Computing Sciences masthead Berkeley Lab Computing Sciences Berkeley Lab logo
Share/Bookmark

Exascale for Energy


Renewable Electricity: Photovoltaic Solar Energy Conversion

Electrical generation by solar energy capture with photovoltaic systems has virtually no environmental impact beyond device manufacturing, is ideal for individual home or other distributed generation, and taps a virtually unlimited resource. Currently, however, it is two to four times more expensive than most residential or commercial rate electricity. So, although photovoltaic has a number of excellent small markets, it currently supplies only a small fraction of total electricity use and requires further cost reductions and efficiency improvements to be able to make a major contribution to meeting electrical demand.

More than 30 years of experimentation was needed for the relatively simple thin-film silicon solar cell to reach its current efficiency of 24%. In order to develop next-generation solar cells based on new materials and nanoscience fast enough to reduce the global warming crisis, a different paradigm of research is essential. Exa­scale computing can change the way the research is done —both through a direct numerical material-by-design search and by enabling a better understanding of the fundamental processes in nanosystems that are critical for solar energy applications.

Alloy simulation
Figure 23. The electron wave functions for an oxygen-induced state (left) and ZnTe conduction band edge state (right) in a ZnTeO alloy with 3% O. The grey, blue, and red dots correspond to Zn, Te, and O atoms respectively.
Source: L.-W. Wang,  Berkeley Lab

Unlike bulk systems, nanostructures cannot be represented by just a few atoms in computational simulations. They are coordinated systems, and any attempt to understand the materials’ properties must simulate the system as a whole. Density functional theory (DFT) allows physicists to simulate the electronic properties of materials, but DFT calculations are time-consuming; and any system with more than 1,000 atoms quickly overwhelms computing resources, because the computational cost of the conventional DFT method scales as the third power of the size of the system. Thus, when the size of a nanostructure increases 10 times, computing power must increase 1,000 times.

Photovoltaic nanosystems often contain tens of thousands of atoms. So one of the keys to unleashing the energy harvesting power of nanotechnology is to find a way of retaining DFT’s accuracy while performing calculations with tens of thousands of atoms.

Researchers at Lawrence Berkeley National Laboratory have demonstrated a way to accomplish this using a divide-and-conquer algorithm implemented in the new Linear Scaling Three-Dimensional Fragment (LS3DF) method. In November 2008, this research was honored with the Association for Computing Machinery (ACM) Gordon Bell Prize for Algorithm Innovation.

In a solar cell, there are a few key steps that determine overall efficiency in the conversion of sunlight to electricity: light absorption, exciton generation, exciton dissociation into separated electron and hole, carrier transport, and charge transfer across nanocontacts. A few aspects of nano solar cells often limit their overall efficiency: weak absorption of light, electron–hole recombination, nanocontact barriers, or large overpotentials. Unfortunately, many of these processes are not well understood. This is one instance where computational simulations can play a critical role.

For example, materials that have separate electron states within the energy band gap, such as zinc tellurite oxide (ZnTeO), have been proposed as next-generation solar cells. Such systems could theoretically increase solar cell efficiencies from 30% to 63%. To test this hypothesis, the Berkeley Lab researchers used LS3DF to calculate the electron wave function of a 13,824-atom ZnTeO supercell on 17,280 cores of NERSC’s Cray XT4 system, Franklin (Figure 23). The LS3DF calculation took just a few hours, compared with the four to six weeks it would have taken using a direct DFT method. The results showed that ZnTeO is a good candidate for photovoltaic applications, with a theoretical power efficiency estimated to be around 60%.

In subsequent tests, LS3DF ran on more than 100,000 supercomputer cores, making it the first variationally accurate linearly scaling ab initio electronic structure code that has been efficiently parallelized to such a large number of processors.

To go beyond nanoscale simulations, algorithmic breakthroughs like LS3DF, combined with multiscale methods and exascale computers, could be used for the integrated simulation and design of entire photovoltaic systems, shortening the cycle for device development and optimization, and improving the efficiency and cost of photovoltaics.

Improved computational methods, such as those described in the scaling sidebar, will play a crucial role in this effort.

 


<< Previous page