1 Introduction

The demand for data storage and the speed of processing and accessing stored information is increasing rapidly. Meanwhile there is a significant research interest in the area of data storage and memory devices. This includes research to improve the fundamental understanding of the materials currently used in these devices, and control of their electronic and optical properties, through to the discovery of new materials with enhanced properties. An area of critical concern is understanding the role of impurities and dopants within these materials. Their use to modify local chemical structure, electronic, optical and thermal properties provides significant scope to create new materials that could contribute to meeting the future demands of the data storage industry.

A key material family utilized for data storage are chalcogenide-based systems such as amorphous germanium telluride (a-Ge1−xTex). These form the basis of Ge–Sb–Te alloys that have proven to be deployable as phase-change memory systems and continue to offer potential for the creation of next-generation rewritable data storage systems [1, 2]. GeTe and its compounds have been widely used in electronic and optoelectronic applications such as memory devices [3, 4], signal modulation [5] and optical imaging systems [6]. a-Ge1−xTex has a glass transition and crystallisation temperature greater than 150 °C, with the ability to reversibly switch between the amorphous and crystalline phase at high speed, displaying high stability in both states [7, 8]. Obtaining an improved understanding of defect engineering (for example via the inclusion of dopants) within amorphous chalcogenides is critical due to the existence of valence alternation pairs (VAPs) within their network-forming structure. The presence of VAPs within these materials explains the p-type conductivity of these materials, which arises from positively charged triply coordinated chalcogens (Ch3+) and negatively charged singly coordinated chalcogens (Ch1−) [9].

Previous work to engineer the structure of GeTe to improve its electronic and optoelectronic performances has been undertaken, for example to reduce the switching power in phase-change memory devices [10]. Other research has attempted to overcome the intrinsic p-type behaviour of these materials which has been (and remains) an important limitation to their application for many years. Previously it was shown that carrier-type reversal (CTR) from p- to n-type in chalcogenide glasses could be achieved in some systems by introducing impurities such as Bi, Pb [11] and Cs [12] into the compound, either as doped or chemically modified systems. Tohge et al. reported the first successfully accomplished CTR using Bi in germanium selenide glasses with a Bi concentration of more than 10% via melt quenching technique [13]. Recently, ion implantation, using the concept of non-equilibrium doping, has demonstrated that obtaining n-type conductivity is possible using ~ 0.6 at.% Bi [14]. Such a small amount of dopant does not result in chemically modified long-range structures. CTR phenomena have many technological applications such as preparation of p–n junction systems for solar cells, light-emitting diodes, photodetectors [15, 16]. CTR phenomena could provide a possibility of development of these p–n junction devices for amorphous chalcogenides [14]. However further investigation is required in order to understand the phenomena of the CTR achieved by this approach and how it can be utilized to fulfil device application requirements.

To determine the majority carrier type in amorphous structures the measurement of a thermally induced Seebeck voltage is used as Hall effect measurements are unreliable as described elsewhere [4, 17, 18]. The measured Seebeck voltage (or determined Seebeck coefficient) has a positive or negative sign depending on the type and the movement of the majority carriers within the material. n-type semiconductors yield a negative voltage, whilst p-type semiconductors yield a positive voltage. Considering an n-type semiconductor, the Seebeck coefficient is determined using the following equation [19]:

$$S_{{{\text{dark}}}} = - \frac{\Delta V}{{\Delta T}} = - \frac{{k_{{\text{B}}} }}{q}\left[ {\frac{{E_{{\text{c}}} - E_{{\text{f}}} }}{{k_{{\text{B}}} T}} + r + \frac{5}{2}} \right]$$
(1)

where Sdark is the Seebeck coefficient in measured in the dark, ΔV is the potential difference across the sample, ΔT is the absolute temperature difference between the points at which the potential difference is measured, kB is the Boltzmann constant, q is the electronic charge, Ec is the valence band energy, Ef is the Fermi level, and T is the average temperature. The parameter r is a scattering parameter equal to 3/2 for impurity scattering and \(\frac{ - 3}{2}\) for lattice scattering of carriers. Through the excitation of carriers under light illumination (photo-Seebeck), the Seebeck coefficient is modified as photoexcited electrons and holes provide a contribution to the resulting Seebeck voltage arising due to the temperature gradient, and accordingly the measured Seebeck coefficient [Slight (λ)] is modified.

