Atomistic simulation of light-induced changes in hydrogenated amorphous silicon

We employ ab initio molecular dynamics to simulate the response of hydrogenated amorphous silicon (a-Si:H) to light exposure (the Staebler– Wronski effect). We obtain improved microscopic understanding of photovoltaic operation, compute the motion of H atoms, and modes of lightinduced degradation of photovoltaics. We clarify existing models of lightinduced change in a-Si:H and show that the ‘hydrogen collision model’ of Branz (1998 Solid State Commun. 105/106 387) is correct in many essentials. (Some figures in this article are in colour only in the electronic version) Solar photovoltaic (PV) devices are an increasingly important source of electrical power. Current PV production is entering the gigawatt regime, so a factor of the order 103 is needed in order to impact the (terawatt scale) energy markets [1]. One material used for PV energy production is hydrogenated amorphous silicon (a-Si:H), which is particularly inexpensive and serves an important niche in energy markets. An impediment to the use of a-Si:H cells is the socalled Staebler–Wronski effect (SWE) [2] (the light-induced creation of carrier traps causing reduced energy conversion efficiency). A salient feature of a-Si:H is that exposure to intense light (as in PV applications) leads to structural change (rearrangements of the positions of the atoms in the amorphous network). These changes have a serious impact on the performance of a-Si:H cells (the PV efficiency drops 15%–20%, and then stabilizes). Such light-induced structural changes are complex and information about the changes is provided through an array of experiments [4, 5]. Key experimental facts are: (1) light-soaking induces large changes in photoconductivity, and defect (carrier trap) formation; (2) light induces H motion [3, 6]; (3) light-soaking preferentially creates protons separated by 2.3 Å in device grade material and a shorter distance in low-quality material [7]; (4) Isoya and co-workers [8] showed that no dangling bond (DB) pairs are formed after light-soaking and that the placement of the light-induced dangling bonds was random [9]. Further, it was shown that metastable dangling bonds were separated by at least 10 Å; (5) studies of defect creation and annealing kinetics in a-Si/Ge:H suggest that there is not a large population of mobile H leading to recapture events of H onto DBs as part of the photo-degradation process [10]. 0953-8984/06/010001+06$30.00 © 2006 IOP Publishing Ltd Printed in the UK L1 L2 Letter to the Editor The most widely accepted model of photo-degradation is the ‘hydrogen collision model’ of Branz [11]. Here, recombination-induced emission of H from Si–H bonds creates both mobile H and vestigial DBs. When two such mobile H atoms join in a metastable Si–H complex, two new DBs become stable. Our work lends support to this view. Associated with Branz’s work is the two-phase model of Zafar and Schiff, which successfully explained thermal metastability data [12, 13], exploited the concept of paired H, and was later merged with the model of Branz and invoked dihydride bonding [14], which emerges naturally in our simulations. A seminal paper emphasized local ‘weak bond’ models [15], which have been ruled out by the nonlocality of rearrangements inferred from electron spin resonance [8]. Recently we combined experimental information with the results of accurate simulation in high-quality a-Si:H to show that a likely consequence of light-soaking is the formation of Si sites bonded to two network Si atoms and two H atoms (we label this class of structures SiH2) [16]. A predictive simulation of these effects must include several ingredients: (1) models of a-Si:H representative of the topology of the network, sufficiently large to faithfully represent short-time (about several picoseconds) dynamics; (2) accurate interatomic interactions (the H energetics are highly delicate in a-Si:H [17, 18]); (3) the electron–lattice interaction must be estimated in a reasonable fashion [4, 19]; (4) the motion of H must be included, as a variety of experiments point to the role of H motion in the SWE. We accommodate all of these requirements in our work, which sets this study apart from other simulations of these effects. Generating a representative plausible model is the first step in incorporating the ingredients stated above. Our starting point in getting this model is a 64-atom defect-free (four-coordinated) a-Si model [20]. To create the hydrogenated amorphous silicon environment, we removed three silicon atoms and added ten more hydrogen atoms to terminate all but two dangling bonds. This procedure generates a 71-atom hydrogenated amorphous silicon model. We relaxed this starting configuration using conjugate gradient for coordinate optimization. We then repeated this supercell surgery at other sites in order to generate an ensemble of models (three configurations). Topological or chemical irregularities in an amorphous network lead to localized electron states in the gap or band tails [21]. If such a system is exposed to band gap light, it becomes possible for the light to induce transitions of electrons from the occupied states to low-lying unoccupied (conduction) states. For the present calculation we do not concern ourselves with the subtleties as to how the EM field induces such transitions; we will simply assume that a photo-induced promotion occurs, by depleting the occupied states of one electron ‘forming a hole’ and moving the electron near the bottom of the unoccupied ‘conduction’ states. Changes in force due to the light-induced transition of carriers will initially be local to the region in which the orbitals are localized, followed by transport of the thermal energy through the network. In general, it is necessary to investigate photo-structural changes owing to a collection of different initial and final states, though we may expect that only well localized states (necessarily near the gap) have the potential to induce structural change [22–25]. To begin unravelling the SWE, we track the dynamics of a-Si:H in the presence of electron– hole pairs. We have performed extensive molecular dynamics simulations of network dynamics of a-Si:H both in the electronic ground state and also in a photo-excited state (in the presence of a simulated electron–hole pair) using our model. The density functional calculations were performed within the generalized gradient approximation (GGA) using the first principle code SIESTA [26] at a constant temperature (300 K). We have used a fully self-consistent Kohn–Sham functional in the calculation of energies and forces within the parameterization of Perdew et al [27]. Norm-conserving Troullier–Martins [28] pseudopotentials factorized in the Kleinman–Bylander [29] form were used. We employed an optimized double-polarized basis set (DZP), where two s and three p orbitals for the H valence electron and two s, six p and five Letter to the Editor L3