Understanding Trap States in InP and GaP Quantum Dots Through Density Functional Theory

Abstract

The widespread application of III-V colloidal quantum dots (QDs) as non-toxic, highly tunable emitters is stymied by their high density of trap states. Here, we utilize density functional theory (DFT) to investigate trap state formation in a diverse set of realistically passivated core-only InP and GaP QDs. Through orbital localization techniques, we deconvolute the dense manifold of trap states to allow for detailed assignment of surface defects. We find that the three-coordinate species dominate trap** in III-V QDs and identify features in the geometry and charge environment of trap centers capable of deepening, or sometimes passivating, traps. Furthermore, we observe stark differences in surface reconstruction between InP and GaP, where the more labile InP reconstructs to passivate three-coordinate indium at the cost of distortion elsewhere. These results offer explanations for experimentally observed trap** behavior and suggest new avenues for controlling trap states in III-V QDs.

For Table of Contents Only

1

Colloidal semiconductor nanocrystals, more commonly known as quantum dots (QDs), have attracted considerable attention as solution-processable materials ^{1, 2} with highly tunable optical properties ^{3, 4, 5, 6, 7, 8}. They have begun to see use in a wide range of applications including photovoltaics ^{9, 10}, photodetectors ^{2, 11}, LEDs ^{1, 12}, lasing ^{13, 14}, drug delivery ^{15, 16}, biological imaging ^{17, 18}, and quantum computing ^{19, 20}. However, the QDs with the best performance to date ^{21, 22, 23} are those composed of highly toxic and internationally-restricted cadmium or lead chalcogenides ^{24, 25, 26}, making the development of a non-toxic alternative material with equally strong optical properties necessary for safe widespread commercialization. III-V QDs, namely indium phosphide (InP), are promising candidates for this replacement due to their low toxicity ^{25, 26} and widely tunable emission range ^{18, 27}. Until recently, their implementation has been held back by generally low quantum yields and broad emission line widths relative to their II-VI counterparts ^{28, 29, 30, 31, 32}. These phenomena are often understood to result from a high density of trap states: occupied or virtual electronic states, usually localized on the surface of the QD, with energies between the valence band maximum (VBM) and conduction band minimum (CBM) ^{31, 33}. While recent advances in control over III-V core/shell heterostructures have led to InP QDs with near-unity quantum yields ^{12, 34, 35}, a complete atomistic understanding of the formation and character of surface traps in III-V QDs remains elusive, especially for core-only QDs.

Trap states are not unique to III-V QDs; a vast body of both experimental and theoretical literature discusses trap states in II-VI QDs ^{33, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45}, IV-VI QDs ^{33, 46, 47, 48}, and lead halide perovskite nanocrystals ^{33, 49, 50, 51}. The most widely accepted origin of trap states is under-coordinated surface atoms ^{31, 38, 50, 51, 52, 53, 54, 55, 56, 57, 42}, although in certain systems excess charge ^{37, 39}, imperfect stoichiometry ⁴⁶, and substitutional defects ⁵⁸ have also been implicated. For CdSe, several studies employing density functional theory (DFT) have shown that trap states arise primarily from two-coordinate Se atoms (Se-2c) but not Se-3c or any under-coordinated Cd ^{38, 40, 41}. No such consensus has been reached for InP QDs, however. Many studies have implied that hole trap** dominates in InP QDs ^{36, 53, 56, 57, 58, 59, 60, 61}, but there is also considerable evidence for the presence of electron traps, especially in the absence of a core/shell heterostructure ^{31, 52, 54, 55}. This disagreement has been compounded by a relative lack of atomistic ab initio studies of trap states in InP QDs ^{52, 54, 55, 56, 57, 58, 62, 63, 64}, many of which only employ less-accurate GGA functionals. Studies have variously emphasized P-3c traps ⁵⁷, In-3c traps ^{55, 64}, both In-3c and P-3c traps but with disagreement on their respective depths ^{52, 54}, traps from the two-coordinate species with additional P-3c traps only in tetrahedral geometries ⁵⁶, as well as studies that find InP QDs with both In-3c and P-3c to be trap free but see traps introduced upon different surface treatments ^{58, 63}. Most of these studies only compute the electronic structure of a single model InP QD, limiting generalizability with respect to shape, size, faceting, and surface passivation. Moreover, very few studies have applied computation to understand trap states in other III-V QDs such as gallium phosphide (GaP) ⁶⁵, a promising but under-studied emissive material ^{66, 67}.

