Intrinsic point defects (vacancies and antisites) in CdGeP₂ crystals Open Access

Cadmium germanium diphosphide (CdGeP₂ or simply CGP) is a tetrahedrally bonded semiconductor analogous to GaP. The CdGeP₂ crystals have a direct bandgap near 1.72 eV, positive birefringence (n_e > n_o), and a large nonlinear optical coefficient.^1,2 They are part of the family of II-IV-V₂ nonlinear optical materials.³ Unlike the ZnGeP₂ and CdSiP₂ crystals widely used in optical parametric oscillators to generate tunable output in the 3–6 μm region,^4,5 the phasematching needed for similar applications is not possible in CdGeP₂ crystals because of a smaller birefringence.^2,4 Other potential applications, however, have emerged for CdGeP₂. Optical rectification of near-infrared laser pulses generates broadband terahertz radiation^6–8 and doping with Mn produces room-temperature ferromagnetism suitable for spintronics.^9–12 

In the present paper, we use electron paramagnetic resonance (EPR), electron-nuclear double resonance (ENDOR), and optical absorption to identify and characterize intrinsic point defects (vacancies and antisites) in a CdGeP₂ crystal. Initially, in the as-grown crystal, the defects are in charge states that have no unpaired spins, and no EPR signals are observed. Exposing the compensated crystal to near-band-edge light while being held at low temperature converts the defects to charge states that are easily monitored with EPR. The light produces spectra from singly ionized cadmium vacancies, singly ionized germanium-on-cadmium antisites, and neutral phosphorous vacancies. Spin–Hamiltonian parameters (g and hyperfine matrices) obtained from the angular dependence of the EPR spectra are used to establish detailed models of the defects. Two photoinduced optical absorption bands observed at low temperatures are attributed to these native defects.

By including CdGeP₂, our present report significantly increases the amount of experimental data in the literature describing native defects in nonlinear II-IV-P₂ crystals. Earlier EPR and optical absorption studies focused on defects in ZnGeP₂, ZnSiP₂, and CdSiP₂ crystals.^13–30 The more complete set of experimental data now available, when coupled with computational modeling efforts using advanced density-functional-theory (DFT) methods,^31–37 is expected to lead to a comprehensive understanding of the device-limiting intrinsic defects in these chalcopyrite materials. This, in turn, should provide guidance to crystal growers attempting to maintain stoichiometry, either during growth or with post-growth diffusion and annealing treatments.

An undoped CdGeP₂ crystal was grown by the horizontal gradient freeze method at BAE Systems (Nashua, NH). The process previously developed at BAE Systems to grow ZnGeP₂ and CdSiP₂ crystals was followed.^5,38 A transparent furnace was used and the starting materials were synthesized from high-purity elements. The melting point of CdGeP₂ is near 800 °C.³⁹ A small rectangular-shaped crystal suitable for optical and EPR experiments was cut from the larger boule. The approximate dimensions of this crystal were 2.7 × 2.0 × 3.5 mm³, with [ $110$], [ $1 1 ¯0$], and [ $001$] directions perpendicular to the faces.

The EPR spectra were acquired with a Bruker EMX spectrometer operating near 9.39 GHz. An Oxford Instruments ESR-900 helium-gas flow system controlled the sample temperature and a Bruker NMR gaussmeter provided corrections for the small difference in the magnetic field at the sample and the Hall sensor located on a magnet pole cap. The intrinsic defects were converted to their paramagnetic charge states with either 700 or 850 nm light from an LED (Thorlabs Models M700L4 and M850L3). The approximate bandwidths (FWHM) of the LEDs were 20 and 30 nm, respectively. The concentrations of defects were obtained by comparing their spectra (intensities, linewidths, and number of lines) to a Bruker standard pitch sample. ENDOR spectra were taken with a Bruker Elexsys E-500 spectrometer operating near 9.48 GHz. Frequency modulation (at 12.5 kHz) of the radio-frequency source gave first-derivative ENDOR signals. A ThermoScientific Nicolet 8700 FTIR spectrometer and an ultrabroadband fused-silica wire-grid polarizer (Thorlabs Model WP25M-UB) were used to take optical absorption spectra in the 700 nm to 2.5 μm region. The white light source and a quartz beamsplitter were used for all wavelengths, while a silicon detector was used for wavelengths less than 900 nm and a DGTS detector was used for longer wavelengths. A cryostat from Cryo Industries (Model 110-637-DND) and a temperature controller (LakeShore Model 335) were used to obtain absorption data below room temperature.

