Ohmic contact formation for inkjet-printed nanoparticle copper inks on highly doped GaAs

One of the main parameters that needs to be evaluated for a metal/semiconductor contact is its contact resistivity. For high performance and consistency of integrated circuits, a low contact resistivity and a small Schottky barrier height at the metal/semiconductor interface are essential. Additionally, these need to be combined with stable contact behavior in terms of chemical structure, adhesion and resistivity. When the metal and the semiconductor are not in contact, their Fermi level energies might differ, which leads to different work functions (Φ) (figure 1(a)). As soon as they are in contact, Fermi levels of metal and semiconductor align at equilibrium. Depending on the work function of metal and n-doped semiconductor, either a Schottky or an ohmic contact can develop. An ideal contact condition for a metal with an n-doped semiconductor occurs when the work function of the metal (Φ_M) is smaller than the work function of the semiconductor (Φ_S)_, resulting in an ohmic junction. At equilibrium, electrons move barrier-free between semiconductor and metal (figure 1(b)). For an n-doped semiconductor, when the Fermi level of the semiconductor is higher than the metal and Φ_M > Φ_S, a Schottky junction is formed. In a Schottky junction, a barrier for electrons to move from metal to semiconductor is formed, which is called the Schottky barrier (φ_B). By decreasing the work function differences between metal and semiconductor, the depletion region at the interface is narrowed, and the barrier height is reduced, which leads to the tunneling of charge carriers between semiconductor and metal. Therefore, in this case, a semi or quasi ohmic behavior of metal on semiconductor might be observed.

Zoom In Zoom Out Reset image size

Among all metals, Cu according to its respective work function, plus its lower cost compared to Au and Ag, is a proper candidate for contacts on GaAs [15]. In contrast to Cu, up to now Au and Ag are usually applied on GaAs in an alloy form or need an additional interface layer to achieve a low contact resistivity [13, 14]. However, the doping level of GaAs might also lead to a low contact resistivity of Ag on GaAs [29]. N-doping in semiconductors provides free electrons in the outer layer of the valence band region. These electrons need less energy to be lifted from the valence band into the conduction band, than the electrons which cause the semiconductivity in semiconductor. Because of the high Fermi-level pinning in GaAs caused by its surface state density, formation of band bending leads to the formation of a depletion region on the surface of the semiconductor similar to a pn junction. A high doping level (n = 1 × 10¹⁸–1 × 10¹⁹ cm⁻³) in GaAs decreases this depletion region and facilitates the movements of charge carriers at the semiconductor-metal interface [30]. In addition to Cu, in this work Ag was depsotited as well as contact material on GaAs in order to investigate the effect of the semiconductor doping level on the electrical contact with Ag.

The energy level alignment of the metal/semiconductor contacts was assessed based on the work function derived from ultraviolet photoelectron spectroscopy (UPS) and the valence region of metals and semiconductor derived from x-ray photoelectron spectroscopy (XPS). Cu and Ag NP inks were deposited on GaAs by inkjet-printing and sintered with required sintering methods. For the proof of concept, both metals were also deposited by electron beam physical vapor deposition (EBPVD), in order to compare the electrical contact of the printed ink to the pure evaporated metals on GaAs.

Figure 2 shows UPS results of the secondary electron cut-off of all samples, which reveals the work function of the metal and the semiconductor. UPS mirrors the broad distribution of states of the conduction band in metal and the valence band of a semiconductor. Therefore, the work functions of metal and semiconductor as well as the energy level alignment at the contact interface can be derived from UPS. XPS of the valence band region of the n-doped GaAs, is shown in figure 2 and illustrates the valence band maximum of the semiconductor. The region of low binding energy in the XPS spectrum is attributed to electrons stemming from the sample surface and thus represents the GaAs valence band in high detail. XPS in that context was selected instead of UPS since a higher penetration depth in the sample can be expected for x-ray photons due to their higher energy compared to UV radiation. Hence, the XPS signal characterizes the actual sample surface better based on a strong signal from the semiconductor with a negligible contribution of residual molecules that were adsorbed on the sample surface during sample preparation under ambient conditions.

