combining high-throughput computing and statistical learning to develop and understand new...

25
Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds Anubhav Jain Energy Technologies Area Lawrence Berkeley National Laboratory Berkeley, CA MRS Fall 2016 Slides (already) posted to http://www.slideshare.net/anubhavster

Upload: anubhav-jain

Post on 19-Feb-2017

124 views

Category:

Science


0 download

TRANSCRIPT

Page 1: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Anubhav JainEnergy Technologies Area

Lawrence Berkeley National LaboratoryBerkeley, CA

MRS Fall 2016

Slides (already) posted to http://www.slideshare.net/anubhavster

Page 2: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Thermoelectric materials convert heat to electricity• A thermoelectric material

generates a voltage based on thermal gradient

• Applications– Heat to electricity– Refrigeration

• Advantages include:– Reliability– Easy to scale to different

sizes (including compact)

2

www.alphabetenergy.com

Alphabet Energy – 25kW generator

Page 3: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Thermoelectric figure of merit

3

• Require new, abundant materials that possess a high “figure of merit”, or zT, for high efficiency

• Target: zT at least 1, ideally >2

ZT = α2σT/κ

power factor >2 mW/mK2

(PbTe=10 mW/mK2)

Seebeck coefficient > 100 �V/K Band structure + Boltztrap

electrical conductivity > 103 /(ohm-cm) Band structure + Boltztrap

thermal conductivity < 1 W/(m*K) •  �e from Boltztrap •  �l difficult (phonon-phonon scattering)

• Very difficult to balance these properties using intuition alone!

Page 4: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Example: Seebeck and conductivity tradeoff

4

Heavy band:ü Large DOS

(higher Seebeck and more carriers)✗Large effective mass

(poor mobility)

Page 5: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Example: Seebeck and conductivity tradeoff

5

Heavy band:ü Large DOS

(higher Seebeck and more carriers)✗Large effective mass

(poor mobility)

Light band:ü Small effective mass

(improved mobility)✗Small DOS

(lower Seebeck, fewer carriers)

Page 6: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Example: Seebeck and conductivity tradeoff

6

Heavy band:ü Large DOS

(higher Seebeck and more carriers)✗Large effective mass

(poor mobility)

Light band:ü Small effective mass

(improved mobility)✗Small DOS

(lower Seebeck, fewer carriers)

Multiple bands, off symmetry:ü Large DOS with small

effective mass✗Difficult to design!

Page 7: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

We’ve initiated a search for new bulk thermoelectrics

7

Initial procedure similar to Madsen (2006)

On top of this traditional procedure we add:• thermal conductivity

model of Pohl-Cahill• targeted defect

calculations to assess doping

• Today - ~50,000 compounds screened!

Madsen, G. K. H. Automated search for new thermoelectric materials: the case of LiZnSb.J. Am. Chem. Soc., 2006, 128, 12140–6

Chen,W.etal.Understandingthermoelectricpropertiesfromhigh-throughputcalculations:trends,insights,andcomparisonswithexperiment.J.Mater.Chem.C 4, 4414–4426(2016).

Page 8: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Going beyond constant relaxation time - AMSET• Fully ab initio mobility and Seebeck

including realistic scattering effects• Previously aMOBT (Washington

University in St. Louis)• Parameterizes the band structure

into 1D– Misses anisotropic effects and doesn’t

fully treat multi-band effects (for now)• Uses scattering expressions derived

by previous work by Rode with DFT parameters– ionized impurity scattering– deformation potential scattering– piezoelectric scattering– polar optical phonon

8

Faghaninia, A., Ager, J. W. & Lo, C. S. Ab initio electronic transport model with explicit solution to the linearized Boltzmann transport equation. Phys. Rev. B 91, 235123 (2015).

