Raül
Perea-Causín
*a,
Samuel
Brem
b and
Ermin
Malic
ab
aDepartment of Physics, Chalmers University of Technology, 412 96 Gothenburg, Sweden. E-mail: causin@chalmers.se
bDepartment of Physics, Philipps-Universität Marburg, 35032 Marburg, Germany
First published on 29th December 2020
Monolayers of transition metal dichalcogenides (TMDs) have been established in the last years as promising materials for novel optoelectronic devices. However, the performance of such devices is often limited by the dissociation of tightly bound excitons into free electrons and holes. While previous studies have investigated tunneling at large electric fields, we focus in this work on phonon-assisted exciton dissociation that is expected to be the dominant mechanism at small fields. We present a microscopic model based on the density matrix formalism providing access to time- and momentum-resolved exciton dynamics including phonon-assisted dissociation. We track the pathway of excitons from optical excitation via thermalization to dissociation, identifying the main transitions and dissociation channels. Furthermore, we find intrinsic limits for the quantum efficiency and response time of a TMD-based photodetector and investigate their tunability with externally accessible knobs, such as excitation energy, substrate screening, temperature and strain. Our work provides microscopic insights in fundamental mechanisms behind exciton dissociation and can serve as a guide for the optimization of TMD-based optoelectronic devices.
During the last years the characteristics of TMD-based optoelectronic devices have been investigated, reporting promising features, such as fast photoresponse,25,26 large responsivities27–30 and high tunability.31,32 To reach optimal operation of these devices, exciton dissociation mechanisms and their tunability need to be understood. Previous studies have investigated exciton dissociation via tunneling to the continuum in presence of strong electric fields,33,34 showing that field-induced dissociation dominates the response at large fields needed to dissociate excitons with large binding energies.35 However, the dissociation of excitons via scattering with phonons, which is expected to play a major role at low electric fields, has been overlooked so far—despite its potential implications on the fundamental efficiency limits of TMD-based photodetectors and solar cells. After an optical excitation, the generated excitons scatter with phonons to reach a thermal equilibrium leading to a finite population of unbound electrons and holes as described by the Saha equation.36 Under the presence of an electric field, these free carriers are dragged away and excitons dissociate to preserve the thermal equilibrium (cf.Fig. 1). Exciton dissociation hence poses an upper fundamental limit on the efficiency of TMD-based photodetectors and solar cells.
The aim of this work is to provide microscopic insights in fundamental processes governing exciton dissociation in TMD materials. Based on a fully quantum mechanical approach, we resolve the dissociation dynamics of excitons in energy, momentum and time. We take into account the Rydberg-like series of bright and dark excitonic states giving rise to a multi-excitonic landscape, which plays an essential role for the dissociation process. We track the pathway of excitons from optical excitation via thermalization to dissociation into free carriers and reveal the underlying microscopic channels (cf.Fig. 1). In particular, we find that intervalley scattering dominates the dissociation dynamics due to its strong exciton–phonon coupling, scattering excitons to higher energies and dissociating them into free electrons and holes. Furthermore, we investigate the tunability of the dissociation efficiency and response time through externally accessible knobs and find the best conditions for optimal operation of a TMD-based photodetector. We find a dissociation-limited response time for WSe2 monolayers that is in good agreement to reported experimental values,35 supporting the accuracy of our model and the crucial role of exciton–phonon scattering at weak fields.