Research interest into the photo-Seebeck measurement of semiconductors has increased recently [20,21,22,23,24], including our previous study of the photo-Seebeck effect in undoped a-Ge1−xTex [25]. In this study we expand on the monochromatic and broad spectral illumination studies previously undertaken, to provide detailed spectral studies of the Seebeck voltage and coefficient for Bi:a-Ge1−xTex. We note that a number of studies have also presented the photo-bolometric effect [26, 27], in which a temperature gradient is induced by non-uniform heating, such as laser light, and which must not be confused with photo-Seebeck effect.

The reported Seebeck coefficient of p-type a-Ge1−xTex, including the bulk form, varies from ~ 0.035 to ~ 0.8 mVK−1 [28,29,30,31,32]. We previously provided a photo-Seebeck coefficient spectrum of oxidized a-Ge1−xTex yielding values of Slight (λ) ranging from ~ 0.413 mVK−1 at 400 nm to ~ 0.409 mVK−1 at 1800 nm with value of Sdark = 0.434 ± 0.011 mVK−1 and conductivity of ~ 5.6425 × 10−10 ± 1.74 × 10−11 Scm−1 [25]. In this work, we provide photo-Seebeck characterisation of Bi-doped a-Ge1−xTex thin films in which CTR is demonstrated to yield n-type electronic behaviour. We discuss the influence of implanted ions on trap states with respect to light power, modification of electronic bands including sub-bandgap states, the optical bandgap and VAPs by comparing a-Ge1−xTex and Bi:a-Ge1−xTex networks.

2 Experimental procedure

The samples studied are from the same batch of previously prepared and reported photo-Seebeck studies of undoped a-Ge1−xTex films [25]. a-Ge1−xTex films were deposited on 1-mm-thick fused silica substrates (resistivity > 1012 Ω cm−1) by R.F. reactive sputtering from a Ge50Te50 target under an Ar atmosphere. Directly after the deposition of the a-Ge1−xTex film the thickness was determined using a stylus profiler to be 100 ± 5 nm. For the purpose of implantation, the samples were mounted onto a silicon carrier wafer using double-sided carbon tape. The a-Ge1−xTex films were then implanted with Bi ions at a doping concentration of 2 × 1016 ions cm−2. During the implantation process, the vacuum pressure was kept at about 6.7 × 10–6 mbar and the sample held at ambient temperature. The Bi-ions were implanted with energy of 190 keV, with the ion beam current being kept below 1 μA to minimise any sample heating. It is noted that the majority of any film oxidation would have taken place prior to ion implantation, as the deposited a-Ge1−xTex films did not have any protective capping layer. The chemical composition of the film after implantation was measured using Rutherford Backscattering Spectrometry (RBS) at the EPSRC National Ion Beam Centre. The composition of the film was determined to be Ge75.8Te24.2 with Bi concentration of 7.0% ± 0.2 as presented in Fig. 1 analysed using SIMNRA as previously described [25]. For RBS analysis a ~ 2-mm-diameter 2 meV 4He+ ion beam was incident perpendicular to the sample with a charge collection of 10 μC. The backscattered He ions were simultaneously collected at Cornell (detector A) and IBM (detector B) geometries with scattering angles of 173.2° and 148.6°, respectively.

Fig. 1
figure 1

RBS measurement of 100-nm a-Ge1−xTex sample containing Bi ions. The left figure presents the data collected using detector A (Cornell), and the right figure presents the data collected using detector B (IBM)

Full size image

Figure 1 clearly indicates that the high-fluence Bi implant has caused Te to be preferentially sputtered, changing the ratio within the a-Ge1−xTex film to x ~ 25% compared to ~ 50% within the sputter target, and Bi ions were found to have penetrated into the silica substrate (Table 1). Films implanted with a lower fluence of Bi (1 × 1016 Bi cm−2) did not change the composition of the similarly prepared film as illustrated in Supplementary Fig. 1.

