Erschienen in: Solid State Phenomena ; 242 (2016). - S /

Transcription

1 Ershienen in: Solid State Phenomena ; 242 (216). - S Konstanzer Online-Publikations-System (KOPS) URL:

2 81

3 82 down the regeneration proess after a firing step, a 16 nm thik layer does. This an be explained by the redued H diffusion from the SiNx:H through the thiker Ah 3 layer. It was also observed, that a deposited PECVD SiNx:H without following firing step to release large ammmts ofh into the -Si bulk does not lead to a signifiant regeneration effet [14]. For time resolved measurements during regeneration on lifetime samples, effetive lifetime 'teff (inluding surfae reombination as well) is normally measured diretly after an annealing step without any BO-related defets in the degraded state (Fig. 1) and then several times during regeneration. Assuming Shokley-Read-Hall reombination to be the dominant proess in the bulk, the measured 'teff an be transformed into a n1malized defet onentration N* (t) via 1 1 N (t)= r eff (t) Teff,annealed (1) beause lifetime of a ertain SRH reombination ha1111el sales inversely to the defet density. A fit of the deaying N* (t) during regeneration yields a harateristi regeneration time onstant to or a regeneration rate 1/t{) via (2) Iii'.8 Iii'.._ :J.. :J...12 l...._... SiNx:H only ;s.1 AI23 (8 nm) I SiNx:H? z AI23 (16 nm) I SiNx:H z.8 all fired.4 :;:: o (,) (,) - -. (,) (,). Q) Q) Regeneration time [min] 6 SiNx:H not fired SiNx:H fired Regeneration time [min] Fig. 2: Effet of ALD Ah 3 layers of varying thikness as H diffusion bani.er during firing on regeneration kinetis (left). Effet of firing of a PECVD SiNx:H layer on regeneration kinetis (right). For both investigations n1malized BO-related defet onentration after anneal (first data point) and dmi.ng subsequent regeneration at 13 C ar1d.6 snns illumination is given. (after [14]) In addition, another method for inorporation of H in the -Si bulk was applied. Samples treated in a mirowave-indued remote H plasma at temperatures <25 C for inorporation of H led also to a signifiant regeneration, even without an additional firing step whih in this ase is not needed to diffuse H into the -Si bulk [15]. Impat of Hydrogen Conentration. If H is a prerequisite for regeneration to our, then the question arises whether regeneration kinetis an be sped up with inreasing [H]. To hek this, again a seri.es of experiments was onduted by varying the PECVD SiNx:H firing step in a onventional fast firing belt furnae [ 16, 17]. Here a higher peak fui.ng temperature leads to more H being released from the SiNx:H layer into the -Si bulk and a faster regeneration. In addition the belt speed was var ied, resulting in different temperature ramps dmi.ng ool-down. For higher peak firing temperature, a fast ool-down speeds up the regeneration proess, while for low peak firi.ng

4 83 temperatures <7 C a fast ool-down does not aelerate regeneration (Fig. 3). This an be explained by a quenhing effet of H: if temperatme is high, H an effuse out of -Si during ooldown faster than additional H an enter the -Si bulk from the SiNx:H soure. This suggests that [H] in the -Si bulk is another imp1iant parameter aprui from temperature and M to desribe and understand regeneration kinetis _ E...12.$.8 :;::;.4 a.> > a.> a::: fast Peak sample temperature [ C].16 E.12 a.> - i.8!.4 / / Peak Temp. [ C] /- 3 T fast I- 3 _..,_ 75 +/- 2 - T- 7 +/ /- 15 medium a3 slow. /.. > t. '--, ' Average ooling rate 7 C - ssoo [Kis] Fig. 3: Dependene of regeneration kinetis on peak firing temperature and belt speed (left). For peak firing temperatures <7 C, higher ooling rates do not lead to faster regeneration anymore. (after [17]) Desription of Defet Kinetis via Reation Rates A vety instmtive model for explaining the observed kinetis during regeneration was published in 21 based on the oulting transitions between the different states of the BO-related defet (Fig. 1) [18]. A set of lineru differential equations desribes the oupation of a respetive staten; and the reation towards or out of this speifi state via reation rates k;j with iij. The respetive reation onstants an be influened by pru ameters like temperature, M, or [H]. An example for the hange in oupation of the degraded state B ns an therefore be given as (3) With this model, the behavior of reombination ativity an be simulated. Under the assumptions that only reation onstants kab (degradation) and ks (regeneration) have values > and kab =1 ks, the resulting oupation of states in the ourse of time is shown in Fig. 4 (left). Struiing after an annealing step (with all defets being in state A), the degraded state B is populated. One there are defets available in state B, the regeneration reation sets in and state C gets populated. As both states A and C show negligible reombination ativity, the lifetime measurement will follow the line representing the sum of defets being in state A and state C. h1 Fig. 4 (right) a lifetime measurement in the ourse of time during regeneration is shown (1 slm illumination at vmying temperatures). The fit aording to the model desribed above leads to exellent mathing and therefore proves that the 3-state model an desribe the situation vety well.

