. The solid columns indicate the VOC of the devices and lined columns indicate radiative VOC. Decrease of nonradiative VOC losses observed from single to binary device and further to ternary device. d. EQE of PV devices based on single, binary and ternary QDs. e. Quantification of density of ingap trap states from TAS analysis. Binary blend shows reduced traps with slight decrease of trap energy (Et). Ternary blend shows significant decrease of traps as well as Et as a result of trap passivation. f. Intensity dependent VOC variation of the devices under study with two different excitation wavelengths (637 nm and 1310 nm). 637 nm laser excites both emitter and matrix PbS QDs whereas 1310 nm light excites only the emitter QDs. The change of VOC from single to binary device is a combined effect of trap passivation and DOS reduction whereas the change from binary to ternary is mainly because of trap passivation. Table 1: EQE values of LEDs varying the PbS QD matrix and the PbS QD emitter bandgaps