TY - JOUR
T1 - Expedited quasi-updated gradient based optimization techniques for energy conversion nano-materials
AU - Hajimirza, Shima
N1 - Publisher Copyright:
Copyright © 2015 American Scientific Publishers All rights reserved.
PY - 2015/4/1
Y1 - 2015/4/1
N2 - Photovoltaic energy conversion has been highly entangled with nano-technology in the past two decades due to the advent of thin film PV technology. Advanced methods of fabrication and material processing have made the design space of PV thin film technology extremely vast. It is now possible to design and fabricate complex nano-scale multi-layer surfaces with textures, gratings, and other types of surface roughness using arbitrary materials (to some extent) and compounds in large scale at a reasonable cost. This freedom of exploration has invoked computational and mathematical techniques such as optimization techniques to help facilitate design problems. However, in comparison to the efforts dedicated to investigating different structure types in nano-scale thin film design, much less has been devoted to developing optimization techniques that are fine-tuned for these classes of complexities. More specifically, inverse design problems dealing with complex partial differential equations such as those of nano-texture designs have two dimensions of complexity: time complexity which is due to expensive forward simulations and, the search space complexity which is due to the complications in the underlying physical system. An efficient optimization technique should be aware of the trade-off between these two, and make use of that optimally. The authors of this paper propose a variation of the first order optimization technique that serves this purpose and expedites inverse optimization of nano-scale designs by a significant factor. The method is based on a modification of quasi-newton search technique that uses a lower dimensional set of sample points at every iteration to estimate an approximation of the gradient vector. A lower dimensional sample size means a significantly lower number of forward simulations and thus a significant saving in the overall design time.
AB - Photovoltaic energy conversion has been highly entangled with nano-technology in the past two decades due to the advent of thin film PV technology. Advanced methods of fabrication and material processing have made the design space of PV thin film technology extremely vast. It is now possible to design and fabricate complex nano-scale multi-layer surfaces with textures, gratings, and other types of surface roughness using arbitrary materials (to some extent) and compounds in large scale at a reasonable cost. This freedom of exploration has invoked computational and mathematical techniques such as optimization techniques to help facilitate design problems. However, in comparison to the efforts dedicated to investigating different structure types in nano-scale thin film design, much less has been devoted to developing optimization techniques that are fine-tuned for these classes of complexities. More specifically, inverse design problems dealing with complex partial differential equations such as those of nano-texture designs have two dimensions of complexity: time complexity which is due to expensive forward simulations and, the search space complexity which is due to the complications in the underlying physical system. An efficient optimization technique should be aware of the trade-off between these two, and make use of that optimally. The authors of this paper propose a variation of the first order optimization technique that serves this purpose and expedites inverse optimization of nano-scale designs by a significant factor. The method is based on a modification of quasi-newton search technique that uses a lower dimensional set of sample points at every iteration to estimate an approximation of the gradient vector. A lower dimensional sample size means a significantly lower number of forward simulations and thus a significant saving in the overall design time.
KW - Nano-scales surface textures
KW - Numerical optimization
KW - Thin film solar cells
UR - http://www.scopus.com/inward/record.url?scp=84930017318&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84930017318&partnerID=8YFLogxK
U2 - 10.1166/jno.2015.1717
DO - 10.1166/jno.2015.1717
M3 - Article
AN - SCOPUS:84930017318
SN - 1555-130X
VL - 10
SP - 140
EP - 146
JO - Journal of Nanoelectronics and Optoelectronics
JF - Journal of Nanoelectronics and Optoelectronics
IS - 1
ER -