Enhanced photoluminescence emission from two-dimensional silicon photonic crystal nanocavities
N. Hauke, T. Zabel, K. Mueller, M. Kaniber, A. Laucht, D. Bougeard, G. Abstreiter, J.J. Finley, Y. Arakawa
EEnhanced photoluminescence emission from two-dimensionalsilicon photonic crystal nanocavities
N. Hauke, ∗ T. Zabel, K. M¨uller, M. Kaniber, A. Laucht, D. Bougeard, G. Abstreiter, J. J. Finley, and Y. Arakawa Walter Schottky Institut, Technische Universit¨at M¨unchen,Am Coulombwall 3, D-85748 Garching, Germany Institute for Nano Quantum Information Electronics,Institute of Industrial Science, The University of Tokyo,4-6-1 Komaba, Meguro, Tokyo 153-8505, Japan (Dated: October 26, 2018)
Abstract
We present a temperature dependent photoluminescence study of silicon optical nanocavitiesformed by introducing point defects into two-dimensional photonic crystals. In addition to theprominent TO phonon assisted transition from crystalline silicon at ∼ .
10 eV we observe a broaddefect band luminescence from ∼ . − .
09 eV. Spatially resolved spectroscopy demonstratesthat this defect band is present only in the region where air-holes have been etched during thefabrication process. Detectable emission from the cavity mode persists up to room-temperature,in strong contrast the background emission vanishes for T ≥
150 K. An Ahrrenius type analysisof the temperature dependence of the luminescence signal recorded either in-resonance with thecavity mode, or weakly detuned, suggests that the higher temperature stability may arise from anenhanced internal quantum efficiency due to the Purcell-effect.
PACS numbers: 42.60.Da 42.70.Qs 78.55.-m 42.50.Ct ∗ Electronic address: [email protected] a r X i v : . [ c ond - m a t . m e s - h a ll ] D ec ue to its indirect electronic bandgap, crystalline silicon is rarely used as the activeemitter in semiconductor optics [1]. Interband light emission is predominantly a phonon-assisted process and silicon has, therefore, a very poor internal quantum efficiency. However,the development of efficient silicon based light emitters would pave the way toward CMOScompatible monolithic optical interconnects and, therefore, signal processing speeds muchhigher than currently provided by silicon micro-electronics [2, 3, 4, 5, 6].Recently, enhancement of the photoluminescence (PL) intensity from crystalline silicon atroom-temperature has been observed using two-dimensional (2D) silicon photonic crystals(PhCs) with photonic point defect nanocavities [7, 8]. Such defect PhC nanocavities modifythe spatial emission profile of light; due to the in-plane photonic band gap a much largerfraction of luminescence is emitted perpendicular to the slab than parallel to it [9]. In addi-tion, the high Q -factors that are attainable using PhC nanocavities [10, 11, 12] together withtheir small mode volumes may lead to an enhancement of the internal quantum efficiencyof the active material due to an increase of the radiative emission rate via the Purcell-effect[13]. Other work demonstrates enhanced PL emission from internal light emitters embeddedin photonic crystal slabs [14, 15].In this letter, we present a detailed investigation of the spectrum and temperature sta-bility of the PL emission from crystalline silicon PhC nanocavities. Analysis of our resultsindicates that both the phonon satellites of the interband silicon emission and surface defectstates are responsible for the luminescence of PhC cavity modes. Most interestingly, wecan detect luminescence from the cavity mode up to room temperature whilst the back-ground emission intensity rapidly reduces below our detection threshold for T ≥
150 K. AnArrhenius type analysis of the temperature dependent data, either in resonance with thecavity mode or spectrally detuned from it, indicates that the local radiative emission rate isenhanced in resonance via the Purcell-effect.The samples investigated were fabricated from silicon-on-insulator (SOI) wafers with a d = 250 nm thick active silicon layer on top of a 3 µ m thick layer of SiO . A 2D PhC ispatterned into the upper silicon layer as illustrated in fig. 1 (a). We defined a triangularlattice of air holes with a period of a = 275 nm in the silicon slab as shown by the scanningelectron microscope (SEM) image in the left panel of fig. 1 (b). This was done using electron-beam-lithography and subsequent SF /C F reactive-ion-etching. These techniques allow usto control the radius r of the air holes with a precision of ± IG. 1: (a) Left panel: schematic cross-sectional representation of the photonic crystal nanocavitystructures investigated. Right panel: layer sequence in the active region. (b) Left panel: SEMimage showing a photonic crystal from top. Right panel: zoom-in to the modified L3 defect wherethe outer holes are shifted by 0.15 lattice constants. step the underlying SiO is removed by hydrofluoric acid. Low mode-volume nanocavities( V Mode ≈ · ( λ/n ) ) were realized by omitting three air-holes in a row and by shifting thelateral holes away from the cavity center by 0.15 lattice constants to form modified L µ PL) spectroscopy setup. The sample was excited by a diode-pumped CW solid-state-laseremitting at λ Laser = 532 nm. We focussed the laser beam using a 50 × microscope objective( N A = 0 .
