Transition metal gettering studies and simulation for the optimization of silicon photovoltaic device processing

Abstract

We use what is known about transition metal (TM) defect thermodynamic driving forces and kinetic responses to make predictive simulation of gettering during solar cell fabrication possible. We have developed a simulator to explore the impact of various device and process parameters on gettering effectiveness. The relevant heat treatments are ramps up in temperature, isothermal annealing, and cools from high temperature down to room temperature. We explore a range of surface conditions, density and size of heterogeneous nucleation sites in the bulk, and the degree of contamination in order to create a framework in which to examine these mechanisms acting in concert. Such simulations enable process optimization for gettering. For solar cell processing, segregation to an Al back contact layer is routine. We have estimated the segregation coefficient between a p-type Si wafer and a molten Al layer by the Calphad method and use these results to estimate the thermodynamic driving force for redistribution of Fe into the Al layer. We simulate gettering treatments of supersaturated levels of Fe contamination in Si samples with FeSi2 and Al contacts and compare these results with data at various temperatures. The gettering data for FeSi2 contacts follow a simple exponential decay and can be simulated with appropriate choice of internal gettering time constant. We recognize that radiative heating dominates the temperature ramp for samples in evacuated quartz ampoules and use reasonable parameters to include this effect in our simulations. Fitting parameters for [Fe] data taken from heat treatments at 755ʻC on samples with FeSi2 and Al contacts successfully predict the gettering data of Al coated samples treated at 810ʻC.

(cont.) Discrepencies in the data for Al coated samples treated at 6950C and data for Al coated samples treated at 755ʻC after long times have exposed a new mechanism dominating internal gettering processes. We propose the existence of a silicide precipitate growth retardation mechanism as a result of supersaturation of the Si vacancy (V). Accumulation of V reduces the ability of precipitates to relax strain free-energy ([Delta]g strain) by further V emission. We performed Cu gettering experiments on p/p+ epitaxial wafers. Photoluminescence measurements revealed significant Cu removal from the epitaxial region compared to similarly doped uniformly doped float zone (FZ) Si wafers. Step etching revealed haze, indicating the presence of silicide precipitates below the epitaxial layer in the heavily doped substrates. Uncontaminated heat treated epitaxial wafers did not demonstrate the presence of haze after step etching. This finding demonstrates that redistribution of Cu from the lightly doped expitaxial layer to the heavily doped substrate as predicted by the dopant enhanced solubility model has occurred. The commonly used T3/2 model for effective density of conduction and valence band states (Nc and Nv respectively) is not accurate for Si, even in the device operation regime, and the available experimentally determined relations of Green do not extend past 500K. We have constructed a DOS model using ab initio calculations and temperature appropriate Fermi-Dirac ...