## Introduction

Doping, the intentional introduction of impurity atoms into a host lattice, is an essential process for developing semiconductor materials and devices. Over the past three decades, researchers have been studying how foreign atoms can be controllably incorporated into semiconductor nanocrystals (NCs) in order to tailor their optical and electronic properties1,2,3,4,5. In bulk semiconductors, substitutional dopant atoms with extra valence electrons can introduce shallow donor levels below the conduction band and when thermally ionized can produce n-type semiconductors that contain excess electrons. These electrons (or holes for p-type doping) are available as mobile charge carriers responsible for current transport. In semiconductor NCs, the collective oscillation of confined charge carriers in response to incident radiation can lead to the so-called localized surface plasmon resonance (LSPR)6,7,8,9,10.

The field of plasmonics can benefit greatly from the development of new nanomaterials with tunable carrier concentration and carrier dynamics6,7,8,11,12. Distinct from conventional metal-based plasmonic materials for which electron density is largely fixed and postsynthetic variation is typically not possible, the carrier density and thereby the LSPR energies of semiconductor NCs made of metal oxides13,14,15,16,17,18,19,20,21,22, chalcogenides9,23,24,25,26,27, phosphides28,29, nitrides30, and silicon31 can be synthetically adjusted to cover the near-infrared (NIR) and the mid-infrared regions. The unique plasmonic properties of semiconductor NCs are being harnessed for new applications and technologies including but not limited to surface-enhanced infrared absorption spectroscopy (SEIRAS)32,33, smart windows10, low-loss optical metamaterials11,34, and bioimaging23,35,36. During the past 5 years, impressive progress has been made toward the syntheses of various plasmonic metal-oxide NCs, such as tin-doped In2O3 (ITO)13,22,37, indium-doped CdO (ICO)16,17, aluminum-doped ZnO (AZO) NCs14,38, to name a few. Colloidal synthesis of doped metal-oxide NCs generally involves co-thermolysis of mixed metal precursors, which requires fine-tuning the relative chemical reactivity of precursors to establish reaction kinetics that favors dopant incorporation6,39. The presence of dopant ions in high concentrations can often significantly alter the nucleation and growth kinetics of host NCs, resulting in the loss of NC size and shape control or unsuccessful doping16,40. A more fundamental issue associated with the co-thermolysis method is the lack of control of the spatial distribution of dopants, which has recently been shown to play a vital role in determining the plasmonic response of metal-oxide NCs41. Alternatively, chemical transformation of pre-synthesized metal-oxide NCs with well-defined structural and spectral properties into derivative NCs could provide improved, simultaneous control of the chemical identity and concentration profile of dopants. Among different “conversion chemistry” strategies, cation-exchange transformations have been extensively studied in various semiconductor NCs for preparation of metastable or doped NCs42,43,44,45. However, to the best of our knowledge, cation-exchange has not been exploited in doped metal-oxide NCs to deliberately vary the carrier concentration and thus the LSPR frequency.

In this report, we describe the use of cation-exchange reactions to tailor the elemental composition and the resulting LSPR characteristics of doped metal-oxide NCs. We demonstrate that several p-type impurities, which have been difficult to incorporate by using the conventional co-thermolysis method, can be doped into various n-type metal-oxide NCs giving rise to predictable LSPR peak redshifts. We further show that rationally designed sequential cation-exchange reactions can lead to enhanced n-type doping. Collectively, our work demonstrates that cation-exchange is an important synthetic tool for creating novel plasmonic metal-oxide NCs via postsynthetic modification of NC composition independently of NC’s size, shape, and crystal structure.

## Results

### Synthesis of Cu:ICO NCs via cation-exchange reactions

ICO and ITO NCs were chosen as model systems because they can support an intense infrared LSPR whose peak wavelength and linewidth are tunable by controlling the concentration of different substitutional dopants13,16,22. This LSPR feature provides a simple optical probe for monitoring reaction progression and for gaining mechanistic understanding of cation-exchange reactions on metal-oxide NCs.

ICO NCs were synthesized via thermal decomposition of cadmium acetylacetonate (Cd(acac)2) and indium acetate (In(ac)3) at elevated temperatures (~316 °C)16. 12 nm ICO NCs synthesized with 10% (atomic percent or at.%) indium input were used as the starting NCs for the majority of cation-exchange reactions (Fig. 1b). Typically, 0.2 mL of an ethanolic solution of CuCl2 was added to 1 mL stirred solution of ICO NCs in toluene (10 mg mL−1). The mixture was maintained at 60 °C for an hour after which NCs were isolated by centrifugation. Transmission electron microscopy (TEM) imaging revealed that NCs retained their initial size and shape uniformity after reacting with CuCl2 over the concentration range of 1–200 mM (Fig. 1c–e, 1m and Supplementary Figs. 1 and 2). This observation is further supported by the small-angle X-ray scattering (SAXS) data for different Cu-exchanged ICO (Cu:ICO) NCs (Fig. 1n). The distinctive ringing pattern and the excellent agreement between experimental and simulated SAXS results indicate that NCs are monodisperse at the ensemble level as they appear in TEM micrographs46. High-angle annular dark field (HAADF) scanning TEM (STEM) imaging proved the single-crystalline nature of the Cu:ICO NCs (Fig. 1f). Clear lattice fringes with measured interplanar spacing of about 0.27 nm can be indexed as (111) planes of the rocksalt CdO phase (Fig. 1f). Powder X-ray diffraction (XRD) and selected-area electron diffraction (SAED) patterns confirmed that the crystal structure of ICO NCs was retained after Cu doping at levels as high as 33% (Fig. 1o, Supplementary Fig. 1, Supplementary Table 1). Elemental mapping using electron energy-loss spectroscopy (EELS) in STEM mode revealed that Cu atoms were incorporated into NC lattice, showing insignificant Cu segregation at the NC surface (Fig. 1g–j). The elemental composition of different NCs was further quantified by using energy dispersive X-ray spectroscopy (EDX) performed on NC ensembles. As shown in Fig. 1k, higher Cu contents were detected when greater amounts of CuCl2 were used during reactions, which is accompanied by lower Cd concentrations and almost unchanged indium contents in NCs (Supplementary Table 1).

