Report

Voltage-tunable circular photogalvanic effect in silicon nanowires

See allHide authors and affiliations

Science  14 Aug 2015:
Vol. 349, Issue 6249, pp. 726-729
DOI: 10.1126/science.aac6275

Chirality from achiral structures

The most common materials used in electronics produce only a simple optical response. Dhara et al. observed a complex circular photogalvanic effect in silicon nanowires, with the magnitude and direction of the induced photocurrent dependent on the polarization of the light. The specifics of the structure and geometry of the component materials are responsible for the effect. It should therefore be possible to engineer the same effect in other achiral materials and thus expand the box of enhanced functional materials for optical applications.

Science, this issue p. 726

Abstract

Electronic bands in crystals can support nontrivial topological textures arising from spin-orbit interactions, but purely orbital mechanisms can realize closely related dynamics without breaking spin degeneracies, opening up applications in materials containing only light elements. One such application is the circular photogalvanic effect (CPGE), which is the generation of photocurrents whose magnitude and polarity depend on the chirality of optical excitation. We show that the CPGE can arise from interband transitions at the metal contacts to silicon nanowires, where inversion symmetry is locally broken by an electric field. Bias voltage that modulates this field further controls the sign and magnitude of the CPGE. The generation of chirality-dependent photocurrents in silicon with a purely orbital-based mechanism will enable new functionalities in silicon that can be integrated with conventional electronics.

In the circular photogalvanic effect (CPGE), the polarity and magnitude of photocurrents can be controlled by the chirality of elliptically polarized optical excitation (Pcirc) in a certain class of materials known as gyrotropic media. The effect originates from the unequal population of excited charged carriers in a preferential momentum direction when excited by light with left (σ = –1) or right (σ = +1) circular polarization. Semiconductors that support the CPGE are traditionally gyrotropic optical media with a strong spin-orbit coupling, so that the effect is ordinarily controlled by angular momentum selection rules for excitation with circularly polarized light. The effect has been observed in different quantum-well (QW) structures (15), for which it is attributed to the k-linear spin splitting of energy bands due to the spin-orbit interaction.

Most materials that have found widespread applications in conventional electronics (e.g., centrosymmetric crystals such as Si and Ge) are not gyrotropic and do not exhibit a bulk CPGE unless quantum-confined to below 10-nm length scales. In Si/Ge QWs, a CPGE due to orbital (6, 7) or valley-orbital interactions (8), appears in the long wavelength range (~100 μm). The underlying mechanism is attributed to intrasubband free carrier absorption pathways interfering with intersubband excitation (6) via polarizability effects. Photogalvanic effects have also been theoretically predicted in carbon nanotubes, without involving the electron spin degree of freedom (9). Because there are similarities between chiral nanotubes and Si nanowires (NWs) (1012), it is desirable to determine whether centrosymmetric crystals of technologically important materials such as Si can also exhibit a CPGE when their shapes are engineered, thus adding novel functionalities.

We demonstrated a mechanism for the CPGE involving only the orbital degrees of freedom that are observed at the surface of Si NWs at the metal-NW junction. The atomic structure of the NW, along with the macroscopic field present at the contact, breaks the bulk symmetries allowing the CPGE. The Embedded Imagesurface of Si is of particular interest because of its high hole mobility (1315) associated with a zigzag chain of atoms running along the 〈110〉 direction. A Schottky electric field along the NW 〈111〉 growth direction breaks the relevant mirror symmetries and produces a chiral structure, producing a CPGE that is tunable with applied bias.

