Main

For most noble-metal NPs, twinning planar defects are difficult to avoid during the nucleation and growth process because of lower surface and volume energies, thus leading to frequently observed thermodynamically stable disordered multiply twinned (MT) structures (manifesting as a decahedron five-fold twin, an icosahedron twin or more complex twin configurations depending on kinetic factors over the growth)14,15,16. Because of different atomic energy sites on the crystallographic facets, the chemical activity of these MT-NPs shows a substantial difference with respect to their single-crystalline (SC) counterparts, or a slight difference among various twin configurations. This could alternatively provide a general synthetic route to allow for chemical activity to direct the nucleation and growth of metal NPs with desired crystallinity17. We show that such control can be enabled by integrating different functional groups into a metal molecular precursor to manoeuvre the chemical environment in a facile one-pot organic single-phase synthesis, and we use silver NPs as an example because of their broad applications as well as general applicability.

We start with air-stable silver phosphine complexes ((PPh3)3Ag-R, where R represents different functional groups) (Fig. 1a) as silver precursor, and amine molecules serving as both reducing agent and NP-capping ligands. Figure 1b and c shows typical transmission electron microscopy (TEM) images of silver NPs with an average diameter of 10.5±0.4 nm synthesized from precursors 1 (R=−NO3) and 2 (R=−Cl), respectively, without subsequent size selection. The difference between these two images can be immediately addressed by the clear inhomogeneous features developed in nearly 100% of NPs presented in Fig. 1b, which can be attributed to the presence of twinning defects16, whereas 100% of NPs in Fig. 1c are homogeneous. This distinct crystallinity is further confirmed by high-resolution TEM (HRTEM) and single-particle electron diffraction (SPED) characterizations, revealing that NPs in Fig. 1c possess a perfect atomic lattice as well as electron diffraction (Fig. 2a, left), whereas the NPs in Fig. 1b show disordered multiple domains and a distorted diffraction pattern (Fig. 2b, left). For both MT- and SC-NPs, their size and size distribution show a strong dependence on reaction conditions such as temperature, concentration of reactants and reaction time, and NPs with size between 8 and 20 nm can be achieved with narrow dispersivity (≤5%) as well as well-defined nanocrystallinity (see Supplementary Information, Figs S1–S4).

Figure 1: Controlled synthesis of silver NPs with well-defined crystallinity.
figure 1

a, Molecular structure of (PPh3)3Ag-R. b, TEM image of 10.5 nm MT-NPs synthesized from precursor 1. The sample is obtained 120 min after the injection of 3.6 mmol oleylamine into 0.19 mmol precursor 1 in 19 ml of o-dichlorobenzene at 140 ∘C. c, TEM image of 10.5 nm SC-NPs synthesized from precursor 2. The sample is obtained 240 min after the injection of 3.6 mmol oleylamine into 0.19 mmol precursor 2 in 19 ml of o-dichlorobenzene at 125 ∘C. d–f, Time evolution of SC-NP growth: 3 min (d) (the inset shows coexistence of small SC (right) and MT (left) clusters at this stage); 60 min (e); 480 min (f). The best size distribution for this synthesis occurs at 240 min, presented in c. Scale bar of large-scale TEM images: 20 nm. Scale bar of inset images of d: 1 nm.

Full size image
Figure 2: Chemical transformation of 10.5 nm silver SC- and MT-NPs to Ag2Se nanostructures at 45 ∘C in o-dichlorobenzene.
figure 2