An interesting question that arises is whether Cu ions replaced indiscriminately the host and the dopant cations (i.e., Cd2+ and In3+ for ICO NCs), or substituted preferentially the host cations. To quantify the ion-selectivity during cation-exchange on pre-doped ICO NCs, the atomic ratios between Cd and In measured from Cu:ICO NCs vs. the concentration ratios between CuCl2 and Cd2+ in the starting ICO NCs (Cu/Cd %) were plotted. It is found that the Cd/In ratio decreases monotonically with increasing amounts of Cu (Fig. 1l). These results imply that Cu ions have a strong preference for replacing Cd2+ over In3+ of ICO NCs, as schematically illustrated in Fig. 1a.

### Optical properties and electronic structure of Cu:ICO NCs

We studied the effect of Cu incorporation on the optical properties of ICO NCs. Pristine ICO NCs (8.6% In) displayed a pronounced LSPR band centered at 2196 nm because of n-type In3+ dopants (Fig. 2a). UV–Vis–NIR spectra showed that as the concentration of CuCl2 increased, the LSPR band of Cu:ICO NCs shifted monotonically to longer wavelengths indicative of gradually decreased free carrier concentration (Fig. 2a, Table 1). A monotonic decrease in the optical direct bandgap energy was also observed at increasing levels of Cu incorporation, providing further evidence for a reduced occupation of low-lying energy levels within the conduction band at reduced carrier concentrations (Fig. 2b).

Because of the isovalency between Cd2+ and Cu2+ ions, substitution of Cd2+ by Cu2+ would result in minimal changes to the electron density of ICO NCs. To elucidate the electronic origin of LSPR redshifts, X-ray photoelectron spectroscopy (XPS) was employed to determine the oxidation state of Cu species in Cu:ICO NCs. As shown in Fig. 2c, two main peaks centered at ~932.8 eV (2p3/2) and ~952.7 eV (2p1/2), respectively, were routinely observed in the high-resolution Cu2p XPS spectra47,48. For reactions at low CuCl2 concentrations (≤25 mM), the lack of shake-up peaks associated with Cu2+ species in the XPS spectra suggested that only Cu+ ions were present in Cu:ICO NCs47,48. The shake-up features appeared for Cu:ICO NCs synthesized under high CuCl2 concentrations (≥75 mM), and deconvolution of the main Cu2p3/2 peak revealed the coexistence of Cu+ and Cu2+ ions (Supplementary Fig. 3, Supplementary Table 2). Meanwhile, XPS spectra of Cd3d and In3d regions remained essentially unchanged, suggesting that reactions between ICO NCs and CuCl2 did not alter the oxidation state of either Cd2+ or In3+ (Supplementary Fig. 4). Electron paramagnetic resonance (EPR) spectroscopy was also used to determine the oxidation state of Cu species within Cu:ICO NCs, given that Cu2+ is paramagnetic ([Ar]3d9) and EPR-active while Cu+ is diamagnetic ([Ar]3d10) and EPR-inactive24,49. Upon reaction with ICO NCs, the characteristic EPR signal from Cu2+ disappeared almost entirely (Supplementary Fig. 5). Collectively, these XPS and EPR results indicate that at low CuCl2 concentrations, Cu2+ ions were completely reduced to Cu+ ions resulting in compensated doping. At high CuCl2 concentrations, excess Cu2+ ions are likely to adsorb onto the surface of Cu:ICO NCs giving rise to the residual Cu2+ signal, although one cannot rule out the possibility of Cu2+/Cd2+ exchange on NC surface.

To establish quantitative correlations between composition and LSPR characteristics for different Cu:ICO NCs, we performed peak fitting of absorption spectra according to the Drude model (Table 1, Supplementary Fig. 6)9,41. The free carrier concentration Ne was found to decrease from 6.29 × 1020 cm−3 for ICO NCs (λinitial = 2196 nm) to as low as 3.80 × 1020 cm−3 for Cu:ICO NCs. The LSPR quality factor (Q-factor), defined as the ratio of the peak energy to the full width at half maximum of the LSPR band, increased slightly at low Cu contents and decreased rapidly at high concentrations of Cu incorporation (Table 1). These results are consistent with the trend observed for the retrieved free carrier damping constant Γ, which shows reduced optical loss at low Cu levels (Table 1). Furthermore, ultraviolet photoelectron spectroscopy (UPS) measurements were performed to determine how Cu doping modifies the electronic structure and shifts the Fermi level of ICO NCs. As shown in Fig. 2d, the secondary electron cut-off shifted gradually to lower binding energies at higher Cu doping levels. As a result, the Fermi level EF has moved from about −3.9 eV with respect to vacuum for initial ICO NCs to about −4.4 eV for Cu:ICO NCs after reacting with 75 mM CuCl2 (Fig. 2e). Meanwhile, the Fermi energy shifted closer to the valence band with higher Cu contents (Supplementary Fig. 7), which is consistent with the interpretation of increasingly compensated n-type ICO NCs upon addition of p-type Cu+ dopants. The Fermi energy values deduced from UPS measurements can be converted into electrochemical potential values relative to the standard hydrogen electrode (SHE), assuming an energy level value of −4.44 eV vs. vacuum for the SHE50. This conversion allows us to compare the redox properties of ICO NCs with various metallic cations. As shown in Supplementary Fig. 8, ICO NCs (8.6% In) with a Fermi level of −3.94 eV have an effective redox potential of −0.50 eV, which is more negative than the standard reduction potential of many cations including the Cu2+ ions. Therefore, it is thermodynamically feasible to reduce Cu2+ to Cu+ with electrons that are close to the Fermi level of ICO NCs. Similar redox reactions between Cu2+ and semiconductor NCs have been reported for CdS and CdSe49,51. However, ICO NCs exhibit a more negative reduction potential than many metal chalcogenide NCs and therefore are expected to be more reducing.