Unintentionally doped Si NWs (diameter, 100 to 150 nm; grown mostly along the 〈111〉 direction and some along the 〈112〉 direction) were used to make two-terminal devices by electron-beam lithography [section 1 in (16)]. Figure 1A shows the schematic of the device and the measurement setup, where the laser [transverse electromagnetic (TEM00) mode, wavelength 680 nm, spot size ~2 μm] is incident at the metal-NW junction [see (16) for details in section 1 and figs. S1 to S6]. The excitation energy (680 nm) of the laser is above the indirect bandgap of Si, and therefore contributions to the photocurrent from interband excitation (corresponding to the bulk of the Si NW) and the surface states on the Embedded Image plane can be simultaneously present, albeit with different selection rules. Because of the geometrical anisotropy of NWs, the bulk contribution to photocurrent has two maxima with respect to linear polarization, owing to the preferential absorption of light polarized along the long axis (TM polarization) (17, 18) and in the perpendicular direction (TE polarization) to the NW at the metal-NW junction due to metal antenna effects (19, 20). In all of our experiments, we observed the usual linear polarization–dependent photocurrent variation as a function of the quarter-wave plate (QWP) [half-wave plate angle–dependent photocurrent variation is discussed in (16), sections 2 and 3 and fig. S8]. However, at the metal-NW junction, the most important contribution to the photocurrent came from the Pcirc dependence, which we observed by noticing that the maximum change in the photocurrent appeared at the left-handed (45°) and right-handed (135°) circular polarization, which has a variation of sin(2α) irrespective of the physical orientation of the NW device [(16), section 4 and fig. S9]. As a result, the observed photocurrent at room temperature of a representative device at an applied bias of 1 V at two different spatial positions, in the metal-NW contact region (Fig. 1B), and on the bare NW (far away from the contacts, Fig. 1C) displayed different polarization dependence. At the metal-semiconductor contact, the photocurrent pattern repeated twice [∝ sin(2α)] as the QWP angle was changed from 0° to 360°, whereas on the NW, the pattern repeated four times [~cos(4α + ϕ), with a phase term (ϕ) depending on the physical orientation of the NW and laser polarization], suggesting a strong Pcirc–dependent photocurrent confined to the contact region. The photocurrents can be fitted (Fig. 1, B and C, solid line) with the expression I(α) = Ic sin 2α + Il cos(4α + φ) + Ιd, where Ic, Il, and Id are the coefficients for circular polarization (CP), linear polarization (LP), and polarization-independent components of the photocurrent, respectively. Figure 1B shows that the CP-dependent photocurrent (normalized with laser intensity) was observed with a value of Ic = 1.0 nA kW–1 cm2 at the metal-NW junction, and it reversed its sign with respect to a background current, Id = 5.2 nA kW–1 cm2, at α = 45° (left circular polarization) and 135° (right circular polarization). The LP-dependent contribution, Il = 0.35 nA kW–1 cm2, was about three times smaller than the CP contribution at the contact region. However, if the light was incident on the bare part of the NW (Fig. 1C), the CP contribution, Ic = –0.1 nA kW–1 cm2, was almost negligible while the LP contribution Il = 2.2 nA kW–1 cm2 was much stronger, which is in agreement with the previous understanding of LP-dependent photocurrent (17, 18) of semiconductor NWs. The polarization-independent contribution (Id = 9.6 nA kW–1 cm2) was also larger at the center, because the absorption of light is more than at the NW-metal junction. In another small-diameter device with negligible Id at the metal-NW junction (Id < Ic), a small contribution from the bulk resulted in a clear reversal in the polarity of the photocurrent with left and right circular polarization (fig. S10).

Fig. 1 CP-dependent photocurrent at two different locations along the NW.

(A) Schematic of the experimental setup (V, voltage source; A, ammeter) along with a schematic of the microscopic view of the metal-semiconductor contact of the Si NW device along with the atomic zigzag chains. The shaded area (x-y and y-z planes) represents the mirror planes on the Embedded Imagesurface. E, direction of the electric field. The inset shows a scanning electron microscopy image of one of the electrodes. Scale bar, 0.2 μm. (B and C) Photocurrent as a function of the QWP angle is shown at two different positions of laser excitation along the NW (the excitation region is indicated by schematics on the corresponding plots). Photocurrent properties (B) when the laser is incident at the metal-nanowire junction labeled 1 and (C) at the center of the NW are shown. In (B) and (C), the solid lines are the fits to the expression of photocurrent I(α) = Ic sin 2α + Il cos(4α + φ) + Id .