5) to a spot size of ≈ µ m. The resulting PL signal was collected through thesame objective and dispersed by a 0 .
32 m imaging monochromator equipped with a 600lines/mm grating and a liquid nitrogen-cooled InGaAs linear diode array.3n fig. 2 (a) we present room-temperature µ PL spectra recorded from a series of L r/a is increased from 0.28 (bottom) to 0.34 (top).The excitation power density used in these measurements was 550 kWcm − . Using thesemeasurement conditions the intensity of the emission from the unpatterned region of thedevice was below our detection sensitivity. In strong contrast, each of the spectra recordedfrom the PhC nanocavities clearly reveals five distinct peaks (marked with arrows on fig. 2(a)) that shift to higher energy with increasing r/a ratio [19]. The extracted peak energiesare plotted in fig. 2 (b), clearly demonstrating that all peaks (labeled M1 to M6) shift in asimilar way to higher energies with increasing r/a , as previously reported in refs. [7, 8]. Infig. 2 (c) we compare the measured PL spectrum for r/a = 0 .
28 with the spectral positions ofmode-emission obtained from photonic bandstructure simulations (vertical red lines). Usingthe geometric parameters extracted from SEM images, we obtain six eigenmodes from thesimulation results. The insets in fig. 2 (c) show the calculated electric field distributions thatreveal three modes with longitudinally oriented field distributions (M1,M2,M6) and threemodes with vertically oriented field distributions (M3,M4,M5). In full accordance withIwamoto et al. [7], we observe the strongest peak intensities for modes that have a highvertical extraction efficiency (M1,M3,M6). Simulation and experiment for the longitudinalmodes are generally in very good agreement. A discrepancy is observed between simulationand experiment for M5, which is not observed in the spectra, probably due to a highersensitivity of mode Q and frequency to structural disorder. This tends to lead to a low-Qand, thus, broad emission from M5. The comparison between simulation and experimentclearly shows that we observe PhC mode-emission in the spectral vicinity of the siliconinterband luminescence and that we are able to clearly distinguish the cavity mode emissionfrom the background.In fig. 3 (a) we compare low temperature ( T = 22 K) µ PL spectra recorded close tothe M1 mode emission from three different spatial positions as indicated on the inset. Themeasurement positions were: on the nano-cavity (blue trace), on the PhC away from thecavity (red trace) and on the unpatterned material (green trace). The peak at ≈ .
10 eVarises from the TO-phonon replica of the interband emission from the silicon [17]. In additionto the TO-line, we observe a broad emission band when detecting on the PhC that extendsfrom ∼ . − .
09 eV on the low energy side of the TO-replica . In bulk silicon one wouldnot expect emission in this range at low temperatures, since all phonon replica are sharply4
IG. 2: (a) Room-temperature PL spectra recorded from a series of L3 PhC cavities with differentnormalized air-hole radii ranging from r/a = 0 .
28 (bottom) to r/a = 0 .
34 (top). (b) Detailedanalysis of the five peaks (labeled M1 to M6, marked with arrows in (a)) shifting to higher energywith increasing r/a . (c) Comparison between experimental PL data (black curve) and simulatedspectral mode positions (vertical red lines) of the six L3 modes for r/a = 0 .
28. Insets showcalculated electric field profiles of the eigenmodes. defined [17]. We investigated the intensity of the broad emission band as a function of theexcitation position on the sample. Selected results of these measurements are presented infig. 3 (b) where we plot the integrated PL intensity from 1 .
06 eV to 1 .
08 eV (highlightedorange in (a)) for a spatial scan across the PhC (compare with fig. 3 (a)-inset). Here, weclearly observe a direct correlation between PL intensity of the emission in the 1 . − .