Based on the density matrix formalism in second quantization,37–40 we make use of the many-particle Hamilton operator and Heisenberg's equation of motion to describe the dynamics of incoherent excitons NνQ = 〈Xν†QXνQ〉. Here, we have introduced the exciton operator Xν(†)Q accounting for the annihilation (creation) of an exciton in the state ν with momentum Q. The Hamilton operator including electron–electron, electron–phonon, and electron–photon interactions is transformed into an excitonic basis resulting in an excitonic Hamiltonian.41 The excitonic Hamilton operator describes the system well for low electron–hole pair densities in an undoped system. Note that doping can result in the appearance of trion features in optical spectra,42 and large doping levels can cause energy renormalization due to many-particle dielectric screening.43 We focus on low-density conditions, where exciton–exciton interaction is negligible.10,21 The central part of our work lies in the exciton–phonon interaction that drives exciton thermalization and dissociation. In general, the exciton–phonon Hamiltonian describing scattering from |μ,Q〉 to |ν,Q + q〉 has the form13
(1) |
Using Heisenberg's equation of motion we obtain an expression for momentum- and time-resolved exciton occupation:
(2) |
The first line in eqn (2) describes the interaction with light. Here, the first term accounts for the absorption of incident light expressed with the microscopic polarization pμ(t), while the second term describes the loss of exciton occupation via radiative recombination.16 A more thorough description of exciton–light interaction can be found in the ESI.† The second line in eqn (2) accounts for exciton–phonon scattering within the second-order Born–Markov approximation.16,37–39 The first two terms describe bound-to-bound transitions with the in- and out-scattering rates and . The appearing scattering matrix reads Here, we have introduced the abbreviation with njq = 〈bj†qbjq〉 being the phonon number. Furthermore, q = Q′ − Q is the momentum exchange, ± accounts for the emission (+) and absorption (−) of phonons, and EνQ and ħΩjq are the exciton and phonon energies, respectively. While the first two terms in the second line of eqn (2) account for scattering within bound exciton states, the last two terms account for the formation/dissociation of excitons from/to free electron–hole pairs with the rates and including a sum over all scattering possibilities.
We now make the assumption that free carriers are immediately extracted from the system after dissocitation due to a finite electric field resulting in a zero contribution from the exciton formation term. The electric field is assumed to be weak enough, so that field-induced dissociation is negligibly small. Previous experiments suggest that this is the case for electric fields below 10–13 V μm−1 in hBN-encapsulated WSe2 monolayers.35 Here, we focus on the microscopic origin of dissociation and its implications on the fundamental limits of real devices. Therefore, we assume a ballistic transport and a complete collection at the leads, so that the current is determined by the dissociation. The latter poses an upper fundamental limit for the maximum photocurrent one can obtain in a real device. Furthermore, in addition to the discussed terms appearing in eqn (2), we also include a phenomenological decay ṄνQ|nr = −τnr−1NνQ to account for non-radiative recombination, which can be significant in real samples due to defects24 and is influenced by doping.48 The decay time τnr has been experimentally and theoretically determined to range from few to hundreds of picoseconds.20,24,35,49
In this study, we discuss exciton dissociation in the four most studied semiconducting TMDs including WSe2, WS2, MoSe2, and MoS2. However, for now, we focus on hBN-encapsulated WSe2 monolayers due to their well established excitonic landscape containing bright (KK) and momentum-dark (KΛ and KK′) excitons.15,19,23,50,51 We consider up to 7 bound s-states for each valley, assigning states with higher main quantum number to the continuum and disregarding states with non-zero angular momentum due to their lower exciton–phonon cross section.16 Moreover, we exclude the influence of spin-dark states, since the timescale of spin–flip scattering processes is considerably slow compared to the spin-conserving ones.19,52 We use input parameters for the bandstructure and electron–phonon coupling from ab initio studies,53–56cf. the ESI.†
Fig. 2 Dissociation dynamics. (a) Dissociation current density jd in hBN-encapsulated WSe2 as a function of time with disentangled contributions from the most important channels. A typical non-radiative decay time τnr = 100 ps has been considered. (b) Dissociation rates ΓdissνQ as a function of momentum and energy. The grey gradient at high energies illustrates free states and the energy axis is shifted, so that the KK continuum is located at 0 eV. (c) Transition rate Γνμ from specific excitonic states into other bound states or the continuum. The size of the circles is proportional to the corresponding rate (see ESI† for the numerical values). Positive and negative values are represented by red and blue, respectively. Exciton occupation as a function of momentum and energy at (d) t = 0.1 ps and (e) t = 200 ps after optical excitation. (f) Relative difference between the exciton occupation with (NQ) and without (NndQ) considering dissociation. |
The efficiency of the dissociation channel KΛbound → KK′cont. is a result of the strong coupling of excitons with M phonons,55 leading to a strong intervalley scattering from KΛ to KK′ excitons. This is manifested in high rates ΓdissνQ for dissociation from KΛ2s compared to other exciton states with a similar energy (cf.Fig. 2(b)). Moreover, the importance of KΛ2s over KΛ1s results from the interplay between occupation and dissociation rates: while lower states are largely occupied (cf.Fig. 2(e)), higher states exhibit faster dissociation rates due to their proximity to the continuum (cf.Fig. 2(b)). Such faster dissociation for higher excitonic states has been already observed in carbon nanotubes.57,58 We also determine the main exciton pathway, starting from the optical excitation at KK1s and ending with the main dissociation channel KΛ2s → KK′cont. identified above (cf. arrows in Fig. 1). For this purpose, we calculate the net scattering rates from ν to μ with , as illustrated in Fig. 2(c) and tabulated in the ESI.† We show that most excitons follow the following path: after optical excitation of KK1s excitons, they scatter to KΛ1s states followed by KΛ1s → KK′2s and KK′2s → KΛ2s transitions, and finally they dissociate through KΛ2s → KK′cont.. Thus, not only is the dissociation dominated by scattering with M phonons, but also the transitions within bound states, which occur mostly between KK′ and KΛ states.