It is known that Si NWs grown along the 〈111〉 and 〈112〉 directions terminate on the [110] facets (2123), which contain zigzag atomic chains (labeled as the z axis in the schematic of Fig. 1A) that are not aligned to the NW growth direction. These states have a different dispersion behavior than the bands in bulk Si and are responsible for its high hole mobility (24, 25). We can understand the origin of the CPGE in our experiments by analyzing the symmetry of this structure. Figure 1A shows the mirror planes (shaded region of the x-y and y-z planes) on the Embedded Image surface, which exist in the absence of any electric field. The emergence of chirality can be understood by introducing an electric field along the NW 〈111〉 direction, which breaks the mirror symmetries of the y-z plane (due to the x component of the field) and x-y plane (due to the z component of the field). Breaking these mirror symmetries implies that the resulting structure is not identical to its mirror image and hence is a structure with distinct chirality.

In order to study the effect of the Schottky electric field at the metal-semiconductor contact on the CPGE, we measured photocurrents at different applied biases as a function of laser polarization (QWP angle). The CP-dependent contribution to photocurrent (Ic) (Fig. 2A) changes as a function of applied bias. At zero bias, the Schottky field determined by the nature of the metal-semiconductor contact produces photocurrent with a nonzero value of Ic = –66 pA kW–1 cm2, in comparison to a lower value of the LP-dependent coefficient Il = 37 pA kW–1 cm2, clearly demonstrating the CPGE at the contacts. As the applied bias is modulated, these coefficients change, and the system’s response can be tuned (Fig. 2A and table S1) from producing a significant CPGE (at ±248 mV) to almost none (at 80 mV), which clearly shows that the Schottky field at the interface is responsible for the observed effect.

Fig. 2 Effect of bias-tunable Schottky field on CP-dependent photocurrents.

(A) Photocurrent as a function of QWP angle (αEmbedded Image at different applied biases. (B) Schematic band diagrams at four representative bias voltages for a p-type NW. (C) I-V curves of the device in the dark (black curve) and upon light excitation at the reverse-biased junction (red curve). (D) Zoomed-in view (close to zero current) of the Ic and Id (obtained from fits for I(α), table S1) as a function of applied bias (V). (E) Plot of Ic and Il as a function of applied bias.

To qualitatively explain the bias-dependent CPGE results, a set of band diagrams is shown in Fig. 2B at different biases. The sign convention of the photocurrent is positive in the direction from junction 1 (J1) to 2 (J2). The NW device can be understood as two back-to-back Schottky contacts; for a p-type semiconductor, applying a positive bias at J1 (J2 is grounded) makes J1 reverse-biased and J2 forward-biased. Under dark conditions, the current through the device is limited by the current in the reverse-biased junction, and as a result a negligible dark current is obtained (Fig. 2C). When light is incident at J2, the magnitude of the photocurrent is large if J2 is reverse-biased (J1 is forward-biased), as the photogenerated carriers produce drift current due to the Schottky field at the reverse-biased junction, resulting in an asymmetric current-voltage (I-V) relation (Fig. 2C, red curve). At zero bias, we observed a CPGE due to the band bending (Fig. 2B). Table S1 summarizes the estimated values of Ic, Il, and Id, as a function of applied bias at a laser intensity of 32 W/cm2. As we increased the bias voltage from zero, J2 was forward-biased, and hence the band bending at J2 decreased, which reduced the CP coefficient; at 80 mV, the CPGE almost vanished (Ic ~ –6 pA kW–1 cm2), and the only remaining contribution was due to Id and Il. When the bias voltage was further increased, the electric field direction at J2 was reversed, which reversed the CP contribution, Ic → –Ic. The sign of the CP-dependent coefficient only depends on the sign of the electric field because of band bending where light is incident and not on the sign of Id (the dominant contribution to the photocurrent), as revealed in Ic – V and Id – V plots (Fig. 2D). Figure 2E shows that the LP-dependent contribution Il varies little in comparison to Ic as a function of applied bias, which implies that only the CP-dependent contribution is sensitive to applied bias among the two polarization-dependent photocurrent contributions.

