Thermoelectric transport at F₄TCNQ–silicon interface

Interfaces play a key role in semiconductor devices. The electron transport at conducting interfaces is of importance in the design of high-electron-mobility-transistors,^1–3 superlattice structure devices,^4–6 and energy conversion materials such as photovoltaics^7–9 and thermoelectric materials.^10–12 In particular, novel hybrid organic-inorganic materials have drawn increasing attention in research for diverse range of applications in energy, environment, and healthcare.¹³ 

This work focuses on the charge transfer at an organic-inorganic interface, which shows promise in designing transfer-doped semiconductor nano-devices. Conventional doping methods (e.g., ion implantation) are oftentimes destructive processes introducing excess defects. The activation process ubiquitously requires solid-state thermal diffusion, which cannot be applied to materials that are heat sensitive. Devices with ultrafine structures would also suffer from non-negligible statistical variation or deactivation of dopants.^14,15 The other main drawback of conventional doping methods is that they all involve incorporating aliovalent impurity ions, which create long range Coulomb potentials that scatter conduction electrons and lower their mobility. Modulation doping or remote doping has been proposed for 2D structures to overcome these shortcomings and has been implemented in transistor technology.¹⁶ Over 22 years (from 1978 to 2000), mobility enhancement by 4 orders of magnitude has been reported in GaAs, which was achieved by spatially separating parent atoms from conduction electrons.¹⁷ Another alternative doping strategy is surface doping of inorganic semiconducting materials by organic molecules, which was proposed fairly recently.^18–21 Both in modulation doping and surface doping, charge carriers are confined to the interface of organic-inorganic materials and form a 2D electron gas (2DEG). To extend the advantages of 2DEG to 3D devices, one can envision forming superlattice structures by alternating organic-inorganic layers. Such hybrid superlattices have been theoretically evaluated for thermoelectric applications.²² Alternative 3D structures such as nanoparticles, nanowires, and holes embedded in host matrices, specifically for thermoelectric applications, have been proposed and discussed in detail.^23–28 

In this work, we fabricated a device to investigate the charge transfer of the tetrafluoro-tetracyanoquinodimethane (F₄TCNQ)–silicon interface. F₄TCNQ is a fluorinated TCNQ derivative and has an exceptionally high electron affinity of E_a = 5.24 eV.²⁹ It is well known as a strong electron acceptor, to dope molecules p-type by forming a charge transfer organic complex, as well as to enhance hole injection by energy level alignment at organic–metal interfaces.^30,31 Therefore, F₄TCNQ has been widely applied in organic light emitting diodes (OLEDs),^32,33 photovoltaics,^34,35 organic field-effect transistors (OFETs),³⁶ etc. P-type surface doping was also achieved at inorganic semiconductor surfaces, such as diamond,³⁷ graphene,³⁸ and VO₂,³⁹ and investigated by photoemission spectroscopy (PES) techniques. However, as the most pervasive semiconductor material for electronic devices, Si has only sparsely been examined experimentally as the active layer for transfer doping by F₄TCNQ. Existing reports mainly focus on electronic structure study by PES^40–42 without details on the resultant transport properties. Therefore, we seek to investigate the electrical transport properties of the F4TCNQ-Si interface. The lowest unoccupied state (LUMO) of F₄TCNQ lies below the silicon valence band maximum (VBM); hence, the electron affinity of F₄TCNQ is larger than the ionization energy of Si [Fig. 1(a)]. This favors electron-transfer from silicon to F₄TCNQ molecules, and Si is thus p-type doped near the interface. The holes are confined in the direction normal to the interface due to the established electrostatic potential but free to move in the parallel direction. XPS study shows that the shift of Si 2p binding energy peak saturates when the in situ deposited F₄TCNQ thickness is 2 nm,⁴³ meaning the F₄TCNQ molecules farther to the interface do not contribute to the charge transfer, as illustrated in Fig. 1(b). Our previous calculation indicated that physisorbed F₄TCNQ self-assembled monolayers can efficiently dope silicon, achieving hole concentrations as high as 10¹³ cm⁻².⁴⁴ 

A combination of enhanced mobility by transfer doping and low thermal conductivity of F₄TCNQ could be utilized to design a high performance F₄TCNQ–nanostructured Si (e.g., nanopatterned holey Si or nanowires) hybrid thermoelectric materials. However, this concept relies on a fundamental understanding of charge transfer and thermoelectric transport at the interface of F₄TCNQ–silicon, which is thereby the main focus of this work.