Zoom In Zoom Out Reset image size

A highly n-doped GaAs wafer with a doping level of n = 1 × 10¹⁹ cm⁻³ (from Azur SPACE) was cleaned by a standard solvent cleaning method [31] to provide a dust-free and clean surface. Solvent cleaned GaAs shows a work function of Φ = 3.8 eV which is lower than the work function for GaAs reported in literature of Φ = 4.2–4.7 eV [32]. This shift in the work function is predictable due to adsorbed solvent molecules on the surface. However, after printing the ink on GaAs, the printed structures are annealed or sintered, which leads to uncapping of the ink and contact formation at the interface. The applied high temperature required for sintering (higher than 250 °C) might cause partial evaporation of the adsorbed solvent on the GaAs surface. Therefore, due to the high temperature of the sintering process, UPS and XPS studies of an annealed surface of GaAs were necessary. A cleaned and annealed GaAs surface (15 min at 250 °C) shows a Φ of 4.2 eV. Evaporated Cu, and Cu ink feature a work function of 4.4 eV. In comparison, for evaporated Ag Φ_m is 4.2 eV and Φ_m is 4.3 eV for Ag ink.

Figure 3 shows the energy level diagram derived from UPS measurements (figure 2) for differently treated GaAs samples, evaporated Cu, laser-sintered Cu ink, annealed Ag ink and evaporated Ag. Cu ink shows a barrier height Φ_B of 0.74 eV on solvent-treated and annealed GaAs. On the same surface, Ag ink showed a barrier height of 0.64 eV. Reported values for barrier heights of pure Cu and pure Ag on GaAs range between 0.88 and 0.99 eV [33]. Therefore, achieving an ohmic contact between NP inks and n-doped GaAs surfaces with low contact resistivity can be possible because of a low barrier height of 0.74 or 0.64 eV. Such ohmic behavior is governed by tunneling of charge carriers at narrow depletion regions due to the impurities in NP inks and a high doping level in GaAs [34].

Zoom In Zoom Out Reset image size

As NP inks are uncapped, residual organic groups may still alter their work function and energy levels. For example, Wang et al investigated the tunability of NP inks' work function by varying the capping groups. Thus, capping molecules which are present in NP inks change the uncapped metal work function leading to a reduction of the barrier height at the metal/semiconductor interface. The literature reported values of 0.88–0.99 eV are presented for pure deposited metals on GaAs and therefore are higher than the values reported for the NP inks [35].

As it is mentioned in section 2.1, a low contact resistivity of the ink on GaAs is a critical factor to provide an ohmic contact. transmission line method (TLM) was used to assess the resistivity and the electrical behavior of the ink on GaAs. Therefore, a series of evaporated and inkjet-printed interdigitated structures of Ag and Cu with channel lengths of 82, 144, 205, 271, 329 and 405 μm were deposited on GaAs in order to measure and interpret their electrical characteristics (current–voltage (IV) measurements) and calculate the corresponding contact resistivity.

As depicted in figures 4(a) and (b), IV characteristics obtained by TLM, for Cu deposited via evaporation as well as Cu ink on GaAs show a close to ohmic contact characteristic. This ohmic behavior shown by evaporated Cu indicates the in principle possibility of an ohmic contact on GaAs for pure Cu. The ohmic behavior of the Cu NP layer can be attributed to a good fusion of Cu NPs at the interface to the semiconductor, induced by the application of the laser sintering process for uncapping of the metal, which enables the NPs to merge well [16].

Zoom In Zoom Out Reset image size

In figures 4(c) and (d), printed Ag ink and Ag deposited via EBPVD on GaAs show rectifying behavior of the current flow which indicates Schottky behavior. Since Ag forms an ohmic contact on GaAs usually when deposited as an alloy or in the presence of intermediate layer, this Schottky behavior of sintered Ag and evaporated on GaAs is expected [13, 14].

According to the results derived from UPS, XPS and electrical behavior evaluations, the Cu ink showed a low barrier contact of 0.74 eV on GaAs with a contact resistivity of 10 mΩ cm² leading to a desired ohmic metal contact on GaAs. Further processing optimizations were investigated to reduce the contact resistivity.

