Effect of oxygen and carbon segregation on the electrical properties of grain boundaries in silicon

Effect of oxygen and carbon segregation on the electrical properties of grain boundaries in silicon

MateriaLs Sciem'e am/Engineering, B4 (1989) 353-358 353 Effect of Oxygen and Carbon Segregation on the Electrical Properties of Grain Boundaries in ...

444KB Sizes 0 Downloads 8 Views

MateriaLs Sciem'e am/Engineering, B4 (1989) 353-358


Effect of Oxygen and Carbon Segregation on the Electrical Properties of Grain Boundaries in Silicon S. PIZZINI. F. BOP, SANI and M. ACCIARRI

l)ipartimento di ('himica Fisica ed Klettrochimica al2d Gruppo Nazionah" di ,~llllllltr(I della Materia, Via (;olgi 19, 1-20133 Milan (Italy) ~Received May 30, 19893


The trans-barrier d.c. conductivity and the 1-V characteristics of several grain boundaries (GBs) of polycrvsmlline silicon samples were dewrmined in order to find possible correlations between the configuration of the impuri O' ehmd at GBs and their eh,ctrical activio'. It has been found that, in good agreement with the predictions, the segregation fi'atures of oxygen and carbon at GBs determine uniquely the height 01' the potential barriers set up in their correspomlence.

1. Introduction

It has already been shown [1-31 that the variety of recombination features found at grain boundaries (GBs) of polycrystalline silicon depend closely on the local configuration of the impurity cloud. A preliminary explanation of these effects was attempted [1, 2] by introducing a qualitative model based on local-stress-induced defects, where the stress-strain relationships are related uniquely to the presence at GBs of impurities exhibiting a size misfit with respect to the atoms or to the interstices of the host matrix. This picture needs to be suitably improved, although this is not an easy task. It is in fact well known [4-6j, for example, that the electrical activity of dislocations in singlecrystal silicon depends on the thermal history of the sample and on the actual temperature of the measurements, which are related to the dopant and non-dopant impurity segregation as well as to the shift with temperature of the acceptor vs. donor characteristics of the defect states. In the case of polycrystalline materials the ability of dopant and non-dopant impurities to segregate at GBs is still more evident. Boron 0921-5107/89/$3.50

segregation has in fact been clearly monitored in close correspondence with GBs using secondary ion mass spectrometry (SIMS)[1] in p-type silicon, while the sole segregation of donors was recognized to occur at GBs in chemical-vapourdeposited n-type microcrystalline silicon [7]. Carbon and oxygen are also found to segregate at GBs in p-type silicon, the width of the impurity cloud being of the order of many micrometers [ 1, 81. These latter impurities are often also unheavily segregated along GBs. These inhomogeneities should depend closely on the set-up of rather undefined but certainly not negligible thermal gradients within the large polycrystalline ingots during the growth cycle. Eventually, redistribution of impurities may occur during further thermal annealing cycles, being possibly influenced, however, by the presence of residual local strain fields and by the presence of dislocations [8]. It is therefore not always possible to distinguish between "clean" and "dirty" GBs and, indeed, between the intrinsic and extrinsic activity of GBs. Moreover, the fact that often the impurity cloud at GBs is many micrometres wide makes it difficult to imagine that simple models like those available in the literature may be entirely appropriate for the description of these extended defects [9], as will become clear when looking at the features of these models. We first consider the double-Schottky-junction model [10-16], which assumes that GBs are associated with deep trap states on which free carriers are trapped, A space charge region is thus created in correspondence to either side of the GB, where the excess GB charge is neutralized by the fixed charge of ionized dopants in the depletion region. According to this model, if quantum mechanical tunnelling is excluded by considering that the © Elsevier Sequoia/Printed in The Netherlands

354 width d of the space charge region is of the order of several hundred nanometres [12], the transit regime is of the over-the-barrier-hopping type and the zero-bias d.c. conductivity is described by an equation of the form [ 16] o = qTA * exp


i 1)

where A* is the Richardson constant, q)~) is the height of the potential barrier and ~ is the position of the Fermi level in the neutral region with respect to the top of the valence band (in the case of p-type materials). As emphasized in ref. 16, the conditions for electrons to cross the barrier in free flight may not be fulfilled in real circumstances since the mean free path is often shorter than the barrier width. In 1 Q cm p-type silicon at 300 K it is in fact close to 0.01 /~m while the barrier width is close to 0.1 ~m for barrier heights of 0.1-0.2 eV. In these conditions Labusch and Hess [16] suggest the use of an alternative equation s=Obq*(,exp( 2dkT ~

q~,t kT]


