Contribution of both bulk and surface states on photothermoelectric transport in epitaxial Bi₂Se₃ thin films

Recently, three-dimensional topological insulators (3DTIs) have been continuously attracting research interest due to their unique properties resulting from the topological protection of their surface states.^1,–7 Although TI surface states have been detected using angle-resolved photoemission spectroscopy (ARPES)^6,8 and scanning tunneling microscopy techniques,^9,10 electrical control over their density, which is required for most transport experiments, remains a challenge. In 3DTI systems, the characterization of the surface states via transport experiments is often hindered by a large bulk charge carrier density.^11,12 Bi₂Se₃ is a promising 3DTI with a bulk bandgap of 0.3 eV and with conducting Dirac-like surface states around the Γ-point in the reciprocal E–K space.^7,9 Furthermore, Bi₂Se₃ is a very good thermoelectric material. The fascinating potential applications of Bi₂Se₃-based devices include spintronic,¹³ enhanced thermoelectric effects,¹⁴ high performance field effect transistor,¹⁵ thermoelectric and infrared applications,¹⁶ and high speed optoelectronic devices.^17,18 Ambipolar electric transport has been reported for arsenic-doped Bi₂Se₃ single crystals via electrical measurements.¹⁹ 

In Bi₂Se₃, an important thermoelectric material, light–matter interactions can induce a significant photothermoelectric effect. Numerous recent studies have been conducted on helicity-dependent photocurrent from Bi₂Se₃ surface states, wherein circular polarized light is used for the selective excitation of spin polarized electrons to enhance the photothermoelectric current.^17,20,21 Previous studies have indicated that the acoustic-phonon-mediated cooling of two-dimensional (2D) Dirac fermions in Bi₂Se₃ implies that long-lived hot photocarriers could be created for use in high efficiency photothermoelectric applications.²² Although studies have been conducted on the selective excitation of Bi₂Se₃ by polarized light, studies on the overall excitation of Bi₂Se₃ via unpolarized-light-induced photothermoelectric effect and its dependency on the electrostatic gating remain lacking. Moreover, the ambipolar electric field effect in pristine Bi₂Se₃ devices has not yet been reported.

Herein, we report the photothermoelectric effect on 30 nm-thick epitaxial Bi₂Se₃ capped with a high K dielectric, Al₂O₃, when illuminated by a 633 nm chopped laser. The photocurrent measured at room temperature with no applied bias exhibited sign reversal at the source and drain contacts, indicating the thermoelectric effect. The photocurrent continuously increased with sweeping gate voltage (V_gs) from 10 to −10 V, indicating the variation of the Fermi level (E_F) with V_gs such that E_F moves inside the band gap. The contribution of the surface state conduction increases when E_F lies inside the bandgap, subsequently increasing the photocurrent. A similar V_gs dependency on the Seebeck coefficient (S), i.e., an increase in S value with sweeping V_gs from 10 to −10 V, was observed at room temperature. The similar correlation of the photocurrent and S value on V_gs signifies the photothermoelectric effect on Bi₂Se₃ with contributions from both bulk and surface states. The surface state conduction was further confirmed with four-probe electrical measurements at 32 K. Furthermore, the ambipolar charge transport was observed in the transfer curve, indicating the surface state transport. The ambipolar transport is explained in terms of band bending-induced 2D surface states.