Small rotations of the four phosphorus atoms surrounding each cadmium and germanium atom are described by the parameter u (see Ref. 28 for an illustration of this effect). These rotations are caused by the different sizes of the cadmium and germanium atoms. In an ABC₂ chalcopyrite crystal where the A and B atoms are the same size, u = 1/4 and c = 2a. For CdGeP₂, however, u is slightly greater than 1/4 and c is less than 2a. Instead of being equally spaced from its four immediate neighbors, each phosphorous atom shifts by an amount Δx = (u − 1/4)a along an a direction toward the two germanium neighbors and away from the two cadmium neighbors. A complete set of bond lengths and bond angles for CdGeP₂, CdSiP₂, and ZnGeP₂ is given in Table I. GeP₄ and SiP₄ units in these materials are very close to regular tetrahedra (with equal P-P separation distances and bond angles near 109.47°). The CdP₄ and ZnP₄ units, however, are distorted tetrahedra, with two of the P-P distances significantly larger than the remaining four P-P distances. Tetrahedral covalent radii (1.173, 1.128, 1.225, 1.225, and 1.405 Å for Si, P, Zn, Ge, and Cd, respectively) provided by Van Vechten and Phillips⁴³ predict, with good agreement, the experimental Cd–P, Ge–P, Si–P, and Zn–P separation distances in Table I.

Density-functional-theory (DFT) calculations of the electronic structure of CdGeP₂ by He et al.⁴⁴ show that Cd–P interactions are dominant at the top of the valence band and Ge–P interactions dominate the bottom of the conduction band. Specifically, the top of the valence band is formed primarily from P-3p and Cd-4p states, with a small contribution from Ge-4p states, whereas the conduction band is formed primarily from P-3p, Ge-4s, and Ge-4p states, with a small contribution from Cd-4p states.

Figure 2 shows the EPR spectra from singly ionized cadmium vacancies (⁠ $V Cd −$⁠) and singly ionized germanium-on-cadmium antisites (⁠ $Ge Cd +$⁠) in the CdGeP₂ crystal. In Fig. 2(a), the magnetic field was along the [001] direction. In Figs. 2(b) and 2(c), the field was along the [100] and [110] directions, respectively. These spectra were acquired at 60 K while the crystal was exposed to 700 nm light (850 nm light produced the same spectra). After removing the light, the intensities of the two spectra decreased by approximately 23% over several minutes, as electrons and holes on spatially close $Ge Cd +$ donors and $V Cd −$ acceptors recombined via tunneling. This left only well-separated $Ge Cd +$ donors and $V Cd −$ acceptors to contribute to the spectra. Although the remaining $Ge Cd +$ donors and $V Cd −$ acceptors are stable in the dark at 60 K, they were destroyed in a few seconds when the crystal was exposed at this temperature to 1064 nm laser light. In Fig. 2, the EPR lines from cadmium vacancies are between 325 and 341 mT. The antisite donors have outer lines near 313 and 355 mT and a third centerline near 334 mT under the cadmium vacancy signals. Lines assigned to the defects are identified in the figure by “stick” diagrams. An estimate of the concentration of $V Cd −$ acceptors in Fig. 2 is 1.5 × 10¹⁸ cm⁻³. A similar number of $Ge Cd +$ donors are present.

There were no detectable EPR signals before exposure to light, indicating that the cadmium vacancies and antisites are in doubly ionized charge states in the as-grown crystal. The following are possible mechanisms that allow the 700 or 850 nm light at 60 K (as in Fig. 2) to produce $V Cd −$ acceptors and $Ge Cd +$ donors. The light may (1) move electrons from the valence band to the conduction band with the “free” electrons and holes then trapped at $Ge Cd 2 +$ donors and $V Cd 2 −$ acceptors, (2) move electrons from $V Cd 2 −$ acceptors to the conduction band with the electrons trapped at $Ge Cd 2 +$ donors, (3) move electrons from the valence band to $Ge Cd 2 +$ donors with the holes trapped at $V Cd 2 −$ acceptors, or (4) move electrons from the valence band to $V P +$ donors (see Sec. V) with the electrons transferring to deeper $Ge Cd 2 +$ donors and the holes trapped at $V Cd 2 −$ acceptors.

