Photoluminescence and electron paramagnetic resonance studies of defect centers in porous silicon

(1)

Photoluminescence and electron paramagnetic resonance

studies of defect centers in porous silicon

Hang-Ting Lue

a

, Bang-Yuh Huang

b

, Juh-Tzeng Lue

b,∗

a_{Department of Electronic Engineering, National Chao Tung University, Hsin Chu, Taiwan} b_{Department of Physics, National Tsing Hua University, Hsin Chu, Taiwan}

Received 25 June 1999; received in revised form 26 November 1999; accepted 30 November 1999

Abstract

Photoluminescence (PL) and electron spin resonance (ESR) spectrometers were exploited to study the Pbdefect centers in porous silicon

(PS) under various heat treatments. The breaking of Si–H_xand Si–O–H bonds in PS by thermal annealing changes the emission wavelength and increases the recombination centers resulting in degrading the PL intensity. Four kinds of dangling bonds on interfaces of silicon (1 1 1), (1,−1,−1), (−1,1,−1), and (−1,−1,1) faces and SiO2with C3Vsymmetry were identified. The skeleton structure of PS collapsed under

thermal annealing. Annealing in hydrogen gas, which is controvertible to the pervasive anticipation, cannot passivate the dangling bonds introduced at high temperatures. © 2000 Elsevier Science S.A. All rights reserved.

Keywords: Photoluminescence; Electron paramagnetic resonance; Porous silicon

1. Introduction

Silicon material, although, has well-developed technol-ogy and widely utilized in electronic devices, the intrinsic indirect-gap behavior limits its use in photonic devices. The discovery of strong photoluminescence [1–3] of porous sili-con (PS) was prospected to have pragmatic uses in light emit-ting devices for the integration of modulaemit-ting light sources, photo-detectors and amplifiers from the implementation of a simple silicon material. In our previous works, we have demonstrated [4] that quantum size effect is not crucial to generate photo-luminescence (PL) in PS. Comparing the PS with siloxane (Si6O3H6) like materials [5], we neither

ob-served [6], the infrared (IR) absorption peak of Si–O–H vi-bration bond nor its Raman peak at 514 cm−1for the freshly prepared PS samples. Only samples stored in air for several days, the PL of the above lines show up with the intensity of other lines being much larger than that of the freshly prepared sample adducing that the IR absorption peaks of Si–H_x, or Si–O–H bonds are found on the oxidized sam-ple. The PL intensity also diminishes with the decrease of Si–H_x bond density due to vacuum annealing informing that the breaking of Si–Hx bonds implies the formation of

non-radiate recombination centers to degrade the PL

inten-∗_{Corresponding author.}

sity. Accordingly, the defects in PS play an essential role of the PL.

In spite of the existence of several reports [7–10] on electron spin resonance (ESR) studies of PS, many itinerant features of PS defects are not revealed, and heat treatment effects on the ESR are not fully studied. In this work, we attempt to study the point defects in PS by electron spin resonance spectrometry. This technique allows us to address the site of a particular defect, the structure symmetry, and the spin density change due to annealing effect.

2. Mechanisms of photoluminescence

The origins of photoluminescence in general arise from the radiative decay of atoms or molecules from the excited energy levels such as in organic dyes and polymers. In direct energy gap semiconductors, the recombination of electrons and holes from the conduction and valence bands, respectively, evoke the luminescence. For silicon, an in-direct energy gap material, the PL is extremely small and can be observed only at very low temperatures. Discrete energy levels split from the continuous conduction band [11,12] showing a direct transition arise due to the quantum confinement of nano-particles. The strong PL as observed even at room temperature of PS suggesting an indirect to direct band gap transition. Unfortunately further

(2)

tions found that almost identical PL and Raman spectra of artificially synthesized silioxane and PS, and the lost of PL for pure silicon nano-particles as small as 4 nm alluding that quantum size effect is not crucial to the PL in PS [4]. In this work, we attempt to study the interplay between the changes of chemical adsorption molecules and defects in PS affecting the PL spectra by thermal annealing, and expect to yield a real consensus on this fact.

3. Defect centers in porous silicon

Native defects generated during crystal growth such as va-cancies, antisites and interstitial control the photonic proper-ties for intrinsic semiconductors. Among the possible intrin-sic defects, the isolated silicon dangling bonds demonstrate themselves to be the dominant interface defects, which cru-cially control the photo-luminescence efficiency.