Oxidation of ICO NCs also occurred after their reactions with cations such as Au3+, Au+, and Hg2+, all of which possess a higher standard reduction potential and are therefore more oxidizing than Cu2+. Sizable redshifts in the LSPR frequency were consistently observed (Supplementary Fig. 9). However, SEM–EDX analysis revealed that negligible amounts of these cations became incorporated into the final ICO NCs (Supplementary Figs. 1012, Supplementary Table 3). Instead, Au NCs were formed when either HAuCl4 or AuCl was used (Supplementary Figs. 10 and 11). Formation of white precipitates (presumably Hg2Cl2) were noticed after reacting ICO NCs with HgCl2. Therefore, the LSPR redshifts observed here were primarily caused by the electron transfer from ICO NCs to cations (i.e., Au3+, Au+, and Hg2+) in solution.

Spectral tuning of LSPR via selective cation-exchange reactions remained effective for ICO NCs doped with different concentrations of In3+ (Supplementary Fig. 13). When starting from 16.2% In-doped ICO NCs with an absorption band centered at 0.643 eV, the LSPR energy was lowered by nearly 200 meV upon reaction with CuCl2, which corresponds to nearly 50% decrease in electron density per NC (Supplementary Table 4).

It was observed that NCs had a strong tendency to agglomerate after reacting with a high concentration of CuCl2 (≥75 mM). This is likely due to the loss of Cd(oleate)2 ligands on NC surface. Displacement of the lead oleate ligands by AgNO3 have also been reported for the cation-exchange reaction between PbSe NCs and AgNO3, causing PbSe NCs to lose colloidal stability when exposed to high concentrations of AgNO34. To overcome this challenge, we tested another cation-exchange protocol that uses a mixture of didodecyldimethylammonium bromide (DDAB), dodecylamine (DDA), and CuCl2 as the Cu-precursor3. The reactions were carried out in a single organic phase (e.g., toluene), and Cu:ICO NCs exhibited excellent colloidal stability even after reaction with high concentrations of CuCl2. Reduction of Cu2+ to Cu+ and redshifts in LSPR were also realized (Supplementary Fig. 14). One drawback of using the DDAB–DDA–CuCl2 precursor is NCs appeared to become partially etched following reaction at high CuCl2 concentrations, limiting the range of Cu-doping level that can be studied (Supplementary Fig. 15).

### Kinetic study of cation-exchange reactions between ICO NCs and CuCl2

Two distinctive features were noticed for cation-exchange reactions between ICO NCs and CuCl2. First, the Cu+/Cd2+ exchange in ICO NCs took place much more slowly (tens of minutes) compared to similar exchange reactions taking place on metal chalcogenide NCs such as CdSe (typically within seconds)42,43,44. Second, there existed a strong preference for Cu+ to replace Cd2+ ions. Since both the overall concentration and the spatial distribution of dopants play an important role in controlling the LSPR characteristics of metal-oxide NCs, a thorough characterization of reaction intermediates will provide further mechanistic insight into the kinetics of dopant incorporation during cation-exchange reactions.

Figure 3a presents the UV–Vis–NIR spectra of a series of aliquots extracted from the reaction mixture containing 100 mM CuCl2 (dissolved in ethanol) and ICO NCs (λinitial = 2222 nm). A redshift in LSPR by about 76 meV was attained within 30 s of reaction. The LSPR continued to redshift by another 47 meV over the next 2 h (Table 2). The change in free carrier concentration retrieved from fitting of absorption spectra followed a similar trend, showing an initial sharp decline followed by a steady decrease at later stages of the reaction (Fig. 3b). Moreover, the shake-up peaks attributable to Cu2+ were not seen in the XPS spectra of NCs isolated within 30 min of reaction (Fig. 3c). These XPS data suggested that a rapid reduction (less than 30 s) of Cu2+ to Cu+ had occurred, and resulting Cu+ ions were subsequently incorporated into ICO NCs. A main peak at the kinetic energy of 915.7 eV was observed on the X-ray-induced Auger spectrum of Cu LMM region, which confirms the presence of Cu+, as well as the absence of Cu0 (Supplementary Fig. 16)47,48. The shake-up features of Cu2+ started to appear in the XPS spectra of NCs reacted for more than 1 h (Fig. 3c, Supplementary Fig. 17 and Supplementary Table 5).

