Theoretical modelling for kesterite photovoltaics

Adam J. Jackson, Aron Walsh

a.j.jackson@bath.ac.uk

Wednesday 18 Nov 2015

1 Introduction

(This presentation is online at https://wmd-bath.github.io/theory4kest)

1.1 About me

  • Adam J. Jackson
  • Undergraduate MEng Chemical Engineering (Bath, UK)
  • Doctoral Training Centre in Sustainable Chemical Technologies (Bath, UK)
  • Currently writing up a PhD thesis
    • "Ab initio thermodynamics for practical kesterite photovoltaics"
  • Based at University of Bath
    • Near Bristol, < 2 hours from London

1.2 University of Bath

bath_view.jpg

1.3 Kesterites at Bath

  • Early experimental work
    • Laurie Peter (my co-supervisor)
      • Jonathan Scragg, Phillip Dale, Diego Columbara
  • Current/recent
    • Aron Walsh [computational]
      • Adam Jackson, Jarvist Frost, Suzy Wallace
    • Mark Weller [crystallography, solid-state synthesis]
      • Mako Ng
    • Laurie Peter, Kieran Molloy [solution processing]
      • Anna Sudlow, Gabriela Kissling
  • PVTEAM partners
    • SPECIFIC, Bristol, Loughborough, Northumbria

1.4 Aims of this talk

  • Introduce the main theoretical methods that are relevant to kesterites research
  • Highlight key existing theoretical work
  • Identify problematic areas and opportunities for collaboration
  • Help with terminology and critical reading of literature

1.5 Not in this talk

  • Nuts and bolts of running calculations
  • Device modelling

2 The "cheat sheet"

TheoryLevels.png

3 Atomistic methods

3.1 Interatomic potentials

  • Simple interaction functions such as Buckingham potentials \((V = Ae^{-Br} - \frac{C}{r^6})\)
  • Polarisation can be included approximately through shell models

Sorry, your browser does not support SVG.

  • Simple to implement, inexpensive calculations
  • Main challenge is in fitting the potentials
  • No information about electronic structure
  • We have access to a parameterisation for CZTS (Chris Eames, University of Bath)

3.2 Ab initio methods

  • Focus on time-independent Schrödinger equation \(H\Psi = E\Psi\)
    • (Time-dependent methods exist, are more complicated…)
    • Solution for a system locates the ground state configuration and energy
    • Ground state energy \(E \approx U(T=0)\)
    • \(H\) is an operator, must be defined for a particular basis
  • Bloch's theorem:
    • The single-electron eigenvectors in a periodic system are the products of a single function with the periodicity of the lattice and all the points of the reciprocal lattice
  • Crystalline systems can be modelled with a periodic boundary condition
    • but the sum over the reciprocal lattice must be converged
      • (integration over the Brillouin zone)

Sorry, your browser does not support SVG.

3.3 Ab initio codes

  • A wide range of packages are available
  • Wikipedia maintains a handy table
  • Most popular codes include
    • VASP (very efficient, if archaic, code for periodic systems)
    • Gaussian (fast and quite user-friendly, non-periodic)

3.4 Density Functional Theory

dft_methods.png

  • Kohn-Sham DFT replaces \(E_{XC}(\Psi)\) with \(E_{XC}(\mathbf{\rho})\)
  • Accuracy of DFT calculation depends on
    • Numerical convergence of iterations
    • Convergence of basis set, k-points
    • Quality of XC-functional
      • Does not "converge" systematically
  • "Jacob's ladder" of methods 10.1063/1.1390175

babel.png

3.5 LDA and GGA

  • These methods are fairly affordable for large (100+ atoms) systems
  • Local (spin)-Density Approximation (LDA)
    • Consider only the local electron density, \(E_{XC} = f(\mathbf{\rho})\)
    • Surprisingly good for metals
  • Generalised Gradient Approximation (GGA)
    • Include local density gradient, \(E_{XC} = f(\rho, \nabla rho)\)
    • Better performance for semiconductors and insulators
    • Popular GGAs are PW91 and PBE
      • We like PBEsol
  • Hugely underestimate bandgaps
    • DO NOT TRUST BANDGAPS FROM LDA/GGA

3.6 Tight-binding

  • Tight-binding models solve a simplified form of \(H\Psi = E\Psi\), where \(H\) contains a minimal set of interactions.
  • Calculations are inexpensive and scalable
  • The interactions must be parameterised
    • "DFTB" refers to automated fitting of a tight-binding model to data from DFT
  • Unlike forcefield methods, this is an electronic structure method

3.7 DFT+U

  • DFT tends to underestimate on-site Coulomb interactions
  • Leads to significant errors for localised d- and f-electrons
  • A simple energy correction, \(U\), is included in SCF cycle
  • The parameter \(U\) must be sourced responsibly!
  • If \(U\) has been fitted to reproduce the experimental bandgap, then good agreement can be obtained between calculated and measured results.
    • This does not mean that good science is being done…
  • If a paper uses DFT+U, check where the \(U\) comes from and bear this in mind.