Exploiting the microscopic character of our theoretical approach, we can resolve the dynamics of excitons in energy, momentum and time. At short times during the start of the optical excitation (t = 0.1 ps, Fig. 2(d)), a small occupation has already been generated at the 1s level of all valleys as a result of ultrafast polarization-to-population transfer and inter-valley scattering, with sharp features appearing due to scattering with optical Γ phonons and intervalley Λ and M phonons.15,16 Higher KK states (2s, 3s) show minor occupation due to off-resonant excitation. At later times, when the stationary state has nearly been reached (t = 200 ps, Fig. 2(e)), the occupation shows a more thermalized distribution, with larger occupation in higher exciton states compared to earlier times, and a strong occupation in the light cone at Q = 0 due to the continuous optical excitation. The equilibration between bound and unbound states described here is opposite to the relaxation cascade16 but occurs on the same timescale of a few picoseconds. This timescale would govern the photoresponse in systems pumped by an ultra-short femtosecond pulse.
Finally, we compare the exciton occupation at t = 200 ps with the case where dissociation has not been considered (NndQ). In order to be able to compare the occupation in these two cases, we normalize them by the total density n, i.e. ÑQ = NQ/n. We compute the relative difference (ÑQ − ÑQnd)/ÑndQ and find that higher states are clearly less occupied when the effect of dissociation is taken into account (cf.Fig. 2(f)). The lower occupation of higher states is a direct consequence of their efficient dissociation (cf.Fig. 2(b)).
In the last section, we have resolved and understood the microscopic channels responsible for exciton dissociation. Here, we investigate their impact on key quantities for a photodetector: EQE, IQE and the response time. EQE and IQE describe the efficiency of dissociation with respect to the rate of incident photons and of optically generated excitons G, respectively, and are directly related through the optical absorption α(ω), i.e. EQE = α(ω)IQE. Moreover, the response time τrp is determined by the dominant decay time, which can be non-radiative decay (τnr), dissociation (τd), or radiative decay (τr), cf. ESI.† Thus, we can write
(3) |
In the following, we reveal how quantum efficiency and response time can be controlled and optimized through externally accessible knobs, such as excitation energy, non-radiative decay time, substrate screening, temperature and strain. We model the effect of defects by investigating two cases of non-radiative decay time: τnr → ∞ and τnr = 100 ps, accounting for a pristine defect-free sample and a more realistic sample with defects,20 respectively. Unless otherwise stated, the excitation energy is centered at the 1s exciton resonance, the dielectric environment corresponds to hBN encapuslation, the temperature is 300 K and the sample is unstrained. We solve eqn (2) in the stationary state, setting ṄνQ = 0, and use the calculated exciton occupations to determine the EQE, IQE, and the response time τrp in eqn (3).