5 84 ;R :::... :+:; ro 4 a. :J 8 \ ' p I... I....,...,. "',... --na(t) /.... ;' I n {t) ;' 8 I / I *' I '. / ; I '-...- : I I I 1. > / n(t) na + n Time [a.u.] C/) 2; 9 l-' Time [min] 14 C.. 12o v 1o Fig. 4: Oupation of different states of the BO-related defet (Fig. 1) aording to the model desribed in the text (left). The measmed minority anier lifetime values during regeneration of an annealed sample (1 sm1 at temperatmes given in the graph) an be fitted vety well aording to the model assuming that both states A and C have negligible reombination ativity (right). (after [18]) Effet of fast and slow Regeneration on long term Equilibrium of Defet States. All fom reations shown in Fig. 1 are thetmally ativated and of Anhenius type, desribing the probability of onversion p onv from one state to another (4) with harateristi trial frequeny Vhar, ativation energy Ea, Boltzmann onstant kb and temperatme T. While anneal and destabilization reation rate depend only on T, the degradation reation rate is influened also by Lill (generated, e.g., by varying illumination intensity) up to illumination levels satmating for values >.1 SllllS [6]. This mem1s that for fixed values oft and illumination levels >.1 Sllll, degradation, anneal and destabilization reation rates show fixed values, whereas regeneration reation rate depends also strongly on speifi values of Lill (adjusted, e.g., by illumination level) and [H]. Table 1 gives an overview of values for Ea and reation rates for T = 2 C. Table 1: Ativation energies and reation rates at 2 C and 1 Sllll illumination for reations shown in Fig. 1. (values from [19-22]) Reation path Ea revl Time onstants rminl Annealing Degradation.4.3 Regeneration 1. vatying Destabilization As all fom reations shown in Fig. 1 ru e ative in parallel, the ratio of the time onstants determines the long te1m equilibrium values of oupany for the three defet states. For a possible industrial appliation, a fast and omplete regeneration proess is desired. As regeneration is sped up by higher T, it is interesting to investigate whether T an be inreased independently of the speifi sample with still all defets ending up in the regenerated state in long te1m equilibrium, or whether there is a peak T that should not be exeeded. At higher T the destabilization reation strut s to ompete with regeneration, therefore the ratio between the time onstants of regeneration and