The elemental composition of NC intermediates was determined by using SEM–EDX and XPS analyses (Table 2, Supplementary Fig. 18). The Cu content in NCs increased gradually over time with a concomitant decrease in Cd content and a nearly unchanged indium content (Supplementary Fig. 18). XPS is a surface-sensitive technique because of the shallow escape depth of photoelectrons, whereas the sampling depth of SEM–EDX measurement is typically above 1 µm. The Cu concentrations measured by XPS for different NC intermediates were found to be consistently higher than the values obtained from SEM–EDX analysis, indicative of the presence of a Cu-rich near-surface region within Cu:ICO NCs (Fig. 3b). The extent of Cu’s surface enrichment can be estimated by taking the ratio between atomic percent values obtained from XPS and EDX measurements. As depicted in Fig. 3d, the Cu(EDX)/Cu(XPS) ratio increased monotonically during the first 30 min of reaction. We speculate that this period corresponds to continued migration of Cu+ ions from the surface of ICO NCs to their cores as the Cu+/Cd2+ exchange reaction progresses. The Cu(EDX)/Cu(XPS) ratio started to decrease after 30 min of reaction, which was attributed to the adsorption of Cu2+ ions onto NC surface when taking into account the XPS results shown in Fig. 3c. Low-temperature (77 K) EPR experiments were also performed to monitor changes in the oxidation state of copper. EPR tubes containing reaction aliquots were plunged into liquid N2 enabling rapid freezing of extracted solutions. More importantly, this step allowed the cation-exchange reaction between ICO NCs and Cu+ ions to be successfully halted (Supplementary Fig. 19). The integral intensity of Cu2+ EPR signal dropped by 98% within the initial 30 s when 100 mM CuCl2 was used, demonstrating that almost all Cu2+ were reduced on the time scale of several tens of seconds (Fig. 3e).

Taken together, a mechanistic picture has emerged for the reaction between ICO NCs and CuCl2. The reaction can be divided into two distinct stages, as schematically illustrated in Fig. 3f. The first stage is the rapid reduction of Cu2+ by electron-rich ICO NCs driven primarily by the difference between their reduction potentials. This redox reaction lowered the electron density of ICO NCs and produced a redshift in LSPRs. The loss of electrons can induce a net positive charge on the NC surface, which are charge balanced via adsorption of Cl ions. This argument is supported by the XPS results of thoroughly cleaned NC intermediates extracted at various time intervals (Supplementary Fig. 20). A rapid rise in chlorine content was detected for NCs retrieved after 30 s of reaction. The second reaction stage is characterized by the Cu+/Cd2+ exchange process, which continues to reduce the free carrier concentration and shift the LSPR to longer wavelengths (Fig. 3f).

It is important to point out that the Cu+/Cd2+ exchange can also be realized by running the reaction with a Cu+ precursor, such as tetrakis(acetonitrile) copper(I) hexafluorophosphate ([Cu(CH3CN)4]PF6) or DDAB–DDA–CuCl (Supplementary Figs. 21 and 22). Elemental analysis revealed that Cu+ ions replaced preferentially the Cd2+ ions of ICO NCs, leaving the concentration of In3+ ions essentially unvaried (Supplementary Fig. 21b). Starting from ICO NCs with λinitial = 2196 nm, the magnitude of LSPR redshift near the end of reaction was smaller (52 meV with [Cu(CH3CN)4]PF6 and 32 meV with DDAB–DDA–CuCl) than that attainable with CuCl2 (123 meV). These results are consistent with the fact that the step of Cu2+ reduction by ICO NCs is missing when using a Cu+ precursor.

### Cation-exchange reactions in other metal-oxide/cation combinations

Having established the feasibility of using cation-exchange reaction to tune LSPR in ICO NCs, we further evaluated the generality of this approach with other combinations of cations and metal-oxide NCs. Similar to Cu+, Ag+ is also a p-type impurity when substitutionally doped into various metal oxides. Starting from ICO NCs with λinitial = 2170 nm, the LSPR peak became more redshifted upon reacting with increasing concentrations of AgNO3 (Fig. 4a). At low AgNO3 concentrations, no obvious changes in NC’s size and shape were observed based on TEM imaging (Fig. 4b). STEM-EELS elemental mapping suggested that a considerable amount of Ag was incorporated into NCs (Supplementary Fig. 23a–e). Furthermore, SEM–EDX analysis indicated that Ag+ ions had replaced preferentially the Cd2+ ions of ICO NCs (Supplementary Table 6). At high AgNO3 concentrations, reactions between ICO NCs and AgNO3 yielded predominantly heterodimer NCs composed of separate Ag and ICO crystalline domains (Fig. 4c, Supplementary Fig. 23f–j), and yet the concentration of Ag within the ICO domain appeared to be much lower than that of Ag-doped ICO NCs shown in Fig. 4b. The formation of Ag NCs was also evidenced by the appearance of the second LSPR band centered at ~440 nm (Fig. 4a). It is plausible that at high AgNO3 concentrations, the fast reduction kinetics of Ag+ to Ag0 promotes the nucleation and growth of Ag NCs. It is also conceivable, given their high mobility in solids, that Ag+ ions are first incorporated into ICO NC lattice through the Ag+/Cd2+ exchange, and later migrate out to the surface and grow into Ag NCs52. Work is currently underway to gain a deeper mechanistic understanding of how the relative kinetics of cation-exchange and redox reactions dictates the reaction pathways between ICO NCs and Ag+ ions.