In order to make a device to measure the interface properties, we built a silicon channel, as shown in Fig. 2. The device Si(100) layer (4″ SOI with 2 µm BOX) was thinned down to ∼100 nm by reactive ion etching (RIE). The subsequent blanket boron implantation yielded a mild hole concentration of ∼10¹⁶ cm⁻³ in the Si layer. Four side contact (5 µm × 20 µm) areas were then exposed by resist and underwent another boron implantation process, which yielded ∼10²⁰ cm⁻³ only in the contact areas. The Si layer was etched down to BOX to form arrays of 30 µm × 200 µm Si mesas for the following metallization. 1 µm Al with a 50 nm Au cap was deposited as contacts for transport measurements. The device configuration is shown in Fig. 2. The heater, thermometers, and contacts are annotated. Finally, rapid thermal annealing was conducted at 500 °C for 30 s to achieve Ohmic contact.

Next, we deposited F₄TCNQ on the silicon device by thermal evaporation. The thermometers and the electrodes are then in between the silicon and F₄TCNQ, allowing us to characterize the interface. F₄TCNQ is insulating. Despite withdrawing electrons from Si upon contact, the electrical conduction of the F₄TCNQ film is negligible compared to the Si layer. Therefore, it was feasible to directly deposit F₄TCNQ onto the metallized device, without shorting the metal lines. Prior to the deposition, the device was dipped in buffered oxide etch (BOE) for 25 s to create a passive H-terminated Si surface, in order to prevent charge confinement by the surface states. The metal pads were then covered by a shadow mask for the ease of wire bonding after deposition. The F₄TCNQ crystal powder (Sigma-Aldrich, 99.9%) and the device were loaded into a homebuilt miniature thermal evaporation chamber with a base pressure of <10⁻⁶ Torr. F₄TCNQ was degassed below 100 °C and then sublimated at 140 °C for deposition. Surface morphology was characterized by atomic force microscopy (AFM). The XTEM sample with an F₄TCNQ/Si interface was prepared by a focused ion beam (FIB) lift-out process, and the interface morphology was then characterized by transmission electron microscopy (TEM) along Si [110]. The electrical resistance was measured using the four-point probe method, both before and after depositing F₄TCNQ. The Seebeck coefficient was measured in a homebuilt 2D transport measurement system incorporated in a Janis cryostat.

It was found that the F₄TCNQ has a relatively low sticking probability on H–Si(100), which is commonly observed for molecules physically adsorbed on the inorganic substrate.⁴⁵ Furthermore, due to the limited wettability, F₄TCNQ adsorbates tend to coalesce into densely packed islands on Si. In order to achieve full coverage and maximize the charge transfer, a relatively high vapor flux is required to create a high nucleation rate. Therefore, we used a higher F4TCNQ temperature of ∼140 °C compared to other PES studies.^40–42 The net deposition rate was calibrated as ∼1.5 Å/s. Figure 3 shows the surface morphology of F₄TCNQ with a root mean square (RMS) roughness of ∼17 nm under this deposition condition.

The interface is investigated by XTEM, as shown in Fig. 4. The active Si layer is ∼100 nm in thickness, on a 2 µm buried oxide layer, so that the subsequent transport measurements are minimally affected by the substrate. As shown in Fig. 4(a), continuous coverage of F₄TCNQ on Si was achieved, with an average thickness of ∼150 nm. In essence, only the first few layers of the molecule would contribute to the charge transfer, leaving the majority as neutral F₄TCNQ bulk. Figure 4(b) is the high-resolution image along the Si [110] axis. It exhibits a smooth interface without observable voids or interlayers. Therefore, an effective charge transfer can be expected across the interface. Nevertheless, the orientation of molecules on the Si surface is still unclear, which could affect the transfer doping efficiency.⁴⁴ Future work using STM⁴⁶ or PES⁴¹ techniques may shed light on orientation of the molecules.