a, Typical HRTEM image (left) and SPED (lower right inset) of SC-NP and schematic diagram of atom diffusion paths (middle); typical HRTEM image (right) and SPED (lower left inset) of SC-Ag2Se NPs with hollow core. b, Typical HRTEM image (left) and SPED (lower right inset) of MT-NP and schematic diagram of different atom diffusion paths (middle); typical HRTEM image (right) and SPED (lower left inset) of solid SC-Ag2Se NPs. Blue and orange spheres represent silver and selenium atoms, respectively. Yellow and red arrows represent outward silver diffusion and inward selenium diffusion through the lattice, respectively. The dark grey spheres (behind some of the orange spheres) highlight silver atoms in the twinning boundaries, providing another fast atom diffusion path for both silver and selenium atoms (white arrows). For the silver SC-NP, the SPED pattern is taken along [110] and indexed to face-centred cubic silver (space group Fm3m), JCPDF no. 04-0783; for Ag2Se, the SPED patterns of hollow and solid NPs are taken along [131] and [111], respectively. Both are indexed to orthorhombic Ag2Se (space group P212121), JCPDF no. 71-2410. Scale bar of HRTEM images: 5 nm.

Full size image

To elucidate the selective mechanism of nanocrystallinity by functional groups of the precursor, we characterize the time evolution of growth of both SC-NPs (Fig. 1c–f) and MT-NPs (see Supplementary Information, Fig. S5). We find that the growth of MT-NPs with precursor 1 follows the conventional synthetic scheme1,9,12, which involves a rapid nucleation and growth event from reduced silver monomers, followed by a self-focusing and slower Ostwald-ripening process. However, when R=−NO3 is substituted by–Cl, precursor 2 not only serves as a silver source but also provides abundant chloride ions after reduction, which was shown to have a dramatic effect on the etching of metal twinning nuclei in the oxidative environment17. Therefore, the existence of chloride ions in our synthesis efficiently inhibits the ongoing growth of small MT-clusters by coordination at the high-energy atom sites provided by twinning defects, and only nuclei without twinning defects can continue to grow and participate in the self-focusing process, ultimately yielding monodisperse SC-NPs. This mechanism is further supported by several experimental observations. We first observe that small (≤3 nm) SC- and MT-clusters coexist (with a ratio of ∼2:3) in the early stage of synthesis of SC-NPs (Fig. 1d); however, only SC-NPs can grow with time (Fig. 1c,e,f); second, the 100% yield of SC-NPs can happen only under certain conditions with precursor 2. A mixture of monodisperse MT- and SC-NPs in a one-pot synthesis could occur if the growth rate of NPs were intentionally increased by higher temperature or concentration of reducing agents, which can be attributed to the inefficient inhibition of MT-clusters by the chloride ions (see Supplementary Information, Fig. S6).

Similar synthetic control can also be available for other metals as well as alloy NPs by using well-defined metal phosphine complexes (Y.T., M.O., unpublished results)18, and offers tremendous opportunities to explore different fundamental chemistry and physics phenomena and novel device engineering at the nanoscale. We choose identical sized (10.5 nm) silver SC- and MT-NPs as examples to examine their tailored functionalities, properties and device performance.

Understanding how nanoscale chemical transformation can be manipulated by defect engineering within the NP template will be vital for its future optimal use for complex structures with novel properties and functionalities19,20. We carry out a direct reaction of SC- and MT-NP templates with selenium suspensions under the same conditions, with results highlighted in Fig. 2. For the MT-NP template (Fig. 2b, left) we observe formation of solid, perfect SC-Ag2Se (Fig. 2b, right), whereas for the reaction started with SC-NPs (Fig. 2a, left) an Ag2Se shell with a well-defined hollow core (typically 20–40% of the final particle size) can be obtained (Fig. 2a, right). We believe the formation of hollow structure after chemical transformation of SC-NPs follows a mechanism analogous to the Kirkendall effect (Fig. 2a, middle)20, in which the diffusion of silver atoms is faster than that of selenium atoms during reaction, leading to vacancy formation and condensation into one hole. However, when twinning boundaries (as line defects) exist within the metal-NP template, the outward and inward transport channels of atoms and vacancies may exist in a hierarchical order (Fig. 2b, middle): (1) along the surface; (2) through the lattice; and (3) along the twinning boundaries; the last one is akin to macroscopic ‘pipe diffusion’21, which could form an additional diffusion network within MT-NPs, balance the outward and inward atom flows and suppress the formation of vacancies via paths (1) and (2). In particular, this effect could be significantly enhanced at the nanoscale owing to large defect–volume ratio.