We further expanded this study to other types of extrinsically doped metal-oxide NCs. ITO NCs represent another widely studied example of degenerately doped metal-oxide NCs13,15,20,41,53. Reactions between ITO NCs and CuCl2 were performed in a manner analogous to the Cu-exchange reactions on ICO NCs. Typically, 0.2 mL ethanol solution of CuCl2 were added to 1 mL toluene solution of ITO NCs (10 mg mL−1), and the reaction was allowed to proceed at 60 °C for 1 h. As shown in Fig. 4d, the LSPR of 6.5% Sn-doped ITO NCs, initially peaked at 1784 nm, became increasingly red-shifted after reacting with higher accounts of CuCl2. TEM and powder XRD showed that the size, shape, and crystal phase of ITO NCs were well preserved (Supplementary Fig. 24a–d). The lack of shake-up features in the Cu2p XPS spectra further indicated that the reduction of Cu2+ by ITO NCs had taken place and Cu+ species were present within Cu:ITO NCs (Fig. 4e). Meanwhile, the oxidation states of both In and Sn remained unaltered (Supplementary Fig. 24e, f). Notably, SEM–EDX analysis of ITO and Cu:ITO NCs suggested a preferential exchange between Cu+ ions and the In3+ host cations (Supplementary Table 7).

Fe-doped ITO NCs (Fe:ITO) were also synthesized via selective Fe2+/In3+ exchange upon reacting ITO NCs (6.5% Sn) with FeCl3. Both XPS (Fig. 4g) and SEM–EDX measurements (Supplementary Table 8) showed that higher Fe contents were measured in Fe:ITO NCs at higher FeCl3 concentrations, accompanied by lowering in LSPR energy (Fig. 4f). Signals corresponding to both Fe2+ and Fe3+ species were detected in the XPS spectra of Fe:ITO NCs, suggesting that Fe3+ ions were partially reduced by ITO NCs to Fe2+ ions54. However, complications such as spectral overlap between the Sn3p3/2 and the Fe2p region (Fig. 4g) and low signal-to-noise ratio of the Fe3p spectra (Supplementary Fig. 25h) had prevented a reliable quantification of the relative population between Fe2+ and Fe3+ species.

Cation-exchange reactions between host cations and Cu+ ions were also feasible when starting from unintentionally doped metal-oxide NCs, such as ZnO and In2O3 (Supplementary Fig. 26). XPS data suggested that the initial Cu2+ were mostly reduced to Cu+ by metal-oxide NCs (Supplementary Fig. 27). The electrons participating in the redox process likely stem from native defects, especially oxygen vacancies that have been shown to induce n-type behavior in unintentionally doped ZnO55,56 and In2O357. The monotonic decrease in the optical direct bandgap for ZnO and In2O3 NCs after reacting with increasing amounts of CuCl2 is consistent with the interpretation that adding p-type Cu+ dopants into NC lattice can lower the Fermi level and enhance NC absorption at longer wavelengths (Supplementary Fig. 27a, b).

### Tuning LSPR energies of metal-oxide NCs via sequential cation-exchange reactions

Selective Cu+/Cd2+ exchange with ICO NCs allows for the introduction of p-type Cu+ dopants into the CdO lattice while leaving the original n-type In3+ dopants largely intact. We hypothesize that if incorporated Cu+ ions can be selectively replaced by In3+ ions in a subsequent reaction step, the net result would be more In3+ are added to NCs, which can lead to enhanced n-type character and thereby shift the LSPR to even higher frequencies than the starting ICO NCs (Fig. 5a). In fact, such In3+/Cu+ exchange has recently been demonstrated in the synthesis of CuInX2 (X = S, Se) NCs from Cu2X NCs58,59, and the synthesis of phase-pure InP NCs from Cu3−xP NCs60.

To test our hypothesis, ICO NCs with λinitial = 2222 nm were first subjected to the Cu+/Cd2+ exchange reaction, generating Cu:ICO NCs with a redshifted LSPR centered at 2598 nm (Fig. 5b). These Cu:ICO NCs were then allowed to react with In(ac)3 dissolved in trioctylphosphine oxide (TOPO) at a certain temperature for 10 min. The atomic percentage of copper for NCs after reaction at 150 °C reached nearly zero according to SEM–EDX analysis (Fig. 5c, Supplementary Table 9), suggesting that the In3+/Cu+ exchange had taken place. The LSPR peak position blue-shifted from 2598 nm (for Cu:ICO NCs) to 2196 nm (Fig. 5b), which is consistent with the expected increase in carrier concentration as a result of indium incorporation. SEM–EDX analysis revealed that In3+ ions had also replaced some Cd2+ ions during the reaction between In(ac)3 and Cu:ICO NCs (Fig. 5c, Supplementary Table 9). Larger LSPR blue-shifts were attained when the reaction was carried out at higher temperatures (Fig. 5b). The magnitude of the LSPR blue-shift is well correlated with the measured indium content that is largely determined by the extent of In3+/Cd2+ exchange at different temperatures (Fig. 5c, Supplementary Table 9). The use of TOPO as the solvent was found to play a key role in preserving the structural integrity of ICO NCs and in suppressing the formation of the In2O3 phase (Fig. 5d–f). When 1-octadecene (ODE) was used as the solvent, ICO NCs appeared to be corroded with concomitant formation of irregularly shaped In2O3 NCs based on XRD analysis (Supplementary Fig. 28). It has been argued that TOPO can bind strongly to In3+ ions, which might increase the nucleation barrier for In2O3 NCs61. Furthermore, trioctylphosphine (TOP) was found to be necessary to facilitate the In3+/Cu+ and the In3+/Cd2+ exchanges via solvation of Cu+ and Cd2+ ions in solution while retaining structural integrity of individual NCs (Supplementary Fig. 29).