In order to evaluate the charge transfer, and its effect on power factor, we performed electrical transport measurements. I-V measurements by the 4-probe method were conducted first on the mildly p-doped, uncoated Si device and then on the same device deposited with F₄TCNQ. As shown in Fig. 5, Ohmic IV curves were achieved in both cases. The measured resistance of F₄TCNQ–Si (401 kΩ) is 10 times lower compared to the resistance of a plain Si device (4.36 MΩ). The result provides direct evidence that charge transfer at the interface occurred and that silicon is surface doped by the physically adsorbed F₄TCNQ.

The Seebeck coefficient was then measured using a microheater and two calibrated thermometers, as shown in Fig. 2. When applying a current to the heater, a temperature difference is built up across the two ends of the Si mesa as a result of Joule heating inside the heater. Therefore, an electrical potential is produced due to the Seebeck effect. Since the thermometers are in direct contact with the Si surface, the Seebeck voltage could thereby be measured. We observe a large drop in the Seebeck voltage from −208.9 ± 12.3 µV for mildly doped, uncoated Si to −73.9 ± 3.7 µV after deposition of F₄TCNQ [see the steady state voltage shown in Fig. 6(a)]. In order to extrapolate the actual Seebeck coefficient (S = $−VSΔT$⁠), the corresponding temperature differential must be measured. The temperature of the device could be extracted from the temperature-dependent resistance of the thermometers. Separate temperature calibrations were conducted for each thermometer. The sample’s temperature was raised using an external heater in the sample holder, and the resistance of the thermometers was measured versus temperature. A minor increase in thermometer resistance after depositing F₄TCNQ was noted due to the presence of the molecules on the gold lines, and therefore their resistances were recalibrated afterwards.

During the Seebeck measurement, when the heater turns on, the resistance (by the 4-point method) of the thermometers ramps up, as shown in Figs. 6(b)–6(e). Using the calibration curves, we could then extrapolate $ΔT(TC1−TC2)$⁠, which reached to a steady state after ∼20 s, corresponding to the Seebeck voltage plateau. The Seebeck coefficient determined by this approach for the uncoated Si device is 594.6 ± 38.0 µV/K and for the F₄TCNQ–Si device is 243 ± 12.3 µV/K. The sign of Seebeck coefficients confirms that both devices are p-type doped. It was found that the power factor (S²/R) of silicon had a 75% enhancement after depositing F₄TCNQ.

Since the screening length of surface-doped holes is estimated to be less than 10 nm, the majority of the coated device is not contributing to the conductance. Given that the resistance had 10 times’ reduction, and carrier mobility should be unchanged in F₄TCNQ–Si, an ∼100 times enhancement in hole concentration can be estimated in the modulation doped region in proximity to the interface. The obtained Seebeck value of 243 ± 12.3 µV/K for the F₄TCNQ–Si sample could be interpreted as the Seebeck coefficient of the 2D hole gas since the transferred charge accumulated near the surface is the dominant contributor to the Seebeck voltage. The thermal conductivity is not affected, and we estimated a similar thermal resistance with and without F₄TCNQ since in this case, measurements are performed for the interface and the lattice thermal conductance of silicon is the dominant heat transfer mechanism.

These results serve as a proof of concept of the potential of organic–Si interfaces with large charge transfer for power factor optimization for thermoelectric device application. The issue of limited conductance enhancement due to the short screening length brings in the necessity of having nanostructured Si with a high surface-to-volume ratio (e.g., nanopatterned holey Si), in which case the transfer doping can be maximized, and the necking size is comparable to the screening length of transferred charges so as to freely transport throughout the material. Future study on manipulating the molecule orientation and surface treatment to minimize surface states as a charge trap is expected to further enhance the power factor.

The authors gratefully acknowledge Petra Reinke, Nathan Swami, Arthur Lichtenberger, and Kyusang Lee for insightful discussions. This work was supported by the National Science Foundation under Grant No. 1723353 (N.L. and M.Z.).