Table 1 Elemental composition of Bi:a-Ge1−xTex film. 1 TFU = 1 × 1015 atoms cm−2
Full size table

High-resolution XPS spectra were obtained using a SPECS NAP-XPS instrument (operating under UHV conditions) to collect information regarding the sample composition and electronic state of the Bi:a-Ge1−xTex films. The use of ultraviolet and X-ray photoelectron spectroscopy techniques to determine the Fermi level of semiconductors is well-established [33, 34] and used in our previous studies of undoped a-Ge1−xTex [25]. XPS measurements were taken using a microfocused, monochromated Al Kα X-ray source (1487 eV), a SPECS Phoibos NAP 150 hemispherical analyser and at a base pressure of ~ 10–9 mbar. Scans were taken at normal emission and a pass energy of 30 eV. The samples were charge normalised using the Fermi edge of a Mo metal contact pad, deposited for electrical measurements following implantation, as a reference. The Fermi edge of the contact pads, valence band maxima of the semiconductor and secondary electron cut-offs were recorded at a pass energy of 5 eV. Core levels were recorded at a pass energy of 30 eV. In order to see the secondary electron cut-off, the samples were biased using a 9-V battery. The XPS data from the surface signal of the sample indicated a high level of carbon and oxygen. Several attempts were undertaken to remove this signal via Ar+ ion sputtering cycles. Although the Ar+ ions clearly etched through the film after 6 min, it was found to not remove the oxide signal that is shown in Fig. 2a and c. As we previously suggested for a-Ge1−xTex, this indicates that the samples could be bulk oxidised. Films consisting of Te oxide are reported to show a peak with a binding energy of 576.6 eV [35]; however, depending on the host environment this peak could be slightly displaced [36, 37]. Te 3d5/2 and 3d3/2 with binding energies of 573.05 eV and 583.4 eV, respectively, are related to Te atoms in a-Ge1−xTex [36]. These peaks are also observed in this study, presented in Fig. 2(c), which indicates the amorphous nature of the films. The corresponding binding energies published for Bi 4f7/2 and Bi 4f5/2 of Bi2Te3 in a single-crystal nanostructure are 157.2 eV and 162.5 eV, respectively, of which the 4f5/2 has a slightly higher intensity [37]. However, Fujimoto reported the binding energies of BiO3 in NaBiO3 with 4f5/2 peak at 164.1 eV and 4f7/2 peak at 158.7 eV [38]. These peaks are related to the formation Bi2O3. For the same formation, the energies of 4f7/2 and 4f5/2 are measured to be 158.5 eV and 163.8 eV, respectively, by Debies et al. [39]. The experimental XPS values in this study are shown in Fig. 2c displaying two peaks at 158.1 eV and 163.4 eV, which are in good agreement with the experimentally measured binding energies by Fujimoto et al. [38], Debies et al. [39] and similarly doped samples reported by Hughes et al. [14].

Fig. 2
figure 2

a XPS spectra with respect to Ar-ion etching time. b Oxygen peak after 6-min etching. c Te3d spectra and d Bi in a-Ge1−xTex XPS spectra with and without etching

Full size image

The Fermi level was determined using XPS to be Ef = − 4.28 eV indicating the top of the valence band as -4.59 eV. This is calculated by taking the difference of the excitation energy (1486.6 eV) and the secondary electron tails (~ 1482.32 eV) which is presented in Fig. 3a. Using this method, the distance of the Fermi level to the top of the valence band for the Bi: a-Ge1−xTex is found to be 0.31 eV (Fig. 3b) defining the top of valence band energy to vacuum level as − 4.59 eV.

Fig. 3
figure 3

a Presents the secondary electron tails detected from Bi:a-Ge1−xTex using XPS. b shows the valence band onset region indicating the top of the valence band as ~ 0.31 eV

Full size image

Cr (100 nm)/Mo (50 nm) top electrode pads were fabricated by RF sputtering to permit the electrical and optical measurements in the same way as it was previously reported for the undoped a-Ge1−xTex [25]. Current–voltage (I–V) measurements were taken using a Keysight B1500A equipped with a high-resolution source measure unit (HRSMU) to ensure the linear ohmic characteristic of the sample (Supplementary Fig. 2). The conductivity of Bi: a-Ge1−xTex is determined to be \(\sim 3.93 \times 10^{ - 11} \pm 6.88 \times 10^{ - 14}\) S cm−1 which is found to be approximately an order of magnitude more resistive than the undoped sample with conductivity of ~ \(5.6 \times 10^{ - 10} \pm 6.65 \times 10^{ - 14}\) S cm−1. This is in stark contrast to previous results of Bi-implanted a-Ge1−xTex, which found that Bi-implanted samples (with similar ion dose) displayed a conductivity that was five orders of magnitude higher. This reduction in the conductivity values compared to the similarly prepared film by Hughes et al. has been observed previously for the undoped sample indicating the oxidation of the films in this study. The Seebeck and conductivity measurements were taken in a base pressure of ~ 2 × 10–6 mbar using the same apparatus [25]. Studies were undertaken at four different values of ΔT = 2, 5, 8 and 10 °C. Conductivity measurements are normalised by the distance between the two electrode contacts (1.8 ± 0.1 cm). During these measurements the voltage range was restricted to between − 3 and 3 V in order to prevent any joule heating within the sample resulting in crystallisation of the structure.

