Table Of Content

Understanding and Control of Bipolar Doping in Copper Nitride Angela N. Fioretti,1,2 Craig P. Schwartz,3 John Vinson,4 Dennis Nordlund,3 David Prendergast,5 Adele C. Tamboli,1,2 Christopher M. Caskey,1,2 Filip Tuomisto,6 Florence Linez,6 Steven T. Christensen,1 Eric S. Toberer,1,2 Stephan Lany,1 and Andriy Zakutayev ∗1 6 1 0 1National Renewable Energy Laboratory, Golden, Colorado 80401 USA 2 2Colorado School of Mines, Golden, Colorado 80401 USA n 3Stanford Synchrotron Radiation Laboratory, SLAC National Accelerator Lab, a J Menlo Park, California 94720 USA 3 1 4National Institute of Standards and Technology, Gaithersburg, Maryland 20899 USA 5Lawrence Berkeley National Laboratory, Berkeley, California 94720 USA ] ci 6Aalto University School of Science, Espoo, Finland s - l r ∗To whom correspondence should be addressed; E-mail: [email protected]. t m . t a m Abstract - Semiconductormaterialsthatcanbedopedbothn-typeandp-typearedesirablefordiode-basedapplicationsand d transistor technology. Copper nitride (Cu N) is a metastable semiconductor with a solar-relevant bandgap that n 3 o has been reported to exhibit bipolar doping behavior. However, deeper understanding and better control of the c mechanismbehindthisbehaviorinCu Niscurrentlylackingintheliterature. Inthiswork,weusecombinatorial 3 [ growth with a temperature gradient to demonstrate both conduction types of phase-pure, sputter-deposited 1 Cu3N thin films. Room temperature Hall effect and Seebeck effect measurements show n-type Cu3N with 1017 v electrons/cm3 for low growth temperature (∼35◦C) and p-type with 1015–1016 holes/cm3 for elevated growth 2 temperatures (50–120◦C). Mobility for both types of Cu N was ∼0.1–1 cm2/Vs. Additionally, temperature- 3 6 dependent Hall effect measurements indicate that ionized defects are an important scattering mechanism in 3 p-typefilms. BycombiningX-rayabsorptionspectroscopyandfirst-principlesdefecttheory,wedeterminedthat 3 0 VCu defects form preferentially in p-type Cu3N while Cui defects form preferentially in n-type Cu3N. Based on . these theoretical and experimental results, we propose a kinetic defect formation mechanism for bipolar doping 1 0 inCu3N,thatisalsosupportedbypositronannihilationexperiments. Overall, theresultsofthisworkhighlight 6 the importance of kinetic processes in the defect physics of metastable materials, and provide a framework that 1 can be applied when considering the properties of such materials in general. : v i X 1 Introduction tors, such as the III-N class of materials. For exam- r a ple, true p-n junctions based on GaN took decades to develop, ultimately leading to a Nobel Prize.1,2 The ability to dope a semiconductor both n-type and Effective p-type doping is also problematic in solar- p-type is a critical factor in determining the viabil- absorbing nitride alloys such as (In,Ga)N1,3 and has ity of that material for diode-based applications and yet to be shown in other novel nitride semiconductors transistor technologies. In cases where bipolar dop- withsolar-matchedbandgaps,likeZnSnN . Giventhe 2 ing is not achievable, selecting a suitable heterojunc- difficulty in obtaining bipolar doping in many cases, tion partner for diode fabrication can generate a host particularly in nitride semiconductors, any material of additional materials challenges that often hinder that exhibits this quality warrants pursuit of a deeper device development. This problem has been partic- understanding of the phenomenon. ularly acute in the n-type nitride-based semiconduc- 1 Incontrasttomanyothernitridematerials, bipolar thesis method paired with high throughput charac- doping has been experimentally demonstrated4 and terization, in addition to advanced characterization externally verified5 in the metastable semiconductor techniques and both bulk and surface defect calcu- copper nitride (Cu N). This behavior has been ob- lations. Finally, the kinetically-driven, point-defect- 3 served as a function of either growth temperature4 or based mechanism developed herein for bipolar doping Cu/N flux ratio,5 suggesting some kind of self-doping in Cu N highlights the important influence of kinetic 3 mechanism. The ability to achieve both conduction processes on the defect physics of metastable materi- types in Cu N is particularly intriguing due to its di- als in general. 