Here, we use DFT to study a large, diverse set of InP and GaP QDs and draw generalized conclusions on the nature of their trap states and the factors that influence trap depth. The six base QD morphologies studied here are summarized in Figure 1. Several important decisions inform the development of our test set. We focus on core-only QDs, carved from the bulk crystal using a well-established construction procedure ^{40, 68}. We create six starting QDs for both InP and GaP, chosen to represent distinct faceting and synthetically realizable shape^{69, 70, 71, 72, 73}. The four larger models, with diameters of 2-2.5 nm, represent the upper limit of computationally realizable QDs. The two smaller models allow for size extrapolation. Surfaces are passivated with X-type F- ligands, representative of the well-established treatment of InP QDs with HF ^{31, 55}. Calculations show that larger halogen ligands create states close to the VBM, potentially interfering with the assignment of trap and bulk states (SI I.I). As all our QDs are cation rich, one can equivalently think of them as stoichiometric InP/GaP cores passivated by Z-type InF3/GaF3 ligands⁷⁴.

Figure 1: Overview of the six base computational InP QD models used in this study. Each InP structure shown here has a GaP counterpart with the same shape and stoichiometry (SI, Figure S2). The colored inserts to the right of each structure provide a visual guide of the corresponding surface facets. Structures (b,c) and (e,f) can be thought of as extended versions of structures (a) and (d), respectively.

The construction procedure employed for these starting QDs is analogous to the ones used in previous ab initio studies of trap states in III-V QDs (SI I.II), and results in some number of three-coordinate In/Ga and P atoms in all structures ^{55, 56, 57, 58, 63}. The difficulty in creating perfectly four-coordinate III-V model QDs arises from charge-orbital balance, in which the formal charge of each atom must add to the total charge of the system to prevent do** ⁷⁵. While some experimental evidence suggests QDs must be strictly charge-neutral ^{76, 77}, it has also been shown that strict charge-balance greatly limits possible model III-V QDs, additionally restricting defects one could induce to these QDs to charge neutrality ⁵⁶. To extend the range of QD shapes and defects available to us without inducing do**, we allow for slight positive charges in our structures ^{37, 78}. The restriction to positive charges serves to avoid exacerbating DFT’s self-interaction error ⁷⁹ and allows our cation-rich systems to have their charge balanced by fluoride counter-ions in solution ⁵³. We observe no do** in any of our systems and no qualitative difference in geometry or electronic structure between our charged and neutral models (SI I.III).

We diversify our dataset and study the effects of surface reconstruction by creating "defective" QDs out of these starting models. These defective structures are created in a similar manner to previous studies, but our lack of charge neutrality affords us a greater variety of available defects ^{52, 38, 33, 78, 56, 54, 50, 41}. When creating defects in a starting structure, we consider all symmetry-unique removals of a single F-, P^3-, InF_x, and InP (SI I.IV). This procedure results in a total library of 160 InP and GaP QDs for consideration.

We then compute the ground state electronic structure of each QD using PBE0, as hybrid functionals are necessary for the accurate reproduction of band gaps ^{80, 81}. Comprehensive identification of trap states requires the prior identification of the first bulk states, i.e. the VBM and CBM. This is made challenging by the dense manifold of intermediately localized states near the band edges (Figure 2). Two techniques are used to visualize the band structure of our models. The first is the projected density of states (PDOS), which visualizes contributions from different atomic species to each band as a function of energy (Figure 2a,c). The second is the participation ratio (PR) (SI II.II), which measures the localization of each electronic state (Figure 2b,d). Analysis of the PDOS and PR alone is insufficient to understand the trap states in the QDs studied here for two reasons. First, the starting structures (Figures 1, S2) have trap states before any defects are induced, arising primarily from pre-existing three-coordinate atoms. Second, the trap states in our QDs are generally shallow because of their high degree of surface reconstruction. Thus, almost all QDs in our dataset display a dense quasi-continuum of trap states at both band edges, which obfuscate both new trap states induced by specific defects and the "true" CBM and VBM.