In Fig. 2(a), the S = 1/2 EPR spectrum from the singly ionized cadmium vacancies (⁠ $V Cd −$⁠) consists of one set of three lines with relative intensities of 1:2:1. Equal hyperfine interactions with two of the phosphorous nuclei adjacent to the vacancy are responsible for the three lines (the ³¹P nuclei have an I = 1/2 spin and are 100% abundant). This establishes a model for the $V Cd −$ acceptor that has the unpaired electron shared by two phosphorous neighbors and not one or four phosphorous neighbors. Similar three-line EPR spectra have been reported for singly ionized zinc vacancies (⁠ $V Zn −$⁠) in ZnGeP₂ and ZnSiP₂ crystals^13,26 and cadmium vacancies (⁠ $V Cd −$⁠) in CdSiP₂ crystals.²⁸ [Note: The small outer lines marked with asterisks in Fig. 2(a) are part of the $V Cd −$ spectrum. They are tentatively assigned to hyperfine interactions with nearby Cd nuclei.] In Fig. 2(b), the $V Cd −$ spectrum consists of two sets of three lines. In Fig. 2(c), the spectrum consists of one set of three lines. The complete angular dependence of the EPR spectrum from $V Cd −$ acceptors is shown in Fig. 3. Spectra were taken at various angles as the direction of the static magnetic field was rotated from [100] to [001] and from [001] to [110]. Overlapping EPR lines were not resolved at some of the angles because of the large linewidths (≥2.0 mT).

As illustrated in Fig. 4, there are four crystallographically equivalent orientations (or sites) adjacent to a cadmium vacancy for the pairs of phosphorous atoms that share the unpaired spin. The P₁–P₂ pair is the labeled site 1, the P₂–P₃ pair is site 2, the P₃–P₄ pair is site 3, and the P₄–P₁ pair is site 4. The unpaired spin is not shared by P₁ and P₃ atoms or P₂ and P₄ atoms. This preferential trapping of the hole on one of the four adjacent pairs, instead of all six, is a result of the compression of the lattice along the c axis (i.e., c/2a being less than 1). As seen in Table I, the P₁–P₃ and P₂–P₄ separations are larger than the separations of the other four pairs of phosphorous atoms surrounding the cadmium vacancy. The increased separation makes these two pairs energetically less favorable for trapping the hole. When rotating from [100] to [001] in Fig. 3, the spectra from sites 1 and 3 are degenerate and give one set of three hyperfine lines (the black set of lines), whereas the spectra from sites 2 and 4 are not degenerate, with each giving a set of three lines (the red set and the blue set, respectively). For rotation from [001] to [110] in Fig. 3, the spectra from sites 1 and 4 are equivalent and give the blue set of lines and the spectra from sites 2 and 3 are equivalent and give the red set of lines.

The spin-Hamiltonian in Eq. (1) was rewritten as an 8 × 8 matrix, and values for the seven parameters were determined using a least-squares fitting program. Input data consisted of 40 of the experimental line positions in Fig. 3 and their corresponding microwave frequencies. The energy eigenvalues of the Hamiltonian were repeatedly calculated as the seven parameters were systematically varied. In the fitting program, the predicted line positions (obtained from the eigenvalues) were compared to the measured line positions. Table II contains the best-fit values for the seven parameters that describe the singly ionized cadmium vacancy $V Cd −$ in CdGeP₂. These final parameters were used to generate the solid curves in Fig. 3. Our g and A parameters for the cadmium vacancy are similar to parameters previously reported for zinc vacancies in ZnGeP₂ and ZnSiP₂ and the cadmium vacancy in CdSiP₂.^15,26,28

The “free” nuclear resonance frequency ν_N is μβ_NB/hI, where I is the nuclear spin, β_N is the nuclear magneton, μ is the nuclear magnetic dipole moment in units of the nuclear magneton, h is Planck's constant, and B is the magnitude of the magnetic field. In experiments, observing a pair of ENDOR lines (at f₁ and f₂) provides values for ν_N and A, with ν_N identifying the responsible nucleus and A representing the hyperfine coupling.