3 Results

In order to experimentally verify CTR resulting from Bi-ion implantation, the Seebeck voltage of the material is measured under dark conditions. The Seebeck coefficient obtained from these measurements has a value of Sdark = \(- 1.34 \pm 0.051\) mVK−1 with the negative sign indicating the n-type characteristic of the doped sample, Fig. 4. This contrasts to the p-type nature of the undoped film which was found to have a Seebeck coefficient of Sdark = 0.434 ± 0.011 mVK−1 [25]. The only other reported value for Bi doped a-Ge1−xTex, at the similar dose and prepared in a similar way, was reported to be an order of magnitude lower than this at − 0.125 mVK−1 [14]. As a result, repeated measurements, using different electronic measuring apparatus, were undertaken to confirm this result. We surmise that the difference between these studies relates to the degree of oxidation present within the films and the Ge:Te ratio which was 3:1 in this study as opposed to the reported 1:1 in [14]. Both of these factors would have a significant effect on the electronic properties of the films studied in each case.

Fig. 4
figure 4

Seebeck voltage measurement (under dark conditions) of Bi:a-Ge1−xTex as a function of temperature gradient across the sample and the calculated Seebeck coefficient value

Full size image

The measurement of Slight (λ) was performed, using the same values of ΔT as for the dark measurement, scanning the incident light from 1800 to 400 nm with the resulting spectrum shown in Fig. 5a. Illumination with light leads to the magnitude of Slight (λ) being reduced compared to Sdark. This implies that whilst both photoexcited carrier types are able to move under the applied temperature gradient holes do so preferentially over electrons (i.e. the hole mobility is higher). This may be the result of electron trapping taking place from the conduction band. Inspection of the variation of Slight (λ) with incident wavelength indicates that as the energy of light is increased, the difference between Slight (λ) and Sdark also increases. This behaviour is dominated by the minority carrier (hole) concentration and thus the hole quasi-Fermi level under optical illumination.

Fig. 5
figure 5

Photo-Seebeck spectrum of Bi:a-Ge1−xTex (a) and a-Ge1−xTex (b) in the region of 400 nm to 1800 nm [25]. The potential optical bandgap region is distinguished with different colour shading

Full size image

Careful inspection of the variation of Slight (λ) in Fig. 5a indicates distinctive regions of similar behaviour from 1770 nm (0.7 eV) to ~ 1010 nm (1.23 eV) and either side of this region as similarly observed for the undoped sample (Fig. 5b). The difference between Sdark and Slight(1800 nm), the lowest photoexcitation energy, is calculated to be 0.35 mV K−1 (a reduction of ~ 26%). We observed similar behaviour within the a-Ge1−xTex sample though with a reduced magnitude (~ 6% reduction), indicating that following doping with Bi the electronic properties are much more sensitive to the effect of illumination with light. This is reflected in the difference between Slight (λ) at 400 nm and 1800 nm which is ~ 0.054 mVK−1 for Bi:a-Ge1−xTex but only ~ 0.0041 mVK−1 in the undoped a-Ge1−xTex sample. We note that the difference in the number of absorbed photons at 1800 nm and 400 nm for the doped sample is about ~ 6.2 × 1012, whilst for the undoped a-Ge1−xTex it is ~ 3.8 × 1011 photons (Supplementary Fig. 3). Though the order of magnitude difference in these values correlates with the order of magnitude difference between the reduction of Slight (λ) between 1800 and 400 nm, the behaviour is strongly influenced by trap states and hence a direct link cannot be immediately assumed between these. We discuss the potential influence of relative light intensity variation during the wavelength scan in the discussion section but note here that we find this not to be a significant effect.

