Application of Bayesian optimization for high-performance TiO /SiO /c-Si passivating contact

Quantitative evaluation of implied open-circuit voltage after metal electrode deposition on TiO x /Si heterostructures by photoluminescence imaging: impact of metallization on passivation performance

“…In our previous work, we have reported that the best passivation performance was observed for the 1 nm thick TiO x /c-Si heterostructure after the HPT. 37) One possible cause of this degradation is due to the UV light soaking generated by HPT, 22) or the insufficient penetration of hydrogen radicals into the TiO x /Si interface when the TiO x film is thick. These results show that HPT under unsuitable conditions could cause considerable damage to the cell performance.…”

“…After the TiO x deposition, hydrogen plasma treatment (HPT) was performed to improve the passivation performance. 37) There are many process parameters for HPT, including process temperature (T HPT ), process time (t HPT ), H 2 pressure (p H2 ), H 2 flow rate (R H2 ), RF power (P RF ), and electrode distance (d). In this study, these parameters were set at T HPT = 373 K, t HPT = 90 s, p H2 = 100 Pa, R H2 = 70 sccm, P RF = 390 W, d = 10 mm based on our previous work.…”

“…In this study, these parameters were set at T HPT = 373 K, t HPT = 90 s, p H2 = 100 Pa, R H2 = 70 sccm, P RF = 390 W, d = 10 mm based on our previous work. 37) After TiO x deposition, Mg and Al layers were subsequently deposited on the TiO x as electrodes by thermal evaporation. The Mg electrode was employed to reduce the contact resistance.…”

High Passivation Performance of Cat-CVD i-a-Si:H Derived from Bayesian Optimization with Practical Constraints

“…BO is an effective method for determining the best value of a black-box objective function with multidimensional parameters. , The objective function is the object to be maximized (or minimized) during optimization, which varies depending on the value of the parameters, and in this study, it is τ eff . For the machine learning method in the BO, we used Gaussian process regression (GPR), which can estimate the mean and variance of the posterior distribution from the prior distribution over the objective function.…”

“…Recently, Bayesian optimization (BO), a machine learning method for sequential optimization, has been widely used to find optimum experimental conditions. − The BO algorithm provides the next experimental condition based on the probabilistic model for the regression of the data. By alternately repeating the experiments and determining the next experimental condition using the BO algorithm, the appropriate condition for maximizing (or minimizing) the object function can be determined with a small number of experiments.…”