Figure 5 shows an ENDOR spectrum from $V Cd −$ acceptors in CdGeP₂, taken at 17 K after exposing the crystal to 700 nm light at this temperature. The microwave frequency was 9.484 GHz, and the magnetic field was along the [110] direction and fixed at 342.4 mT, a position that corresponds to the highest field EPR line in Fig. 2(c). Using this fixed value of the magnetic field and the known nuclear magnetic moments from Ref. 51, the corresponding free nuclear resonance frequencies ν_N are 5.907, 0.510, 3.105, and 3.248 MHz for the ³¹P, ⁷³Ge, ¹¹¹Cd, and ¹¹³Cd nuclei, respectively. In Fig. 5, lines from three ³¹P interactions and one ⁷³Ge interaction are present as well as “distant” ENDOR lines from uncoupled ^111,113Cd and ³¹P nuclei. Figure 6 shows the locations around the Cd vacancy of the nuclei responsible for the ⁷³Ge interaction and the three ³¹P interactions. The phosphorous sites are labeled P_A, P_B, and P_C.

The pair of ENDOR lines, near 44.07 and 56.49 MHz in Fig. 5, corresponds to the case described by Eq. (2) and is due to ³¹P nuclei at P_A positions in Fig. 6. These two atoms share much of the unpaired spin and are responsible for the resolved hyperfine observed in the EPR spectra of the $V Cd −$ acceptor (see Fig. 2). The small barely resolved 0.4 MHz splitting within each of these lines in the high-frequency region of Fig. 5 is a result of the indirect coupling of equivalent ³¹P nuclei.⁴⁹ The lines are separated by 12.42 MHz, which is only slightly greater than the 2ν_N value of 11.81 MHz for the ³¹P nuclei. Second-order effects resulting from a large A value (greater than 100 MHz) are responsible for the increased separation.⁴⁹ The ENDOR lines, at 15.44 and 10.80 MHz in Fig. 5, are caused by ³¹P nuclei at sites labeled P_B and P_C, respectively, in Fig. 6. These assignments are supported by an earlier ENDOR study of $V Zn −$ acceptors in ZnGeP₂.¹⁶ The ³¹P line at 15.44 MHz is part of a pair described by Eq. (2), with an A value near 23.63 MHz. Its companion line near 3.5 MHz is not clearly seen in Fig. 5 because of partial overlap from the uncoupled ^111,113Cd nuclei. The ³¹P line at 10.80 MHz is part of a pair described by Eq. (3), with an A value near 9.79 MHz. This places its companion line near 1.00 MHz and, thus, is below the range of the rf sweep in Fig. 5.

The pair of ENDOR lines at 18.52 and 19.58 MHz in Fig. 5 are due to ⁷³Ge nuclei (7.76% natural abundance) located at the two Ge sites shown in Fig. 6. The lines are described by Eq. (2) where ν_N is less than A. Unlike ³¹P and ^111,113Cd nuclei with nuclear spins I = 1/2, the ⁷³Ge nuclei have a nuclear spin I = 9/2 and are expected to exhibit nuclear electric quadrupole splittings in an ENDOR spectrum. Only the m_I = +1/2 to m_I = −1/2 transition for the two values of M_S, however, are seen in Fig. 5 because either the electric field gradient at the Ge nucleus is too small to produce measurable splittings or the quadrupole-split lines are too broad to observe. The two lines are separated by 1.06 MHz, which is only slightly greater than the 2ν_N value of 1.02 MHz for the ⁷³Ge nuclei (when the magnetic field is 342.4 mT). The A value obtained from the mid-point of the pair of Ge lines is 38.1 MHz. Nuclei responsible for additional ENDOR lines in Fig. 5, near 17 and 21 MHz, are not identified in the present study.