3 rect bandgap of 1.4 eV and indirect bandgap of 1.0 eV, which are well-suited for photovoltaic (PV) ap- 2 Methods plications.4,6 Growth temperature would be expected to affect the defect balance (and thus doping type) in 2.1 Thin Film Deposition Cu N, since nitrogen chemical potential is a function 3 of both nitrogen pressure and substrate temperature ThinfilmsofCu3Nweredepositedviaradiofrequency under the ideal gas approximation.6 However, chang- (RF)sputterdepositionon50x50mmEagleXGglass ing Cu N growth temperature may not only control substrates using a combinatorial sputter deposition 3 its doping but can also lead to its decomposition or chamber with 1 x 10-6 Torr residual water base pres- changethesurfacekineticsduringgrowth,sinceCu N sure and an RF-plasma atomic nitrogen source. Two 3 isathermodynamicallymetastablematerial.6–8 Thus, sputter guns with 50 mm elemental Cu targets were a sound mechanistic understanding of how bipolar inclined at 45◦ with respect to the substrate, and the doping occurs in this material is necessary in order atomic nitrogen source was positioned at normal in- to manipulate this property with skill, and such un- cidence angle, to achieve a nominally constant film derstanding is currently lacking in the literature. thickness and to avoid a gradient in nitrogen activity (left side, Fig. 1). Prior work on Cu N has focused on characteriz- 3 ing properties such as decomposition temperature,7–9 reflectivity,10–12 and conductivity11 as a function of changing deposition parameters. Cu N was found to 3 exhibiteithern-typeconduction13,14 orquasi-metallic behavior,11,15 with carrier concentration or conduc- tivitybeingdirectlyproportionaltosubstratetemper- ature during growth. Theoretical calculations later showed that Cu N has a band structure with anti- 3 bondingstatesatthevalencebandmaximum(VBM), opposite to that of conventional semiconductors with Figure 1: Diagram depicting the combinatorial cham- bonding states at the VBM. Such an unusual band ber geometry used to obtain a gradient only in tem- structurecausesdefectstoformasshallowstatesnear perature (left) and the layout of the 44-point com- the band edges rather than in the mid-gap, leading to binatorial grid used for all mapping characterization a property known as defect tolerance.4 This finding described in this work (right). The asterisks denote suggests that intentionally introducing native point regions of the film that were removed and subjected defects into Cu N during growth could be a possible 3 to non-mapping-style measurements. technique for tuning its doping without deteriorating its performance. Target-substrate distance was kept at 13 cm for all In the following manuscript, we demonstrate n- samples. Foramoredetaileddescriptionoftheatomic type and p-type conduction in Cu N as a function nitrogen source and of the sputter deposition proce- 3 of substrate temperature during growth, and propose dures used in this work, see Ref. [6]. Target power a mechanism that explains both the origins and the density for each gun was maintained at 1.5 W cm-2 temperature-dependence of this bipolar doping be- andtheatomicnitrogensourcewaspoweredat250W. havior. The findings presented in this work create Nitrogen gas was flowed in through the nitrogen RF a new framework for understanding bipolar doping in source at a rate of 10 sccm and balanced by 10 sccm Cu N, which is strongly dependent on a balance be- argon gas for a total chamber pressure of 20 mTorr. 3 tween surface and bulk kinetic processes. This un- One side of the glass substrate was kept in contact derstanding is achieved using a combinatorial syn- with a resistive heating unit maintained at constant 2 temperature, so that a continuous temperature gradi- electron yield (TEY), which has a penetration of ∼10 ent was induced across the substrate during growth. nm.20 For a more detailed description of how the combina- Additionally, both n- and p-type Cu N sam- 3 torial temperature gradient was induced in this work, ples were analyzed using positron annihilation spec- see Ref. [16]. Hot-side set point temperatures ex- troscopy. The positron irradiation energy was tuned plored in this study were 120◦C, 90◦C, and no ac- from 0.5 to 25 keV to probe the samples from the tive heating. These set point temperatures represent surface up to a mean implantation depth of 1.3 µm a continuous growth temperature spread from 120– (i.e. into the substrate). After implantation and ther- 50◦C in addition to a growth temperature of 35◦C, malization in the layer, each positron diffuses and an- as calibrated by a thermocouple on the substrate sur- nihilates with an electron. Each such event results face. in two photons with energy of 511 ±∆E keV, which aremeasuredbyahigh-purityGedetectorwithanen- ergyresolutionof1.3keVfullwidthathalfmaximum. 2.2 Characterization TheDopplerbroadening,∆E,dependsonthemomen- tum distribution of the annihilating electrons. Typ- Eachdepositionperformedasdescribedaboveyielded ically, the Doppler broadened energy spectrum nar- a library of samples grown at a range of temperatures rows when positrons are trapped by vacancy defects. dictated by the hot-side temperature. Long-range The shape of the spectrum is characterized using S phase purity was determined using X-ray diffraction and W shape parameters referring to the fraction of (XRD). Conductivity of each library was measured positron annihilation with low-momentum electrons using four-point probe sheet resistance measurements (p < 0.4 a.u.) and high-momentum electrons (1.6 L combined with thickness measured by profilometry.17 