The gradient of Slight (λ) with wavelength [∆S (λ)] is found to be constant between ~ 1770 and ~ 1010 nm with a value of \(5.817 \times 10^{ - 8} \pm 9.67 \times 10^{ - 10}\) mVnm−1. We previously observed a change in ∆S (λ) for the a-Ge1−xTex film at ~ 1410 nm (0.88 eV), attributed to the onset of band-to-band transitions and defining the energy of the bandgap. However, for the Bi:a-Ge1−xTex sample we do not see this feature in ∆S (λ), indicating that the optical bandgap has been modified. Electronic doping with impurity ions is expected to modify the optical bandgap through shifting and/or broadening the valence and/or conduction bands. This is well established in previous reports, where the introduction of impurities within the band states, either by ion implantation or other doping methods, leads to a broadening or narrowing of the bandgap [40,41,42,43,44,45]. Here we note that the behaviour of ∆S (λ) above 1010 nm is similar to that observed for a-Ge1−xTex, and at wavelengths below ~ 1750 nm a change in behaviour may also be observed.

In common with the a-Ge1−xTex film, the magnitude of the gradient, |∆S (λ)|, reduces as wavelength is decreased for the Bi:a-Ge1−xTex film (at ~ 950 nm and ~ 1010 nm, respectively). Thus the data in Fig. 5 can be divided into two (or potentially three) regions, with a different gradient either side of ~ 1010 nm. As discussed previously, it might be expected that either side of the optical bandgap a different dependence of Slight (λ) on wavelength is found. If the change in gradient corresponds to this, then Fig. 5 indicates a bandgap of ~ 1010 nm. This is however too high in energy with the bulk material known to have an optical bandgap of ~ 0.88 eV (~ 1410 nm). Furthermore, upon doping, the bandgap is reduced and the XPS studies indicated a bandgap energy of ~ 0.7 eV. We therefore propose that the region between ~ 1770 and 1010 nm corresponds to direct band-to-band excitation of the sample. At energies below 0.7 eV (1770 nm) the number of data points is limited by the poor transmission of the optical fibre used in the set-up. Nonetheless, analysis indicates that there is a subsequent change in the gradient of Slight (λ) here which is not unexpected given the high density of trap states typically found close to the band edge would support trap-to-band and band-to-trap transitions.

Figure 6 provides further information regarding the photogenerated carriers based on the measured photo-Seebeck voltages normalised by the number of absorbed photons at each wavelength [Vlight (λ)] (Supplementary Fig. 3). Vlight (λ) is plotted for the different values of ΔT = 2, 5, 8 and 10 °C as were used within Fig. 4.

Fig. 6
figure 6

The measured photo-Seebeck voltage [Vlight (λ)] normalised by the number of incident photons for Bi:a-Ge1−xTex at values of ΔT = 2, 5, 8 and 10 °C

Full size image

Due to different optical gratings being used either side ~ 1050 nm, within this region additional data points were obtained to ensure an overlap. The peak appearing at ~ 1380 nm is as result the strong attenuation of the optical fibre in this region and therefore an artefact of the system (similarly observed for a-Ge1−xTex [25]). The general variation of Vlight (λ) from 730 to 1800 nm indicates that as the energy of the incident light is decreased (towards 1800 nm), the value of Vlight (λ) increases to higher values for each value of ∆T. This behaviour contrasts that of a-Ge1−xTex, where an increase in Vlight (λ) was only observed for wavelengths above 1560 nm. The estimated bandgap for the Bi:a-Ge1−xTex sample is around 0.7 eV (1770 nm) at which energy there appears a clear peak in Vlight (λ). Further investigation of Vlight (λ) for the energies below 1800 nm was not possible. Careful inspection of Vlight (λ) in Fig. 6 shows the presence of a shoulder at ~ 1620 nm with a second at ~ 1730 nm. These are both above the bandgap (~ 1770 nm) and therefore indicate a variation in the density of states in this region. Further peaks (~ 650 nm and 830 nm) in Vlight (λ) are found at much higher energies above the bandgap with the former closely resembling a similar feature found at ~ 680 nm for the undoped a-Ge1−xTex sample. A similar strong wavelength dependency at wavelengths above 470 nm has been found in both a-Ge1−xTex and Bi:a-Ge1−xTex.

It is important to characterise the type of carriers generated upon illumination of the device with light. The absorption of light yields an excited electron (hole) population of magnitude ∆n (=∆p). The fact that this magnitude is proportional to the intensity of the light [I0 (λ)] needs taking into account when considering Slight (λ). The results of measuring Slight (λ) at 1200 nm, 1400 nm, 1600 nm and 1800 nm, as a function of the incident light intensity for Bi:a-Ge1−xTex, are shown in Fig. 7. As the power of incident light is reduced, the photo-Seebeck coefficient approaches Sdark. The measurement was taken, starting from the highest wavelength (1800 nm) to the lowest wavelength, with the signal allowed to stabilise for an hour in the dark between changes of wavelength. Supplementary information Fig. 4 shows the incident light spectrum at 100% intensity.

