Cadmium nitrate hexahydrate (Cd(NO₃)₂·6H₂O, 99.99%), cadmium acetate dihydrate (Cd(CH₃COO)₂·2H₂O, 98.0+%), anhydrous acetonitrile (CH₃CN, 99.8%), anhydrous N,N-dimethylformamide (DMF, 99.8%), chloroform (CHCl₃, 99.8%), diethyl ether ((CH₃CH₂)₂O, 99.0+%), technical grade diphenyl ether ((C₆H₅)₂O), tetramethylammonium chloride (C₄H₁₂NCl), thiophenol (C₆H₅SH, 97%) and triethylamine ((C₂H₅)₃N, 98%) and were purchased from Sigma-Aldrich. 2-hexyldecanoic acid (C₁₆H₃₂O₂, 98.0+%) was purchased from TCI America. Selenium shot (99.99%) and powder (99.5%), and tri-n-octylphosphine (TOP, 97%) were purchased from Strem Chemicals. Technical grade methanol was purchased from Fischer. Anhydrous and air-sensitive solvents were stored in a glove box. Cadmium acetate was recrystallized from hot acetic acid, while all chemicals were used without further purification.

Precursors were injected through the microfluidic reactor system using a PHD 2000 syringe pump from Harvard Apparatus. The reagents were first mixed through PFA high-pressure tubing (0.020″ ID) that was connected to polyimide tubing (0.024″ ID) wrapped around one (single-stage synthesis) or two (dual-stage synthesis) toroidal aluminum heating block(s). The polyimide reaction tubing includes both the nucleation stage (30″, 240 µL volume) and a growth stage (16″, 80 µL volume). The material was chosen due to its thermal stability (−269°C → 400 °C according to the manufacturer, Nordson Medical, PN# 141–0036). Each metal block was machined in-house; the temperature was set using a controller that regulated power coupling from a variable transformer to a TUTCO cartridge heater. Tubing connectors were PEEK/stainless steel collect-lock system ferrule, ETFE flat-bottom fittings, PEEK Y-connectors, polypropylene union body connectors, and ETFE shut-off valves from IDEX Health and Science and Nordson Medical. Plastic Luer-lock sterile syringes were purchased from Norm-Ject.

The organometallic clusters used in this study have been prepared using literature procedures. Characterization data are provided in the Supplementary Material, Supplementary Figure S1.

(NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] clusters (10 mg) were dissolved in 10 mL DMF. Samples were monitored by UV/Vis spectroscopy for ∼7 h to quantify aggregation into magic size (CdSe)₃₂. Another portion was applied into the microfluidic reaction system via PFA (0.03″ ID, 16″ length) tubing at a 3 mL/h flow rate over a 25°C → 100°C before collection. All samples were diluted 4× before optical characterization, see Supplementary Figure S2.

1M TOPSe was prepared in a glove box by mixing selenium shot (0.7896 g, 10 mmol) in 10 mL of TOP until the Se dissolved (∼4 h). Cd(2-hexyldecaonate)₂ was prepared by dissolving cadmium acetate (1.0 mmol) and 2-hexyldecanoic acid (5 mmol) in 10 mL diphenyl ether in a three-neck round bottom flask under an inert gas using a Schlenk line. The reaction mixture was stirred under vacuum at 100°C for 2 h. The flask was backfilled with N₂ gas and heated to 250°C until the presence of a yellow color was detected. The mixture was cooled slowly to room temperature.

Cadmium hexyldecanoate and TOPSe were added to diphenyl ether to create 0.1 M solutions that were degassed and loaded into individual syringes. Various cluster seeds (0 µmol → 4 µmol) were dissolved in anhydrous DMF (0 µL → 100 µL) and were added to the cadmium precursor solution. The flow rates were set to 2 mL → 20 mL per hour. Nucleation and growth temperatures varied over 100°C → 200°C and 150°C → 220°C, respectively. The products were collected in hexane for characterization.

Absorption spectra (optical and IR) were collected with a Bio 300 Cary spectrophotometer and a Thermo Scientific Nicolet iS 5, respectively. Fluorescence spectra were recorded on a Horiba Jobin Yvon FluoroMax-4 spectrofluorometer using 400 nm excitation, while time resolved PL was characterized with an Edinburgh FLS1000 PL spectrometer. ¹H NMR spectra were characterized on a Bruker DPX 400 MHz. PXRD diffractograms were collected using a D8 Advance ECO Bruker XRD diffractometer with Cu Kα (λ = 1.54056 Å) radiation. The sizes and concentrations of CdSe QD products were estimated from previously published regressions (Yu et al., 2003; Jasieniak et al., 2009). The regression of Jasieniak et al. was used for large bandgap QDs (<450 nm) due to greater accuracy when applied to smaller CdSe nanoparticles. Such analyses are only possible if a well-defined first exciton absorption feature is observed.