a.u. < p < 4 a.u.), respectively.21 L XRD, four point probe, and profilometry measurements were taken using a combinatorial 44-point grid (right side, Fig. 1). Film microstructure was im- 2.3 Computation aged at representative regions via scanning electron microscopy (SEM). Carrier concentration and mobil- Thedefectformationenergies∆H forthecopperva- D ity were measured on selected regions of each library cancy (V ), copper interstitial (Cu), the N vacancy Cu i using room temperature Hall effect in the van der (V ), and substitutional oxygen (O ) were deter- N N Pauw configuration and room temperature Seebeck minedusingtheresultsofpreviouslypublishedsuper- effect. Regions selected for Hall and Seebeck mea- cellcalculations,22 usingtheGGA+Ufunctionalwith surements are indicated in Fig. 1 with white aster- U = 5 eV for Cu-d. In order to determine the ∆H D isks, and were removed from the original libraries in for the actual growth conditions, we assumed here 7x7 mm pieces. Temperature-dependent Hall effect a quasi-equilibrium of Cu N with Cu-metal (∆µ 3 Cu measurements were performed on a region of another = 0), and an activated source of N (∆µ = +0.76 Cu p-type library grown isothermally at ∼160◦C. Error eV).Forthechemicalpotentialofoxygen,weassumed bars shown for the conductivity data give the stan- ∆µ = -0.84 eV, which corresponds to the ideal gas Cu dard deviations among the four repeated rows of each law chemical potential for T = 200◦ C (upper T limit combinatorial sample. Error bars shown for the Hall forCu Ngrowth)andpO =10-8 atm(basepressureof 3 2 effect data give the standard deviations among 5–10 thegrowthchamber). Inordertocalculatethedepen- repeated measurements of each of the selected pieces dence of the V and Cu defect formation energy as Cu i measured from each library. Further information on a function of the distance from the surface, we used a the experimental methods used herein can be found slab supercell with 248 atoms modeling the partially in prior publications.4,17,18 oxidized (001) surface of Cu N. This surface termi- 3 ExperimentalX-rayabsorptionspectra(XAS)were nation, formed by substituting every other N atom takenonrepresentativep-typeandn-typeCu Nsam- of the (001) surface by an O atom was previously 3 ples at beamline 8-2 at the Stanford Synchrotron Ra- foundtoyieldanexceedinglylowsurfaceenergy.4 The diation Lightsource (SSRL). A nominal resolution of chargecompensationwasmaintainedbychangingthe ∆E/E of 10-4 is provided by this beamline.19 Data O/Nratioatthesurfaceuponintroducingthedefects. was collected both by using largely bulk-sensitive to- Thesesupercellcalculationswereperformedusingthe tal fluorescence yield (TFY), which has a penetration VASPcode,23 withotherwiseidenticalcomputational of ∼100 nm, and by the more surface sensitive total settings as in Ref. [24]. 3 The modeled XAS structures were generated as de- tfaacileedstrpurcetvuiorue.s4lyT, hweitphretvhieouexslcye-pctailocnuloaftetdhebuClukOstrsuucr-- (a) sity (arb.)1100107 120ºC Cu (b) n tduerfeecstcsoninssisetrtoefdapinptrooxtimheamte.lyT10h0eastuormfascweistthruspcteucirfiecs Inte 104 001 011 111 002 100Grow cwCoauns3sNigsteaenndedroarfteeladatxbilneyagspttlha5ec0isntagrtuoCcmtuusOr.eoTwnhitehtoiCnpuVoOAf SasPtrs.ul2ac4bturoef Intensity1111000010789 8 600 th Temperatu (c) re bhooTltehh(elaXXrCg-erHlay)ydscepatleaccuitllreaadtipworneersvewioceuarsellcypu.el2ar5ft,2oe6rdmFoierndsltyi,nXattw-roathywecaoynrsie-, y (arb.)1100107 35ºC Cu 40 (ºC) trogen K-edge. Within an ultrasoft pseudopotential ensit104 formalism, the excitation is simulated by replacing Int 20 23Θ0 (degr4e0es) 50 the excited atom with a 1s core hole, such that the excited nitrogen now is represented as 1s12s22p4. K- Figure 2: (a) Cu N XRD patterns as a function of 3 point sampling is used to converge the spectra. The growth temperature spanning the full temperature unoccupied p-type orbitals are then projected onto gradient. All peaks shown correspond to Cu N, and 3 the excited atom and the resultant stick intensity as no contribution from copper metal can be seen. The a function of energy are broadened by 0.05 eV and dashed white line horizontally separates the 120◦C the process is repeated for every atom. data set from the 90◦C data set. (b-c) SEM images Second, calculations based on solving the Bethe- of characteristic Cu N samples grown on glass at (b) 3 Salpeter equation (BSE) were performed within 120◦Cand(c)noactiveheating. Thefilmgrownwith ocean for Cu L-edge as previously detailed, using no active heating exhibits pyramidal grain termina- DFT as a basis.27 The electronic structure was cal- tion with 50–100 nm grains while the film grown at culated within the local-density approximation using 120◦Cexhibitsspheroidgrainterminationwith10–50 the Quantumespresso code,28 and sampling of the nm grains. Brillouin zone was improved through use of the OBF interpolation scheme. The number of bands included hundreds (>1000) of the first conduction