The point defects in PS, can be a Pbo-like (i.e. Si≡Si),

a Pb1 (i.e. Si≡SiO2), and a Pb-like center. The ESR of Pbo

yields a broad and small signal-to-noise ratio signal, while the Pb1 is essentially capricious with thin oxides [8]. The

most intricate Pb defects are associated with silicon

dan-gling bonds and can be classified to three types. They are positive charged D+, the negative charged D−, and the neu-tral D◦. The D◦has unpaired electrons with a net spin mo-ment and can be readily detected by an ESR spectrometer. The rapture of one of the tetrahedral sp3valence electrons leaves one unoccupied dangling bond. The annealing of PS in hydrogen gas can passivate the dangling bonds by hop-ping process [13]. On the other hand, heating the sample in vacuum or in air, the hydrogen atoms may escape from the weak Si–H bonds and restore to the dangling states. Porous silicon, which is modified from crystalline silicon, has a lower symmetry of C3v than its original diamond Td

struc-ture. The porous structure implies a large surface area to be oxidized when stored in air, which invokes dangling bonds at the interface between silicon and SiO2.

The commonly specified Pb centers [3,14,15] are the

point defects concealed at the interface of Si(1 1 1)/SiO2as

shown in Fig. 2. There are four paramagnetic centers belong-ing to the Pbdefects named as S1=(1 1 1), S2=(1,−1,−1),

S3=(−1,1,−1) and S4=(−1,−1,1) corresponding to the

dangling bonds at the interface of the specified Si surfaces and oxides.

4. Theggg-tensors in ESR studies

For a defect center with a free spin ES, the spin Hamiltonian in a magnetic field EH is [16–18]

R = β ESg · EH, (1)

whereβ is the electron Bohr magneton, Eg is a dyadic tensor. Taking a Cartesian coordinates (xyz), the magnetic field H has components of

E

H = H (`xˆx + `yˆy + `zˆz), (2)

where,`_x,`_y and`_z are the cosines along the ˆx, ˆy, and ˆz axes, respectively. The spin Hamiltonian Eq. (1) then can be written as R = β · H (`x,`y,`z)  g_gxx_yx g_gxy_xy g_gxz_yz gzx gzy gzz    s_sx_y sz   , (3)

where (s_x, s_y, s_z) are the components of the spin along the Cartesian coordinates. We can define an effective geff such

as g2 eff= (`x, `y, `z)(g ∼_{· g}∼₎  `_`x_y `z   = `x, `y, `z    g2 xx g2 xy g2 xz g2 yx g2 yy g2 yz g2 zx g2 zy g2 zz     `_`x_y `z   (4)

where ¯¯g∼is the transpose conjugateg.

In this case, the Zeeman splitting energy 1E for a spin 1/2 transition can be readily written as1E=βgeffH with

(1E)2_{= β}2_g2

effH2= β2( EH · g

∼_{) · (g· E}_{H )}

= β2_{H · g}E 2_{· E}_H ₍₅₎

The elements (g2)_ij can be determined from the ESR rotat-ing spectra as addressed below. If the y-axis of the sample surface (the xy plane) is rotated with respect to the magnetic field ˆH with an intersection angle θ, then

g2

eff = (g2)xxsin2θ + 2(g2)xysinθ cos θ + (g2)yycosθ (6)

We can readily determine (g2)_yy atθ=0◦, (g2)_xx atθ=90◦, and (g2)_xy at θ=45◦, respectively. In the same way, with different rotating plane, we can evaluate the six tensor ele-ments (g2)ij=(g2)_jiby which the g2is diagonalized to yield the principle values such as

g2₌  g 2 x 0 0 0 g2_y 0 0 0 g2_z   (7)

For crystals have axial symmetry such as hexagonal, tetrag-onal, and trigtetrag-onal, then

g =  g0⊥ 0g_⊥ 00 0 0 g_k   (8)

whereg_kand g_⊥are the g-values obtained for the magnetic field H to be parallel or perpendicular to the symmetry axis. In this case, the experimental resonant position occurs at

H = hv

gβ, with g = (g

2_cos2_{θ + g}2

⊥sin2θ)1/2 (9)

whereθ is the angle between the dangling bond ES_ii and the magnetic field EH . In this experiment, we rotate the PS

(3)

surfaces with respect to the c-axis by several angles δ to detect the spectra.

