Banner
    Evolution Algorithm Inspires Organic Solar Cell Design
    By News Staff | January 27th 2013 09:59 AM | 3 comments | Print | E-mail | Track Comments

    The sun has terrific energy potential but harnessing its electricity with silicon solar cells is too expensive; at times 10 times the price of coal energy it is fine for wealthier people but unrealistic in developing nations, where CO2 emissions are rising the fastest.

    Organic solar cells may be a better solution. These  polymer solar cells use organic materials to absorb light and convert it into electricity but current designs have poor electrical properties. Instead of attempting to increase efficiency by altering the thickness of the solar cell's polymer layer, a tactic that has preciously garnered mixed results,  researchers at Northwestern University sought to design the geometric pattern of the scattering layer to maximize the amount of time light remained trapped within the cell - all by using a mathematical search algorithm based on evolution.

     The algorithm pinpointed a specific geometrical pattern that is optimal for capturing and holding light in thin-cell organic solar cells, resulting design exhibited a three-fold increase over the Yablonovitch Limit, a thermodynamic limit developed in the 1980s that statistically describes how long a photon can be trapped in a semiconductor.
    The resulting pattern will be fabricated with partners at Argonne National Laboratory.
     

    In their newly organic solar cell, light enters a 100-nanometer-thick scattering layer, a geometrically-patterned dielectric layer designed to maximize the amount of light transmitted into the cell. The light is then transmitted to the active layer, where it is converted into electricity.

    "We wanted to determine the geometry for the scattering layer that would give us optimal performance," said Cheng Sun, assistant professor of mechanical engineering at Northwestern and co-author of the paper. "But with so many possibilities, it's difficult to know where to start, so we looked to laws of natural selection to guide us."

    So the researchers employed a genetic algorithm, which they say mimics the process of natural evolution (minus random mutation and genetic drift and natural selection and everything else in actual evolution, it seems) - "Due to the highly nonlinear and irregular behavior of the system, you must use an intelligent approach to find the optimal solution," said  Wei Chen, professor of mechanical engineering. "Our approach is based on the biologically evolutionary process of survival of the fittest."

    They consider it evolution because they began with dozens of random design elements, then "mated" and analyzed their offspring to determine their particular light-trapping performance. This process was carried out over more than 20 generations and also accounted for evolutionary principles of crossover and genetic mutation.


    Published in Scientific Reports.


    Comments

    Gerhard Adam
    They consider it evolution because they began with dozens of random design elements...
    Why wouldn't it be considered evolution? 
    So the researchers employed a genetic algorithm, which they say mimics the process of natural evolution (minus random mutation and genetic drift and natural selection and everything else in actual evolution, it seems)...
    What would random mutation, genetic drift, etc. have to do with whether or not this is an evolutionary process?  It seems like a lot of biologically related terms are being used, as if there is some sort of preferential "space" in which these terms are applicable.

    Anyone that considers evolution as being only biological is clearly missing the point.  Without non-biological evolution, there is no biology.

    Mundus vult decipi
    MikeCrow
    Genetic algorithm's are pretty cool, and the ones I'm familiar with have mutations and selection.

    But humans can't debug the resulting code, so you either get code that works, you keep mutating and selecting fitness of your results, or you pick an earlier results, and restart from there.
    Never is a long time.
    There are actually a lot of different flavors of evolutionary algorithms. I'm using some of them to do near real-time optimization of autonomic computing systems.

    The challenge with these sorts of algorithms is that they have a lot of parameters/functions that matter. For example with genetic algorithms you have to decide population size, mutation probability, crossover probability, fitness selection vs linear selection, etc., type of genomic string, use gray code or not. All of these things have the potential to have a big impact on how quickly you converge and what kind of optimum you converge onto.

    For my work I'm working on automating meta-optimization of these search parameters.

    FWIW, genetic algorithms can genetic drift. And it's not always a good thing.