## Discussion

Our results clearly demonstrate that cations including Cu+, Ag+, among others, can be successfully incorporated into various pre-doped n-type metal-oxide NCs via preferential cation-exchange with NC’s host cations. The vast majority of newly added cations are p-type dopants and the process bears close resemblance to compensation doping in bulk semiconductors. Reactions between NCs and Cu2+ are characterized by two distinct stages: the initial rapid reduction of Cu2+ to Cu+ followed by the cation-exchange process between Cu+ and host cations. Both steps cause a reduction in electron density, producing a greater LSPR redshift than that attainable from reactions starting with a Cu+ precursor. The ion-selectivity during cation-exchange reactions can be rationalized by considering the bond dissociation energy (BDE)43,62. Taking ICO NCs as an example, the average BDE value of Cd–O bond (236 ± 84 kJ mol−1) has been determined to be considerably smaller than that of the In–O bond (346 ± 30 kJ mol−1). This difference in BDE suggests that thermodynamically it would be more difficult to break the In–O bond than the Cd–O bond, and the resulting lower energy barrier for removal of Cd2+ vs. In3+ is in good agreement with experimentally observed preference for the Cu+/Cd2+ exchange over the Cu+/In3+ with ICO NCs. A similar argument can be made for ion-selectivity with ITO NCs by considering the even higher BDE value for the Sn–O bond (528 kJ mol−1). Future work to validate this mechanism includes cation-exchange reactions on doped oxide NCs for which the BDE of dopant–oxygen bond is smaller than that of the host–cation–oxygen bond.

Colloidal synthesis of doped NCs often demands carefully balanced reaction conditions to facilitate the incorporation of dopants into host lattice. Although a wide range of compositions for doped plasmonic oxide NCs have become accessible via the most commonly used co-thermolysis method, the preparation of Cu-doped and Ag-doped plasmonic metal-oxide NCs has remained particularly difficult. Taking Cu:ICO NCs as an example, our attempts to directly synthesize these NCs through thermal decomposition of Cd(acac)2 and different Cu precursors had merely produced sub-µm-sized CdO NCs with no signatures of Cu-doping (Supplementary Fig. 30). In contrast, the cation-exchange method developed here offers a facile “conversion chemistry” route to create NCs with previously inaccessible compositions and topologies. The Ag–ICO heterodimer NCs with dual LSPR bands represent another example that further illustrates this point (Fig. 4a, c). Although previous work has shown that a variety of metal–ICO heterodimer NCs can be made by growing ICO NCs onto preformed seed NCs including Au, Pt, Pd, and FePt, Ag NCs do not survive the high temperature at which ICO NCs are formed (~316 °C) and hence only ICO NCs are produced in the presence of added Ag NCs63.

Cation-exchange reactions also enable sophisticated postsynthetic tuning of LSPR wavelength for metal-oxide NCs. We have found that varying either the concentration of dopant ions exposed to preformed metal-oxide NCs or reaction time allows the magnitude of LSPR shifts to be controlled with milli-electron volt precision. Since variations in LSPR peak energy are routinely found for metal-oxide NCs synthesized under nominally identical input of dopants, we argue that the combination of direct thermolysis of mixed precursors and cation-exchange reactions constitutes a powerful toolkit for the targeted synthesis of metal-oxide NCs with desired LSPR characteristics. The ability to readily tune LSPR frequency to match a vibrational mode of interest without compromising the LSPR Q-factor can be beneficial for boosting the sensitivity in SEIRAS.

In summary, we have developed an effective cation-exchange method for the incorporation of various cationic dopants into preformed metal-oxide NCs while preserving their size, shape, and crystal structure. We demonstrate that successful addition of p-type dopants, such as Cu+, Ag+, and Fe2+ into n-type metal-oxide NCs, which may not be feasible using alternative synthetic methods, leads to programmable LSPR redshifts due to dopant compensation. We further show that enhanced n-type doping can be realized via rationally designed sequential cation-exchange reactions mediated by the Cu+ ions, yielding metal-oxide NCs with the ensemble LSPR energy surpassing that of starting NCs. Cation-exchange transformations allow doped metal-oxide NCs synthesized by existing methods to be used as templates to create compositionally tunable NCs that exceed existing performance limits in LSPR wavelength range. The ability to postsynthetically alter the doping profile of individual NCs opens the door to a multitude of new opportunities to deepen understanding of the relationship between local structural details and plasmonic properties of metal-oxide NCs.

## Methods

### Chemicals

Cadmium(II) acetylacetonate (Cd(acac)2, ≥99.9%), zinc stearate (ZnSt2, technical grade), tetrakis(acetonitrile)copper(I) hexafluorophosphate ([Cu(CH3CN)4]PF6, 97%), copper(I) chloride (CuCl, 99%), silver nitrate (AgNO3, 99%), hydrogen tetrachloroaurate trihydrate (HAuCl4·3H2O, 99.9%), gold(I) chloride (AuCl, 99.9%), mercury(II) chloride (HgCl2, 99.5%), iron(III) chloride (FeCl3, 97%), oleic acid (OLAC, 90%), ODE (90%), 1-dodecanol (DDOL, 98%), TOP (97%), oleylamine (OLAM, 70%), tetrachloroethylene (TCE, 99%) were purchased from Sigma-Aldrich. Indium acetate (In(ac)3, 99.99%), copper(II) chloride, (CuCl2, anhydrous, >98%), was purchased from Alfa Aesar. DDAB (98%) and oleyl alcohol (>60%) were purchased from TCI America. DDA (98%) and TOPO (99%) were purchased from Acros Organics. Anhydrous solvents were obtained from a column solvent purification system.