Figure 1(a) shows the SEM image of the multipurpose top-gated field effect transistor device from which most of the data were collected. Bi₂Se₃ thin films were grown via molecular beam epitaxy on a GaAs (111) substrate.²³ The films were patterned using standard e-beam lithography and plasma etching. First, the marker was deposited on the as-grown substrate via platinum (Pt) sputtering. Then, using the pattern reversal technique, the Bi₂Se₃ channel was defined by Argon plasma (150 W, 50 SCCM, 2 min) etching followed by 50 nm of Pt electrode deposition for the metallization. For the top-gate dielectric, a 30 nm layer of Al₂O₃ was deposited via atomic layer deposition. For the electrical contact area of the Al₂O₃ covering, the Pt pad was wet etched on a 300MIF developer for 18 min. Finally, 20 nm of Pt was deposited for the top-gate electrode. The 20 nm-thick Pt is semitransparent and is believed to transmit about ∼15% of the laser incident on it; thus, the photocurrent from Bi₂Se₃ can be measured in the top-gated structure. The photocurrent was measured using a 633 nm red laser with a spot size of ∼2 μm at room temperature and ambient environment. The Bi₂Se₃ thickness was set as 30 nm since the penetration depth of the 633 nm laser on Bi₂Se₃ film is 25.5 nm.²⁴ Furthermore, to avoid the effect of excitation on GaAs, a low laser power of 10 μW was used such that about 1.5 μW of laser power reaches Bi₂Se₃. A low noise current preamplifier (SR570), a Keithley source-measure unit 2400, and an SR 830 lock-in amplifier were used to measure the photocurrent of the chopped light with a chopping frequency of 411 HZ. A Keithley 2400 was used to apply the gate voltage, and a preamplifier was used to collect and convert the photocurrent into a voltage signal that was subsequently fed into the lock-in amplifier; the data were collected through the lock-in amplifier. The left and right electrodes in Fig. 1(a) were used as the source and drain electrodes, respectively.

Figure 2(a) shows the photocurrent as a function of gate voltage, which was measured at six different points of the device in Fig. 1(a) with no applied bias. The inset of Fig. 2(a) displays the measurement positions of the photocurrent. The photocurrent was positive on the source contact side and negative on the drain contact side. The sign reversal of the photocurrent at the source and drain contacts indicates the photothermoelectric effect. Photo-induced current generation includes several mechanisms, such as the photothermoelectric effect,^20,21 photoconductive effect,²⁵ photos bolometric effect,²⁶ and spin-polarized current (for TIs).²⁷ The photoconductive and photo bolometric effects only occur in biased samples. Since no bias is applied in our experiment, these two effects can be omitted. Due to the spin-momentum locking of topological surface states (TSS), the polarized light affords a photocurrent due to the spin imbalance. Herein, photocurrent could not be generated by unpolarized light using this mechanism. Thus, we exclude mechanisms other than the photothermoelectric effect in our device. The photothermoelectric effect arises when the laser causes local heating and a temperature difference occurs at the two electrodes. This temperature difference creates a potential difference (ΔV_PTE) across the Bi₂Se₃ channel, which drives the diffusion of photogenerated carriers from the laser excited spot to the two contacts. The potential difference can be written as ΔV_PTE = −SΔT; more specifically, ΔV_PTE = −(S_BiSe + S_Pt) (T_source − T_drain). Here, ΔV_PTE is the voltage measured by the external circuit, i.e., lock-in amplifier; S_BiSe and S_pt are the S values of Bi₂Se₃ and platinum, respectively; and T_source and T_drain are the temperatures at the source and drain contacts, respectively. Since S_pt ≪ S_BiSe, the photothermoelectric voltage can be written as