Figure 7 shows EPR spectra taken at 18 K from the CdGeP₂ crystal. The microwave frequency was 9.392 GHz, and the magnetic field was along the [110] direction. The spectrum in Fig. 7(a) was obtained with 700 nm light on the crystal. The light was then removed and the spectrum in Fig. 7(b) was taken within a few seconds. In Fig. 7(a), signals from both singly ionized cadmium vacancies (⁠ $V Cd −$⁠) and neutral phosphorous vacancies (⁠ $V P 0$⁠) are present as well as lines from germanium-on-cadmium antisites (⁠ $Ge Cd +$⁠). [Note: The $V Cd −$ acceptors and the $Ge Cd +$ donors are easily saturated with microwave power at 18 K because of long spin-lattice relaxation times. The $V P 0$ donors, with much shorter relaxation times, are not saturated at this temperature.]

After removing the 700 nm light, all the neutral phosphorous vacancies quickly decay at 18 K and only the $V Cd −$ and $Ge Cd +$ spectra are left in Fig. 7(b). A small portion of these $V Cd −$ acceptors and $Ge Cd +$ donors also decayed when the light was removed, similar to the decrease that occurred at 60 K after removing the light (see Sec. IV A). Two possible mechanisms could explain the decay of neutral phosphorous vacancies at 18 K. If they are sufficiently close to $V Cd −$ acceptors, then direct donor–acceptor recombination could account for the decay of $V P 0$ donors. It is also possible that $V P 0$ donors are shallow and revert to singly ionized $V P +$ donors at 18 K by thermally releasing an electron to the conduction band. The report by Höglund et al.⁵² that the neutral phosphorous vacancy in GaP has a state close to the conduction band supports the second mechanism.

To isolate the spectrum from the $V P 0$ donors in CdGeP₂, the difference spectrum in Fig. 7(c) was generated by subtracting the spectrum in Fig. 7(b) from the spectrum in Fig. 7(a). Before subtraction, the spectrum in Fig. 7(b) was multiplied by a factor of 1.2 to account for the decrease in the number of $V Cd −$ acceptors when the light was removed. For this [110] orientation of the magnetic field, the difference spectrum in Fig. 7(c) contains two equally intense lines from the $V P 0$ donors, identified by the lower stick diagram. The concentration of $V P 0$ donors in this spectrum is approximately 8.7 × 10¹⁶ cm⁻³. This concentration of photoinduced phosphorous vacancies is a factor of five less than the concentration of photoinduced cadmium vacancies in Fig. 7(a). These results, however, do not necessarily suggest that there are fewer phosphorous vacancies in our crystal, as not all the vacancies may have been converted to their paramagnetic form.

The angular dependence of the $V P 0$ EPR spectrum is shown in Fig. 8 for rotation of the static magnetic field from [100] to [001] and from [001] to [110]. There are four distinguishable orientations of the defect. At each angle where data were obtained, a spectrum was first taken with 700 nm light on, then a spectrum was taken after removing the light, and finally, a difference spectrum was generated to identify the $V P 0$ lines. Large linewidths and overlapping lines prevented usable data from being acquired at some of the angles. Since the EPR spectrum showed no resolved hyperfine, an S = 1/2 spin-Hamiltonian containing only an electron Zeeman term (⁠ $H= β S⋅ g⋅ B$⁠) was used to analyze the angular dependence in Fig. 8. The g matrix has six independent parameters, three principal values and three Euler angles (α, β, and γ) that describe the directions of the principal axes. After writing the spin Hamiltonian as a 2 × 2 matrix, values for these parameters were determined using a least-squares fitting program. Input data consisted of the 30 experimental line positions in Fig. 8 and their corresponding microwave frequencies. Table III contains the best-fit values for the parameters that describe the neutral phosphorous vacancy (⁠ $V P 0$⁠). These parameters were used to generate the solid curves in Fig. 8. To allow the principal axes to be more easily visualized, the directions are expressed in polar and azimuthal angles in Table III. The polar angle θ is defined relative to [001], and the azimuthal angle ϕ is defined relative to [100] with a positive rotation from [100] toward [010] in the plane perpendicular to [001]. Our g-matrix parameters for the phosphorous vacancy in CdGeP₂ are similar to those obtained for the phosphorous vacancy in ZnGeP₂.¹⁴ In ZnSiP₂ and CdSiP₂, only the g_c value was reported for the phosphorous vacancy.^26,28