The quality of the sintered metal contacts depends on the quality of the printed layer, the sintering conditions and laser fluence applied [7]. The small size of NPs in the range of 20–100 nm enhances the printability of these inks [36]. Additionally, it decreases their melting point to a lower temperature compared to their bulk forms due to melting temperature depression in NPs [37]. Lower melting temperature facilitates the sintering process at a moderate working temperature being less than 300 °C, which very likely reduces related manufacturing costs [38].

In addition to the principle electrical contact formation, the metal/semiconductor contact quality depends on the adhesion and merging of grain boundaries of metals near to the interface to a semiconductor. In that context, Niittynen et al investigated the laser sintering method for the uncapping of Cu NP inks. They discovered that the laser fluence level affects the quality of Cu particles near to the interface of metal and substrate [16]. A similar behavior exists for Ag ink and was investigated by Xiao et al [39]. They showed that different curing temperatures influence the Ag grain boundary formation and that organic NP capping species improve the adhesion of the Ag ink on the substrate. Moreover, ink components and sintering conditions might improve the electrical quality of the sintered metal and reduce the contact resistivity of metal on GaAs.

Thus, in order to evaluate the efficiency of the sintering process, Cu NP structures with defined form and thickness were inkjet-printed on GaAs. Then laser sintering was performed with a picosecond laser to convert the Cu ink into the final desired metallic Cu layer. Laser fluence was varied in order to realize a Cu layer with an optimally low contact resistivity.

To achieve a low contact resistivity at the metal/semiconductor interface and to realize a highly conductive metal layer, an optimal level for the applied fluence needs to be found. Application of this optimized fluence level means exposing the sample to a laser power, which leads to a high amount of uncapped NPs and simultaneously avoids ablation of the printed structure. This amount of uncapped NPs improves the conductivity of the contact grids and reduces the contact resistivity. However, fluence values higher than this optimized fluence value lead to ablation of the ink resulting in a non-ohmic electrical contact.

In order to find the optimal level for the laser fluence, in a first step, the sheet resistance of the sintered Cu layer at different fluences was measured via a four-point probe method to evaluate the conductivity of sintered Cu. The structures were printed on glass slides to omit the influence of the metal/semiconductor contact resistance on sheet resistance. Figure 5 shows the results for the respective sheet resistances of these structures sintered at different fluence levels. It can be seen that with increasing laser fluence the sheet resistance decreases from 0.45 Ω/□ at 1.80 mJ cm⁻² to 0.23 Ω/□ at 2.46 mJ cm⁻². With further increase of the laser intensity, the sheet resistance increases again (0.26 Ω/□ at 2.83 mJ cm⁻²).

Zoom In Zoom Out Reset image size

The minimum sheet resistance achieved defines the optimum fluence level for the sintering process.

After the printing on glass, in a second step, a series of Cu ink TLM structures, with finger distances of 34, 102, 170, 238, 306 and 408 μm (figure 6), were inkjet-printed and laser-sintered on GaAs with a thickness of 400 nm. To evaluate the contact resistivity of the Cu layer on GaAs, the resistances of these laser-sintered TLM structures were calculated from IV measurements and plotted versus the finger distances (figure 6).

Zoom In Zoom Out Reset image size

The contact resistivity then was calculated based on the respective formulae [40] and resulted in a value of 8 mΩ cm² at a fluence of 8.3 mJ cm⁻². This low contact resistivity provides a low-barrier electron transfer between metal and semiconductor. The processing window for fluence values resulting in an effective and efficient laser sintering for 400 nm Cu on GaAs ranges from 3.9 to 8.3 mJ cm⁻², resulting in a decrease of contact resistivity from 10 mΩ cm² for 6.3 mJ cm⁻² to 8 mΩ cm² for 8.3 mJ cm⁻². It has to be noted that the optimum fluence level of the NP ink is different on GaAs compared to glass, which is explained by the different heat capacity of these two materials.

SEM images (figure 7) reveal that the increase of the fluence values for sintering the Cu NP ink deposited on GaAs up to the optimal fluence level (below 8.3 mJ cm⁻²) increases the fusion of NPs. As it is shown in figure 7(d), a fluence level higher than 8.3 mJ cm⁻² leads to ablation and deformation of Cu NPs. Fluence levels less than 3.9 mJ cm⁻² lead to a Cu layer with less conductivity, high contact resistivity and bad adhesion of the metal on the semiconductor.