Fig. 7
figure 7

Photo-Seebeck coefficient [Slight (λ)] of Bi:a-Ge1−xTex obtained at 1200 nm, 1400 nm, 1600 nm and 1800 nm as a function of relative light intensity

Full size image

The intensity of the incident light was controlled by the use of neutral density filters starting with the highest attenuation to the lowest. For the highest attenuation (\(1 \times 10^{ - 6}\)% of full power) all values of Slight (λ) are found to be closest to Sdark (~ − 1.34 mV K−1) as expected. However, even at such a low light level the Seebeck coefficient is significantly modified from Sdark and is further modified as the incident light intensity increases. For each measured wavelength a similar logarithmic dependence on incident light intensity is observed. Upon reduction in intensity to 10% of the initial value a linear fit of the data in Fig. 7 yields a gradient of ~ 20 μV K−1 W−1 which reduces to ~ 5 μV K−1 W−1 as the intensity is reduced further.

Conductivity measurements were taken as a function of light intensity at excitation wavelengths of 1200 nm, 1400 nm, 1600 nm and 1800 nm (Fig. 8). The Bi:a-Ge1−xTex sample shows a reduction in conductivity [σ (λ, I)], as the light intensity (I) is increased for each wavelength. A significant reduction in σ (λ, I) occurs immediately upon illumination and continues up to 15% light intensity level. As the light intensity is further increased, the value of σ (λ, I) is only slightly reduced further, indicating a reduction in the process which leads to the effect is occurring. This behaviour indicates negative photoconductivity is occurring (\(\sigma_{{{\text{meas}}}} \; = \;\sigma_{{{\text{dark}}}} + \;\sigma_{{{\text{light}}}}\)) with light intensity and also the energy of the incident light.

Fig. 8
figure 8

Conductivity [σ (λ, I)] of Bi:a-Ge1−xTex as function of incident wavelength and light intensity

Full size image

The measurement of σ (λ, I) was the repeated for several different applied temperature gradients and is presented in Fig. 9. After keeping the sample under dark conditions for an hour, held at a given value of [ΔT, σ (λ, I)] was measured starting from the smallest temperature gradient (2 °C).

Fig. 9
figure 9

Photoconductivity as function of light intensity for various ΔT

Full size image

Whilst the data obtained at ∆T = 8 °C, with the exception of the 1800-nm dataset, also show negative photoconductivity, the variation is an order of magnitude less than that seen at ΔT = 10 °C for example. Similar fluctuations at certain applied temperature gradients were also observed for a-Ge1−xTex [25]. Unlike the undoped sample, the photoconductivity of Bi:a-Ge1−xTex does not show any reproducible saturation behaviour for high-intensity incident light. In general, the negative photoconductivity observed in this sample could be due to the increase of the trap states between the valence and conductions bands which is experimentally shown in Fig. 5 and 7. As the number of electron trap states is increased, the ratio of free minority carriers (holes) to free majority carriers will increase.

4 Discussion

XPS measurements determined Ef = − 4.28 eV and Ev = − 4.59 eV resulting in Ev–Ef to be 0.31 eV. The undoped a-Ge1−xTex films were previously found to have values of Ev(a-Ge1−xTex) = − 4.21 eV, Ef(a-Ge1−xTex) = − 4.05 eV, Ec(a-Ge1−xTex) = − 3.41 eV (presented in Fig. 10). This change in the valence band structure indicates Bi ion implantation has clearly modified the electronic structure. Previous studies of the influence of ion beam sputtering in hydrogenated amorphous silicon‐germanium films have shown changes in both valence band maximum and conduction band minimum resulting in a decrease of bandgap as the concentration of germanium is increased [46]. Moreover, the increase of germanium content causes more damage to the structure of silicon, which results in the density of dangling bonds increasing along with trap states within the valence and conduction bands. However, depending on the dopant, its concentration and the composition of the host material, there can be a change in the valence band either towards the vacuum level or away from it [47,48,49]. The measured 0.31 eV valence band offset (a decrease in the valence band compared to the undoped sample) in Bi-doped a-Ge1−xTex with less than 2% Bi concentration in this sample (reported by Hughes et al. [14]) is significantly larger than that previously observed. Its origin may therefore be due to compositional change (e.g. oxidation or bonding) or other undesired effects occurring (e.g. charging) during XPS measurement.