The transform matrix R that rotates the spin ES_i by an angleδ to yield

θ = cos−1H · R · ESE i (10)

can be derived as follows. Firstly, we decompose the spin vector ES_i into components of ES_i` and ES_it to be parallel and perpendicular to the c-axis, respectively. From ES_it, we then take a third vector ES_it0 to be perpendicular both to ES_it and ˆc, such as

ES0

it = ˆc × ESit = ˆc × ( ESi − ( ESi· ˆc)) and ESit0 =ESit . (11)

On rotating ES_i with respect to the c-axis by an angleδ, the

ESi`retains its original value while ES_it moves to a new value

ESit(δ) = ESitcosδ + ES0_itsinδ. (12)

The new spin direction ES_i(δ) becomes

ESi(δ) = ESi`+ ESit(δ) = (ESi· ˆc)ˆc + (ESi− (ESi· ˆc)ˆc)cos δ +(ESi− (ESi· ˆc))sin δ = R · ESi (13) The transform matrix is explicitly written as

R(δ) =

 c

2

x(1 − cos δ) + cos δ cxcy(1 − cos δ) − czsinδ c_xc_z(1 − cos δ) + c_ysinδ cxcy(1 − cos δ) + czsinδ c_y2(1 − cos δ) + cos δ c_yc_z(1 − cos δ) − c_xsinδ cxcz(1 − cos δ) − cysinδ c_yc_z(1 − cos δ) + c_xsinδ c_z2(1 − cos δ) + cos δ



 ₍₁₄₎

where c_x, c_y, c_zare the direction cosines of ˆc-axis. After a tedious manipulation of Eqs. (10)–(13), we obtain the effec-tive g values in terms of the rotating angleδ for the defect centers S1, S2, S3, S4. With the magnetic field EH to be along h1, 0, 0i, we can derive

geff{S1} = 1 √ 6 h 3(g2_k+ g_⊥2) + (−g2_k+ g2_⊥)cos (2δ) −2√2 g2 k− g2⊥ sin(2δ) i1/2 geff{S2} = 1 √ 6 h 3 g2 k+ g⊥2 +−g2 k+ g⊥2 cos(2δ) +2√2 g2 k− g2⊥ sin(2δ) i1/2 geff{S3} = 1 √ 6 h g2 k+ 5g⊥2 + (g2k− g2⊥)cos (2δ) i1/2 = geff{S4} (15)

The measured g-values at various rotation angles can be deconvoluted to yield the principle g-values.

5. Electron spin resonance measurement

The p-type (1 0 0) and (1 1 1) silicon wafers with

resis-tivity of 0.01–0.02 cm were vacuum evaporated with

aluminum on the backside and rapidly thermally annealed (RTA) at 500◦C for 10 min to form an ohmic contact. The anodization method followed the same procedures as previ-ously reported [6]. The heavily doped PS film can be easily detached from its original Si substrate by sharply increasing the current to 500 ma for 30 s at the final anodization pro-cess. The fragile thin PS was then tailored in a proper size of 3×10×0.1 mm3and loaded in a TE104 dual microwave

cavity. A Brucker EMX-10 ESR spectrometer operating at x-band was exploited for this detection. The data were assessed at every 5◦ rotating angles from 0 to 180◦. The experimental data for rotating the crystal face (0,−1,1) with respect to the magnetic field ˆH at various angles are plotted in Fig. 1.

The experimental data are simulated to yield the true resonance positions for the four Pb point defects by

least-mean-square curve fitting. Rotating the crystal axis

h0,−1,1i, the C3v symmetry conveys that the defect

cen-ters at S3=(−1,1,−1) and S4=(−1,−1,1) do have the same

angular dependence on the ESR spectra.

Three resonance centers S1, S2 and S3=S4 have ESR

intensity ratio of 1:1:2. Assuming a homogenous broad-ening, the ESR line-shape can be accessed by taking the

composition of three Lorentzian lines at various field x such as Y (x) = − 3 X i=t 8I_i(−λ_i+ x) 0i1+ (4(−λi+ x)2_{)/ 0}_i2 (16)

whereλ_i are the resonance magnetic fields and0_i are the full-width at half-maximum (FWHM) of each spectrum.