In a DFT study, Jiang et al.³³ predicted that the unpaired spin associated with the neutral phosphorous vacancy (⁠ $V P 0$⁠) in ZnGeP₂ is shared by the two neighboring Ge atoms. These investigators³³ also found that the formation of the neutral anion vacancy was accompanied by a significant relaxation of the surrounding lattice, with the Zn atoms moving away from the vacancy and the Ge atoms moving toward the vacancy. In CdGeP₂, the experimentally determined directions of the principal axes of the g matrix in Table III suggest that the phosphorous vacancy's unpaired spin is localized on one adjacent Ge atom. The direction of the principal axis associated with g₁ is close to the line from a Ge atom to the center of the vacancy (i.e., the direction of the sp³ lobe extending out from the Ge atom). The polar and azimuthal angles (θ and ϕ) for this latter direction in the unrelaxed lattice are included in Table III. If the unpaired spin were shared by two adjacent Ge atoms, one of the principal axes should be close to the line joining the two Ge atoms. None of the g-matrix principal-axis directions, however, is close to the Ge–Ge internuclear axis.

Figure 2 shows the photoinduced EPR spectrum from the singly ionized germanium-on-cadmium antisite (⁠ $Ge Cd +$⁠) in the CdGeP₂ crystal. Similar spectra have been reported for cation antisite donors (group IV atoms on group II sites) in ZnGeP₂, ZnSiP₂, and, CdSiP₂ crystals.^17,26,28 In the II-IV-P₂ chalcopyrites, these antisite donors are easily saturated (especially at lower temperatures), and, thus, are best observed with low microwave power. The $Ge Cd +$ spectra in Fig. 2 consist of three lines caused by equal hyperfine interactions with two adjacent phosphorous nuclei (the center line is under the $V Cd −$ spectrum). Just as with the $V Cd −$ spectra in Sec. IV, equal interactions give ratios of 1:2:1 for the lines. The germanium atom is smaller than the cadmium atom it replaces, thus allowing relaxation of the surrounding lattice. A similar situation occurs in CdSiP₂ where the silicon atom is considerably smaller than the cadmium atom. It remains unclear, however, as to why the unpaired spin is primarily localized on two phosphorous neighbors, instead of being equally shared by all four phosphorous neighbors in these antisite defects. In ZnGeP₂, the zinc atom is the same size as the germanium atom, but the unpaired spin in the $Ge Zn +$ donor is still shared by only two adjacent phosphorous atoms.¹⁷ In Fig. 2, the widths of the $Ge Cd +$ lines are approximately 6.1 mT. Much of this large width is unresolved hyperfine caused by overlap of the unpaired spin onto the neighboring atoms.

The EPR spectrum from the $Ge Cd +$ donors exhibits very little angular dependence (see Fig. 2), and a detailed analysis of g and hyperfine parameters was not attempted. With the magnetic field along the [001] direction, the g value is 1.996 and the hyperfine parameter for each of the two primary ³¹P nuclei (spacing between adjacent lines) is 615 MHz. For the [100] direction, these values are 2.001 and 579 MHz. For the [110] direction, they are 2.004 and 568 MHz. Slightly smaller hyperfine values were reported for $Ge Zn +$ and $Si Cd +$ donors^17,28 in ZnGeP₂ and CdSiP₂, and values approximately 50% smaller were reported for the $Si Zn +$ donor²⁶ in ZnSiP₂.

For completeness, we draw attention to another antisite donor (a group V atom on group IV sites) that was observed with EPR in an early study of ZnGeP₂ and CdSiP₂ crystals.⁵³ These $P Ge 0$ and $P Si 0$ donors (i.e., PP₄ clusters) have a large hyperfine interaction (∼2380 MHz) with the center phosphorous atom and have smaller equal hyperfine interactions (∼244 MHz) with the four surrounding phosphorous atoms. Their neutral charge state was formed at low temperature with near-band-edge light (and decayed in a few minutes at 20 K after removing the light). The $P Ge 0$ and $P Si 0$ donors have not been detected in crystals grown in recent years by the horizontal gradient freeze technique.