bands, and thescreeningofthecore-holepotentialincludedthou- Cu N and a phase-separated mixture of Cu N and 3 3 sands (>4000) bands. The k-space and mesh sam- Cu0. pling were checked for convergence. The broadening was 0.2 eV. The spectra were aligned based on the As shown in Figs. 2b-c, two distinct film mor- Fermi level and potential on the core. phologies were observed for the thin films prepared in this study. Films grown without actively heating the substrate (i.e. the only source of substrate 3 Results heating was from the impinging sputtered atom flux), exhibited pyramidal grain termination with 50-100 3.1 Structure nm grains (Fig. 2b). For films grown at slightly ele- Crystalline Cu N films with long-range phase-purity vated substrate temperature (i.e. 50–120◦C) spheroid 3 were achieved at all growth temperatures explored in grain termination was observed with grain size on this work. Long-range phase-purity can be seen in the order of 10–50 nm (Fig. 2c). Observing larger Fig. 2a,whichdisplaysXRDpatternsforCu Ngrown grains for unheated growth and smaller grains for 3 at various temperatures. Based on comparison to the slightly elevated growth temperature is related to the peak intensities of the Cu N powder diffraction pat- metastable nature of Cu N. At elevated growth tem- 3 3 tern from the Inorganic Crystal Structure Database perature, slightly below the decomposition tempera- (ICSD), these films exhibit no preferential orienta- ture of Cu N, it is possible that partial evaporation 3 tion. The absence of peak intensity at 50.5◦ and at leads to spherical grain termination such that sur- 43.3◦ confirms the absence of any long-range metallic face area is minimized. A more detailed description copper second phase, indicating the films in this work of this counterintuitive grain size phenomenon and represent a comparison between phase-pure Cu N its relation to metastability can be found in previous 3 films grown at different conditions and not between works.6,15 4 3.2 Transport 1) 0.1 -m (a) 4 c T = 25°C Fig. 3 shows the electrical transport properties of S 2 Meas ( 0.01 Cu Nfilmsasafunctionofgrowthtemperature. Des- 3 y vit 4 n p ignations of n-type and p-type in all panels of Fig. 3 cti 2 arebasedonthesignoftheHallcoefficientsandcorre- u 0.001 d sponding Seebeck coefficients for the samples in ques- n 4 o tion. The average n-type and p-type carrier density, C 2 0.0001 mobility, and Seebeck values are summarized in Ta- 30 60 90 120 ble 1. Films from the 120◦C region were unable to Growth Temperature (°C) be reliably measured by Hall effect, due to contact 18 3 ) 10 10 degradation with repeated measurement. The same 3 (b) -cm 1017 oM effect was not observed while collecting Seebeck effect sity ( 1016 102 ytilib dbaetda,eswighnicahteadllpow-teydpet.heThoigfihll-ttehmepgearpatuinrethsaemHpalellsetfo- er Den 101145 101V mc(2 fpercetvdioautsa,waorskam4 pwlaesgurosewdnfiosrotohbetraminailnlygattem16p0e◦rCatuinrea- ri 10 1- dependent Hall effect data. The data from that sam- Car 1013 p-type TG = 160°C 100s 1- ple is indicated with an asterisk in Table 1. ) 2 4 6 8 Hall and Seebeck effect measurements of the three -1 -1 -3 T (K x10 ) Cu N libraries reveal a switch from n-type to p-type Meas 3 5 conductionasgrowthtemperatureincreasedfrom35– ) (c) 1-1 s 4 T = 85-50°C 1ci6d0e◦sCw. iTthhias Usw-sihtcahpeidncmonadjuorcittiyvitcyartrrieenrdtympeeascuoriend- -V G 2 3 TMeas = 25°C as a function of increasing growth temperature in m the 35–120◦C range (Fig. 3a). This correlation sug- c ( 2 y gests competing formation of donor and acceptor de- t bili 1 fects, with donor defects dominating during unheated Mo p-type growth and acceptor defects dominating at elevated 0 temperature. Consistent with this hypothesis, evi- 2 4 6 2 4 6 1014 1015 1016 dence for ionized defect scattering was observed via -3 Carrier Density (cm ) temperature-dependentHalleffectforp-typesamples. This can be seen in Figs. 3b-c, in which carrier den- Figure 3: Cu3N conductivity determined by four sity and mobility were inversely proportional to each point probe as a function of growth temperature for other as a function of decreasing temperature for p- bothlowandelevatedgrowthtemperatureCu3Nfilms type Cu3N grown at 160◦C (Fig. 3b). For low tem- isshownin(a). T-dependentHalleffectdataisshown perature p-type samples, it was found that mobility in (b) for the 160◦film. An inverse relationship be- exhibited an exponential decrease as a function of in- tweencarrierdensityandmobilityastemperaturede- creasing carrier density; again, indicative of ionized creases indicates ionized defect scattering dominates defect scattering and consistent with a competing de- the room temperature mobility. For low temperature fect formation hypothesis. Collectively, these find- p-typefilms, roomtemperaturemobilitywasfoundto ings shed light on the underlying transport properties decreaseexponentiallywithincreasingcarrierdensity, of Cu N and also provide a reliable method