heat treatments, which are expected to enhance the electrical activity by impurity segregation. On this boundary a non-negligible density of states :LDOS) within the gap was found [111, with a broad maximum of 2× 1I~~' ~:V ~ cm : at 0.59 eV above the valence band. These results were obtained by a deconvolution of the 1 - c u r v e s obtained by d.c. biasing the boundary via a couple of ohmic electrodes and determining first the barrier height from the equation J = A * 7 : expi - / 3 ( ~ , + ~/q:~il - e x p - ( / 3 V )

where J is the density of the current flowing through the barrier, @~ is the voltage-dependent barrier height under non-zero-bias conditions in the negative grain, fl = q / k T and V is the applied voltage, and then the electron concentration trapped at the barrier, nv, from the equation n.r= (2se0ND)I/2~t t2 + (~1 #-

)j -~


and the position of the quasi-Fermi level (QFL) E 0 using the equation E~, * =

I~'(~ - q q ) l - ~ - A


where where S is the conductance per unit area, o~ is the bulk conductivity and d is the barrier width. It should be noted that for highly doped samples and at low temperatures, phonon-assisted tunnelling cannot instead be excluded [13]. Until recently, no firm information about the intrinsic or extrinsic nature of the GB states responsible for the setting up of a potential barrier at GBs has been available. It is generally assumed, however, that in silicon the GB states are only acceptor in character [15], considering the pure acceptor properties of dangling bonds. As suggested by Stuetzler and Queisser [8], intrinsic trap states should result instead from a variety of point and line defects in addition to dangling bonds, since apparently a contribution from paramagnetic species is still present in electron spin resonance spectra after hydrogen passivation. This last suggestion supports also the existence of potential barriers at GBs in p-type polycrystalline silicon, on which, otherwise, no specific hindering of carrier transport should be observed, contrary to experience. Recent experiments [11, 15] on near-Z25 GBs in n-type float zone silicon bicrystals proved that this GB is electrically active without any previous

A = - k T In

(l+exp(-fiV,) ¢

Since n l = f N~:Ji'j~.L,/.7i dE l: t

and assuming that the DOS changes only slightly at the position of the QFL, that • r < V and that 1~ is constant, the derivative dnT - Nt,.


is just the DOS at the position of the QFL. Incidentally, as already remarked in ref. 14, the 1-V characteristics of a double Schottky barrier show an ohmic range at low bias (eV4. k T ), a barrier breakdown region and an ohmic postbreakdown region at large enough applied voltages, where the current-limiting transport is the finite conductivity of the grains. The second model, the dopant segregation model, suggests that the charge transport is con-


trolled by the segregation of dopant atoms at GBs, where they become electrically inactive. Although the thermodynamics of this model is well supported by theoretical considerations E7], the physics of the process leading to the electrical deactivation of the dopants at GBs has been not considered in sufficiently close detail. Furthermore, since we have evidence from SIMS experiments [1] and from local IR experiments [17] that boron segregated at GBs does increase the concentration of free carriers, the proof exists that, contrary to the model expectations, dopants segregated at GBs may be electrically active. The aim of this present study was therefore that of analysing the influence of the segregation of non-dopant and dopant impurities on the electrical properties of model GBs in order to implement, both experimentally and theoretically, our knowledge of correlations between the segregation of impurities at GBs and their electrical activity. 2. Experimental details Trans-barrier d.c. conductivity measurements were carried out in the range 70-300 K on p-type (boron content around 101~, cm 3) polycrystalline samples of known oxygen and carbon concentration. Before the conductivity measurements these samples were heat treated at 850 °C for about 30 min. The features of oxygen and carbon segregation at GBs as well as the local recombination characteristics in these samples have already been determined [1, 2]. The samples selected for measurements were as follows. One carbon-rich sample on which GBs were decorated with carbon and which showed substantial GB recombination [1]. As demonstrated in ref. 1, in carbon-rich samples the carbon segregation is associated with boron segregation. The carbon content of this sample was 7.3 ppma and that of oxygen 1.4 ppma. One oxygen-rich sample (oxygen content 13.7 ppma, carbon content 7 ppma) which presented GBs decorated with oxygen and which showed reduced recombination features. One sample which was equiconcentrated in oxygen and carbon (oxygen content 7.7 ppma, carbon content 7.1 ppma) and which presented, mostly, slightly recombining GBs. On these GBs both oxygen and carbon were observed to segre-

gate, albeit spatially separated [2] one from the other. GBs which appeared decorated only with carbon showed instead strong recombination features. A p-type single-crystal sample of comparable resistivity to the reference one. Before the d.c. conductivity measurements the samples were mildly etched in an anisotropic {CP4 etch, rinsed with doubly distilled water and contacted using an In-Ga alloy (with a macroscopic contact area of less than 1 ram:), which is known to behave as an ohmic contact on p-type silicon. On the same samples the dark 1-V characteristics were determined at 298, 209 and l14K. 3. Results