08 eVband and the number of air-holes in the probed region. This strongly indicates that thebroad emission band arises from phonon-mediated recombination from surface defects onthe sidewalls of the air holes created by the RIE etching process. A large number of surface5 a) (b)(c)
FIG. 3: (a) PL at T = 22 K for three different detection positions (indicated by the inset) onthe sample structure. (b) Integrated PL of broad emission band at the low-energy side of the TO-replica (range highlighted orange in (a)) when scanning across the PhC. Inset shows correlation withetched air-hole structure. (c) Red data: integrated mode intensity of M3 as a function of emissionenergy for r/a ranging from 0.25 to 0.20. Black line: reference PL spectrum for r/a = 0 .
25 at T = 24 K. defect states with different trapping energies would lead to the inhomogeneously broadenedemission band on the low energy side of the TO-phonon replica. To investigate the origin ofthe emission in the PhC cavity modes we controllably shifted M3 through the TO-phonon lineand the surface defect band by progressively decreasing the air-hole radius from r/a = 0 . r/a = 0 .
20 and extracted the intensity of the M3 emission. The results are summarizedin fig. 3 (c), where we plot the integrated intensity of M3 as a function of peak position (reddata points). For comparison, we plot the PL spectrum recorded from the r/a = 0 .
25 sample(black line). The mode intensity clearly tracks the spectrum of the TO replica and follows6lso the intensity trend of the surface defect band. These observations indicate that thecavity modes are pumped predominantly via the phonon replica and also more weakly viathe surface defects introduced by the fabrication process. This arises from the fact that theluminescent defects are close to the air-silicon interface, where the electric field amplitudeof the eigenmodes as shown in fig. 2 (c) are weak, leading to a weak coupling between modeand surface defects.We continue to analyze the temperature stability of the PL spectrum recorded from thePhC cavity site. Typical temperature dependent data recorded from the M1 mode emissionfor a PhC with r/a = 0 .
34 from 24 K to room-temperature (295 K) is presented in fig. 4 (a).We observe a clear and rapid reduction in the intensity of the TO-phonon replica, the M1mode emission and the surface defect band, respectively. This is caused by an increasingimportance of non-radiative decay channels, like Auger and free carrier recombination [18].For further quantitative analysis we describe the temperature dependent PL intensity I P L ( T )as a function of the external photon detection probability P , the excitation rate R and theinternal quantum efficiency η int ( T ): I P L ( T ) = P · R · η int ( T ) , (1a)with η int ( T ) = Γ Rad Γ Rad + Γ NR ( T ) (1b)and Γ NR ( T ) = Γ NR · exp (cid:18) − E A k B T (cid:19) , (1c)In these equations Γ Rad is the interband radiative recombination rate and Γ NR ( T ) is thenon-radiative recombination rate. The temperature-dependence of Γ NR ( T ) is described bydistributing excitation amongst the two levels, | (cid:105) and | (cid:105) separated by E A according toBoltzmann statistics. A schematic few level-diagram of the system is presented in fig. 4 (b).State | (cid:105) is pumped at an excitation rate R and decays either via a radiative channel to | (cid:105) ornon-radiatively over | (cid:105) . Eqn. 1a to eqn. 1c are directly obtained from rate equations gov-erning the steady state populations of | (cid:105) and | (cid:105) using the approximation Γ NR (cid:29) Γ ∗ NR ( T ).This condition ensures that the non-radiative channel dominates carrier recombination atelevated temperature as expected for silicon. By reformulating eqn. 1a we obtainln (cid:18) I P L ( T K ) I P L ( T ) − (cid:19) = − E A k B T − ln (cid:18) Γ rad Γ NR (cid:19) , (2)where I P L ( T K ) is the extrapolated PL intensity for T → I P L ( T K ) /I P L ( T ) − IG. 4: (a) PL spectra as a function of temperature for a PhC with r/a = 0 .
34. (b) Schematiclevel diagram illustrating the decay from state | (cid:105) via a radiative channel to state | (cid:105) and via anon-radiative channel over state | (cid:105) . (c) Arrhenius-type analysis for the TO-replica (black squares),defect band emission (blue circles) and M1 mode emission (red triangles). The activation energies E A extracted from linear fits label the three curves .
07 meV to 1 .