As initially described by Shay et al.,¹ CdGeP₂ crystals have a direct energy gap near 1.72 eV at room temperature, corresponding to a transition from the highest of the three valence bands to the conduction band. Selection rules for this transition require that light polarized with E ‖ c is preferentially absorbed (i.e., the optical absorption coefficient when approaching the edge from the long wavelength side is larger for E ‖ c than for E ⊥ c). Absorption spectra show an apparent shift of the band edge when using different polarizations of the probe light. Shay et al.¹ also confirmed that the energy gap widens when the crystal is cooled below room temperature, an important behavior for the present study, when 700 nm (1.77 eV) light is used to produce EPR spectra.

Figure 9(a) shows the optical absorption spectra obtained at 80 K from the CdGeP₂ crystal. Polarized light (E ⊥ c) was used. Corrections were not made for reflection losses. A spectrum was taken before exposing the crystal to 850 nm light (only light from the FTIR spectrometer was present). Then a spectrum was taken with 850 nm light from the LED on the crystal. The additional 850 nm light further increased the absorption across the region from 2.0 μm to 700 nm. We attribute much of the absorption seen in this region before applying the LED to the effects of the spectrometer's white light source. Figure 9(b) shows the difference between the “with 850 nm” and the “before” spectra. This difference spectrum is composed of two broad overlapping bands, as illustrated by green and blue curves in the figure. The green curve is a simulation that has a Gaussian shape with a peak at 1.18 eV (1.05 μm). It was generated by matching the simulation to the difference curve in the 0.7–1.0 eV region. The blue curve in Fig. 9(b), with a peak near 1.64 eV (756 nm), was then obtained by subtracting the green curve from the difference curve. Although not shown, a less intense absorption band is seen near 2.0 μm with light-polarized E ‖ c.

The two photoinduced optical absorption bands in Fig. 9(b), at 1.05 μm and 756 nm, are associated with native point defects. In ZnGeP₂, a broad band peaking near 1.2 μm has been attributed to singly ionized zinc vacancies,^18,20 and in CdSiP₂, several broad bands peaking in the 800 nm to 1.0 μm region have been attributed to singly ionized silicon vacancies.²⁹ By analogy with ZnGeP₂, we suggest that the 1.05 μm band in CdGeP₂ is a transition associated with singly ionized cadmium vacancies (EPR shows that cadmium vacancies are the dominant acceptors in CdGeP₂). Possible models are either an intracenter transition from the ground state to a localized excited state of the $V Cd −$ acceptor or an acceptor-to-donor transition involving $V Cd −$ acceptors and $Ge Cd +$ donors. The defect(s) responsible for the 756 nm band in CdGeP₂ are not identified in the present study.

There is a consistency among the native defects observed in various II-IV-P₂ chalcopyrites. Crystals grown for nonlinear optical applications are usually compensated, with group II and group IV vacancies being the dominant acceptors and antisites (group IV atoms on group II sites) and P vacancies being the dominant donors. The energies of formation for these defects have been calculated by Jiang et al.³³ Low levels of unintentional impurities such as Fe, Mn, Ni, and Cu, coming from starting materials, are also often present but normally play minor roles.

Group II vacancies: These vacancies are very common in the II-IV-P₂ chalcopyrites. They are present in the as-grown crystals as doubly ionized acceptors but are converted to paramagnetic singly ionized acceptors (⁠ $V Zn −$ and $V Cd −$⁠) by exposure at low temperature to near-band-edge light. When there is an insufficient concentration of donors, a portion of the group II vacancies will be in the singly ionized state in the as-grown crystals. The singly ionized acceptors have their unpaired spin localized on two of the phosphorous atoms adjacent to the vacancy. This gives a characteristic three-line hyperfine spectrum that is easily monitored with EPR at low temperatures. An intracenter transition (ground state to a localized excited state) of these singly ionized acceptors is responsible for a broad optical absorption band peaking near 1.0 μm.