3.1. Ternperamre dependence of the conductivity The results of the d.c. conductivity measurements are displayed in Fig. 1 as an ln(R ~7') vs. reciprocal absolute temperature plot according to eqn. (2) where R is the experimentally measured resistance. It appears that only the carbon-rich sample (GBs AC3 and AC7) exhibits a temperature-activated behaviour (E~,~,t=0.098 meV for the GB AC3 and 0.197 meV for the GB AC7) while the other polycrystalline samples display a behaviour which is very close to that of the singlecrystal silicon. It turns out that under the conditions of the experiments a double-Schottky-barrier-limited carrier transfer occurs only on GBs on which carbon segregates and which appear strongly 2.6 r - - T -







1.0 F[E =





reciprocal temperature (103~K-~)

Fig. 1. Temperature dependence of the electrical conductivity of different silicon samples: ca) high carbon sample, GB AC7; (b) high carbon sample, GB AC3; (c) high oxygen sample, GB TTO; (d) equiconcentrated sample, GB SG; (e) equiconcentrated sample, GB NR: (f) single-crystal silicon.


recombining [1]. Apparently, and in good agreement with previous experiments which showed a wide range of recombination activities of different GBs belonging to the same sample, the activation energy for the carrier transport process also varies from one GB to another. It should also be remarked that a saturation of the conductivity occurs in the limit of high temperatures and that a transition to a second transport regime occurs at temperatures close to 140 K.

3.2. I- V characteristics Figure 2 displays the I- V characteristics of one GB of the carbon-rich sample (the AC3 one) and Fig. 3 those of the equiconcentrated sample (the SG one). The most striking difference between the behaviour of the two samples is the occurrence, at temperatures lower than atmospheric, of ohmic, sub-ohmic and barrier breakdown ranges in the carbon-rich sample, as theoretically predicted, while only an ohmic range is observed in the other sample. 10-2 - -




~ A~




-=, 10-5 10-6

10-7 ,









t0 -1

applied voltage (V) Fig. 2. I - V characteristics of the G B A C 3 (high c a r b o n sample): o, T = 2 9 8 K: A, T = 209 K;t3, 7 = 114 K.






~o c ~ l Q e ~ l p -




= 10-6


10-8 10-5

A more careful inspection ot Fig. 3 shows, however, that on the equiconcentrated sample (on which, incidentally, no barrier-limited t~ansport was observed from the d.c. conductivity measurements) barrier breakdown definitely occurs at very low voltages, It appears therefore that the d.c. conductivity was measured in the post-breakdown range on these last samples, the absence of a thermally activated regime being only an experimental artefact. 4. Discussion

It appears first from our experiments that all the polycrystalline samples examined exhibit a barrier breakdown regime, which calls for a barrier to be set up at GBs. However, while on carbon-rich samples barrier breakdown occurs at relatively high values of the applied voltage, barrier breakdown occurs at very low applied voltages in the other samples. The height of GB barriers in carbon-rich samples is therefore larger than in oxygen-rich samples and calls for a larger density of charge trapped at GBs. As a second conclusion, a transition from a high 7 conductivity regime to a low 7 regime occurs in carbon-rich samples, which calls for the setting up of tunnelling conditions [13] or for localized hopping conditions at low temperatures. A deeper insight to these results comes from the analysis of the polarization curves, performed according to the procedure reported in Section 1, after the preliminary evaluation of the crosssection of the conductivity channel, which needs to be known in order to use eqn. (3). This analysis has been applied only Io measurements performed at 209 K, since at room temperature the purely ohmic behaviour indicates an overall conductivity dominated by the impedance of the conductivity channel within the bulk of the grains while at 114 K tunnelling or localized hopping conditions could prevail. The cross-section A of the conductivity chain nel may be evaluated from the pre-exponential term of eqn. (2), using its experimental value obtained by interpolation of the straight portions of the curves of Fig. 1, which is just


I 10-4

i 10-3

I t0 -2


applied voltage (V) Fig. 3. l - V characteristics of the G B S G (high oxygen sample): o, T = 298 K; A, T = 209 K; tz, T = 114 K.