08 meV (blue circles)and the M1 cavity mode emission (red triangles) [21]. For all datasets we observe a straightline as predicted by eqn. 2 supporting the vailidity of our analysis. From the slopes of thesecurves we extract an activation energy E A of 8 ± ± ± I P L ( T ) for T <
100 K where sufficiently intense signalfrom mode, defect band and TO-replica are available to make a valid comparison. Fromthe similar values of E A for the cavity mode and TO-replica we conclude that the modeis predominantly excited via the TO-replica, as already deduced from the data shown infig. 3 (c).Comparing mode and TO-replica in fig. 4 (c), we see from eqn. 2 that the lower valuesof ln( I P L ( T K ) /I P L ( T ) −
1) from the mode emission can only be explained via the termΓ
Rad / Γ NR , since both emission peaks exhibit the same activation energy. Hence, the datasuggests that the presence of the cavity mode results in a ratio of Γ Rad / Γ NR which is largerthan for the TO-replica. As shown in refs. [7, 9], an enhanced PL signal can also be causedby a redistribution of the spatial emission profile via the photonic crystal structure. However,this effect does not increase the radiative recombination rate. Therefore, the results indicatethat the enhanced photoluminescence from the cavity mode may be caused by an enhancedinternal quantum efficiency η int ( T ) due to a larger radiative carrier recombination rate causedby the Purcell-effect [22]. This would explain the observed temperature stability of the modeemission up to room-temperature, in strong contrast to the vanishing background emissionfor T ≥
150 K.In conclusion, we presented a temperature dependent investigation of PL in Si PhCnanocavities. We suggest two mechanisms being responsible for the luminescence of cavitymodes, namely phonon-mediated recombination from charge carriers in the electronic bandstates and recombination from charge carriers trapped in surface defect states. Furthermore,we observed an enhanced internal quantum efficiency in spectral resonance with the cavitymode emission.We acknowledge financial support from the German Excellence Initiative via the Nanosys-tems Initiative Munich (NIM), the TUM International Graduate School of Science and En-9ineering (IGSSE) and the TUM Institute for Advanced Study (IAS). [1] L. Pavesi,
Silicon Photonics (Springer, Berlin, 2004).[2] B. Jalali, Nature Photonics , 193 (2007).[3] B. Jalali and S. Fathpour, J. Lightw. Technol. , 4600 (2006).[4] R. Soref, IEEE J. Sel. Topics Quantum El. , 1678 (2006).[5] N. Izhaky, M. T. Morse, S. Koehl, O. Cohen, D. Rubin, A. Barkai, G. Sarid, R. Cohen, andM. J. Paniccia, IEEE J. Sel. Topics Quantum El. , 1688 (2006).[6] L. Kimerling, Appl. Surf. Sci. , 8 (2000).[7] S. Iwamoto, Y. Arakawa, and A. Gomyo, Appl. Phys. Lett. , 211104 (2007).[8] M. Fujita, Y. Tanaka, and S. Noda, IEEE J. Sel. Topics Quantum El. , 1090 (2008).[9] M. Kaniber, A. Kress, A. Laucht, M. Bichler, R. Meyer, M. C. Amann, and J. J. Finley, Appl.Phys. Lett. , 061106 (2007).[10] Y. Akahane, T. Asano, B.-S. Song, and S. Noda, Opt. Express , 1202 (2005).[11] B. S. Song, S. Noda, T. Asano, and Y. Akahane, Nature Materials , 207 (2005).[12] T. Asano, B. S. Song, Y. Akahane, and S. Noda, IEEE J. Sel. Topics Quantum El. , 1123(2006).[13] E. M. Purcell, Phys. Rev. , 681 (1946).[14] M. El Kurdi, X. Checoury, S. David, T. P. Ngo, N. Zerounian, P. Boucaud, O. Kermarrec,Y. Campidelli, and D. Bensahel, Opt. Expr. , 8780 (2008).[15] J. Xia, R. Tominaga, S. Fukamitsu, N. Usami, and Y. Shiraki, Jap. J. Appl. Phys. , 022102(2009).[16] Y. Akahane, T. Asano, B. S. Song, and S. Noda, Nature , 944 (2003).[17] R. Hull, Properties of crystalline Silicon (INSPEC, 1999).[18] W. P. Dumke, Phys. Rev. , 1559 (1962).[19] The fixed peak at 1 .
165 eV (marked with dashed line) arises from the 1064 nm emission ofthe laser.[20] Equal to I P L ( T ) in absence of a non-radiative channel.[21] The intesity data for the TO-replica and defect band was obtained when detecting on the PhCregion but away from the cavity.
22] An unambiguous prove of Purcell-effect requires time-resolved PL measurements.22] An unambiguous prove of Purcell-effect requires time-resolved PL measurements.