If the laser excitation either at the source or drain electrode or inside the channel yields a temperature gradient across the channel, i.e., at the source and drain, the photovoltage (V_ph) and hence the photocurrent can be measured using Eq. (1). The S value of our Bi₂Se₃ was negative, signifying that it is an n-type material at room temperature. Hence, the sign of the photocurrent in Fig. 2(a) agrees with that obtained using Eq. (1): the photocurrent is positive when shined at points P1, P2, and P3 and negative when shined at points P4, P5, and P6. The magnitude of the photocurrent was maximum when shined at points P1 and P6 since a larger temperature gradient (T_source − T_drain) is induced at these points compared to that at the other four points. Similarly, the photocurrents at points P2 and P5 were larger than those at points P3 and P5. All the six curves in Fig. 2(a) exhibit a similarity: photocurrent continuously increases with sweeping V_gs from 10 to −10 V. This implies the tuning of the Fermi level inside the bandgap with V_gs sweeping. The contribution of the surface states to the conduction improves with E_F movement inside the bandgap. Due to the high mobility of the surface states carrier, the photocurrent increases with V_gs sweeping as E_F moves inside the bandgap. Thus, the simple photocurrent measurement revealed the presence of surface states in our sample. Moreover, according to Eq. (1), V_gs modulation on photocurrent should stem from the V_gs modulation on the S value. Therefore, we measured the S value for the device in Fig. 1(a) as a function of V_gs inside a closed cycle refrigerator (CCR). To measure S, the resistance and hence the temperature of the two thermometers (electrodes) were measured using two lock-in amplifiers, a heater current of 20 mA was supplied by a Keithley 2400, a gate voltage was applied by another Keithley 2400, and the resulting voltage between the two electrodes was measured using a Keithley 2182A nanovoltmeter. Figure 2(b) displays the measured S values at 32 and 295 K. At 295 K, S increased from −109.3 to −111.7 μV/K, and at 32 K, it increased from −24.4 to −25.6 μV/K on sweeping V_gs from 10 to −10 V. The photocurrent variation in Fig. 2(a) is similar to the S variation in Fig. 2(b). At all V_gs, the transport is dominated by bulk states, resulting in the unipolar sign of S. When V_gs is negative, the Fermi level cuts the bulk and surface conduction levels; hence, the transport is accomplished solely by electrons. When V_gs moved toward positive values, the Fermi level cuts the bulk conduction level and the surface valence bands. That is, the bulk conducts with electrons, whereas the surface states conduct with holes. In other words, there is mixed conduction in this condition. The mixed conduction leads to a decrease of S values; therefore, the monotonic decrease of S is observed when V_gs is moved from −10 to +10 V. The band diagram for this situation is described later in Fig. 3(c). Hence, we conclude that the V_gs-modulated photocurrent with the V_gs-modulated S value stem from both the surface and bulk states. Small variations in the S value in this V_gs range indicate the large bulk density of our Bi₂Se₃. The comparatively larger modulation in the photocurrent than that in the S value with V_gs sweeping is probably due to the AC measurement (lock-in technique) of the photocurrent, wherein large bulk currents (dark currents) are suppressed. The estimation of the temperature increase by laser irradiation with the measured photocurrent, S value, and resistance yields a maximum value of 4.8 mK when shining at the electrode–channel interface, i.e., points P1 and P6.

After the photocurrent measurements, we performed electrical measurements on the same device; the obtained results further support the surface state conduction in our sample. Figure 3(a) shows the temperature dependence of the resistance measured in CCR using the Keithley 2400 source meter. The square resistance measured by the four-probe geometry linearly decreased with the temperature up to 32 K, exhibiting typical metallic behavior, wherein phonon scattering is dominant.²⁸ The unavoidable presence of intrinsic defects, like Se vacancies in the growth process wherein electrons are doped in the Bi₂Se₃ film,^19,29 cause the metallic conductivity in Bi₂Se₃. The inset of Fig. 3(a) displays the I–V measurements between the source and drain contacts in the two-probe geometry. The linear I–V indicates the ohmic contact between Pt and Bi₂Se₃. Figure 3(b) shows the transfer curve measured using the four-probe geometry at 32 K using two 2400 source meters. The curve displays clear ambipolar charge transport, but the resistance modulation is very small. The ambipolar charge transport implies the surface state conduction in the topological insulator. As explained above, the surface state conduction is due to band bending at the surface–bulk interface, accompanied by a high K dielectric, i.e., Al₂O₃, and the gate–sensitive interface of Al₂O₃–Bi₂Se₃. Additionally, selenium vacancies on the Bi₂Se₃ surface can modify the topological surface states (TSS) but cannot remove them.³⁰ Interface TSS located at the interface between the gate dielectric and Bi₂Se₃ are gate sensitive, contrary to bulk states.³⁰ Thus, the possibility of Fermi level tuning of interface TSS exists with a strong gate field. When the interface TSS are gate sensitive, V_gs can deplete the bulk charge carriers at the interface, causing the band bending of bulk toward the surface. Although the Fermi level at the bulk remained almost constant, the Fermi level at the surface varied with the vertical shift of the surface bands [as shown in Fig. 3(c)]. At V_gs ∼ −1.25 V, the Fermi level at the surface was at the Dirac point [Fig. 3(c) II)], and hence, the maximum resistance was observed at this V_gs [Fig. 3(b)]. When V_gs < −1.25 V, the Fermi level moved below the Dirac point toward the valence band of the surface state, resulting in hole conduction. Figure 3(c-III) displays the schematic for V_gs = −10 V. In contrast, when V_gs > −1.25 V, the Fermi level moved above the Dirac point toward the conduction band of the surface state, resulting in electron conduction. At V_gs = 10 V [Fig. 3(c) I], some band bending still occurred; otherwise, the curve became saturated when the conduction was fully dominated by the bulk. With this simple electrical measurement, our study confirms the presence of surface states, although the observed states may simply be either 2D electron gas (2DEG) states or topological states. The coexistence of topological and 2DEG states on the Bi₂Se₃ surface has been previously reported via the ARPES measurements.³¹ The very small gate modulation observed in resistance without the compensating Selenium vacancies via atomic doping well agrees with the ∼2% resistance modulation in arsenic-doped Bi₂S₃.¹⁹ Compared with the ambipolar charge transport, i.e., sign changes of charge carriers observed in transport measurement at 32 K as shown in Fig. 3(b), and the S measurement at same temperature did not exhibit the ambipolar nature as shown in Fig. 2(b). To observe the sign change in the Hall coefficient and S value, the Fermi level in the bulk should be tuned from the conduction band to the valence band,³² which was not possible in our sample. This illustrates why the peak of the photocurrent was not observed in Fig. 2(a) around V_gs = 0 V as one may expect the maximum of photocurrent and S value for V_gs = 0 V referring to the band bending scheme in Fig. 3(c).

