Figure 1 displays the key fabrication steps to create Si micropillar arrays. Deep reactive ion etching (DRIE) is a well-established fabrication technique to produce 3D structures by repeating the cycle of a plasma ion etching and a conformal polymer coating (Fig. 1a). These etching/coating processes, however, introduce unintended porous surface structures and electrically-active defect centers. An example of the scalloping profile of the etched structures is shown in Fig. 1b. To remove this structural damage, we used two successive thermal oxidation (1000 °C for 25 min; ≈ 300 nm thick SiO2; Fig. 1c) and oxide removal processes (10% HF for 1 min). A representative SEM image in Fig. 1d confirms the dramatically enhanced surface smoothness of the Si micropillars after the rigorous cleaning and oxidation/strip processes. We formed the radial PN junction using gas diffusion of n-type phosphorous dopants. An estimated surface doping concentration (Nd) is ≈ 1020 cm− 3 with a junction depth (xj) of ≈ 0.3 μm (Neudeck 1989). The complete Si micropillars measured ≈ 30 μm in height and ≈ 7 μm in diameter with a distance between the pillars of approximately 4 μm (Fig. 1e). The current-voltage (I-V) curves of the pillar array in Fig. 1f showed a good diode behavior. We estimated the turn-on voltage of 0.63 V, the leakage current of 10 nA, and the ideality factor of about 1.7. Comprehensive dark and light I-V characteristics of the radial junctions in various geometrical parameters (e.g., diameter, height, pillar-to-pillar-distance) can be found elsewhere (Yoon et al. 2010, 2011). While extremely informative, I-V curves reflect the overall PN junction performance and do not capture the local junction properties.
EBIC microscopy allows a direct access to the local PN junction in 3D with an adjustable probe size from 10’s nm to several μm, in accordance with the interaction volume between the incident electron beam and the semiconductors. To visualize the local junctions at the level of individual pillars, we carefully cleaved the Si pillar array using a fine scriber and exposed the cross-section. The sample was mounted on an EBIC holder, where the metal contacts of the emitter (i.e., indium dots on n+-Si) and the collector (i.e., Al on p-Si) were connected to the external EBIC circuit. The electron beam was injected from the n-Si emitter shell to the PN depletion region and the p-Si pillar core. Figure 2 displays an SEM and the corresponding 5 keV (Fig. 2b) and 10 keV (Fig. 2c) EBIC images. Figure 2d shows the overlaid SEM and 5 keV EBIC, indicating a continuous PN junction formed across the 3D geometry. The overall EBIC intensities of the individual pillars at 5 keV and 10 keV are relatively uniform, suggesting the presence of conformal radial junctions along the individual micropillars.
For a quantitative analysis, we extract the EBIC line scans along/across the pillars. The line scan plot across the pillar diameter (Fig. 2e) shows the highest EBIC value near the pillar center (≈ 80 nA) that decreases gradually with the electron beam probe moving away to the perimeter of the pillar (≈ 50 nA). Considering the direction of the incident electron beam to the curved pillar surface, as illustrated in the inset of Fig. 2e, we suggest that this EBIC change is mainly attributed to the shape of the pillar rather than due to inhomogeneous junction properties. With the increase of the backscattered electrons (BSE) and the decrease of the effective electron-hole pair (EHP) generation volume at the curved pillar surface, the reduction of the EBIC magnitude is evident near the pillar perimeter.
Figure 2f displays the EBIC line scans along the length of the pillars (i.e., axial-direction), showing a relatively constant EBIC value within the pillars at a fixed accelerating voltage. The mean EBIC value of the pillar increases from 80 nA at 5 keV to 480 nA at 10 keV. We observe the highest EBIC values are present in the area of the pillar base, which is associated with the direct local carrier generation within the depletion region. When the electron beam is directly injected to the cross-sectional PN junction (i.e., mechanically cleaved junction area), the generated EHPs are separated quickly without diffusion owing to the built-in electric field (C. J. Wu and Wittry 1978; Yakimov 2015). In contrast, the electron beam irradiated on the pillar sidewalls generates the EHPs in the depletion region as well as the charge-neutral regions (i.e., n-Si shell, p-Si pillar core). The excess carriers must travel to the junction (i.e., ambipolar diffusion) before they are separated and collected in EBIC. Since the n-Si emitter layer (Nd ≈ 1020 cm− 3, xj ≈ 300 nm) of our pillars is conductive, yet highly defective, the EHPs generated in this region tend to be recombined, decreasing overall EBIC values as compared to the direct electron beam injection at the cross-sectional PN junction. As the EHP generation volume increases with the accelerating voltage (i.e., 5 keV to 10 keV), the portion of the EHPs generated in the n-Si decreases, resulting in a comparable EBIC value of the pillar and near the base.