We further investigate fundamental electron–phonon (e–ph) interactions within MT- and SC-NPs, which is of central interest for understanding many new properties of materials and represents the key for fully exploiting quantum coupling effects. Extensive studies have been carried out on metal NPs, yielding, for example, information on the intrinsic e–ph coupling, but with contradictory results22,23,24. Using two-colour time-resolved spectroscopy, we reveal for the first time that fundamental electron and phonon properties of NPs are dramatically modified by their nanocrystallinity, with typical data shown in Fig. 3a. For both samples, inter-correlated energy transfer processes exist at different timescales with distinct scattering mechanisms (see the Supplementary Information). To further quantify e–ph interaction within NPs and gain more physical insight, we estimate the e–ph coupling constant (G) based on the two-temperature model by relating the e–ph decay constant (τe−ph) to the experimental absorbed pump-laser energy density (EAED) as25: τe−ph=(γ T0/G)+(1/(T0G))EAED, where γ is the electron specific heat and T0 represents room temperature. Our experimental data fit the predicted linear dependence well (Fig. 3b) and yield G values of (1.49±0.22)×1016 J K−1 m−3 s and (3.66±0.54)×1015 J K−1 m−3 s for MT- and SC-NPs, respectively. The G value of MT samples is in good agreement with the range in previous reports24,25, and thus provides an additional consistency check in our work. However, our measured G of silver SC-NPs is about one-quarter of that of MT-NPs, which suggests e–ph scattering is significantly enhanced owing to the existence of twinning defects (see the Supplementary Information).

Figure 3: Time-resolved measurements of 10.5 nm silver MT- and SC-NPs.
figure 3

a, Normalized optical transmission change (ΔT/T) as a function of time delay (Δt). Red and blue curves are experimental data from SC-NPs recorded at EAED=1.38×106 J m−3 and from MT-NPs acquired at EAED=1.44×106 J m−3, respectively. The inset shows the equilibrium electron distribution of NPs (blue line) and the perturbed one after absorption of pump pulse energy (Ep) (orange dashed line); Ef and NE represent the Fermi energy and the density of states, respectively. b, Experimental EAED dependence of τe−ph for both MT- and SC-NPs. τe−ph is obtained by theoretical fitting to the time-resolved ΔT/T trace (see the Supplementary Information). Red and blue lines are linear fits to data for SC- and MT-NPs, respectively. c, Coherent acoustic-phonon oscillation modulation of ΔT/T data shown in a on an enlarged scale, after removing residue exponential background contributed from the e–ph scattering process. Black lines are fits to experimental data. d, Experimental EAED dependence of acoustic-phonon oscillation period for both MT- and SC-NPs. Red and blue lines represent the averaged period for SC- and MT-NPs, respectively. Red and blue stars are data acquired from SC- and MT-NPs dispersed in dichlorobenzene, respectively. The error bars in b and d are derived from best fit to the experimental dataset.

Full size image

Our time-resolved measurements simultaneously provide valuable information on the low-frequency acoustic vibration modes of NPs that bear a unique signature of their structural and mechanical properties. Figure 3c highlights a portion of the data (Δt≥5 ps) presented in Fig. 3a, clearly revealing oscillation characteristics associated with the fundamental breathing mode of acoustic-phonon vibration in both MT- and SC-NPs24,25. Accordingly, we fit the oscillation traces with a phenomenological response function: A(t)=A0e(−t/τm)cos((2πt/Tbr)+φ), where A0 is the amplitude, τm is the decay time of the energy transfer from the NPs to the local environment, φ is the phase and Tbr is the oscillation period. As shown in Fig. 3d, oscillation periods extracted from both samples show no dependence on EAED or environmental medium (solvent), but yield distinct averages of 4.16±0.02 and 3.55±0.02 ps for MT- and SC-NPs, respectively. We further relate observed Tbr to the elastic modulus (Y) of NPs as , where ξ0 is 2.85 for silver, ρ is the silver density of 10.5 g cm−3 and R is the radius of the NPs25, and we find that Y increases by about 37% from MT-NPs (81±5 GPa) to SC-NPs (111±7 GPa). Although theoretical simulations have predicted how twinning defects affect nanoscale mechanical characteristics26,27,28, our result provides direct experimental evidence revealing the dependence of nanomechanics on defects.