In a pristine defect-free sample, dissociation is the fastest decay mechanism and hence all optically generated excitons dissociate, yielding an IQE of 100%. The EQE follows the absorption spectrum for varying excitation energies (cf. blue solid line in Fig. 3(a)), with maxima at the exciton resonance energies (1s, 2s, 3s). When including a realistic non-radiative recombination time of 100 ps, the IQE drops down to 50%, and the EQE decreases by the same factor (cf. red lines in Fig. 3(a)). The IQE is constant for a wide range of energies where the 1s exciton is excited. As the excitation energy approaches the 2s resonance, excitons in this state are excited as well. Since dissociation from higher states is more efficient (cf.Fig. 2(b)), the IQE increases with excitation energy, though showing again a plateau around 2s and 3s resonances (cf. the red dashed line in Fig. 3(a)). The most optimal performance is thus obtained by exciting at excitonic resonances, where the optical absorption is maximized, with a trade-off between lower absorption and faster dissociation at higher states. Note that above-bandgap excitation would result in a large fraction of photoexcited free electron–hole pairs,59,60 giving rise to an efficient IQE. Moreover, we expect the large absorption of the B and C exciton absorption peaks above the bandgap to boost the EQE as well.
The effect of defects can be studied by varying the non-radiative decay time τnr (cf.Fig. 3(b)). When non-radiative recombination is much faster than dissociation, most excitons are lost via this decay channel and the IQE is 0%. As τnr progressively approaches the dissociation time τdiss ≈ 100 ps, dissociation starts to take over and the IQE increases. At even longer τnr > 103 ps dissociation dominates yielding an IQE of 100%. Nevertheless, there is a trade-off between efficiency and response time. For short τnr, the response time will be dominated by this value and can be tuned, while the quantum efficiency becomes low (cf. inset in Fig. 4(a)). Remarkably, similar response times on the order of 100 ps have been also reported in experimental studies for the same system at low electric fields in ref. 35, where τnr ∼ 300 ps. It was shown that the measured response times deviate from the field-assisted tunneling model at low electric fields, showing saturation at values around 100 ps. This deviation is consistent with our phonon-induced dissociation model and indicates that exciton–phonon scattering indeed limits the photoresponse at weak fields. Our theoretical predictions thus have important implications on the limitations of real TMD-based optoelectronic devices.
Another experimentally accessible knob is dielectric engineering via substrate screening that influences the material's characteristics in many different ways. First, stronger screening results in lower exciton binding energies, i.e. the excitonic levels are closer to each other and to the continuum, leading to a more efficient exciton dissociation. This is manifested in a significant increase of IQE and EQE (cf.Fig. 3(c)) and a decrease of the response time (Fig. 4(a)). On the other hand, screening also affects the absorption and thus the EQE. The optical absorption is directly proportional to the inverse dielectric function, α ∝ εs−1. Moreover, the oscillator strength is proportional to the exciton probability |ϕν(r = 0)|2, which also decreases for stronger screening due to more delocalized excitonic wavefunctions. Both facts contribute to a reduction of the absorption, resulting in the decrease of the EQE at large dielectric screening constants, cf. Fig. 3(c). The reduction of the oscillator strength also weakens radiative recombination, causing the response time to slightly increase for low εs in a pristine sample (cf. blue line in Fig. 4(a)). For weak screening, dissociation is so inefficient that the main decay channel is radiative recombination in a defect-free sample. Including a finite τnr of 100 ps results in an overall reduction of the efficiency (cf. red lines in Fig. 3(c)). However, while dissociation is very inefficient on a SiO2 substrate (τd ∼ 4 ns), the IQE increases to approximately 50% on an hBN-encapsulated sample and the dissociation can compete with non-radiative recombination. The most optimal quantum efficiency is thus obtained for substrates with a stronger screening (such as hBN) giving rise to a more efficient exciton dissociation.