To assess the local junction quality of the micropillar array, we collected the baseline EBIC characteristics of a commercial planar device (Solar Made). This planar PN junction (n+-p) was built on a high-purity single crystalline Si substrate, and it showed a carrier collection efficiency close to 100% in the depletion region obtained in a normal collector EBIC configuration (Yoon et al. 2014). Figure 3 displays a representative SEM image of the planar PN device and the corresponding EBIC maps collected at 5 keV, 10 keV, and 20 keV. The large dark area of the EBIC images is associated with the metal contact, highlighted in yellow in the SEM image (Fig. 3a). The injected electron beam (1 keV ~ 30 keV) does not penetrate this thick metal layer (a few mm thick Ag paste), producing negligible EBIC signals (Fig. 3b-d). The dark speckles in the 5 keV EBIC image are likely attributed to thin organic residue or dust particles on the sample, of which EBIC contribution becomes insignificant with the higher beam energies (> 10 keV). Qualitatively, the bright contrast increases with a higher keV, showing similar behaviors as those observed with the pillar array radial junction (Fig. 2).
Figure 3f through g show the representative line scans extracted from the EBIC images (Fig. 3b-d). A relatively constant EBIC was observed in the device area, a stack of p-Si collector, n-Si emitter, and SiN passivation layer. A notable current fluctuation near the metal contact is mainly attributed to the spread of the metal paste. By aligning the line scans, we find that a low keV EBIC is much more sensitive to the surface features than higher keV. For instance, two distinct peaks observed in the 5 keV EBIC line scan (marked with a green box) conform to the sample topography shown in SEM (Fig. 3a). This feature becomes less distinguishable with increasing incident beam energy, as the electron beam penetrates deep in the sample with a larger EHP generation volume. We used the EBIC images from 5 keV to 30 keV and calculated mean EBIC values for the Si area, shown in Fig. 3e. The increase of EBIC with a higher keV is evident in the line scans. The average EBIC value increases from 127 nA at 5 kV to 3.55 μA at 30 kV, increasing over one order of magnitude. Interestingly, the EBIC values observed in the planar junction are slightly higher than those in the radial junction in Fig. 2: 127 nA (vs. 83 nA of the radial junction) at 5 keV, 574 nA (vs. 492 nA of the radial junction) at 10 keV.
The experimental results qualitatively suggest that EBIC magnitude near the PN junctions is strongly influenced by the EHP generation by the incident electron beam and the local carrier separation/collection properties. To gain a deeper understanding of the local radial junction characteristics, we estimate the carrier generation profile using Monte Carlo simulations and calculate the local carrier collection efficiency for the planar and the radial junctions. Figure 4a (top) displays an example of the simulated electron trajectories, where a ray of 5 keV electron beam is irradiated onto a Si substrate. The blue lines represent the collision events of the primary electrons with Si until they lose their initial energy (i.e., 5000 V in this example) to 50 V or lower. The red lines represent the paths of the backscattered electrons. A corresponding energy contour plot is shown in the bottom image. For instance, the 95% contour represents the sample area where the injected primary electrons have lost 95% of their initial energy. Figure 4b plots the estimated interaction bulb size at different accelerating voltages (1 kV ~ 30 kV). The overall ratio of depth to diameter (depth/diameter) is comparable for higher lost-energy contours (> 75%), yet slightly higher for low energy contours (< 50%), indicating a pear-shape of the interaction bulb. The calculated bulb size at 1 keV is approximately 19 nm, inferring that the spatial resolution of the EBIC image for flat Si devices can be achieved as high as < 20 nm. The inset of Fig. 4b shows the increase of the penetration depth with the beam energy, which was extracted from the 95% energy contour of each simulation. The numerical fit overlaid on the datasets confirms the bulb size is proportional to \( {E}_b^{1.78} \) (Eb is the beam energy), showing an excellent agreement with the analytical prediction of \( \approx {E}_b^{1.7} \) by Wittry et al. (Wittry and Kyser 1967). By controlling the incident beam energy, the size of the EHP generation bulb can be tunable from 10’s nm to several μm, offering versatility to study local carrier dynamics in optoelectronic semiconductor materials and devices.