Zoom In Zoom Out Reset image size

Apart from that, the thickness of the printed layer obviously strongly influences the electrical behavior of the sintered Cu NP layer. From the IV characteristics of TLM structures shown in figures 8(a) and (b), it can be seen that a 250 nm thick layer of Cu of leads to a Schottky contact behavior compared to an ohmic contact behavior achieved for a layer with a thickness of 400 nm. The main cause for this differing behavior can be explained by SEM images of the surface of the two differently thick contacts (figures 8(c) and 9(d)). It can be seen clearly that at a thickness of 250 nm (figure 8(c)), distinct cracks form in the copper layer. In contrast to that, the copper layer shows a continuous coverage in the case of a thickness of 400 nm (figure 8(d)). Furthermore, the fusion of the NPs seems to be much more pronounced in the case of 400 nm. Thus, for the Cu NP ink presented in this work, a minimum layer thickness of 400 nm can be defined as a requirement for the deposition of a conductive contact grid on GaAs.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As a final step of evaluation, XPS measurements were performed on the Cu NP layer before and after laser sintering (figure 9). Peak fitting of the corresponding Cu 2p3/2 peak before sintering revealed the contribution of seven different components (figure 9(a)). These contributions can be assigned to metallic Cu bonds as well as to Cu–O and Cu–C bonds in the binding energy region between 933 and 940 eV. Associated satellite peaks were observed in the region between 940 and 945 eV, resulting from different bonds between Cu as well as from NP capping groups in the sample. A metallic Cu peak and Cu₂O peaks appear between 933 and 934 eV and are not easily distinguishable due to many different groups in the ink formulation (figure 9(a)). The fitted peak at 935 eV is attributed to CuO. From 935 to 940 eV, a broad peak can be attributed to the bond between Cu and carbon groups.

After laser sintering, XPS shows only Cu metallic bonds in the fitted Cu2p3/2 peak at a binding energy of 933 eV (figure 9(b)) and all of the satellites have disappeared.

The presence of only Cu metallic bonds indicates an efficient uncapping and subsequent merging of the Cu NPs resulting in the conversion to metallic Cu.

The XPS signal that can be attributed to carbon based components before sintering shows the contribution of C–C at 284 eV, Cu–C between 289 and 290 eV as well as C–O peaks at 286 eV for single- and 288 eV for double bonds (figure 9(c)).

After laser sintering, the contribution of C–C bonds at 284 eV is clearly reduced, and no Cu–C peak can be detected. The remaining carbon contribution, stemming from single- and double carbon–oxygen bonds at 286 and 288 eV, is assigned to residual uncapped organic groups on the sample surface (figure 9(d)).

A comparison of the achieved resistance values to those in literature further supports our results on the efficiency of the presented printing and sintering process for the Cu NP ink. With respect to the sheet resistance, for example, Soltani et al applied inkjet printing to deposit a similar commercial Cu NP ink on the surface of silicon wafer [7] and, in contrast to our work, used a continuous wave laser with a wavelength of λ = 808 nm for annealing. The lowest sheet resistance achieved was 1.2 Ω/□ which is five times higher than the optimum sheet resistance of 0.23 Ω/□ presented above. As this commercial ink features a composition similar to the ink presented herein, it is possible to compare these two laser sintering methods directly [7]. Jun et al recently published inkjet printing results on a synthetic NP ink which was uncapped via thermal annealing. The lowest sheet resistance reported was 1 Ω/□ with a resistivity of 1.2 × 10⁻⁶ Ω m. By adding some copper complexes to that ink, a lower sheet resistance of 1 × 10⁻³ Ω/□ with a resistivity of 1.2 10⁻⁹ Ω m was realized [27]. In that context, also Kim et al showed that addition of a reducing agent to the NP ink improves the quality of printed copper layer after sintering [26].