Group IV vacancies: These vacancies are less common in the II-IV-P₂ crystals. The $V Si −$ acceptors are seen with EPR in as-grown CdSiP₂ crystals, and their concentration can often be increased by illuminating at low temperatures with near-band-edge light. In contrast, the $V Ge −$ acceptors are not seen in as-grown ZnGeP₂ crystals but have been formed by displacement events when a crystal is irradiated at room temperature with 2 MeV electrons. Thus far, there have been no reports of group IV vacancies in ZnSiP₂ and CdGeP₂ crystals. Singly ionized group IV vacancies have a characteristic five-line EPR hyperfine spectrum because the unpaired spin is equally shared by the four phosphorous neighbors. In CdSiP₂, the $V Si −$ acceptors are responsible for broad optical absorption bands peaking near 800 nm and 1.0 μm.

Antisite donors: The group-IV-on-group-II antisites are a major donor in II-IV-P₂ crystals, being seen with large concentrations in all the materials studied thus far. Because of the presence of acceptors, these antisites are usually doubly ionized (⁠ $Ge Zn 2 +$⁠, $Si Zn 2 +$⁠, $Si Cd 2 +$⁠, and $Ge Cd 2 +$⁠) in the as-grown crystals. They are converted to paramagnetic singly ionized donors with near-band-edge light, as electrons move from acceptors to donors. The unpaired spin is equally shared by two of the adjacent phosphorous atoms, thus giving a widely separated three-line EPR hyperfine spectrum that has very little angular dependence.

Phosphorous vacancies: These vacancies are the other common donor in II-IV-P₂ crystals. They are present as singly ionized donors (⁠ $V P +$⁠) in as-grown crystals and are only converted to neutral donors (⁠ $V P 0$⁠) during illuminations at temperatures below 15–20 K. After removing the light, the neutral donors quickly decay at these low temperatures, either by direct transfer of an electron to a nearby acceptor or by the thermal release of an electron from the shallow neutral state to the conduction band. The photoinduced neutral state is paramagnetic but has proven difficult to observe with EPR, primarily because its spectrum is always obscured by the acceptor spectrum that is unavoidably present. There is no resolved hyperfine to aid in identification, and only the anisotropy of the g matrix allows models to be established.

Electron paramagnetic resonance (EPR) has been used to identify and characterize native defects in CdGeP₂ crystals. Spectra from cadmium vacancies (⁠ $V Cd −$⁠), germanium-on-cadmium antisites (⁠ $Ge Cd +$⁠), and phosphorous vacancies (⁠ $V P 0$⁠) are observed, and their production conditions are described. The analysis of the angular dependence of these EPR spectra gives spin-Hamiltonian parameters and allows models to be established. In the case of the $V Cd −$ acceptor, ENDOR gives additional hyperfine information that is unresolved in the EPR spectra. These ENDOR results verify the significant delocalization of the unpaired spin density around the cadmium vacancy. Broad photoinduced optical absorption bands peaking near 1.05 μm and 756 nm in CdGeP₂ are attributed to native defects. A final section in the paper provides an overview of the dominant donors and acceptors experimentally observed in II-IV-P₂ nonlinear crystals.

One of the authors (T.D.G.) was supported at the Air Force Institute of Technology by an NRC Research Associateship Award. Any opinions, findings, and conclusions or recommendations expressed in this paper are those of the authors and do not necessarily reflect the views of the United States Air Force.

The authors have no conflicts to disclose.

T. D. Gustafson: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Validation (lead); Writing – review & editing (equal). N. C. Giles: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Resources (equal); Visualization (lead); Writing – review & editing (equal). P. G. Schunemann: Conceptualization (equal); Resources (equal); Writing – review & editing (equal). K. T. Zawilski: Conceptualization (equal); Resources (equal); Writing – review & editing (equal). K. L. Averett: Conceptualization (equal); Funding acquisition (equal); Project administration (equal); Writing – review & editing (equal). J. E. Slagle: Conceptualization (equal); Funding acquisition (equal); Project administration (equal); Writing – review & editing (equal). L. E. Halliburton: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Writing – original draft (lead); Writing – review & editing (equal).

The data that support the findings of this study are available within the article.