The cluster seed approach was applied to the synthesis of CdSe QDs in the microfluidic system shown in Figure 1 to evaluate the dependence on product quality (stoichiometry, average size, and homogeneity) on the reaction parameter space (temperature, time, and concentration). QD sizes and concentrations were estimated from absorption spectra using previously published regressions. The main goal of synthesizing QDs at low temperatures (100°C) in a single-stage microfluidic reactor was successful as shown in Figure 1A, where the presence of (NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] resulted in products with typical CdSe nanoparticle absorption spectra. The first absorption feature at 389 nm reveals that the QDs are small, estimated to be between 1.4 nm and 1.5 nm. Larger 1.9 nm diameter products are prepared at 120°C as shown in Figure 1B. Quantum dot features are absent with no catalysts are used, which was expected given the low temperatures employed.

Figure 2 shows the normalized absorption spectra of cluster seeded CdSe QDs as a function of temperature in a single-stage reactor, from 100°C → 200°C. The red shift in absorptions (401 nm → 525 nm) is due to an increase in size from 1.5 nm → 2.8 nm (see also Supplementary Figure S3 and Supplementary Table S1). This indicates that the QDs both nucleate and grow despite the modest residency time in the single-stage reactor. Furthermore, the size distribution appears sub-optimal and does not improve at higher temperatures, suggesting that the QDs are not experiencing focused growth. The emission was also found to be broad; however this is not directly indicative of the size distribution as the photoluminescence mostly originates from deep trap emission as shown in Supplementary Figure S3. An attempt was made to enhance monodispersity by synthesis through a temperature gradient from low to high; note the reversal from convention. CdSe QDs were nucleated at 120°C in a first stage reactor, producing a modest size product (1.6 nm diameter) with a first absorption feature at 419 nm (Supplementary Figure S3). These materials were connected to a second reactor to grow further over a 120°C → 180°C temperature range. New QDs were not nucleated regardless of the temperature as the concentrations of products were invariant. Figure 3 shows that the first absorption features red-shifted from 424 nm → 478 nm due to the increase in QD size from 1.7 nm → 2.2 nm (Supplementary Table S2) as the second stage reactor temperature was increased. While conventional wisdom suggests these are good conditions for focused growth, in fact, there was no noticeable improvement in monodispersity as revealed by the absence of sharp exciton features in Figure 3. Furthermore, adding a growth stage does not provide the same level of size tunability compared to that realized by simply varying the first stage nucleating reactor temperature (Figure 2).

These results demonstrate that the cluster seed method enables the synthesis of CdSe QDs in a microfluidic system at significantly lower temperatures than in conventional rapid injection and in other examples of microfluidic synthesis. To provide contrast, the synthesis was examined under identical reaction conditions without clusters; characterization data are shown in Supplementary Figures S4, S5 in the Supplementary Material. A minimum reactor temperature of 150°C was necessary to observe some absorption, although definitive excitonic features are not evident from reactor temperatures less than 180°C. When the temperature was high enough for spontaneous nucleation (>150°C), the nanomaterials were larger than that observed from cluster seeded synthesis albeit at a lower concentration. This is likely due to the formation of a smaller number of nuclei compared to the quantity of clusters added, resulting in a higher precursor-to-QD ratio and larger products. Regardless of the temperature, the absorption features are indicative of materials with poor size distributions to the extent that characterization of the bandgap is problematic. To summarize, the presence of cluster seeds has a profound effect on the size, quantity, and quality of products synthesized in a microfluidic system. And although cluster seeded QDs do not have ideal monodispersity, the results are superior to preparations without cluster catalysts.

Seeding was investigated using (NMe₄)₄[Zn₁₀Se₄(SPh)₁₆] to study the flexibility of the method concerning the identity of the cluster. The absorption spectrum of the materials prepared using the zinc selenide cluster is shown in Figure 4. The results reveal successful CdSe QD nucleation; furthermore, the QD size as characterized by the first absorption feature (462 nm) is nearly identical to that realized from the use of (NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] seeds (460 nm, Supplementary Figure S4B). However, the excitonic features are not as distinct, suggesting less homogeneity.