6 85 destabilization beomes impmtant. If this ratio is 1/1, almost all defets (99%) an reah the regenerated state in long tenn equilibrium (Fig. 5, left). If this ratio is only 1/2, then only 6% of the defets end up in the regenerated state, and also the annealed state stays populated (Fig. 5, right). As the only differene between the two senru ios shown in Fig. 5 is the time onstant of regeneration (.4 min vs. 33 min, other time onstants fi:om Tab. 1), it beomes lear that for a omplete regeneration at high T (2 C or above) the srunple has to have a shmt regeneration time onstant. 1. -,--.. -<1l annealed -C/) t> 2 <1> u.5 regenerated C/) 2 1., !! C/) t> <1>.5 annealed <1l Q. ::J (.).+--= E-4 1 E Time [min] annealed Fig. 5: Oupany of BO-related defets with time under illmnination (1 sun) at 2 C, struting after annealing. For fast regenerating samples (assumed regeneration time onstant:.4 min) all defets end up in the regenerated C in long term equilibrium (left), while a regeneration time onstant of 33 min leads to inomplete regeneration with a fration of defets ending up in the (unstable) annealed state A (right). (after [19]) High Speed Regeneration. The regeneration time onstant at a given T an be influened by Llli and [H]. This means that aprut from adjusting illmnination to speed up regeneration, the main parruneter detennining whether regeneration is omplete at given T and Llli is the amount of H in the -Si bulk present in the fmm needed for the regeneration reation to om [14]. The ombination of hosen Llli and distribution of H therefore detetmines the peak T allowed for reahing omplete regeneration. Fig. 6 visualizes what this means in pratie by plotting the ompleteness of regeneration in long tenn equilibrium in dependeny of regeneration time onstants with proess temperatme as parameter. For regeneration temperatme of 2 C it an be onluded that a regeneration time onstant <1 min is needed to end up with almost all defets in the regenerated state C, whereas a time onstant of 5 min will lead only to 5% of the defets in the regenerated state. At 16 C, a time onstant of 5 min will lead to almost omplete regeneration, and a time onstant of 5 min at 13 C to omplete regeneration.

7 86 (]) m 1 C/) " 2 8 me... '- (]).Q 6 (])+-' O>ro (]) '-..._ '-.3 ro 4 C/) >- '- (.) (]) 2 CQ: m ro a. ::3 (.) \ \ , - 13 C \ C 2 C \ C - 28 C \ 1 min \ \. \. \ ' \ 5 min. ' _', (.) 1" 2 1" Regeneration time onstant [s) Fig. 6: Dependeny of ompleteness of regeneration in long term equilibrimn on regeneration time onstants with T as parameter. (after [19]) For fast regenerating samples, regenerating temperature an be inreased (with also illumination intensity being inreased), so that regeneration time onstants in the range of 1 san be reahed [14, 23, 24). Generation of Llli is also possible via laser illumination leading to regeneration time onstants <1 s [26]. Model of Regeneration A general model explaining the observed dependeny of regeneration kinetis on [H] has been developed [14, 15]. At temperatures <1 C, His normally bound to impurities in -Si (e.g. dopants or other impurities/defets) or to H forming moleular H. Assuming that at least pa1t of the H is bmmd in a onfiguration so that dissoiation is enhaned tmder artier injetion, e.g., in BH-pairs with a dissoiation energy being redued to 1.1 ev under injetion onditions [14], this might explain the observed neessity for artier injetion for regeneration to our and the ativation energy for the regeneration reation of 1. ev (Table 1). Slow ool-down rates after peak firing temperature an result in less overall [H] in the -Si bulk and to H being botmd to other defets than B [14, 25], showing a more stable onfiguration with higher dissoiation energy. This applies also to temperatm e steps ar otmd 1-4 C being anied out after ool-down as shown in [14, 25]. Therefore, the temperature history after peak firing (introdution of H into the -Si bulk) determines the regeneration rate at given regeneration temperature and &1. Depending on Llli, the harge state of H an be hanged in p-type -Si material. E.g., upon illmnination part of the normally positively harged W an be present in the neutral state H. The importane of the H har ge state for mobility of H in -Si and the possible bom1ding to and therefore passivation of defets like, e.g., the BO-related defet was highlighted reently by several authors [24, 26, 27). ill ombination with the model presented in the paragraph above, this might explain the observed need for hydrogen, temperature ar1d Llli for regeneration to our. Within the model also the observed dependeny of regeneration rate on the amount of dopants, with higher dopant density leading to lower regeneration rates [28], an be explained at least qualitatively. The more dopants present in the -Si bulk, the more H is trapped at these defet sites and the slower is the trap limited diffusion of H towards the BO-related defet. This behavior was also observed in ompensated -Si, where for ompensation of the present P more B (and possibly also Ga) is present, signifiantly inreasing the overall p-type dopant onentration and slowing down regeneration rate despite of the same net doping onentration [19, 28-3). A longer regeneration time onstant then allows only lower regeneration temperatures at onstant Llli for omplete regeneration (see above) or leads to inomplete regeneration as observed experimentally [19).

8 87

9 88

10 89