for con- 3 as shown in (c), again indicating the influence of ion- trolled bipolar self-doping in this material. izeddefectscatteringforp-typefilms. Designationsof Demonstration in this work of bipolar doping n-type and p-type in all panels are based on the sign in Cu N confirms the reproducibility of this phe- 3 of the Hall coefficients and corresponding Seebeck co- nomenon as reported elsewhere,4,5 but does not offer efficients for the samples in question, summarized in insight into the specific origins of this property nor Table 1. to the identity of the defects giving rise to each ma- jority carrier type. Arguably, a deeper understanding of bipolar doping in copper nitride will be necessary if this desirable property is to be manipulated 5 Growth T / ◦C Carrier Type Carrier Density / cm-3 µ / cm2 V-1 s-1 S / µV K-1 35 electrons 1.62 x 1017 1.63 -69 50–85 holes 1.87 x 1015 1.45 +575 90–120 holes – – +421 160* holes 5.56 x 1016 0.10 +200 Table1: SummarizedHallandSeebeckeffectdatafromphase-pureCu Nfilmsdepositedatincreasingsubstrate 3 temperatures. CarriertypewasdeterminedbasedonthesignsoftheHallandSeebeckcoefficients. S represents the Seebeck coefficient for each sample and µ represents the mobility determined by Hall effect for each sample. *Data taken from Ref. [4] with precision. Recognizing this necessity, we used that only V defects could contribute to p-type dop- Cu different computational methods of defect identifica- ing in this material, as this is the only defect with a tion coupled with X-ray absorption spectroscopy and positiveslopeastheFermilevelmovesdowninenergy positron annihilation spectroscopy to investigate the fromtheconductionbandminimum(CBM)totheva- atomic-scale origins of bipolar self-doping in Cu N. lence band maximum (VBM). This is consistent with 3 other CuI+-based semiconductors, such as CuInGaS 2 and Cu O, that exhibit p-type conductivity. For n- 2 type doping, three stable donor defects were found: V , O , and Cu. Oxygen substitutional defects were N N i included in these calculations, even though O is not N a native defect, because oxygen incorporation from background pressure is common in nitride films. The formation enthalpy of V was found to be 1.5 eV N higher than O and Cu for the entire range of Fermi N i energies plotted, making it the least favorable donor defect to form and therefore an unlikely contributor to n-type doping in copper nitride. Considering defect formation enthalpies alone, the results presented in Fig. 4 suggest that V defects give rise to p-type Cu doping and O plus Cu defects give rise to n-type Figure 4: Calculated defect formation enthalpies as N i doping. a function of the Fermi level for Cu N, assuming a 3 The formation enthalpies shown in Fig. 4 suggest quasi-equilibrium between Cu N, Cu-metal, and acti- 3 that even when the incorporation of O is suppressed vated N. V defects were found to be the only na- Cu by reducing or eliminating O from the growth process tive point defects that could lead to p-type doping in (effectively increasing the ∆H of O ), a true quasi- Cu N.V defectswerefoundtobethehighest-energy D N 3 N equilibriumsituationwouldleadton-typedopingdue defect to form among the native donor defects, indi- to an excess of Cu over V . However, such a situa- cating it is an unlikely contributor to n-type doping i Cu tionmightbedifficulttoachieve,becauseitrequiresa in this material. O defects were included in the cal- N sufficiently high temperature to enable bulk diffusion, culation due to oxygen being a common impurity in which at the same time also facilitates the decom- nitride films. position of the metastable Cu N compound. There- 3 fore, we employed additional computational and ex- perimentalmethodstogathermoreinformationabout 3.3 Bulk Defect Formation Enthalpy the atomic-scale origins of bipolar doping. Fig. 4showsthedefectformationenthalpiesasafunc- tion of the Fermi level, assuming a quasi-equilibrium 3.4 X-Ray Absorption Spectroscopy between Cu N, Cu-metal, and activated N. Only de- 3 fectsthatarestableandcouldcontributeton-typeor Using Near Edge X-ray Absorption Fine Structure p-type doping are shown. It is apparent from Fig. 4 (NEXAFS) measurements and modeling performed 6 Cu 2p L-edge N 1s K-edge To arrive at the above conclusions, it was neces- (a) p-type Cu N (b) sary to use a formalism for modeling NEXAFS spec- 3 p-type Cu N 3 trathatwasnotbasedonself-consistentfieldmethod- a.u.) n-type Cu3N a.u.) n-type Cu N ologies, which are often limited specifically to K-edge y ( y ( 3 absorption. These methods, including FCH, X-ray sit sit en Perfect Xtl en Perfect Xtl corehole(XCH),andHCH,self-consistentlysolvethe nt nt electronic structure with a core hole in place. How- I I CuO Surface CuO Surface ever, the ability to model L-edge excitations using these methods is much more limited. For this reason, (c) (d) it was possible to model the N K-edge using XCH Cu but not the Cu L-edge. To overcome such limita- i Cu u.) u.) i