Cluster concentrations were varied over a ∼70 μM → 350 μM range in a single-stage reactor at two temperatures (100°C and 120°C) to study the effect on QD product properties. The effect of the addition of clusters is to produce smaller QDs as evident from product spectra (Figure 5), where it can be seen that the absorptions blue shift with increasing cluster concentration regardless of the temperature. This is consistent with our previous proposal that the application of too many nucleating seeds lowers the precursor activity such that the ability of the products to gain additional mass is impaired (Jawaid et al., 2013). These data also reveal a linear increase in absorptivity and a corresponding rise in the number of products as a function of cluster concentration (Supplementary Figure S6). It is tempting to ascribe the effect as akin to simply adding more reagents; however, the mass range of (NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] added varies from 5 mg → 20 mg in these experiments (Supplementary Table S3). This is too low to account for a near tripling of the product yields as seen in a regression of product vs. cluster seed concentrations shown in Supplementary Figure S6. What is perplexing in these data is the fact that the regression in Supplementary Figure S6 has a non-zero y-intercept, which means that a minimum ∼55 μM catalyst concentration is necessary to observe product seeding. This threshold behavior suggests a loss of catalytic activity that can occur due to a variety of reasons including aggregation of the clusters as discussed in Section 3.3. Poisoning of the catalysts may also play a role, (Norrish and Russell, 1947; Kumar et al., 2014), especially by oxygen as we cannot vigorously prevent O₂ diffusion through the plastic components of the microfluidic reactor.

While the non-zero y-intercept of the regression of Supplementary Figure S6 can be attributed to sensible factors that may be addressed with additional engineering controls, the fact that the slope is less than 1:1 is problematic. This is because the number of clusters and the number of QDs produced should equate in a seeded mechanism, as was observed in previous studies (Jawaid et al., 2013). An examination of all the data collected in this study reveals that, while the products are responsive to seed concentration, the effect is non-stoichiometric (generally the product concentrations are ∼15% of that of the cluster). This could be the result of the limited synthesis time (<7 min), and to investigate the concentration of QD as a function of flow rate was measured (Supplementary Figure S7). These data reveal that increasing the residency time of precursors in a single-stage reactor enhanced the product population by ∼50%. Other potential contributing factors include Ostwald ripening which consumes the smaller QD populations. That the products are ripening is evident from the absorption spectra that reveal sub-optimal size distributions. While these issues are problematic, they cannot fully account for the discrepancy between catalysts and products. Further confounding the issue is the fact that one dataset was collected that revealed a near quantitative agreement between product and cluster concentration in one of our temperature-dependent studies, see Supplementary Figures S8, S9. The correlations between reagent and product were equal up to a reactor temperature of 200°C, whereupon the number of QDs rose above the line as seen in Supplementary Figure S8. Given the observation of spontaneous nucleation >150°C for unseeded growth, a simple explanation is that both spontaneous and seeded nucleation events occur and the resulting QD concentration at 200°C is a sum of both.

These issues are likely resolved by the ability of (NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] clusters to act as both seeds and reagents given that these species have been used as precursors for CdSe QDs albeit at much higher temperatures (Cumberland et al., 2002). As a result, the time and temperature dependence to the evolution of the neat clusters in solution was investigated, which revealed a rich cache of chemical dynamics as discussed below.

The inconsistent yields on a per catalyst basis prompted an investigation into the mechanism of cluster seeding. Our working hypothesis is that partial decomposition via ligand dissociation of the cluster instigates dynamics that result in nucleation of a single QD; furthermore, this must occur at a temperature lower than that necessary for spontaneous nucleation. This mechanism was proposed earlier to describe the evolution of (NMe₄)₄[Cd₁₀S₄(SPh)₁₆] into higher order clusters and (NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] (acting as reagents) into CdSe QDs by Bendova and Cumberland, respectively (Bendova et al., 2010; Cumberland et al., 2002). To investigate, the changes in the absorption of (NMe₄)₄[Cd₁₀Se₄(SPh)₁₆] clusters in DMF were measured over time with and without thermolysis in the microfluidic system over a 25°C → 140°C temperature range. It was found that a ∼400 nm feature slowly appears at room temperature as shown in Supplementary Figure S2A. This is likely magic sized (CdSe)₃₂ (Landes et al., 2001). This transformation eventually ceases; however, at temperatures >50°C additional conversion is observed as evident from the growth of the ∼400 nm absorption as shown in Supplementary Figure S2B. Past 100°C the absorptions’ red shifting and the loss of less well-defined excitonic features can be attributed to Ostwald ripening.

In most preparations some time would elapse before the clusters were fully dissolved in DMF. As a result we attribute the issues with the observation of a consistent 1:1 QD to catalyst ratio to the cluster’s propensity to aggregate before injection into the flow system. In the future, we hope to minimize this by enhancing the cluster solubility, perhaps through variation of the counterion. And while these results may tempt the employment of clusters with greater stability, this may be counterproductive as we hypothesize that ligand displacement of a ligand, or perhaps via the loss of the apical [Cd(SPh)₃]^–, is essential to the cluster seeding process; further investigations will examine this potential.