### Synthesis of ICO NCs

ICO NCs were synthesized according to previously reported method under nitrogen atmosphere using standard Schlenk line techniques16. In a typical reaction, 3 mmol of metal precursors (e.g., for 10% ICO, 2.7 mmol of Cd(acac)2, and 0.3 mmol of In(ac)3), 15 mmol of OLAC and 100 mL of ODE were loaded into a 250 mL three-neck flask. After degassing under vacuum for 1 h at 100 °C, the mixture was heated rapidly to reflux (~316 °C). The color of the solution turned dark brown after about 30 min of refluxing, indicating the formation of ICO NCs. The reaction flask was air-cooled to room temperature 10 min after the initial color change, and NCs were isolated by precipitation with isopropanol and centrifugation at 6000 rpm for 5 min. The NCs were re-dispersed in 30 mL of hexane in the presence of 0.2 mL of OLAM and 0.2 mL of OLAC, and centrifuged at 3000 rpm for 3 min to remove large aggregates and Cd NC byproducts. Following another round of isopropanol precipitation and centrifugation, the ICO NCs were finally dispersed in anhydrous toluene (10 mg mL−1). It is worth mentioning that the incorporated In content or In/(In + Cd) atomic ratio for ICO NCs determined by elemental analysis does not precisely follow the input In content within mixed metal precursors. Taking 10% In input as an example, the In/(In + Cd) atomic ratio for ICO NCs from different batches was found to vary in the range from 8.5% to 11.0%.

### Synthesis of ITO and In2O3 NCs

ITO and In2O3 NCs were synthesized according to previously reported method22. In a typical reaction, 1.8 mmol of In(ac)3 and 0.2 mmol of Sn(ac)4 were mixed with 4 mL of OLAC in a 25 mL flask. After degassing under vacuum for 1 h at 100 °C, the mixture was heated at 150 °C for 30 min to promote the formation of metal oleate precursors. In another reaction flask, 26 mL of oleyl alcohol was vacuum dried at 100 °C for 1 h before being heated to 290 °C under N2 atmosphere. Afterwards, the metal precursor solution was added dropwise via a syringe pump (~ 0.35 mL min−1) into the flask containing oleyl alcohol. After completion of injection, the reaction mixture was kept at 290 °C for another 10 min before being cooled to room temperature. A similar procedure was employed for the synthesis of In2O3 NCs using 2 mmol of In(ac)3 as the metal precursor.

### Synthesis of ZnO NCs

ZnO NCs were synthesized according to previously reported method38. In a typical reaction,1.89 g of ZnSt2, 4.8 g of DDOL, 2.65 g of OLAC, and 21 mL of ODE were loaded into a 100 mL three-neck flask. After vacuum degassing at 60 °C for 10 min, the mixture was heated at 140 °C for 30 min to promote complete dissolution of ZnSt2. The solution temperature was subsequently raised to 250 °C at a heating rate of 4 °C min−1 and was kept at 250 °C for 3 h before being cooled down to room temperature.

### Cation-exchange reactions

Cation-exchange reactions of metal-oxide NCs were carried out inside a N2-filled glovebox using anhydrous solvents. Typically, 0.2 mL of ethanolic solutions of the appropriate metal salts, such as CuCl2 with different concentrations (from 1 to 500 mM) were prepared, to which 1 mL toluene solution of metal-oxide NCs (10 mg mL−1) was added. The mixture was then heated to 60 °C under stirring using a hotplate. The reaction was allowed to proceed for 1 h at 60 °C, and was halted by the injection of 4 mL of methanol. The NCs was isolated via centrifugation at 6000 rpm for 3 min and were dissolved in TCE or toluene for further characterization. One exception to this general procedure is methanol was used to dissolve [Cu(CH3CN)4]PF6 because of its poor solubility in ethanol.

For Cu-exchange reactions that involves DDAB–DDA, 2 mmol of CuCl2 or CuCl together with 1.2 g of DDAB and 0.8 g of DDA were dissolved in 10 mL of toluene at 80 °C under stirring. The cation-exchange process was performed by adding a certain amount of the DDAB–DDA–CuCl2 (or CuCl) solution into 1 mL of toluene solution of ICO NCs (20 mg mL−1). The reaction mixture was heated at 60 °C under stirring for 1 h after which NCs were isolated via centrifugation.

### Sequential cation-exchange reactions with ICO NCs

Cu:ICO NCs were first synthesized via a cation-exchange reaction between ICO NCs and the DDAB–DDA–CuCl2 precursor, as detailed in the last section. The Cu:ICO NCs were isolated by methanol precipitation followed by centrifugation and were re-dispersed in 2 mL of TOP. In a separate 25 mL three-neck flask, indium precursor was prepared by dissolving 0.2 mmol of In(ac)3, 0.5 mmol of OLAC (0.16 mL) in 5 g of TOPO at 120 °C. After degassing under vacuum for an hour, the solution mixture was heated to 150 °C under nitrogen and Cu:ICO NC solution was rapidly injected. The reaction was allowed to proceed for another 10 min at a given temperature (150, 200, 250, 300 °C). Afterwards, NCs were isolated by isopropanol precipitation and centrifugation at 6000 rpm for 5 min and were dissolved in TCE for further study.

