Hello, A yearly allocation for the LCRC cluster has been requested with the following updated information: Submitter/PI: Fatih Sen Project Name: grbdft Division: NST Project title: Atomistic modeling of grain boundaries in CdTe Associated funding: DOE-EERE Other Systems: NERSC (100,000 core hours) Science: The efficiency of poly-CdTe solar cells has benefited from mostly engineering optimization routines in recent years. This has slowed performance gains and leaves current state-of-the-art cells far below their theoretical limits [1-3] unlike III-V and IV group semiconductors. There are three primary factors for CdTe’s low performance: Shockley-Read-Hall (SRH) [4, 5] recombination at grain boundaries and within grains, difficulty in doping, and difficulty in forming ohmic and transparent contacts. Grain boundaries can have drastic effects on CdTe’s photovoltaic efficiency via SRH recombination, but the precise mechanism is unknown. Much insight on the role of grain boundaries can be gained by combining DFT calculations with experiments to determine why certain behaviors are observed in poly-CdTe and why cells are far from their theoretical limit. With underlying mechanisms revealed, new techniques can be developed which may eliminate recombination centers and allo w for high-performance CdTe devices. The goal of this project is to determine defect and impurity thermodynamics and their effects on electronic structures at grain boundaries in CdTe. Density functional theory (DFT) calculations will be performed on atomic structure models of grain boundaries in CdTe. The grain boundary structures will be built based on STEM images using image analysis. The atomic structures of experimentally fabricated 4.8˚ tilt (110)||(110) grain boundary was built using image analysis methods on STEM images. The constructed atomistic model will be further used to understand the electronic structure changes in the presence of impurity atoms at the grain boundary. Thereby, we will be able to determine candidate alloying elements to CdTe grain boundaries, that will potentially reduce the recombination by eliminating the midgap states. 1. C. Ferekides and J. Britt, Solar Energy Materials and Solar Cells 35, 255 (1994). 2. X. Wu, Solar Energy 77, 803 (2004). 3. W. Shockley and H. J. Queisser, Journal of Applied Physics 32, 510 (1961). 4. W. Shockley and W. T. Read, Jr., Physical Review 87, 835 (1952). 5. R. N. Hall, Physical Review 87, 387 (1952). Project description: The proposed research aims to investigate simplified model structures of representative grain boundaries in CdTe, towards a detailed understanding of complete grain boundary networks present in poly-CdTe solar cells. In our previous project term we have created 4.8˚ tilt (110)||(110) grain boundary, and density of states (DOS) analysis identified that dislocation cores at this grain boundary induce midgap states that are detrimental to photovoltaic efficiency. We showed that some dopants can eliminate these midgap states. In this project term, we are going to carry out systematic screening of possible alloying elements to the 4.8˚ tilt (110)||(110) grain boundary that can eliminate the midgap states and effectively passivate this grain boundary. We are going to select different elements from different regions of the periodic table and substitute these elements in place of Te and Cd atoms the dislocation core in addition to interstitial sites. Subsequen tly, we will carry out detailed electronic structure analysis and estimate the site- and angular-momentum projected DOS to determine the atomic orbitals from which mid-gap states arise. The electronic structures of the lowest-energy models will be also investigated using hybrid functionals, which give more accurate values of the band gaps and band edges than local and semilocal functionals. To estimate the charge state of the impurities, we will carry out charged defect calculations in a super cell of bulk CdTe. The charge transition levels of impurities in the bulk CdTe will be used as a guide to estimate the charge levels of these impurities in the grain boundary. Overall, these calculations will allow new passivants to CdTe grain boundaries to be developed to exploit the behavior and encourage doping while reducing recombination and increase the photovoltaic efficiency. All DFT calculations will be carried out using a parallel plane wave code Vienna Ab Initio Simulation P ackage (VASP), which is already available in LCRC resources. Accordingly, our proposed work can be listed as: i) Possible dopants to CdTe and their charge state transition levels will be computed in a large supercell of bulk CdTe. For each impurity, we will consider both substitutional and interstitial sites. We will carry out defect calculations at 4 different possible charge states. Each calculation takes approximately 10 hours using 64 cores for a 3x3x3 supercell of CdTe containing 216 atoms. We will consider 30 different dopants. As a result the time requested for these calculations will be: 15 dopant * 4 (charge state) * 2 (defect site) * 64 (cores) * 10 (hours) = 153,600 core hours. ii) Dopants will be placed at 4.8º tilt (110)||(110) grain boundary. We are planning to try 30 different dopants. Each atomic relaxation and electronic structure calculation is expected to take 10 hours using 512 cores. For each dopant we will also consider four different charged states. As a result the time requested for these calculations will be: 30 dopants * 4 charge states * 10 hours *512 cores = 614,400 SUs. Total computation time requested: 800,000 core hours. Industry partnership: Project URL: Current FY Hours Used: undetermined amount New FY Requested allocation: 800000 Q1: 200000 Q2: 200000 Q3: 200000 Q4: 200000 Justification: We carried benchmarks on CdTe grain boundary atomic structures with different number of atoms. VASP parallel efficiency: Total core hours for 5 electronic minimization: For 128 atom system: # core: Time(s): Efficiency 64 5.1 1.0 128 7.1 0.73 256 6.3 0.82 512 10.3 0.86 For 620 atom system # core: Time(s): Efficiency 128 481 1.0 256 518 0.93 512 737 0.65 1024 1310 0.37 We also had a VASP benchmark in KNL nodes for a 201 atom system. For a single electronic minimization step It took 222 sec using 2 Nodes (128 cpu) 105 sec using 4 Nodes (256 cpu). Consequently, VASP is working ~100% effiicent in KNL nodes. We could not run a single CPU for any of those calculations, since the systems are too large to run in a single core. Storage requirements: 1 TB Thank You, The LCRC Accounts System