Figure 2: (a,c) Projected density of states for two fluorine defects in the truncated InP cuboctahedron. Colored lines indicate contributions from different elements. (b,d) Participation ratio for the same two structures. Blue lines represent KS eigenstates that are clearly localized while purple lines represent intermediately delocalized states that become localized upon Pipek-Mizey localization. Green lines represent the first bulk state identified by our algorithm. The orange line represents the highly delocalized quantum-confined S-like state. (e) Pipek-Mizey orbital localization for the VB edge of the truncated InP cuboctahedron. Three states with intermediate localization are mixed with eight clearly localized states to form eleven localized states on different P-3c. Orbitals shown at an isosurface level of 0.03.

A naive analysis of the PR is problematic because localization is not an intrinsic property of DFT Kohn-Sham eigenstates ^{82, 83}. In fact, any linear combination of degenerate eigenfunctions is also a solution to the Kohn-Sham equations, and in many systems these linear combinations will be more delocalized than what experiment and chemical intuition would suggest, especially when the eigenstate spectrum is particularly dense ⁸⁴. Confronted with this problem, we can utilize orbital localization methods, such as Foster-Boys ⁸⁵ or Pipek-Mizey ⁸⁶, to perform unitary transformations on a selected subset of molecular orbitals to maximize their localization. We find applying Pipek-Mizey localization to the band edges of our QDs reveals that many states with intermediate delocalization reduce to linear combinations of clearly localized trap states (Figure 2e). Combined with the observation that “true” bulk states fail to localize into clear surface states, orbital localization gives us a powerful tool to test for the location of the VBM and CBM. Our procedure is described in detail in the Supporting Information (II.III) alongside data highlighting the consistency of our predictions between related structures. As an example, the VBM is chosen to be the highest energy delocalized occupied state without a disproportionate contribution from under-coordinated P, and all occupied states above it in energy should be localizable into clearly surface-bound trap states. All such higher energy states are then taken to be hole traps, with trap depth equal to the difference between their energy and the energy of the VBM. An analogous definition identifies the CBM and associated electron traps.

A complication arises when considering the conduction band due to the intermittent presence of a low-energy, highly-delocalized state which likely corresponds to the $1S_{e}$ ground exciton state observed in experiment and the S-like envelope state predicted by "particle-in-a-sphere" theories ^{3, 87, 88, 89}. While delocalized, this quantum-confined state cannot be considered the CBM due to its energetic isolation from the quasi-continuous conduction band. Furthermore, the state is present in less than 30% of our structures. Figures 2b,d display two QDs with identical shape and stoichiometry but where the former displays the quantum-confined state, the latter does not. We have found no chemical justification for this intermittency. Nevertheless, our algorithm produces consistent VBMs, CBMs, and trap depths if we exclude the quantum confined state from consideration (SI II.IV). We observe no general qualitative difference between the trap states in QDs with and without the quantum-confined state.

Recently, Snee et al. found that DFT predicts non-singlet ground states for certain halide-passivated model InP QDs - a situation that would have major implications for the magnetic properties of these structures ⁹⁰. We were unable to demonstrate a similar effect in the structures studied here. For all 12 starting QDs in Figures 1 and S2, we find the singlet to be the lowest energy spin state, as would naively be expected for charge-balanced QDs. We thus conclude that there must be some important geometric differences between our QDs and those of Snee et al. While it would be interesting to understand the nature of those differences, for now we focus on the non-magnetic ground states of these structures.

The defective QD structures reveal an immediate difference in surface reconstruction between InP and GaP (Figure 3a,b). While GaP is relatively rigid with little reconstruction after defect induction, InP reconstructs heavily both in the immediate vicinity of the defect and further afield. Representative of all QDs studied here, we use the smaller tetrahedral QD in Figure 3 as a particularly clear example of this effect. Degree of reconstruction can be quantitatively measured though reorganization energies, which we find to be over three times larger on average in InP than in GaP (SI III.I). We hypothesize the origin of this contrast to be In’s stronger, more ionic bonds with anionic ligands than Ga, due to In’s lower electronegativity. We note that the observed trends in reconstruction still hold when Cl ligands are employed instead of F (SI III.II). These results are supported by prior experimental findings for InGaP, where Ga is found to reside disproportionately at the surface, and In-to-Ga substitution is found to be thermodynamically favorable, increasing QD stability and narrowing X-ray peaks ^{91, 92}.