tion, we instead used a method based on the Bethe- a. a. y ( y ( Salpeter equation (BSE), specifically ocean,25,29 to nsit ON nsit ON model both the N K-edge and the Cu L-edge NEX- e e nt nt AFS spectra. I I VCu VCu Upon inspection of Figs. 5a-b, which shows the total electron yield of the NEXAFS experiments for n- 930 940 950 960 970 395 400 405 410 415 420 Energy (eV) Energy (eV) and p-type Cu N, it is apparent that the n-type and 3 p-typesamplesproducedsimilarspectrawithonlymi- Figure 5: NEXAFS spectra for n-type (black curves) norintensitychanges. InboththeCuedgeandNedge and p-type (blue curves) Cu N at the Cu L-edge (a) spectra, the p-type signal has greater intensity than 3 and N K-edge (b) show very similar features other the n-type signal. Furthermore, the modeled spec- than higher intensity for p-type spectra. Modeling tra for a perfect Cu N crystal (dashed black curves) 3 performed using a BSE-based method shows that does not fully capture the features of either the Cu the experimental spectra exhibit features of a perfect L-edge or the N K-edge. Instead, the perfect crystal Cu N crystal (dashed black curves) and of a CuII+ simulation effectively models one leading edge peak 3 surfacestateconsistentwithaCuOsurfacelayer(gray and misses the other leading edge peak in each set of curves). Modeled defect structures for Cu N (pan- spectra. It is only with the addition of a CuII+ sur- 3 els (c) and (d)) indicate that O substitutional de- face state, consistent with a CuO surface oxide (gray N fects (red curves) likely are not major contributors to curves), that both leading edge peaks in both sets of the experimental NEXAFS signal, due to the shift to spectra can be effectively modeled, indicating that a lower energy observed for the Cu L-edge O curve. CuO surface oxide layer is likely present. It is im- N portant to note that this surface oxide layer is prob- ably formed after growth due to exposure to air, and not formed as a native surface termination inside the using ocean,25,29 it was found that O defects are growth chamber. This is in contrast to the partially N unlikely to be dominant contributors to bipolar dop- oxidized (001) surface of Cu3N described in Ref. [4], ing in Cu N. This finding is in addition to the find- which is anticipated to form even with minimal pres- 3 ing in Section 3.3 that V defects are the most en- ence of O during growth. Both n-type and p-type N ergetically unfavorable donor defects to form in this Cu3N exhibit peaks consistent with a CuO surface material. Taken together, these results suggest that oxide layer, and also exhibit features consistent with Cui andVCu defectsmake themostsignificantcontri- a perfect Cu3N crystal. butions to Cu N bipolar doping behavior. NEXAFS Three defect structures for Cu N were simulated 3 3 spectra for the Cu L-edge and the N K-edge for rep- using the BSE-based method described above: Cu, i resentative n-type and p-type samples are shown in V , and O . These defect structures were chosen Cu N Fig. 5, along with modeled spectra corresponding to based on the point defect formation enthalpies pre- Cu, V , and O defect structures. Both Cu and sented in Section 3.3. The resulting modeled spectra i Cu N i V defects were found to be present in both p-type for each edge are displayed in Figs. 5c-d. A verti- Cu andn-typeCu N,whichsupportsthehypothesisthat cal gray line is overlaid on each set of panels to indi- 3 competing donor and acceptor defect formation is the catetheexperimentalpeakpositionofthefirstleading mechanism behind bipolar self-doping in this mate- edge peak that is not due to the CuII+ surface layer. rial. LookingspecificallyattheCuL-edge(Figs.5aandc), 7 theonlydefectstructurethatisunlikelytobeamajor 2.0 contributor to the NEXAFS signal is ON substitution S u (red curve), due to the modeled spectrum being sig- r f a nificantly red-shifted compared to experiment. This 1.5 ce is not to say that ON defects are not present in the Bulk ) films in this work, only that O substitutions in the V N e Cu bulk must exist at a far lower concentration than ei- ( 1.0 i D V ther Cu or V . Turning to the N K-edge (Figs. 5b H Cu i Cu ∆ and d), it is clear that none of the other defect struc- 0.5 tures are shifted enough to be reasonably ruled out as contributing to the overall spectra. It should be noted that the resemblance of the simulated defect 0.0 spectra in panel (d) to the CuO surface layer spec- 16 12 8 4 0 trum in panel (b) is due to the s-p transition of the Depth from Surface (Å) nitrogenK-edgebeinghighlysensitivetobrokenlocal symmetry.30,31 This explains why the simulated de- Figure 6: Depth-dependent defect formation en- fect spectra shown in panel (c) more closely resemble thalpiesforthedonordefect,Cu,andfortheacceptor the simulated perfect crystal spectra shown in panel i defect,V ,showthatV defectsformpreferentially (a),ratherthanresemblingthesimulatedCuOsurface Cu Cu atthefilmsurfaceandarehighenergydefectstoform layer: because