Fig. 10
figure 10

Energy bands of a-Ge1−xTex and Bi:a-Ge1−xTex. The conduction bands are calculated from XPS experimental results

Full size image

The reduction in the valence band energy could be due to the increase of the bonding states in the system as the valence band is directly related to the electronic energy levels that are occupied, whereas the conduction band is the corresponding unoccupied states [50]. In the case of a Bi-doped GeTe network, the increase in the bonding states could be related to the increase of glassy Bi2Te3 clusters. It has been previously reported that the n-type defects (resulting in the n-type conductivity of the solid) are mainly associated with the Bi2X3 (X = Se, S and Te) ionic bonds (Bi2Te3 in this case) [14]. The formation of ionic bonding between Bi and chalcogen atoms has been reported for different structures such as Ge20Se80−xBix [51].

From the XPS results alone, it is difficult to experimentally distinguish if there is a change in the position of the conduction band level and/or a change in the optical bandgap. However, the energy of the Fermi level is shifted from − 4.05 to − 4.28 eV. Considering the fact that Bi implantation with ion dose of \(2 \times 10^{16}\) ions cm−2 results in majority carrier-type reversal from holes to electrons (confirmed via the Seebeck measurements), it is expected that the Fermi level should be closer to the conduction band than the valence band following doping. The Fermi level may be calculated from the dark Seebeck coefficient (Eq. 1) by assuming an optical bandgap is about ~ 0.7 eV that results in a conduction band energy of ~ − 3.9 eV as it is provided in Fig. 10. This yields a Fermi-level energy of ~ − 4.257 eV which is ~ 0.023 eV (approximately kBT at ~ 300 K) higher than the value determined by the XPS results. Therefore the distance of the Seebeck-calculated Fermi level [Ef(Seebeck)] to the top of the valence band for the n-type a-Ge1−xTex is found to ~ 0.333 eV. Based on the XPS results, this distance [Ef(XPS)] should be about 0.31 eV which is almost twice that of the value of valence band maximum for the undoped a-Ge1−xTex (0.16 eV ± 0.02). This experimentally indicates that the density of negatively charged centres (\(Ch_{1}^{ - } )\) has been increased which results in the attraction of holes. The consequence is the subsequent formation of \(Ch_{1}^{0}\) centres \((Ch_{1}^{ - } + h \leftrightarrow Ch_{1}^{0} + e)\), thereby increasing the number of electrons in the conduction band. It is important to indicate that the Fermi level in this structure is not pinned in the middle of the gap (as is found for the a-Ge1−xTex sample in which no Fermi-level pinning has been observed). A similar argument has been provided for the Ge20Se80 system in which Bi-doping (at a concentration of 4% to 10%) results in n-type conduction and Fermi-level unpinning [52]. In Bi-doped GeSe glass networks, Bi doping lowers the conduction band minimum towards the Fermi level due to the formation of antibonding states based on Bi-Se compound [52]. Both the XPS experimental results and the (dark) Seebeck coefficient value indicate reduction of the bandgap to ~ 0.7 eV (~ 1770 nm) from that of ~ 0.88 eV (1410 nm) for the a-Ge1−xTex film. Though there are limited data below ~ 1770 nm, there is an indication of change in ∆Slight (λ) for the Bi:a-Ge1−xTex film that is not present for the a-Ge1−xTex sample. The thin nature of the films (100 nm) prevented direct optical studies of the bandgap via common optical spectroscopies (absorption and ellipsometry).

Returning to examine Fig. 5, it is observed that the gradient of Slight (λ) in each case is very consistent either side of ~ 1000 nm for both films, before it deviates at the bandgap of a-Ge1−xTex (1410 nm). Within spectral the region between the a-Ge1−xTex bandgap and that of Bi:a-Ge1−xTex, deviation is expected due to the different excitation mechanisms available for each sample. For Bi:a-Ge1−xTex we note that at wavelengths greater than 1770 nm, though the data are limited, it appears that the gradient changes.