3.8 Hybrid functionals

  • GGA functionals tend to under-estimate the "exchange" portion of EXC.
  • The Hartree-Fock (HF) method calculates exact exchange, but no correlation.
  • Calculated energies can be improved by mixing in some HF exchange
  • The main implementations are PBE0 (below), B3LYP (older, very succesful) and HSE06 (PBE0 with range-screening)
\begin{equation} E^\text{Hybrid}_\text{XC} = E^\text{GGA}_\text{XC} + a(E^\text{HF}_\text{X} - E^\text{GGA}_\text{X}) \end{equation}
  • Unfortunately, the scaling of HF is worse than GGA, so calculations become slower
  • Generally hybrid functionals give "reasonable" results, approaching "chemical accuracy"

3.9 "Beyond-DFT" methods

  • Quasi-particle methods
    • \(G_{0}W_{0}\), sc\(GW\)
  • Time-dependent DFT (TD-DFT)
    • Useful for studying transition states, optical properties
  • Post-Hartree Fock methods
    • coupled cluster [CCSD, CCSD(T)]
    • quantum monte carlo
    • configuration interaction (CI)
  • Some of these methods are very expensive, poor scaling
    • Used for "gold standard" reference calculations

4 Theoretical modeling for kesterites

4.1 Total energy calculations

  • Geometry optimisation required
  • DFT calcs were key for confirming kesterite ground structure
    • Conventional unit cells with LDA, GGA sufficient
    • Several papers in 2009

paier2009structures.jpg (Paier et al. 2009)

papers2009.png

4.2 Defect energies

  • By studying larger supercells, simple defects and defect clusters can be studied
    • VCu, VCu + CuZn, CuZn + ZnCu clusters shown to have low formation energy

ChenDefectsHeader.png

  • Defect formation energies depend on energies of other phases; and, ultimately temperature
  • Kosyak et al. (2013) combines lattice vibrations with defect formation; finds S vacancies are also likely under annealing conditions

kosyak2013header.png

defects.png

4.3 Bandgaps

  • LDA and GGA fail badly
  • It is possible to apply an empirical correction or "scissors operator" but there is no guarantee of accuracy
  • LDA+U or GGA+U can improve matters, and was used by Persson (2010) for high-quality electronic structure calculations.
    • Optimised with LDA for GGA+U calcs
    • Est. \(E_g\) 1.5 eV for CZTS, 1.0 eV CZTSe
    • From band structure obtain effective masses, dielectric constant, optical absorption coefficient
  • Hybrid DFT calculations by Paier et al. (2009) found \(E_g\) 1.49 for kest CZTS, 1.30 for stannite
    • Followed up with G0W0 quasiparticle calculation on HSE results, showing little change.
    • Expensive!

bottibands.png

  • Same paper shows how sensitive bandgap is to anion displacement
  • Shaded regions are experimental values

anions.png

4.4 Vibrations

  • Many theoretical papers say "enthalpy" and mean "ground-state energy"
  • Thermodynamic potentials (H, S, \(\mu\) …) are temperature-dependent
    • In the solid state this is primarily driven by lattice vibrations
  • Some lattice vibrations can be directly observed by IR and/or Raman spectroscopy

4.5 Harmonic approximation

  • Assume simple harmonic motion of ground-state lattice
  • Two main methods:
    • Density functional perturbation theory (DFPT)
      • Often quicker, built into many DFT codes. Only gives information at Γ-point.
    • frozen phonon method (AKA "direct method" and "supercell method")
      • Obtain whole phonon band structure
  • Diagonalize matrix of force constants to obtain normal modes and frequencies.
  • Results tend to agree qualitatively with experimental frequencies, with errors ~ 10 cm-1

kest_gamma.gif

4.6 Beyond the harmonic approximation

  • The first step is the quasi-harmonic approximation
    • carry out harmonic approximation at several lattice expansions to form Equation of State
    • Accounting for thermal expansion tends to "soften" frequencies
  • Phonon-phonon interactions
    • Very demanding, requires many displacement calculations and slow post-processing
    • Allows peaks to be broadened using phonon lifetimes
    • For a recent test case we used… CZTS!
    • Skelton et al. (2015) improves agreement with IR and Raman data.
      • Also shows that CZTS has very low thermal conductivity

5 Some other useful techniques

5.1 Projected density of states

  • Energy is mapped to atom-centred functions
    • Choice of procedures, no "correct" method
  • Application: ZnS/Se resonant Raman spectroscopy
    • HSE06 hybrid functional used for "good enough" bandgap
    • Main challenge was converging k-point mesh

ResRamanHeader.png

ResRamanFig.png

5.2 Surfaces and work functions

  • Alignment of valence and conduction bands is important for device design
  • Ideal reference point is the vacuum level
  • Typical approach is look at electrostatic potential of "slab" models
    • Slab cannot be a dipole
      • dipole + periodic boundary = infinite energy

sns_slab.png

bandalignment.png

  • Band alignment has been done by modelling junctions with CdS and using CdS as the reference. Chen et al. (2011)

5.3 Molecular dynamics

  • Calculate forces and follow Newtonian mechanics
  • Various methods to introduce temperature
  • Many time steps needed for good quality data
    • No problem for forcefield models, expensive for ab initio
  • We've done some simple tests
  • https://www.youtube.com/watch?v=BuvtoyDF4vU
  • Activation barrier to disorder is low, but still requires long MD runtimes

6 Wrapping up

6.1 Key work so far

  • Ab initio calculations have provided qualitative and quantitative information about the electronic structure of CZT(S,Se)
  • Due of scaling challenges we know a lot more about highly-crystalline phases than disordered ones

6.2 Within reason, you can trust

  • Lattice parameters (~1%) from LDA or better
  • Elastic properties from GGA or better
  • Optical properties from GGA+U or better
  • Bandgaps from hybrid DFT or better

6.3 Beware

  • Optimisation with one method and electronic structure from another
  • Conveniently fitted scissors operator or \(U\)

6.4 Some good areas for collaboration

  • Γ-point vibrations
  • Values for effective mass, absorption
  • Properties of well-defined hypothetical materials

6.5 It's hard to help with

  • "Fractional" structures
  • Bandgaps of disordered, heavily-defective systems
  • Surfaces and interfaces - but we must try!

cu2sns3.jpg

6.6 Acknowledgements

  • Centre for Sustainable Chemical Technologies
  • PVTEAM
  • Computer time from EPSRC, STFC and University of Bath
  • Walsh Materials Design group