A°BqOP°-- 128 Q- 1 2dk, where ~ comes as well from the d.c. conductivity


experiments and equals 0.098 eV, and d=

2 e e ~ - - NA q

From the k n o w n values of o 8 = 1.04 f2-~ c m and NA = 1.5 X 10 ~(' c m -3 (which is the acceptor

concentration) we obtain d = 9.25 x 10-s cm and A = 1.99 x 10-5 cm 2, which closely agree with the experimental values reported in the literature [18] for a GB of a commercial polycrystalline silicon sample contacted in a similar way. The dependence of the barrier height ~ on the voltage may then be calculated from eqn. (3) using for the Richardson constant A* a value of 46.32 A c m 2 K-2 and for the position of the Fermi level ~ a value of 0.091 eV at 209 K. The results of this calculation are reported in Fig. 4, which shows a good agreement between the experimental values of ~0 obtained from the d.c. conductivity experiments (0.098 meV) and


• ~

• AA~

• •'LA~



those independently obtained from the analysis of the polarization curves, considering that the zerobias conductivities measured by the d.c. conductance experiments were actually obtained with a d.c. bias of 0.1 V in our experimental conditions. These curves are then converted to nT(Eo*) curves, from which, after derivation, the DOS is obtained, as reported in Fig. 5. From these results it is therefore possible to draw the following conclusions. (1) In contrast with our previous considerations, the double-Schottky-barrier model is appropriate, at least down to about 150 K, for the description of GBs in both carbon-rich and oxygen-rich samples. (2) In carbon-rich samples GBs are characterized by a distribution of traps lying at about 0.15 eV above the valence band, which are responsible both for the barrier-limited current and for the strong recombination activity of GBs. (3) The behaviour of GBs in oxygen-rich and in compensated samples deserves further investigation, although it is known that also on these samples barrier-limited conditions occur but that the charge trapped at these GBs is less than that trapped at GBs of carbon-rich samples. (4) The nature of these traps is unknown but is certainly correlated with the segregation of oxygen and carbon at GBs.



Acknowledgments 0.03 1(55


e 10-3


i I 102 1(51 Applied voltage IV)


Fig. 4. Dependence of the barrier height on the applied voltage for the GB AC3 (high carbon sample); T = 209 K.

2.8 I


This work was entirely supported by the Progetto Finalizzato Energetica of the National Research Council. We acknowledge further the cooperation of all the members of the laboratory, with a particular mention to Drs. S. Farina, N. Butta and E Allegretti.





08 i 0.3 Q84









0.96 eo(eV)

Fig. 5. DOS determined from the I-V characteristics as a function of E,~ at T= 2(19 K for the GB AC3.

1 S. Pizzini, F. Borsani, A. Sandrinelli, D. Narducci, M. Anderle, R. Canteri, in H. J. Moeller, H. R Strunk and J. H. Werner (eds.), Polycrv'smlline Semiconductors, Springer-Verlag, Berlin, 1989, p. 115. 2 S. Pizzini. F. Borsani, A. Sandrinelli, D. Narducci, E Allegretti, in G. Benedek, A. Cavallini and W. Schroeter (eds.), Point, Extended and Surface DeJ~,cts in Semiconductors, Plenum Press, New York, 1989, in the press. 3 W. Seifert and M. Kittler, Phys. Status. Solidi (a), 99 (1987) K11. 4 H.J. Queisser, J. Phys. Soc. Jpn., 18(1963)142. 5 I.E. Bonderenko, E. B. Yakimov and N. A. Yarykin, Acta


Phys. Polon. A, 69(1986) 393. 6 H. Ono and K. Sumino, J. Appl. Phys., 54 (1983) 4426. 7 M. M. Mandurah, K. C. Saraswhat, C. R. Helms and T. 1. Kamins, J. Appl. Phys., 5l (1980) 5755. 8 F.J. Stuetzler and H. J. Queisser, J. Appl. Phys., 60 (1986) 3910. 9 S. Pizzini, Mater. Chem. Phys., 23 (1989) 349. 111 G. Baccarani, B. Ricco and G. Spadini, J. Appl. Phy,v, 49 11978) 5565. 11 C.H. Seager and T. G. Casmer, ,/. Appl. Phys., 40(1078) 3879. 12 G. E. Pike and C. H. Seager, J. Appl. Phys., 50 !1979) 3414.

13 C. H. Seager and G. E. Pike, ,,:tpp/. Phys. Lett., 40 '~1982 } 471. 14 G. Blatter and F. Greuler. l'h3.~. Rev. B, 3.¢(1986) 3952. 15 G. Peterman, Phys. Status Solidi ia~, 106 (1988) 535. 16 R. Labusch, J. Hess, in G. Benedek, A. Cavallini and W. Schroeter (eds.), Point, k.'xlettded and ~'urJ~ce Defects in .Semiconductors, Plenum Press. New York, 1989. in the press. 17 A. Borghesi, M. Geddo, G. i.}uizzetti, S Pizzini, I). Narducci, A. Sandrinelli and G. Zachman, Solid .State (ommun., 69(1989) 457, 18 M. Boehm, H. C. Scheek, W. Seilcrt and H. G. Wageman, personal communication, 1988