Photodetector devices should importantly possess stability and repeatability of V_ph upon laser shining. Thus, we conducted the cyclic exposure of laser on another device of similar thickness that was not covered by Al₂O₃ [Fig. 1(b)] to study the stability and repeatability of V_ph. The laser was signed at five different positions, and the induced voltage signal was measured using a Keithley 2182A nanovoltmeter with no applied bias [Fig. 1(b)]. V_ph was obtained by subtracting the voltage measured in the dark from the voltage measured with the laser shining. Figure 4(a) displays the measured V_ph with time in a cycle of the laser off and on, which is depicted in blue and orange colors, respectively. As observed, V_ph sharply increased when the laser was turned on and decreased to nearly zero immediately when the laser was turned off. This sequence of increase and decrease of V_ph for the laser on and off states was observed for multiple cycles, demonstrating the suitability of the device for photoswitching or the high quality of photodetectors. Furthermore, the positive and negative V_ph as a function of laser position, i.e., positive at P1 and P2, negative at P4 and P5, and nearly zero at P3, indicates a photothermoelectric effect similar to that observed for the gated device in Fig. 1(a). The quality of a photodetector device is determined in terms of the rise and decay times of the signal as the light turns on and off. Compared to the reported rise and fall times of Bi₂Se₃ nanowires of 520 and 730 ms, respectively,³³ the measured both rise and fall times of 0.029 s (29 ms) in Fig. 4(b) indicate the high quality of our device.

In summary, we fabricated a multipurposed top-gated Bi₂Se₃ FET and measured the photocurrent, S value, R–T curve, and transport curve. The measured photocurrent was thermoelectric in nature and gate dependent. An increase in the photocurrent toward the negative gate voltage indicated the tuning of the Fermi level inside the bandgap and the enhanced contribution of the surface states. The gate dependency of the S value followed that of the photocurrent curve, further confirming the photothermoelectric effect in our sample. The low temperature electrical measurement showed the ambipolar transport, indicating the presence of surface states. Combining both the electrical transport curve and photocurrent, we conclude that both bulk and surface states contributed to the observed photothermoelectric current. Furthermore, we showed the high quality of our device, which could be used in photo switches or photodetectors in terms of fast rise and decay times.

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean Government (MIST) (Grant No. 2017R1E1A1A01074650).

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.