Figure 3: InP (a) and GaP (b) converged structures highlighting differences in surface reconstruction after corresponding F and InF defects are created in the smaller tetrahedra. Red colored atoms indicate the metal atom from which a fluorine anion is removed, and cream colored atoms indicate the phosphorus from which InF cations are removed. Distribution of trap depths for all electron (c) and hole (d) traps for all QDs in the dataset combined. Traps from InP and GaP, as well as 3-coordinate and structural traps, are colored separately. Discrete trap depths are broadened by normalized Gaussian functions with RMS width 0.1 eV. Non-trap** 3c atoms are assigned a depth of 0.

The greater reconstruction in InP leads to differences in the depth distribution of electron and hole traps between the two materials, shown in Figures 3c and 3d. Across our data set there are fewer In-3c in InP than there are Ga-3c in GaP, and similar numbers of P-3c in InP and GaP (SI III.I). The cost of the passivation of In-3c is evidenced in the formation of additional trap states localized around distorted In-4c and P-4c atoms, here denoted “structural” traps. While localized with mid-gap energies, these structural traps appear to arise not from under-coordinated surface species but rather from structural deformations caused by extensive surface reconstruction. We do not delve deeply into the nature and origin of these structural traps here; for now, we simply note their presence, even before orbital localization is applied. Despite these additional structural traps, electron traps in InP are generally less deep than those in GaP, with many In-3c being non-trap** or very shallowly trap**. This finding agrees with experimental results for InGaP nanowires, where nonradiative recombination increases with increasing Ga concentration ⁹³. P-based hole traps, on the other hand, have similar depths across InP and GaP QDs (Figure 3d). A structure with each defect clearly labeled can be found in the Supporting Information (III.III).

Close analysis of electron traps in InP and GaP QDs reveals two distinct geometries of the three-coordinate cation, which lead to distinct distributions of trap depths (Figure 4a,b). We designate these geometries as planar, when the cation is coplanar with its coordinated atoms, and pyramidal, when the cation is out of said plane. While both geometries are present in both InP and GaP, planar Ga-3c is far more prevalent relative to pyramidal Ga-3c than planar In-3c is to planar In-3c. This further explains the surface reconstruction in InP, where planar In-3c is more likely to convert to the pyramidal geometry so ligands can bridge to additional In-3c. In both materials, traps arising from the pyramidal geometry are significantly deeper, on average, than those from the planar geometry. In InP this difference is such that most planar In-3c are effectively non-trap**. By separating the two geometries, we see that, averaging over all QDs in our dataset, planar Ga-3c (0.23 eV) forms deeper traps than planar In-3c (0.14 eV), and that pyramidal Ga-3c (0.84 eV) forms deeper traps than pyramidal In-3c (0.52 eV).

Figure 4: Distribution of trap depths for In-3c (a) and Ga-3c (b) traps for all QDs in the dataset combined, with contributions from planar (blue) and pyramidal (orange) cations displayed separately. Without loss of generality, we only show the depth of the first trap on each 3c cation to better highlight trends. Discrete trap depths are broadened by normalized Gaussian functions with RMS width 0.1 eV. Non-trap** 3c atoms are assigned a depth of 0. (c) Change in LUMO energy when InCl3 and GaCl3 are interpolated from planar to pyramidal. Plots of initial and final InCl3 LUMO orbitals are inset with an isosurface level of 0.06. (d) Cartoon illustrating the difference in electron trap depths between InP and GaP arising from GaP’s larger band gap.

To understand the origin of these differing trap depths, we performed interpolations between pyramidal and planar InCl3 and GaCl3 (Figure 4c). We find that the differences between pyramidal and planar defects can be understood using simple molecular arguments. As we interpolate from planar to pyramidal (SI IV.I), we observe that the LUMO, which corresponds to the trap state, decreases in energy by around 0.7 eV in both InCl3 and GaCl3. This is accompanied by a shift of electron density onto the metal in the LUMO, as can be seen in the orbital plots in Figure 4c and through ChELPG charge analysis (SI IV.II) ⁹⁴. This charge makes the LUMO more lone-pair-like, lowering the energy of the anti-bonding state resulting in a deeper trap. This analysis does not explain the difference between the electron trap depths in InP and GaP, which most likely arise from GaP’s wider band gap ^{56, 92} where shallow traps in GaP become non-trap** as the CBM decreases in energy (Figure 4d). We find support for this idea through additional interpolations between four-coordinate and three-coordinate InCl4 and GaCl4 (SI IV.III).

