Figure 1a shows the schematic of our pump-probe s-SNOM setup. A 400 nm pump beam and an 800 nm probe beam with a controlled delay time D*t* are directed to an atomic force microscope (AFM) tip with an oscillating frequency Ω (see Methods and Supplementary Fig. 1 for more details on the setup). The backscattered probe beam is detected and demodulated at higher harmonics of the tapping frequency (nΩ) to suppress background noise and amplify the near-field signal23. The samples in this study are SiNWs synthesized via an electroless etching method (Inset in Fig. 1i)24. Figure 1b and Figs. 1c-g show the topography of a SiNW and the time-resolved s-SNOM imaging at different time delays, respectively. The high spatial overlap of the height and s-SNOM profiles indicates the successful detection of near-field signals (Fig. 1h). The s-SNOM time snapshots distinctly exhibit a dynamic change in the intensity of scattered light (Fig. 1j). Specifically, the s-SNOM signal amplitude undergoes an intense increase (Fig. 1d), followed by a decay process (Figs. 1e,f) and a slow rise at longer delay times (Fig. 1g). This trend can also be quantitatively visualized in the transient s-SNOM signal presented in Fig. 1i.

To understand the evolution of the s-SNOM signal over the increasing delay time, we adopt a point-dipole model to analyze the scattered light (see Supplementary Note 1)22,25,26. Briefly, the scattered field **E**sc by the tip is derived from the dipolar near-field interaction between the AFM tip and the sample with the dielectric function *e*, yielding25,26

\({\mathbf{E}}_{\text{s}\text{c}}\propto A{e}^{i\phi }{\mathbf{E}}_{\text{i}\text{n}}=\)\(\frac{\alpha \left(1+\beta \right)}{1-\frac{\alpha \beta }{16\pi {\left(a+z\right)}^{3}}}{\mathbf{E}}_{\text{i}\text{n}}\) (1)

Here the measured scattering intensity \(S\propto {A}^{2}\), \(\phi\) is the phase of scattered light, *a* is the AFM tip radius, *z* is the tip-sample distance, \(\alpha =4\pi {a}^{3}\) is the polarizability of the tip, and \(\beta =(\epsilon -1)/(\epsilon +1)\), respectively (Fig. 2a). The dielectric function *e* is further calculated as a function of free carrier density *N* based on the Drude model (Supplementary Fig. 2)27

$$\epsilon (\omega ,N)={\epsilon }_{\infty }-\frac{{{\omega }_{p}\left(N\right)}^{2}}{\omega (\omega +i{\Gamma })}$$

2

where \({\epsilon }_{\infty }=11.7\) for silicon, \({\omega }_{p}\) is the plasma frequency and a function of the carrier concentration *N* (Supplementary Note 2), and \({\Gamma }\) is the damping rate. In addition, the effect of temperature is considered by the Jellison-Modine model28 (Supplementary Fig. 3, also see Supplementary Note 2 for more details).

Figure 2b shows the calculated s-SNOM intensity with an increasing free-carrier concentration *N* from 1018 to 1022 cm− 3 at the wavelength of 800 nm. As *N* increases, the calculated s-SNOM signal exhibits a minimum at *N* ~ \(6\times {10}^{19}\) cm−3 and then reaches its maximum at *N* ~ \(1.2\times {10}^{20}\) cm− 3. This behavior arises from the resonant near-field interaction between the AFM probing tip and the free carriers in SiNWs25. The experimental transient s-SNOM signal can be well fitted by the point-dipole model assuming a biexponential decay of photoexcited carriers, i.e., \(N\left(\varDelta t\right)=N\left(0\right)\times ({A}_{1}{e}^{-\frac{\varDelta t}{{t}_{1}}}+{A}_{2}{e}^{-\frac{\varDelta t}{{t}_{2}}})\) (Fig. 2c). The biexponential kinetics (*t*1, *t*2) are attributed to the carrier recombination and diffusion29, and the average carrier lifetime *t*avg can be calculated by \({t}_{\text{a}\text{v}\text{g}}=({A}_{1}+{A}_{2})/({A}_{1}/{t}_{1}+{A}_{2}/{t}_{2})\). The corresponding phase change is also consistent with the experimental results (Supplementary Fig. 4). We exclude the effect of laser heating as the temperature increase is negligible with respect to the s-SNOM amplitude (Supplementary Note 3 and Supplementary Fig. 5). In addition, the variation of tip-sample distance due to thermal expansion of SiNWs only leads to slight changes in the s-SNOM intensity (Supplementary Note 4 and Supplementary Fig. 6). These results verify the dominant role of carrier density in the transient s-SNOM measurements. We further validate the point-dipole model by measuring the same SiNW with different pump power (Fig. 2d), where two curves can be well fitted with different initial carrier densities and the same decay kinetics (Supplementary Table 1).

We then apply pump-probe s-SNOM to investigate the carrier dynamics in individual SiNWs with varying geometries. Figure 3a shows the transient s-SNOM signals measured in different nanowires with various widths. The corresponding scanning electron microscope (SEM) images are shown in Fig. 3b. All experimental curves are well fitted with the point-dipole model, and the decay times are summarized in Fig. 3c. The carrier lifetime shows a linear increase with the increasing size of the SiNWs, indicating that surface recombination dominates in semiconductor nanowires30. The surface recombination velocity (SRV) can be calculated from carrier lifetime *t*avg as SRV = *d*/4*t*avg, where *d* is the SiNW width31. The linear fitting gives a surface recombination velocity of 2.2 \(\times\) 104 cm/s, which is consistent with previous reports29,32.

We further explore the capability of pump-probe s-SNOM to probe the spatially resolved carrier dynamics with a nanoscale resolution. A large, nonuniform silicon nanowire is selected as the test sample (Fig. 4a). Transient s-SNOM signals measured at different locations (P1-P4 in Fig. 4a) show distinct temporal evolution (Fig. 4b), corresponding to carrier lifetimes of 460.3, 379.9, 294.6, and 338.0 ps, respectively. The time-resolved s-SNOM images also exhibit spatially nonuniform dynamics of the nanostructure (Supplementary Fig. 7). Figure 4c presents the time-resolved s-SNOM map along the nanowire (dashed line in Fig. 4a). The carrier lifetime extracted from the spatiotemporal mapping is strongly correlated to the topography as evidenced by the qualitatively similar profiles (Fig. 4d). The probe of carrier dynamics at a spatial resolution of 35 nm is demonstrated.