the p-d transitions probed at the Cu in the bulk. In contrast, Cu defects are high energy L-edge are not as sensitive to broken local symmetry, i to form at the film surface but require lower energy and therefore the defect spectra are not dominated to form in the bulk compared to V . by these effects as much as the N k-edge spectra are. Cu TheconclusionthatO defectsarenotmajorcontrib- N utors to the NEXAFS spectra is consistent with the films in this work being grown under activated nitrogen atmosphere, in which substitution of oxygen im- 3.5 Depth-dependent Defect Formation puritiesonnitrogensitesismitigatedbythedramatic improvement in nitrogen incorporation.6 Overall, the combination of calculated defect formation enthalpies To build upon the conclusions presented above, point andmodeledNEXAFSspectrapointstoCu andV i Cu defect formation enthalpies for Cu N were recalcu- 3 defects as being the most prominent point defects in lated as a function of depth from the film surface. both p-type and n-type Cu N. 3 Only formation enthalpies for Cu and V defects i Cu were recalculated, because both V defects (Fig. 4) N Conceptually, if both Cu and V defects are and O defects (Fig. 5) were already identified as not i Cu N present in both conduction types of Cu N, then it being major contributors to the bipolar doping phe- 3 follows that the donor defect (Cu) must dominate in nomenon. The results of these calculations, shown i n-type copper nitride and the acceptor defect must in Fig. 6, indicate that V formation energy is Cu dominate in p-type copper nitride. What is not clear sharply lowered towards the surface, whereas the Cu i from this analysis is why the donor defect would form formation energy is only slightly lowered, and even preferentiallyatlowgrowthtemperature(∼35◦C)and increased for a Cu adatom on top of the surface. This the acceptor defect (V ) would do so at elevated finding implies that the prevalence of Cu over V in Cu i Cu growthtemperature(120◦C).Toanswerthisquestion, thebulk(Fig.4)couldbeinvertedclosetothegrowing and thus propose a complete mechanism for bipo- surface, creating a net p-type doping. It would then lar doping in Cu N, an additional set of first princi- dependonkineticfactorssuchasdiffusivity, tempera- 3 ples calculations were performed. Specifically, depth- ture, and growth rates whether the defect concentra- dependentdefectformationenthalpieswerecalculated tionsatthesurfacewouldbedominatedbyCudefects i to determine whether each defect type (Cu and V ) (i.e. excessCuadatoms)orbyV defects(i.e. excess i Cu Cu couldformpreferentiallyatthesurfaceorinthebulk. nitrogen adatoms that do not recombine and desorb). Based on these surface calculation results, a kinetic Toexplorethishypothesis,wetookakineticapproach mechanismwasproposedtoexplaintheobservedtem- todescribingthesurfaceconcentrationofCu andV i Cu perature dependence of the doping type in Cu N. defects as a function of temperature during growth. 3 8 Figure 7: Energy-dependent positron annihilation experiments with p-type and n-type Cu N samples. As the 3 positron penetration depth increases, the n-type sample’s response becomes more similar to the p-type samples’ response, before both of them trends towards the substrate. The dashed lines represents the plateaus defined in the text. 3.6 Defect Kinetics fusion may still occur. Hence, surface contributions (∆H ) to the bulk defect formation enthalpies Surf From the discussion above it is evident that the ob- (∆H ) need to be taken into account to deter- Bulk served p-type character of Cu N at elevated deposi- 3 mine the resulting carrier concentration: ∆H = Tot tion temperature, and the switch to the n-type char- ∆H +∆H , where ∆H is the resulting de- Bulk Surf Tot acter at low temperature (Fig. 3), both cannot be ex- fect formation enthalpy that defines the carrier con- plained by either bulk (Fig. 4) or surface (Fig. 6) de- centration.32 Subtracting the calculated 1.5 eV sur- fectthermodynamics. Hence,weinvokehereasurface face contribution (Fig. 6) from the bulk formation en- kinetic model to explain the experimentally observed ergy of the V acceptor (Fig. 4), and adding the cor- Cu Cu N bipolar doping behavior. Before describing the 3 responding contribution for the Cu donor, makes the i proposedsurfacekineticmodel,itisimportanttonote V acceptor the most energetically favorable defect Cu that Cu N would not exist in the first place in the ab- 3 under elevated growth temperature conditions. This senceofrelativelyhighbulkkineticbarriers,sinceitis enhancedV acceptorformationandsuppressedCu Cu i a thermodynamically metastable material with +0.8 donor formation explains the measured p-type con- eV/f.u heat of formation.6 Thus, it is plausible that ductivity in Cu N in the 50–120◦C substrate temper- 3 there are other lower-height kinetic barriers that con- ature range (Fig. 3a). trol doping behavior below the bulk decomposition temperature, such as surface diffusion barriers. This