The least-mean-square fitting [19] is exploited to simulate the experimental data and extract the resonance fieldλ_i. The smooth curves adjoined with the experimental data points as plotted in Fig. 1 yield the resonant g-values at each rotation angles for the four Pb centers. Fig. 2 portraits the variation

of g-value with rotation angles δ with respect to the four Pbdefect centers. The four Pbdefects in ideal should have

the same principle g-value, but on account of the slight deviation of bond angles for dangling bonds on different faces, they imply three g-values as depicted in Table 1. The average g-values are given by g_k=2.0016±0.0003, g_⊥=2.0089±0.0003 for Pb centers along the h1 1 1i axis.

The angular dependence of g-values of the Pbcenters turns

out that the virgin PS belongs to the C3vsymmetry.

6. Photoluminescence measurement

The PL spectra of PS crucially depend on the surface absorbed molecules such as Si–H_x, Si–O–H, and Si–O–Si

(4)

Fig. 1. Typical ESR spectra at various anglesδ for the rotating axis along h0,−1,1i and the magnetic field ˆH along h1 0 0i at angles of 30◦ _and 90◦.

Fig. 2. The g-values at various rotating anglesδ for the theoretical calcu-lation by exploiting Eq. (11) (the smooth curves), and the experimental data fitting for the䊉, S1, O, S2,∇, S3=S4 of Pb centers.

Table 1

The principle g-values for different Pbdefect centers

Pb defects gk g⊥

S1 2.0015±0.0003 2.0089±0.0003

S2 2.0016±0.0003 2.0088±0.0003

S3=S4 2.0016±0.0003 2.0087±0.0003

while the immersing environment determines the surface chemical bonds and the crystalline size determining the in-teraction field strength between radicals which yield the shift of the PL peak. In order to immunize from its direct contact with air, we embedded the PS powder detached from the Si wafers into the silica glass by sol–gel method [20,21] and then analyzed the changes of PL with heat treatment.

The complex solution composed of 20 ml of tetra-ethyl-ortho-silicate (Si(OC2H5)4) with 20 ml of ethanol, 16 ml of

de-ionized water and the PS powder were mixed uniformly by a magnetic stirrer for 4 h. The mixed solution was poured into a plastic disc with a cover cape and waited to gel for 2 days. The gel was dried in air at room temperature for 1 week. The PL was measured at various temperatures by mounting the sample on a heating head. For temperatures below 120◦C, the spectra as shown in Fig. 3 saliently in-creases from room temperature to 62◦C and then gradually decreases until 162◦C. Above which the PL peak shifts rig-orously from its original 700 to 530 nm, and the intensity increases with temperatures again. An adducing from the vanishing of some of the Fourier transform infra-red (FTIR) spectral lines after annealing, we may tacitly assume that the peak near 700 nm is attributed from the Si–H_x, and 600 nm from the Si–O–H attached on the PS walls. At temperatures below 162◦C, the transformation of SiH_xand Si–O–H is re-versible. For temperatures further increasing above 270◦C, the PL peak shifts to 530 nm and the intensity decreases with

Fig. 3. The PL of PS powder embedded in silica glass measured at various temperatures. The spectra measured between 24 and 162◦C are reversible.

(5)

Fig. 4. The PL of PS powder embedded in silica glass measured from 24 to 440◦C.

temperatures again as shown in Fig. 4. This evidence might be due to the decomposition of Si–H_x and Si–OH bonds to form Si–O–Si bonds and yielding the 530 nm line. This de-composition is irreversible. As temperature increases, the re-combination centers so called dangling bonds are formed by the decomposition of the adsorbed chemical radicals result-ing in the decrease of PL intensity. For temperatures above 440 C, the PL vanishes.