One important application of metal NPs is their potential for chemical and biological sensing due to extremely sensitive and stable localized surface-plasmon resonance (LSPR) spectral response to the local environment. To illustrate the effect of twinning defects on the LSPR dielectric sensitivity, we evaluate and compare the LSPR spectra of SC- and MT-NPs collected in various solvents (see the Supplementary Information), and the results are summarized in Figure 4a. For all the solvents investigated, the NPs show good dispersion without any form of aggregation from our TEM characterization. However, we observe that the full-widths at half-maximum (FWHMs) of LSPR peaks of SC-NPs (with an average of 0.29±0.01 eV) are consistently smaller than those of MT-NPs (with an average of 0.50±0.01 eV) (Fig. 4a, inset), which suggests that twinning defects play a critical role in the damping of LSPR of NPs and significantly enhance the scattering process on the basis of the frame of Mie theory23, also agreeing with our time-resolved measurements (Fig. 3). The overall linear dependence of LSPR energy on refractive index of solvent, n, for both SC- and MT-NPs is similar to the tendency seen in previous reports7. However, linear regression yields different slopes, m, of (−0.55±0.02) eV RIU−1 and (−0.33±0.04) eV RIU−1, for SC-NPs and MT-NPs, respectively, where RIU is the refractive index unit. To further assess the device performance of such a chemical sensor we further evaluate the figure of merit (FOM) of devices as follows7: FOM=|m| (eV RIU−1)/FWHM (eV), and find that the FOM is improved about threefold for the SC-NP-based devices (1.89 RIU−1) as compared with the MT-NP-based devices (0.66 RIU−1).

Figure 4: LSPR response of SC- and MT-NPs as molecular sensors.
figure 4

a, Dependence of LSPR peak energy on refractive index of solvent. Red and blue lines are linear regression fits to data for SC- and MT-NPs, respectively. Inset, FWHM of LSPR measured in different solvents. Red and blue lines represent averaged FWHM for SC- and MT-NPs, respectively. b, Linear relationship between LSPR response and the alkyl chain length of molecules adsorbed to the NPs. Red and blue lines are linear regression fits to data for SC- and MT-NPs, respectively.

Full size image

We also characterize the use of SC-NPs and MT-NPs as molecular rulers, and compare the LSPR energy response to straight-chain alkanethiol molecular adsorbates (CH3(CH2)xSH) with different lengths (x=9–17) (see the Supplementary Information). Figure 4b shows the linear dependence of the LSPR energy shift on alkanethiol chain length29; however, the sensitivity of SC-NPs of 0.037 eV for each additional methylene unit in the adsorbed molecules is almost three times higher than that of MT-NPs of 0.014 eV. One mechanism for the higher sensing capability of SC-NPs is that the packing density of capping ligands and molecular adsorbates on the surface of MT-NPs might be different from that of SC-NPs, resulting in different effective n immediately surrounding the NP core. However, this effect will induce only a constant LSPR peak energy shift instead of sensitivity changes. Further systematic experimental and theoretical studies will be necessary to provide more insight into sensing mechanisms, which is vital for future optimization of plasmonic metal-NP-based devices.