A similar behaviour is observed when exploring temperature as a potential knob to tune the performance of a photodetector. As intra- and intervalley scattering of excitons is governed by phonons in the considered low-excitation regime, the strength of these scattering channels strongly depends on temperature.11,13,61,62 Here, the boost in the IQE (Fig. 3(d)) and the drop in response time (Fig. 4(b)) are caused by an increase in the phonon number njq with temperature. As more phonons are available, exciton–phonon scattering becomes stronger, populating higher excitonic states and increasing the dissociation efficiency. Note that at very low temperatures there is a 0% plateau in the IQE and EQE at which radiative recombination is more efficient than dissociation. While the IQE displays a monotonic increase with temperature, the EQE reaches a maximum and then decreases due to lower absorption at higher temperatures. The increase in the phonon-induced linewidth of excitonic resonances11 results in a broader absorption. Since the oscillator strength is conserved, the absorption peak decreases causing a clear reduction of the EQE at the exciton resonance (cf. solid lines in Fig. 3(d)). The inclusion of a finite τnr shifts the IQE plateau to higher temperatures, at which exciton dissociation can compete with non-radiative decay, giving rise to an overall decrease of the EQE at low temperatures (cf. red lines in Fig. 3(d)). Overall, the optimal quantum efficiency is found for temperatures slightly above room temperature for a realistic finite non-radiative decay.
Finally, we study the effect of strain, which is known to shift the electronic K and Λ valleys in opposite directions with respect to the valence band.56 These shifts have a direct impact on the linewidth and thus on the EQE of a pristine sample. For increasing tensile strain at approximately +1%, the KΛ1s state approaches the KK1s state, suppressing the scattering channel KK1s → KΛ1s induced by phonon emission. Similarly, further increasing strain towards +2% results in a shift of KΛ1s above KK1s and the inhibition of the intervalley scattering induced by phonon absorption. As a result, the exciton linewidth decreases and the absorption peak increases with tensile strain resulting in a step-wise increase of the EQE whenever a scattering channel is blocked (cf. solid blue line in Fig. 4(e)). The slight increase in the EQE at negative (compressive) strain can be traced back to a larger spectral distance between the KK and KΛ excitons, which shifts the final state |μ,q〉 to larger momenta, resulting in a smaller wavefunction overlap for the intervalley scattering. In order to understand the effect of strain in a more realistic sample with defects (τnr = 100 ps), we study the changes in the IQE, which reflects the efficiency of dissociation with respect to the non-radiative decay. For tensile strain, the KΛ valley is lifted up so that the KΛ1s state lies closer to the continuum of the KK′ valley. This optimal situation favours the dissociation of excitons from KΛ1s to KK′cont., yielding an enhancement of the IQE (cf. dashed red line in Fig. 4(e)). In contrast, when compressive strain is applied, the KΛ valley shifts down. As a result, KΛ1s has by far the largest occupation. Since this state moves further away from KK′cont. as compressive strain increases, the dissociation becomes less efficient and the IQE decreases.
Considering realistic values for non-radiative decay time, the quantum efficiency for exciton dissociation is the highest and the response time the fastest at the following conditions: temperatures slightly above 300 K, substrates with strong dielectric screening, excitation energies resonant to exciton states, and tensile strain values around 1%. For short non-radiative decay times, there is a trade-off between a fast response time and a low quantum efficiency.
In order to understand why dissociation is more efficient in some TMDs, we need to consider their complex excitonic landscape. The long response time in pristine MoS2 on any substrate compared to the other TMDs is a consequence of the slow exciton dissociation in this material (cf.Fig. 5(c)). This can be traced back to the fact that the largest occupation is found in ΓK and ΓK′ excitons exhibiting a large effective mass and hence large exciton binding energies. Thus, it is difficult to scatter into higher exciton states, which are energetically far away. Note however that the excitonic landscape and in particular the question which states is the lowest is still under debate in the case of MoS2.12,64–66
The TMD material with the fastest dissociation on a SiO2 substrate is WSe2 (cf. light blue boxes in Fig. 5(c)) due to its optimal band structure favouring the dissociation channel KΛ2s → KK′cont.. Under hBN encapsulation, however, it is MoSe2 that shows the fastest dissociation due to the proximity of KΛ1s excitons to the KK′ continuum. In both dielectric environments, the dissociation in WS2 is not as fast as in WSe2 and MoSe2. While the band structure of WS2 is similar to the one of WSe2, the main state for dissociation KΛ2s is too high in energy exhibiting a very low occupation and thus giving rise to a longer dissociation time. Note that the dielectric environment and in particular hBN encapsulation will also modify the transport properties affecting the performance of a real device. Our calculations provide an upper fundamental limit of the performance limited by dissociation.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/D0NR07131E |
This journal is © The Royal Society of Chemistry 2021 |