Bulk defect characterization in metalized solar cells using temperature-dependent Suns-Voc measurements

Deep Learning Extraction of the Temperature-Dependent Parameters of Bulk Defects

“…Each lifetime curve has unique noise, but the noise generation parameters are kept constant across all the curves for consistency. SRH lifetime curves are presented as lifetime as a function of Δn for one (IDLS) 8,49 or multiple (TIDLS) 7,13,15 temperatures. In this paper, we present an alternative method to present the lifetime: using temperature and injection lifetime maps (TILMs).…”

“…The n 1 and p 1 terms govern how the defect energy level E t impacts the SRH lifetime. As discussed, often k is used as an alternate defect parameter: k : = σ normaln σ normalp = ν normalp · τ normalp 0 ν normaln · τ normaln 0 which is usually reported alongside E t in the literature. ,,− …”

“…When measuring the SRH lifetime curves through TIDLS, measurement noise impacts the extraction of the defect parameters, usually by impacting the DPSS resolution, leading to uncertainties in predicting the defect parameters. As discussed, in the DPSS approach, noise often results in uncertainty as to whether the defect energy level is in the upper bandgap half ( E t > 0 eV) or lower bandgap half ( E t < 0 eV). ,,, To mimic the noise seen in measurements, a Gaussian noise that scales inversely proportional to Δ n (low Δ n have higher noise) can be added to the lifetime: τ normalS normalR normalH , noise ( Δ n ) = τ normalS normalR normalH ( Δ n ) · ( 1 + N false( 0 , scriptϵ n 2 false) · ln true( normalΔ 0 normalΔ n true) ) where the noise is drawn from a normal distribution scriptN (0,ϵ n 2 ) for each Δ n point (ϵ n is the noise scale parameter) before being scaled by the logarithmic factor, and Δ 0 is the excess minority carrier concentration with zero noise. At Δ n = Δ 0 , the logarithmic factor becomes 0, and the SRH lifetime has no added noise.…”

“…SRH lifetime curves are presented as lifetime as a function of Δ n for one (IDLS) , or multiple (TIDLS) ,, temperatures. In this paper, we present an alternative method to present the lifetime: using temperature and injection lifetime maps (TILMs).…”

“…The true defect parameter lies at the intersection of all these curves. Due to the presence of noise in TIDLS measurements, at least two potential solutions are commonly identified (usually one in the upper half bandgap and one in the lower half bandgap). ,, The DPSS method, referred to as the “traditional” approach, has been refined over the years: linearization of the SRH equation under specific conditions has been proposed to facilitate the fitting procedure; , faster convergence techniques, such as the Newton–Raphson method, have been adapted; and defect parameter contour mapping was developed to visualize the potential solutions . Recently, a machine learning (ML) based framework was introduced to analyze TIDLS measurements. , Multiple ML algorithms were trained to successfully predict the defect parameters (regression) and the half-bandgap location (classification).…”