Simulation tools in advanced chalcogenide PV

Carmen M. Ruiz-Herrero,

Université Aix-Marseille, France

• Why we need modeling on PV

• Developing a physical model

• 1D simulations

• 2D and 3D simulations• 2D and 3D simulations

• Hybrid models

• Conclusions

Why we need modeling on PV

PV are complex systems:

Different materials

Optical and electrical problems

Technological limitations

Difficult to understand what it is happening

Modeling allows:

A more analytical aproach to the problem

Better design of experiments

Optimization of parameters

Better control of certain situations ( diffusion , second phase generation…)

Easy validation of diagnostics

Developing a physical model

A model has to be realistic

Mathematics and computers allows anything

Need of materials characterization:

- Optical

-- Electrical

Clarify the cell structure:

- Presence of layers

- secondary phases

- gradients- gradients

Simulation tools:-Dimensions

- Equations and algorithms

- electrical vs optical potential

- limitations

Developing a physical model Material parameters

Some basical parameters :

Bandgap

Optical absorption

Doping density

Thickness

Mobility

Not so basical parameters :

Electronic affinity

Dielectric permittivity

Defect structure

Carrier lifetime

Developing a physical model Clarify the cell structure

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

-20

-10

0

jtot(

mA

/cm

2)

v(V)

Importance of a good model: difference between ideal and realistic

-40

-30

-20

Modelled valuesVoc = 0.6918 VoltJsc = 35.74 mA/cm2FF = 81.03 %eta = 20.0 %

Voc = 0.691953 VoltJsc = 35.72509443 mA/cm2FF = 79.2841 %eta = 19.5991 %

jtot(

mA

/cm

2)Experimental values

Developing a physical model Simulation tools

1D vs multidimensional

- Speed

- Easy definition

- Variation of definitions

and parameters

1D Multidimensional

- Structural complexity

- Second order optical and

electrical effects

- Powerfull analysisand parameters

- Most codes are freeware- Powerfull analysis

-Effects such as parallel resistance or

optical interference not taken into account

-Rugosity, grain boundaries

- Not homogeneus distribution of phases

-Time and resource consuming

- Codes are expensive

- Dificult definition files

1D simulations

Software choice

PC1D: first developed 1 dimensional code. Very good for thick cells and silicon

AMPS: open source. WX-AMPS specially developed for thin films

AFORS-HET: conceived for amorphous silicon solar cells

SCAPS-1D: Initially developed for CdTe, very powerful for CIGS systems,

potential for kesteritas

All of them are easy to run, freeware and with a big community of users.

Example: NREL 2009 CIGS record

0

10

20

30

40

50

Ga

cont

ent (

at%

)

1,0

1,2

1,4

Eg(

Ev)

NREL CIGS recordNREL CIGS record Reproduction Reproduction withwith SCAPSSCAPS

0,0 0,5 1,0 1,5 2,00

Absorber thickness from the back contact (µm)

1,0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

-40

-30

-20

-10

0

Modelled valuesVoc = 0.6918 VoltJsc = 35.74 mA/cm2FF = 81.03 %eta = 20.0 %

Voc = 0.691953 VoltJsc = 35.72509443 mA/cm2FF = 79.2841 %eta = 19.5991 %

jtot(

mA

/cm

2)

v(V)

Experimental values

Example: diagnostics of doping problems

V(mV) →Small differences in J

VVococ(mV)(mV) JJscsc(mA/cm(mA/cm22)) FF(%)FF(%) Eff(%)Eff(%)

Standard cellStandard cell 816 21.261 75.17 13.04

Bad CISBad CIS 632 20.788 55.14 7.24

Bad MoSBad MoS22 633 21.378 64.06 8.66

Bad Bad CdSCdS 776 21.485 69.88 11.65

Bad CISBad CIS--ββββββββ 713 22.164 63.31 10.01

SCAPS

0 200 400 600 800

-25

-20

-15

-10

-5

0

J sc(m

A/c

m2 )

V(mV)

Standard cell Bad CIS layer Bad MoS

2 layer

Bad CdS layer Bad CIS-ββββ

→Small differences in Jsc

→ When the problem is on the back of the device,

there is an increase of the Voc losses

→ Problems on the CIS doping level lead to the

bigger losses (up to22.5%)

→ The barrier generated by the MoS2 layer

generates losses of 22.4%

→ The CIS-β layer presents the smallest losses onVoc (12.6%), but FF is severely affected

→ The p-i-n structure generated by an isolant CdS

does not affect very mcuh to Voc

C.M. Ruiz et V. Bermúdez (2009) Proceedings of the 34th IEEE Photovoltaics Specialist Conference

Example: implementing defect levels

Selenium vacancies

Donor at 110 meV detected by Admittance Spectroscopy

-0.2 0.0 0.2 0.4 0.60

5

10

15

20

25

30

35

J(m

A/c

m2 )

Experimental values Fitted curve

1E18

DOS Error

DO

S (

1/cm

3 eV)

-0.2 0.0 0.2 0.4 0.6

-5V(V)

-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6

-10

0

10

20

30

jtot(

mA

/cm

2)

v(V)

Medium density High density

Voc Losses : increase of resistance in the layer

Jsc loses: charge trapping in the SCR

Loss of efficiency

0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

1E17

E (eV)

Multidimensional simulations� Finite Element methode

� 1D, 2D and 3D simulations

12

_ZnO

Cds

� Space (Δx, Δy, Δz) discretization

� Grain structure

� Rugosity

�Grain boundaries+

CZTS

Mo

Glass

�Steady state resolution of the coupled

Poisson’s and continuity equations

� Newton and Gummel iterations

[1,2]

[1] H. K. Gummel, IEEE Trans. Electron Devices 11, 455 1964.

[2] D. L. Scharfetter and H. K. Gummel, IEEE Trans. Electron Devices, 16, 64 1969

�Grain boundaries

�Secondary phases

Hybrid modelUsing SPICE+ 1 dimensional curves A multidimensional electrical model

CIGS and kesterites can not simply be simulated by a diode:

we extract IV curves from 1D softwares and generate a « paralel matrix » of curves

It allows an interesting aproach to problems such as front contact design, grain boundaries

or cell interconnection

Hybrid modelImpact of lateral resistances in ZnO and Mo:

From 1D simulation to a 1000 cells matrixImpact of front grid design:

100x100 cell matrix

Impact of grain boundary capacitance in a 20x20 cell matrix

WHAT WE CAN AND CANNOT EXPECT FROM MODELING

� Qualitative information about impacts of variations on an standard structure

� Better understanding of electrical and optical phenomena involved

� Rapid testing of parameters variation on a provided and reliable model

� Information proportional to the quality of the model used

� Absolute quantitative information

� Insights in process steps for cell optimization

� Actual device behavior explanations

� Crackpot intelligence with basic parameters

� Actual results without a reliable model