To understand the microscopic origin of the symmetry breaking, we adopt the model shown in Fig. 1A. The Embedded Image plane contains the zigzag chains of Si atoms along the 〈110〉 direction, which makes an angle of ~35° to the growth direction and the Schottky field of the NW. Figure 3A schematically shows a simplified linear chain, where the valence band is formed by p orbitals (px, py, and pz) and the conduction band by s orbitals. For small crystal momentum kz along the z direction, the Schottky electric field is a perturbation that mixes the Bloch states with orthogonal orbital polarizations, Ψ1(k,r) = (1/Embedded Imagej exp(ikzZj)px(rZj) and Ψ2(k,r) = (1/Embedded Imagej exp(ikzZj)pz(rZj), so that the perturbed states can be written as Embedded Image [(16), section 5]. An energy-splitting ~±|kzγa| is obtained between the states Ψ+ and ΨEmbedded Image where a is the lattice constant and γ is the nearest-neighbor overlap integral between the px and pz orbitals in the presence of a Schottky field, whereas the ± sign depends on the sign of kz and the Schottky field. Figure 3B shows the splitting of Ψ+ and Ψ energy bands for small values of kz induced by the Schottky field. Two valence bands Ψ± have unequal orbital population for a nonzero kz, and as a result, circularly polarized light propagating along the y direction with σ = ±1 can excite electrons only from the initial state with Embedded ImageEmbedded Imageto final state |S > with ly = 0 (Fig. 3B). This generates an asymmetry in the population of momentum distribution of the excited electrons at ±kz and hence produces photocurrents, which reverse sign when the chirality of the CP light is changed.

Fig. 3 Orbital-based linear chain model for the CPGE in Si nanowires.

(A) Schematic of a linear chain of atoms along the z direction with the valence band formed by p orbitals. Light is normally incident along the y direction. The direction of the electric field (E) due to Schottky contact and the growth direction of the NW (along 〈111〉) lie in the x-z plane. (B) Plots of calculated energy band dispersion using the linear chain model, showing the splitting of valence bands due to the mixing of |px〉 and |pz〉 orbitals, in the presence of the Schottky field [(16), section 5]. (C) The power dependence of the CPGE (open circles) as well as the polarization-independent background (solid squares) shows linear dependence with laser power.

Because a 〈111〉 or 〈112〉 grown NW can also terminate at facets other than the Embedded Imageset of planes (22), there is a possibility that the electrically contacted plane is different from Embedded Image and hence should not produce any CPGE. Statistically, we observed from our measurements that ~40% of NWs did not show any CPGE at the metal-NW contacts but always had the LP-dependent photocurrent. We sometimes observed an asymmetry in the CPGE response (fig. S12) between the two contacts on the same NW, suggesting that contacts may not be symmetric, or the Embedded Image plane may not run uniformly along the entire length of the NW. This observation is in agreement with the observed stacking faults and other defects typically observed in ~25 to 30% of the as-grown Si NWs, which can disrupt the chains. This statistical analysis implies that the quality of metal-NW contact at the surface of the NWs is important to observe the CPGE, which is further corroborated by the observation that the CPGE phenomenon decreases as the device ages, probably due to the diffusion of the metal or other impurities that can disrupt the chains.

Although our model explains the bias-dependent results, we performed more experiments, such as the power and energy dependence of the photocurrent, for further verification. Figure 3C shows a linear dependence of the CP contribution (Ic) with light intensity, which is required for the CPGE as Embedded Image, where ℰx and ℰz are complex electric field components of light. The polarization-independent contribution (Id) is also linear with the light intensity, and hence we used this quantity to normalize the Ic to study the wavelength dependence of the CPGE.