Even accounting for the differing depths of pyramidal and planar traps, the distributions of electron trap depths in Figure 4a,b remain quite broad. These shifts can be explained primarily through two electrostatic effects in the trap centers’ local environment. First, we observe that In-3c and Ga-3c bound to P-3c have shallower traps on average than those not bound to P-3c (Figure 5a,b). This can be understood by recognizing P-3c as having an excess of negative charge, which destabilizes the nearby trap state and raises its energy towards the CBM. A similar effect exists in reverse for hole traps, where P-3c bound to In-3c have shallower traps as the excess positive charge of the In-3c stabilizes the trap state (SI V). The second effect arises from the internal dipole moment of the QD. In both real QDs and our models, any asymmetry will lead to the creation of an internal dipole moment which can shift the depth of trap states, as previously noted for perovskite NCs ⁵¹. We show this effect for In-3c and Ga-3c in Figure 5c,d, computing the dipole overlap as the dot product between the dipole moment (pointing toward the positive charge) and the position vector of the trap center. We observe that a positive dipole overlap deepens trap states, whereas a negative dipole overlap almost always makes In-3c and Ga-3c nontrap**. Again, a similar effect exists in reverse for P-3c (SI V).

Figure 5: (a,b) Trap depth distribution for In-3c and Ga-3c, respectively, across all QDs in the dataset combined with contributions from cations bound to P-3c separated from those not bound to P-3c. (c,d) Trap depth distribution for In-3c and Ga-3c, respectively, across all QDs in the dataset combined with contributions from cations with different dipole overlaps separated. Dipole overlap is calculated as the dot product of the position vector of the trap center with the dipole moment (pointing toward positive charge). A cutoff of (-40,40) is used to define the "near zero" region. Only the depth of the first trap on each 3c cation is shown to highlight trends. Discrete trap depths are broadened by normalized Gaussian functions with RMS width 0.1 eV. Non-trap** 3c atoms are assigned a depth of 0.

Note that we have not included traps from two-coordinate atoms in the above discussion. Such species appear in our dataset, and moreover cause deep traps when they appear (SI VI). However, we find that surface reconstruction in InP is sufficient to passivate most two-coordinate defects when created, and that structures with two-coordinate atoms are highly unstable relative to structures without. We thus conclude that two-coordinate species are at best minor contributors to trap** in real InP and GaP QDs.

In conclusion, we have investigated the prevalence, character, and depth of trap states in InP and GaP QDs using DFT. Our results are kept generalizable through the study of 160 QDs with variable size, shape, and surface defects. Through orbital localization, we deconvolute the dense band edge to identify otherwise evasive trap states including as yet unexplored “structural” traps tied to distorted fully-coordinated atoms. We leverage our dataset to analyze trends in trap depth arising from cation species, cation geometry, and local electrostatic effects which provide avenues for trap control. These results yield important insights into trap states in InP and GaP QDs, as well as informing guidelines for effective simulation of trap states in general QDs. Future directions include detailed investigation of the origin and character of the structural traps observed here, as well as the study of trap states in other III-V materials such as InGaP and III-V core-shell heterostructures. Investigation into the effects of surface oxidation and other impurities on trap state formation, as well as the efficacy of different surface passivation schemes, such as co-passivation with L-type ligands, is necessary to gain a complete picture of trap states in III-V QDs. Finally, it would be instructive to investigate the excited state electronic structure of III-V QDs directly using, for example, time-dependent DFT.

This work was supported by the German Research Foundation (DFG, KI 2558/1-1). We would like to thank the MIT School of Science, MIT Chemistry Department, and the Office of Graduate Education for support of this work through the Dean of Science Fellowship.

The Supporting Information is available free of charge at XXX

• Ligand choice, construction procedure, charge effects, defect induction procedure, computational details, participation ratio, procedure for identification of 1st bulk state, discussion of quantum-confined state, reconstruction energies, reconstruction with Cl ligands, labeled defects, interpolation details, ChELPG charge analysis, interpolation from four-coordinate to three-coordinate, broadening of hole trap depths, two-coordinate traps (PDF)

• All QD .xyz files (.zip)

• All trap depths (.xlsx)