TY - JOUR
T1 - Extremely efficient design of organic thin film solar cells via learning-based optimization
AU - Kaya, Mine
AU - Hajimirza, Shima
N1 - Publisher Copyright:
© 2017 by the authors. Licensee MDPI, Basel, Switzerland.
PY - 2017/12
Y1 - 2017/12
N2 - Design of efficient thin film photovoltaic (PV) cells require optical power absorption to be computed inside a nano-scale structure of photovoltaics, dielectric and plasmonic materials. Calculating power absorption requires Maxwell's electromagnetic equations which are solved using numerical methods, such as finite difference time domain (FDTD). The computational cost of thin film PV cell design and optimization is therefore cumbersome, due to successive FDTD simulations. This cost can be reduced using a surrogate-based optimization procedure. In this study, we deploy neural networks (NNs) to model optical absorption in organic PV structures. We use the corresponding surrogate-based optimization procedure to maximize light trapping inside thin film organic cells infused with metallic particles. Metallic particles are known to induce plasmonic effects at the metal-semiconductor interface, thus increasing absorption. However, a rigorous design procedure is required to achieve the best performance within known design guidelines. As a result of using NNs to model thin film solar absorption, the required time to complete optimization is decreased by more than five times. The obtained NN model is found to be very reliable. The optimization procedure results in absorption enhancement greater than 200%. Furthermore, we demonstrate that once a reliable surrogate model such as the developed NN is available, it can be used for alternative analyses on the proposed design, such as uncertainty analysis (e.g., fabrication error).
AB - Design of efficient thin film photovoltaic (PV) cells require optical power absorption to be computed inside a nano-scale structure of photovoltaics, dielectric and plasmonic materials. Calculating power absorption requires Maxwell's electromagnetic equations which are solved using numerical methods, such as finite difference time domain (FDTD). The computational cost of thin film PV cell design and optimization is therefore cumbersome, due to successive FDTD simulations. This cost can be reduced using a surrogate-based optimization procedure. In this study, we deploy neural networks (NNs) to model optical absorption in organic PV structures. We use the corresponding surrogate-based optimization procedure to maximize light trapping inside thin film organic cells infused with metallic particles. Metallic particles are known to induce plasmonic effects at the metal-semiconductor interface, thus increasing absorption. However, a rigorous design procedure is required to achieve the best performance within known design guidelines. As a result of using NNs to model thin film solar absorption, the required time to complete optimization is decreased by more than five times. The obtained NN model is found to be very reliable. The optimization procedure results in absorption enhancement greater than 200%. Furthermore, we demonstrate that once a reliable surrogate model such as the developed NN is available, it can be used for alternative analyses on the proposed design, such as uncertainty analysis (e.g., fabrication error).
KW - Neural networks
KW - Organic photovoltaics
KW - Plasmonics
KW - Surrogate-based analysis and optimization
KW - Uncertainty analysis
UR - http://www.scopus.com/inward/record.url?scp=85044527412&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85044527412&partnerID=8YFLogxK
U2 - 10.3390/en10121981
DO - 10.3390/en10121981
M3 - Article
AN - SCOPUS:85044527412
VL - 10
JO - Energies
JF - Energies
IS - 12
M1 - 1981
ER -