To verify the hypothesis that [110] surface states contribute to the observed CPGE, we performed laser wavelength–dependent photocurrent measurements in the energy range from 500 to 800 nm (Fig. 4A) and plotted a normalized quantity, I/IdEmbedded Image as a function of the QWP angle. It is expected that the absorption in the bulk Si NW should mostly affect the background and LP-dependent photocurrent, whereas the surface states, because of their particular dispersions, may have a different response. Previous studies via angle-resolved photoemission spectroscopy (24, 25) and scanning tunneling microscopy (26) measurements have demonstrated that the Si [110] plane contains many surface states with varying energy gaps. An identification of these with the states at the NW surface is only approximate, because the surface states are sensitive owing to surface reconstruction. Nevertheless, the energy difference between some of the reported states matches (within the thermal energy) the excitation energy of the laser at which a strong signal is recorded. For example, the energy difference of surface states C3 (2.6 eV) and S3 (0.75 eV) (24) matches the laser excitation at 680 nm, where we see a larger Ic/Id as compared to 500 and 710 nm. We also measured the response at lower excitation energies and observed that the CPGE response is stronger at 800 nm in addition to 680 nm (Fig. 4A), which matches the energy difference between the C3 (2.6 eV) and S4 (1 eV) surface states (24). The quantity Ic/Id that characterizes the relative strength of the CPGE over the background contribution as a function of excitation wavelength (Fig. 4A) changes from a maximum value of ~0.19 observed at 800 nm to 0.02 at 500 nm (the minimum obtained). We studied the polarization-dependent photocurrents at 77 K and observed an overall decrease in the photocurrent, which is expected because of the reduction of phonon-induced absorption in bulk Si as well as the reduction of thermally generated carriers; however, the overall magnitude of normalized Ic at 77 K was similar to the room temperature results (Fig. 4, B and C). More importantly, we observed a blue shift of the excitation energy corresponding to a maximum in Ic/Id (Fig. 4, B and C), which suggests that the energy bandgaps of the surface states increase at lower temperature. The above observations imply that our experiments support the idea of extended one-dimensional states at the surface of Si NWs grown along the 〈111〉 or 〈112〉 direction, responsible for the observed CPGE and not dependent on metal antenna, plasmonic, or hot electron effects (19, 20).

Fig. 4 Wavelength and temperature dependence of the CPGE.

(A) Normalized photocurrents as a function of QWP angle α for different excitation wavelengths in the range of 500 to 800 nm at room temperature along with fits (solid lines). Comparisons between polarization-dependent normalized photocurrents obtained at room temperature and at 77 K for (B) 800-nm and (C) 750-nm laser excitation are shown. (D) Trend of Ic per unit of Id given in percentage Embedded Image as a function of laser excitation wavelength at room temperature and at 77 K.

Si NWs support the CPGE where the propagation direction of the current can be controlled by the sense of circular polarization of light, which is also tunable with an external bias. The CPGE traditionally occurs in gyrotropic optical media and hence does not occur in bulk Si because of its diamond structure. In contrast, we showed that the CPGE in Si NWs is a purely orbital effect, and it arises from the geometrical effect in the NW that reduces the bulk mirror symmetries. Because the effect can be engineered by a combination of shape, crystal anisotropy, and applied fields, many applications, as well as the ability to encode more information in a device by using the orbital degrees of freedom, are possible.

Supplementary Materials

www.sciencemag.org/content/349/6249/726/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S12

Table S1

REFERENCES AND NOTES

  1. Materials and methods are available as supporting materials on Science Online.
  2. ACKNOWLEDGMENTS: This work was supported by the U.S. Army Research Office (grants W911NF-09-1-0477 and W911NF-11-1-0024) and the seed project support from the Laboratory for Research on the Structure of Matter, NSF Materials Research Science and Engineering Center grant DMR-1120901. E.J.M. is supported by the U.S. Department of Energy–Basic Energy Sciences under grant DE FG02 84ER45118. Si NWs were provided by B. Tian (Chicago) and CINT, a U.S. Department of Energy, Office of Basic Energy Sciences User Facility at Los Alamos National Laboratory (contract DE-AC52-06NA25396) and Sandia National Laboratories (contract DE-AC04-94AL85000). The data described in the paper are archived by the Agarwal Group at the University of Pennsylvania.
View Abstract

Navigate This Article