In that context, it should be noted that the thickness of deposited conductive thin films considerably influences their resistivity. In particular, layers with a thickness in the nm regime in principle show a higher resistivity than bulk Cu [41, 42], due to a strong dependence of the charge carrier mobility on the crystallinity and nanostructure of polycrystalline materials [41, 42]. With respect to that Kang et al [25], for instance, reported a resistivity of 3.6 × 10⁻⁸ Ω m for an inkjet-printed and thermally-sintered reactive Cu ink layer on glass with a thickness of 3.2 μm. In our study, we achieved a resistivity of 1.1 × 10⁻⁷ Ω m for a laser-sintered Cu layer with a thickness of 400 nm. For the sake of completeness, bulk Cu shows a resistivity of 1.7 × 10⁻⁸ Ω m [25].

Compared to literature at first glance, the value of 8 mΩ cm² for the achieved contact resistivity seems to be high, in particular when comparing it to reported values for Cu in the range of 10^–2–10^–4 mΩ cm² [43–47]. However, these low resistivity values have not been realized with Cu NP inks but were found for Cu alloys combined with an additional barrier layer, the latter of which is needed due to the high diffusion rate of Cu in GaAs [15]. In addition, it needs to be noted that being in the nanometer regime the thickness of the printed Cu NP layer combined with impurities stemming from remaining organic groups in the NP Cu ink can increase its contact resistivity [7, 48]. Thus to the best of our knowledge, the presented contact resistivity values of Cu NPs on GaAs belong to the lowest so far reported in the literature for Cu NP inks.

In order to deposit the selected Cu Dycotec ink (DM-CUI-5002) Ag Sigma Aldrich ink (Sigma Aldrich 736473) for this experiment and achieve the required results, the following method was determined and carried out in a cleanroom environment (ISO class 6) as follow. 2 × 2 cm pieces of GaAs wafer with a doping level of, were first cleaned by standard wafer cleaning methods with different polar and non-polar solvents [31]. For Cu printed sample, the GaAs wafers were subsequently dip-coated for 60 s in a 5% acidic solution composed of acetic acid + hydrogen peroxide 1:1 and de-ionized H₂O and then blown optically dry with an N₂ gun. They were then placed on a hot plate for 5 min at 120 °C. Immediately prior to printing, each substrate was again placed on a hot plate for 120 s to ensure minimal H₂O on the substrates. For Ag printed samples, only standard cleaning was applied for GaAs wafers. Using a PiXDRO LP50 printing platform, in combination with a FUJIFILM Spectra SE 128 print head, Cu ink and Ag ink were deposited and fully covered each substrate. The print head, as well as printing platform parameters, are presented in table 1. After printing, the substrates were dried in a vacuum oven at 800 mbar below ambient pressure, and at 50 °C for 2 h.

For laser sintering, a diode-pumped, picosecond laser with a wavelength of 1064 nm was used. The used EdgeWave PX200-P1-GF laser is based on a Nd:YVO4 crystal, has a maximum pulse repetition rate of 50 MHz and reaches a power of up to 100 W. Series of TLM structures were laser-sintered with an adjusted fluence on squares of 500 μm × 500 μm on interdigitated structures with different distances. Table 2 shows the laser sintering parameters.

Ag printed samples were dried for 2.5 h at 50 °C and then annealed for one hour at 250 °C in a convection oven.

The XPS and UPS data were determined by using a 'Multiprobe' ultra-high vacuum surface analysis system from Omicron Nanotechnology with a base pressure of around 5 × 10⁻¹⁰ mbar. The excitation source for the XPS was a DAR400 x-ray tube with Al anode and an XM500 quartz crystal monochromator providing an excitation energy of 1486.6 eV (Al Kα-line) with an energy broadening of around 0.1 eV. The photoelectron electron energies were measured with an EA125 5 channel puls-counting channeltron hemispherical electron analyzer with a pass energy of 80 eV for the survey scans and 20 eV for the detailed spectra. The total energy resolution for the detailed scans (excitation source, electron analyzer) was 0.5 eV. For the peak analysis we used the Unifit2017 spectra processing software. The fit procedure was a convolution of Gaussian and Lorentzian peak profile (Voigt profile). The valance band edge was fitted using the square root model with Gaussian broadening. The UPS-spectra were measured in the same vacuum system with the EA125 electron analyzer (1 eV pass energy) and a HIS13 UV-discharge-lamp with an excitation energy of 21.2 eV (He I line).