It was previously observed that the photo-Seebeck spectrum of a-Ge1−xTex was split into three regions for the energies above 1.34 eV (wavelengths below ~ 925 nm) based on small changes in gradient in this region. These three regions are highlighted in Fig. 5 for Bi:a-Ge1−xTex, though the changes in the gradient of Slight (λ) here are much less pronounced compared to a-Ge1−xTex results. As these regions were linked to the binding energies of Te, lone-pair and Ge bonds, similar behaviour (i.e. energy dependence) is expected for both samples. The more noticeable changes of gradient in this region for a-Ge1−xTex therefore point to a modification within the structure of the bonding with Bi doping. For example, the gradient changes at the energies above 2 eV (yellow area in Fig. 5) are expected to be due to Te atoms forming non-bonding orbital with lone pairs. The breaking of these (via optical excitation) would lead to an increase in loan-pair states and therefore p-type behaviour, thereby modifying Slight (λ). A much reduced (almost negligible) change in the gradient is observed in the doped sample however. One possible explanation is that as Te atoms have formed stronger bonds with Bi ions such that excitation with energies above 2 eV is not able to break these and lead to an increase in the number of Te lone pairs. This is in line with the previous study in which the Bi doping results in inducing disorder in the material and tuning the Fermi level [53]. Yet at higher energies (~ 2.7 eV, ~ 460 nm), there does appear to be a modification of Slight (λ) in the n-type sample (highlighted as green area in Fig. 5). Shevchik et al. reported that bonding p-orbitals of Ge and Te are about 3 eV [36]. It could be assumed that the illumination of light could contribute in breaking and/or forming Ge-Te bonds equally resulting in anomaly behaviour of the Seebeck voltage, which requires more investigation at this region.

In considering the above-discussed wavelength dependence of Slight (λ) shown in Fig. 5 we are able to exclude the variation in relative light intensity as the wavelength is scanned from having a significant effect. Using supplementary information Figure 4, we note that the incident intensity at ~ 600 nm and ~ 1200 nm is equivalent; hence, if Slight (λ) in Fig. 5 was dominated by relative intensity variation, we would expect to measure a similar value, yet this is not the case. Furthermore, as we move to wavelengths shorter than ~ 600 nm using this argument, we would expect Slight (λ) to decrease, whilst Fig. 5 clearly shows that in fact the opposite occurs. This is not surprising given that Fig. 7 shows that to achieve substantial changes in the value of Slight (λ) the light intensity must be reduced by orders of magnitude.

Figure 11 plots Slight(1800 nm) for both the a-Ge1−xTex and Bi:a-Ge1−xTex samples as a function of the number of incident photons. Inspection of Figs. 7 and 11 indicates that at higher photon fluxes (i.e. above ~ 10% light power in Fig. 7) a perceptible change in the gradient of Slight (λ) is found. This light power (at 10% light power) is equivalent to \(2.9 \times 10^{9}\) photons cm−2, thus placing a maximum value on the magnitude of ∆n. Importantly it is noted that Slight(λ) continues to decrease as the light intensity is further increased and does not saturate as found for the undoped a-Ge1−xTex sample. In the case of a-Ge1−xTex sample, the saturation point was found to occur at ~ 0.1% light power (~ \(2.7 \times 10^{5}\) photons.cm−2). This is further direct evidence of the narrowing of the bandgap when doping with Bi with an accompanying increasing density of states.

Fig. 11
figure 11

Comparison of photo-Seebeck coefficients at 1800 nm for n- and p-type sample. Black data line presents Bi:a-Ge1−xTex (n-type), and red line shows p-type a-Ge1−xTex

Full size image

5 Conclusion

In conclusion, we successfully demonstrated the effect of Bi doping and carrier reversal in the electronic and optoelectronic properties of an amorphous GeTe via photo-Seebeck measurements. We have presented a detailed study of the resulting thin-film composition and electronic structure using surface science techniques enabling the position of the valance and conduction bands to be determined along with the fermi-level postdoping. The measurement of the Seebeck coefficient and its dependence upon monochromatic light irradiation provided an independent method corroborates the electronic structure. Comparing the photo-Seebeck behaviour of the n-type Bi:a-Ge1−xTex and the undoped p-type a-Ge1−xTex provides insight into the underlying factors at play in CTR in these material systems. The presence of Bi ions and oxygen within the amorphous network has shown a large Seebeck coefficient value with a negative sign. Meanwhile the bandgap is narrowed as the density of states is increased. It is important to note that the photo-Seebeck technique has shown to be a reliable technique to probe the trap states in the structure of amorphous materials.