7. Discussion

In this work, the ESR spectra are exploited to study the change of the density of defect centers by thermal annealing. The ESR line-shapes almost unchanged at low temperatures, while after 500◦C thermal annealing in vacuum, the signals survive in three resonance lines with intensity ratio chang-ing to nearly 1:1:1 and cannot be simulated by the C3v

sym-metry. Possibly due to the collapse of skeleton structure of PS, a newly created defect center which is almost isotropic with the rotation angleδ can be detected. We have plotted the ESR spectra for the virgin and the annealed sample on the same chart to compare their shifts in resonance position as shown in Fig. 5. The spin density of PS can be readily calculated by comparing the spectral area with that of the referenced DPPH signal, which was simultaneously detected with a dual cavity. The thermal annealing also increases the dangling bonds by breaking the covalent bonds. The higher the temperature, the more the defect centers were created as shown in the isochronal annealing curves of Fig. 6. The Fourier transform infrared (FTIR) spectra as shown in Fig. 7 also reveals the presence of dangling bonds arising from the Si–OH, Si–O–R and Si–O2bonds after annealing in air.

The oxygen defects are smeared out after 400◦C annealing in H2but without convincing evidence for vacuum annealed

Fig. 5. The ESR spectra for virgin (---), and 500◦C annealed for 10 min (. . . ) PS at rotating angles of (a) 30◦and (b) 90◦, respectively.

Fig. 6. The change of spin densities for isochronal annealing of PS at various temperatures

(6)

Fig. 7. The FTIR spectra at different annealing conditions.

samples. The degradation of the PL intensity almost follows the increase of defect centers as measured by the ESR spec-trometer.

In conclusion, we have studied the point defects of PS arising from the dangling bonds by ESR. The four Pbdefects

are identified to be S1· · · S4 with C3v symmetry. Thermal

annealing at 200–300◦C increases the spin density vastly in-forming that most of the weak bonds of PS have a strength of

∼49 meV which will be broken into dangling bonds above

this temperature. Annealing in H2gas can passivate the

dan-gling bonds but not for bond broken at high temperatures. The dangling bonds play an essential role of recombination centers and degrade the photoluminescence intensity.

This work was supported by the National Science Council of the Republic of China under contract NSC 88-2112-M007-019.

References

[1] A. Uhlir, Bell System, Technical J. 35 (1956) 333. [2] D.R. Turner, J. Electrochem. Soc. 105 (1958) 402. [3] L.T. Canham, Appl. Phys. Lett. 57 (1990) 1046.

[4] S.K. Ma, J.T. Lue, Thin Solid Films 304 (1997) 353.

[5] H.D. Fuch, M. Statzman, M.S. Brandt, Phys. Rev. B. 48 (1993) 2172. [6] G.S. Chang, J.T. Lue, Thin Solid Films 259 (1994) 275.

[7] H.J. von Bardeleben, D. Stievenard, A. Grosman, C. Ortega, J. Siejka, Phys. Rev. B 47 (1993) 10899.

[8] Y. Uchida, N. Koshida, H. Kayama, Y. Yamamoto, Appl. Phys. Lett. 63 (1993) 961.

[9] F.C. Rong, J.F. Harvey, E.H. Poindexter, G.J. Gerardi, Appl. Phys. Lett. 63 (1993) 920.

[10] V.Ya. Bratus, S.S. Ishchenko, S.M. Okulov, I.P. Vorona, H.J. von Bardeleben, Schoisswohl, Phys. Rev. B 50 (1994) 15449.

[11] W.C. Huang, J.T. Lue, Phys. Rev. B 49 (1994) 17297. [12] W.C. Huang, J.T. Lue, J. Phys. Chem. Solids 58 (1997) 1529. [13] P.J. Caplan, E.H. Poindexter, B.E. Deal, R.R. Razonk, J. Appl. Phys.

52 (1981) 879.

[14] A. Stesmans, Appl. Phys. Lett. 48 (1986) 972. [15] A. Stesmans, Phys. Rev. B 45 (1992) 9051.

[16] E.H. Poindester, P.J. Caplan, B.E. Deal, R.R. Razonk, J. Appl. Phys. 52 (1981) 879.

[17] J.T. Lue, Nuovo cimento IL, 31B (1976) 372

[18] J.E. Wertz, J.R. Bolton, Electron Spin Resonance, McGraw-Hill, New York, 1986.

[19] C.F. Gerald, Applied Numerical Analysis, Addison-Wesley, 1983, p. 465.

[20] J.T. Lue, W.C. Huang, S.K. Ma, Phys. Rev. B 51 (1995) 14570. [21] S.K. Ma, J.T. Lue, Solid State Commun. 97 (1996) 979.