Next, we address n-type conductivity at lower sub- is particularly true in this case of Cu N thin films for strate temperature (∼35◦C, Fig. 3a). It is likely that 3 which the surface to volume ratio is larger than in the in this temperature regime, equilibration of the sur- bulk. face defects as discussed above does not fully occur. First, we explain p-type conduction in Cu N at el- In this case, a non-equilibrium defect distribution be- 3 evated substrate temperatures (50–120◦C, Fig. 3a). comes frozen in the bulk when the next monolayer Since this temperature range lies immediately below of Cu and N atoms arrives. Note that Cu atoms the decomposition temperature of bulk Cu N, diffu- are more likely to be frozen in the bulk than N 3 sion in the bulk must be frozen, whereas surface dif- atoms, because the Cu-Cu diatomic species proba- 9 thiskinetically-limiteddepositiongrowthprocess. In- N N Cu 2 Cu 2 deed, the rate dependence of the Cu N phase forma- N 3 2 tion has been previously reported for films grown at p-type n-Cu3N p-Cu3N Cu elevated substrate temperatures.6 As for the lower- layer temperature phase-pure Cu3N thin films reported SiO SiO SiO 2 2 2 here, the results of positron annihilation spectroscopy (PAS) suggest the presence of a p-type layer at the n-Cu N p-Cu N Cu+N ) 3 3 2 film/substrate interface of nominally n-type films. 3 -m Fig. 7 shows the W and S parameters plotted as a c y, Cu function of positron irradiation energy for n-type (red sit i Net Defect circles) and p-type (blue squares) Cu3N films, respec- n e Density tively. These results indicate that the positron anni- D hilation characteristics are homogeneous throughout t c e thep-typesamples’depth,whileadefectinhomogene- f e D ity profile is observed through the thickness of the n- g ( VCu type samples. The first 150 nm of the n-type sample o L at the interface with the substrate are different than the rest of the 500 nm thick film. The S and W val- 50°C 180°C ues form a plateau (shown with a dashed line) at the Deposition Temperature, °C top of the film, then move toward the S and W val- Surface slow fast fast ues of the p-type films when proceeding towards the Kinetics film/substrate interface. Hence, the defect structure Bulk of the nominally n-type sample resembles that of the slow slow fast Kinetics p-type samples at the interface with the substrate. The presence of this interfacial p-type layer can be attributed to longer time spent at the growth tem- Figure 8: Schematic illustration of the current under- perature than the rest of the n-type film, supporting standingofbipolarself-dopinginCu N.Thiskinetics- 3 the hypothesis that the n-type doping is enabled by drivendopingmechanismexplainsthephenomenonas sluggishkineticsatthislowerdepositiontemperature. arising from a balance between surface and bulk ki- It should be noted that detailed positron netic processes. When surface diffusion is dominant and bulk diffusion is frozen (50–120◦C), p-type Cu N annihilation-based identification of the vacancy-type 3 defects in the films in this work was not possible due is observed. When both surface and bulk diffusion is to the lack of a reference sample with low defect den- kinetically limited (no active heating), n-type Cu N 3 sity. ForanadditionalrepresentationofthePASdata, is observed. See manuscript text for further explana- showing the variation of W with S for n-type and p- tion. type samples, please see Fig. S1 of the Supplemental Information. bly has a longer residence time at the surface than Our current understanding of bipolar self-doping the gaseous N-N species (i.e. N2 molecule). From a in Cu3N is summarized in Fig. 8. At lower sub- kinetic point of view, atom concentration (c) at the strate temperatures, Cu N shows n-type conductiv- 3 surface of the growing Cu3N film at any given time is ity due to Cui defects frozen in the bulk as a re- given by c ≈ c0e−Kt, where K = K0e−Ea/kbT is the sult of surface diffusion barriers. As the growth tem- T-dependent desorption rate determined by activa- perature increases, such that the thermal energy be- tion energy Ea (proportional to residence time of the comes comparable to the surface diffusion barriers, atoms at the surface). This inherent Cu vs. N asym- the conductivity switches from n-type to p-type due metryleadstoenhancedformationofCui donors,and to the decreased formation enthalpy of VCu, and in- thus produces n-type conductivity at these lower sub- creased formation enthalpy of Cu, at the growing i strate temperatures. Cu N surface (Fig. 6). Finally, at the highest sub- 3 The kinetic character of the hypothesis proposed strate temperature, Cu N decomposes into metallic 3 above to explain the temperature-dependence of Cu and N ,6 which corresponds to bulk equilibra- 2 Cu N conductivity type would imply that the depo- tionofthismetastablesemiconductingmaterial. Note 3 sition time (or rate) is another important variable in thatthismechanisticunderstandingisconsistentwith 10