Our study opens up exciting opportunities, ranging from fundamental solid-state chemistry and physics explorations to ultrasensitive sensor applications at the nanoscale. We show that twinning-defect-mediated chemical transformation enables a rational design of complex nanostructures if nanoscale structural disorder is sensibly introduced to suppress or enhance competing mechanisms. Consequently, this nanocrystallinity engineering will provide a means for precisely tailoring nanoscale electronic and mechanical properties as well as controlling classical and quantum coupling interactions (significant enhancement of e–ph interaction by twinning defects can eventually lead to sizeable modulation of electronic and heat conductivity and/or superconductivity of metals at the nanoscale30), ultimately leading to the optimization of device performance of crystallinity-engineered NPs. More generally, more than one functional group R can be simultaneously incorporated into metal phosphine precursors18, which should enable finer kinetic control of chemical environment in metal-NP synthesis and further distinguish different MT species (see the Supplementary Information).

Methods

General synthetic procedure of MT and SC silver NPs

All chemicals are used as received without further purification. In a typical synthesis, (PPh3)3Ag-R molecular precursor (see the Supplementary Information) is dissolved in anhydrous o-dichlorobenzene under helium protection, and after the clear solution is heated to the desired temperature a certain amount of oleylamine is quickly injected into the solution with vigorous stirring. Oleylamine serves as both the reducing agent and the capping ligand of the NPs. Once the concentration of reduced silver monomers in solution is high enough, homogeneous nucleation initiates and grows into NPs. The formation of silver NPs is evidenced by a slow colour change of the solution from colourless to bright yellow. The NPs can be precipitated out by adding methanol after the reaction is stopped and cooled to room temperature, and can be redispersed in other non-polar organic solvents, such as toluene, for later characterizations and measurements.

Transmission electron microscopy characterizations

Samples for TEM characterization are prepared by adding one drop of NP solution in toluene onto 300-mesh copper grids with carbon support film (no. 01820, Ted Pella). All HRTEM images are taken from JEOL 2100F and 4000FX field-emission TEMs. The SPED pattern is obtained by using the nanobeam diffraction mode with the JEOL 2100F, in which the nearly-collimated-electron-beam size is adjusted to be comparable to or smaller than the single-NP size and the electron beam is aligned directly on top of a single NP. To unambiguously identify distinct crystallinity between SC- and MT-NPs, additional characterizations, including angle-dependent imaging and through-focus series imaging, are applied (see the Supplementary Information).

Two-colour time-resolved measurement

All time-resolved experiments are carried out at room temperature on silver NP solution sealed in a 1-mm-thick fused silica liquid cell (Starna Cells) under nitrogen-atmosphere protection. For 10.5-nm-sized silver NPs in toluene the LSPR energy (∼3.1 eV) is well separated from the interband transition (∼4.0 eV for the 4d→5s p transition), enabling the measurements at the off-resonance condition to achieve the weak-perturbation regime. Two independent homemade optical parametric amplifiers with a 250 kHz repetition rate are applied for pump–probe measurements, enabling tunable pump energy between 1.72 and 2.59 eV and probe energy near 3.12 eV (a monochromator is applied to further select the probe energy). Output laser pulses are further compressed by prism pairs to produce ∼100 fs pulse width. The time evolution of ΔT/T measured by the probe directly reflects the absorption change induced by the laser pump pulse.

Measurements of LSPR spectra

All LSPR spectra are taken either from a homemade spectrometer with a water-cooled PMT detector or from a commercial UV–Vis Recording Spectrophotometer (Shimadzu, UV-2501PC) with a sealed 1-mm-thick fused silica cell. All chemicals are used as received without further purification. In a typical molecular-ruler experiment an alkanethiol molecule is injected into the NP solution in chloroform under nitrogen protection, and the NPs are incubated in the solution for at least 10 h to achieve stable and full coverage of self-assembled alkanethiol molecules on the NP surfaces. The LSPR peak shift in Fig. 4b is defined relative to the peak energy of the NPs before the addition of the alkanethiol molecule.