Based on the Monte-Carlo simulations and EBIC modeling (Leamy 1982; Haney et al. 2016; Yakimov 2015), we estimated the local carrier separation/carrier efficiency of the radial junction and compared it to planar PN controls. The EBIC collection efficiency (ηEBIC) is defined as the ratio of the measured current (IEBIC) to the EHP generation rate (β). Here, e is the unit charge (1.6 × 10− 19 C).
$$ \eta (EBIC)=\frac{I_{EBIC}}{e\cdot \beta}\kern0.5em $$
(1)
The generation rate, which is the total number of EHPs created by the injected electron beam, can be calculated as below.
$$ \beta =\frac{I_b\cdot {E}_b\cdot \alpha }{e\cdot {E}_{EHP}} $$
(2)
Eb is the incident electron beam energy, α is the fraction of beam energy absorbed inside the material (i.e., Si in our case), and EEHP is the average energy to create an electron-hole pair. The Ib of our SEM was measured in the range of 250 pA (Eb = 5 kV) to 300 pA (Eb = 20 kV). We calculated the magnitude of α using the backscattered coefficient obtained from the Monte Carlo simulation (e.g., 0.152 at 5 keV, 0.142 at 20 keV). The EEHP was estimated using an empirical relation of EEHP = 2.596 Eg + 0.174 (Kobayashi et al. 1972), giving EEHP ≈ 3.621 eV for Si (Eg = 1.12 eV). The EBIC currents (Ib) extracted from the line scans in Figs. 2 and 3 were used for the pillar array and the planar device, respectively. We note that a typical uncertainty in our EBIC measurement and analysis is about 10% associated with the fluctuations of the baseline e-beam current (Ib) and the signal-to-noise ratio of the EBIC preamplifier. Also, the parameters extracted from the Monte Carlo simulations (e.g., backscattered coefficient, mean EHP generation rate (β), empirical parameter (EEHP) to generate EHPs) contribute to the uncertainty.
Finally, we plot the resulting EBIC efficiency of the devices at different incident beam energies in Fig. 4c. The EBIC efficiency increases with the incident beam energy, reaching close to unity at Eb > 15 keV for the planar PN junction device. A similar trend was observed for the radial junction of the pillar array, yet the overall EBIC efficiency is slightly lower than that of the planar device (about 10%). In both cases, EBIC was measured in the depth-dependent configuration. The injected electrons travel from the highly-doped emitter (a few 100 nm thick) to the depletion region (< 1 μm) and the p-Si collector, generating the EHPs in three different layers. The low EBIC efficiency at 5 keV (50% for the planar device; 30% for the pillar array) is likely attributed to the EHP production in the highly-doped emitter region. Our Monte Carlo simulation shows an interaction bulb size of (300 nm)3 at 5 keV, suggesting that most EHPs were produced in the highly-defective (i.e., high-density of recombination centers) emitter region that promotes excess carrier recombination. At higher keV, most EHPs are generated in the strong built-in electric field region and the collector, increasing the EBIC efficiency. Our observation indicates that the surface damage introduced on the pillars by DRIE could be effectively removed by rigorous cleaning and oxidation/strip processes. The magnitude of the EBIC efficiency of our pillar array is about 70% at 10 kV, slightly lower than that of the planar device (≈ 81%). We speculate that a slightly higher EBIC efficiency for the planar junction is associated with the surface passivation (e.g., SiN) that decreases the surface recombination of EHPs. Detailed EBIC studies of the surface passivation based on 3D continuity equations together with Poisson equations (Yakimov 2015; Haney et al. 2016; Zhou et al. 2020) will provide additional insight on the excess carrier dynamics and general guidance to improve their device performance.