### Drude fitting of LSPR absorbance

Drude fitting of LSPR peaks in the UV–Vis–NIR spectra was performed by using a custom written Matlab script. The absorbance (A) of NC solutions is determined by the Beer−Lambert law53:

$$A = \frac{{N\sigma _{\mathrm{A}}l}}{{{\mathrm{ln}}{\mathrm{(10)}}}}$$
(1)

where N is the NC concentration in solution, and l is the path length of the cuvette. According to Mie theory25, the absorption cross section (σA) of spherical NCs can be calculated as:

$$\sigma _{\mathrm{A}} = 4{\mathrm{\pi }}{\mathbf{k}}r^3{\mathrm {Imag}}\left( {\frac{{\varepsilon (\omega ) - \varepsilon _{\mathrm{m}}}}{{\varepsilon (\omega ) + 2\varepsilon _{\mathrm{m}}}}} \right)$$
(2)

where k is the wave vector of the incident light and $${\mathbf{k}} = \sqrt {\varepsilon _{\mathrm{m}}} \omega /c$$, r is the NC radius, ε(ω) is the dielectric function of NCs, and εm is the dielectric constant of the medium (for TCE, εm = 2.268). The frequency-dependent dielectric function is given by:

$$\varepsilon _{(\omega )} = \varepsilon _\infty - \frac{{{\omega _{\mathrm{p}}}^2}}{{(\omega ^2 + {\mathrm{i}}\omega \varGamma )}}$$
(3)

where ε is the high-frequency dielectric constant (for ICO, ε = 5.5)34, ωp is the bulk plasma frequency, and Γ is the free carrier damping constant. By fitting the LSPR absorption spectrum with the Drude model, ωp and Γ can be retrieved. The free carrier concentration Ne can then be calculated using the formula9,41:

$$\omega _{\mathrm{p}} = \sqrt {\frac{{N_{\mathrm{e}}e^2}}{{m^ \ast \varepsilon _0}}}$$
(4)

where e is the elementary charge, m* is the effective mass of electrons (for ICO, m* = 0.27m0)64, and ε0 is the vacuum permittivity.

### Structural and optical characterization

Low-magnification TEM images and SAED patterns were acquired on a JEOL JEM 1400 plus microscope operating at 120 kV with a LaB6 filament. TEM samples were prepared by drop-casting ~10 µL of NC solution onto a 300-mesh carbon-coated copper grid (Ted Pella). STEM-HAADF images (spatial resolution of ~1 Å) and STEM-EELS elemental maps were acquired on an aberration-corrected Hitachi HD 2700C STEM microscope equipped with a parallel EELS detector (Gatan Enfina-ER). Samples were prepared using holey 400-mesh Cu grids coated with ultrathin carbon films (Ted Pella). The STEM-EELS mapping was performed at a collection angle of 26.7 mrad with a dispersion of 1.25 eV per channel. The spectrum images are 72 by 72 pixels with a pixel size of 5.6 Å. EDX analysis was conducted with an X-Max 50 mm2 silicon drift detector (Oxford Instruments) interfaced either to a Carl Zeiss Auriga SEM or to a JEOL JEM 1400 plus TEM via the AZtec software package. For EDX analysis, NCs were thoroughly cleaned with the precipitation–dissolution procedure for 4–6 times before being deposited onto a silicon substrate (for SEM–EDX) or a TEM grid (for TEM–EDX).

Solution-phase UV–Vis–NIR absorption spectra were recorded on a Varian Cary 5000 spectrometer. Fourier-transform infrared spectroscopy (FTIR) spectra were recorded on a Bruker Vertex 70V FTIR spectrometer in transmission mode using a Harrick demountable liquid cell. Determination of the cadmium content for ICO NCs was carried out on an Agilent 7700 ICP-MS. Samples for ICP-MS measurements were prepared by digesting NCs with aqua regia followed by dilution with ultrapure water. Powder XRD patterns were collected on a PANalytical Empyrean diffractometer with a copper target. SAXS experiments were performed at beamline station 12-ID-C of Advanced Photon Source (Argonne National Laboratory, Argonne, IL) with the incident X-ray energy of 12 keV (λ = 1.0332 Å). SAXS data was collected via a 2D CCD detector at a sample-to-detector distance of 2 m with exposure times ranging from 0.01 to 1 s. Analysis and modeling of SAXS data were performed by using the Irena software package65. Models were fit using the size distribution model with the lognormal distribution. SAXS data for NCs reacted with high concentrations of CuCl2 (75 and 100 mM) showed a slight distortion in the low q range below 0.04 Å−1, indicating NC aggregation46. To compensate for the effects caused by particle aggregation, models were calculated using interferences as the structure factor46. The structure factor parameter of 156 (141) for 75 mM (100 mM) CuCl2-reacted samples was used to describe the relative positions between NCs.

### XPS and UPS

XPS analysis was performed on a PHI 5000 VersaProbeTM II instrument equipped with a focused monochromatic Al K(alpha) source. The instrument base pressure was about 8 × 10–10 Torr. A beam size of about 200 μm and an X-ray power of 50 W at 15 keV was used for all experiments. The C 1s peak calibrated at 284.4 eV was used as an internal standard for the binding energy scale. The PHI’s dual beam charge neutralization system was used for all XPS measurements. UPS spectra were recorded with a He(I) and He(II) excitation source (21.2 and 40.8 eV, respectively) and a pass energy of  0.58 eV in the UHV chamber of the XPS instrument. NC samples were drop casted onto a piece of gold foil for UPS testing.

### EPR spectroscopy

EPR spectra were acquired on a Bruker EMX X-band EPR spectrometer with an ER 4119 HS cavity (9.840 GHz at room temperature) using 100 kHz field modulation (modulation amplitude: 6 G). A 150 mL Suprasil offset liquid nitrogen dewar flask (Wilmad-LabGlass) was used for low-temperature measurements. Individual EPR tubes were filled with ~0.2 mL of solution inside a glovebox and were placed at the same position of the resonant cavity for EPR spectral acquisition.