Page 9: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Transport database

9

All data will be made available via upcoming publication as well as on Materials Project• Seebeck• conductivity/tau• effective mass• electronic thermal conductivity

Page 10: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

New Materials from screening – TmAgTe2 (calcs)

10

Zhu, H.; Hautier, G.; Aydemir, U.; Gibbs, Z. M.; Li, G.; Bajaj, S.; Pöhls, J.-H.; Broberg, D.; Chen, W.; Jain, A.; White, M. A.; Asta, M.; Snyder, G. J.; Persson, K.; Ceder, G. Computational and experimental investigation of TmAgTe 2 and XYZ 2 compounds, a new group of thermoelectric materials identified by first-principles high-throughput screening, J. Mater. Chem. C, 2015, 3

Page 11: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

TmAgTe2 (experiments)

11Zhu, H.; Hautier, G.; Aydemir, U.; Gibbs, Z. M.; Li, G.; Bajaj, S.; Pöhls, J.-H.; Broberg, D.; Chen, W.; Jain, A.; White, M. A.; Asta, M.; Snyder, G. J.; Persson, K.; Ceder, G. Computational and experimental investigation of TmAgTe 2 and XYZ 2 compounds, a new group of thermoelectric materials identified by first-principles high-throughput screening, J. Mater. Chem. C, 2015, 3

Page 12: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

YCuTe2 – friendlier elements, higher zT (0.75)

12

• A combination of intuition and calculations suggest to try YCuTe2

• Higher carrier concentration of ~1019

• Retains very low thermal conductivity, peak zT ~0.75

• But – unlikely to improve further

Aydemir, U.; Pöhls, J.-H.; Zhu, H.l Hautier, G.; Bajaj, S.; Gibbs, Z. M.; Chen, W.; Li, G.; Broberg, D.; Kang, S.D.; White, M. A.; Asta, M.; Ceder, G.; Persson, K.; Jain, A.; Snyder, G. J. YCuTe2: A Member of a New Class of Thermoelectric Materials with CuTe4-Based Layered Structure. J. Mat Chem C, 2016

experiment

computation

Page 13: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Bournonites – CuPbSbS3 and analogues

• Natural mineral• Measured thermal conductivity for

CuPbSbS3 < 1 W/m*K– Stereochemical lone pair scattering

mechanisms• Measured Seebeck coefficient in

the range of 400 µV/K• BUT electrical conductivity likely

requires improvement – can calculations help?

• Total of 320 substitutions into ABCD3 formula computed

• Experimental study is next

13

Faghaninia A., Yu G., Aydemir U., Wood M., Chen W., Rignanese G.M., Snyder J., Hautier G., Jain, A. A computational assessment of the electronic, thermoelectric, and defect properties of bournonite (CuPbSbS3) and related substitutions (submitted)

Page 14: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Variation of properties with substitution

14

Page 15: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Variation of properties with substitution

15

B and C groups (lone pair sites) require heavier elements for stability (low Eh) – Si and N are very unstable!

Page 16: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Variation of properties with substitution

16

As expected, band gaps tend to decrease with heavier anionsThis is due to shifting up of the VBM level

Page 17: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Variation of properties with substitution

17

Page 18: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Variation of properties with substitution

18

Cu has lowest bandgap because Cu1+ also tends to be very high up in the valence band

Page 19: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Variation of properties with substitution

19

Jain,A.,Hautier,G.,Ong,S.P.&Persson,K.Newopportunitiesformaterialsinformatics:Resourcesanddataminingtechniquesforuncoveringhiddenrelationships.J.Mater.Res. 31, 977–994(2016).

Page 20: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Interesting bournonites and effect of scattering

20

AMSET indicates interband scattering is extremely significant – need to confirm

Substitutions listed here are close to thermodynamic stability (<0.05 eV /atom unstable)

Page 21: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Defects – selenide looks slightly better than sulfide

21

(a) (b)

• Multiple defects prevent n-type formation• p-type limited by SbPb defect. Situation slightly better in selenide because CuPb can help

compensate• Extrinsic defects calculations (not shown) do not provide clear paths forward

Faghaninia A., Yu G., Aydemir U., Wood M., Chen W., Rignanese G.M., Snyder J., Hautier G., Jain, A. A computational assessment of the electronic, thermoelectric, and defect properties of bournonite (CuPbSbS3) and related substitutions (submitted)

CuPbSbS3 CuPbSbSe3

Page 22: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Open data and software

22

www.materialsproject.org

www.pymatgen.org

www.github.com/hackingmaterials/MatMethods

www.pythonhosted.org/FireWorksNote: results of 50,000 transport calcs will eventually be posted here

Coming soon: AMSETComing soon: MatMiner

Page 23: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

MatMiner (coming soon)MatMiner’s goal: help enable data mining studies in materials science

23

Page 24: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Interactive demo of MatMiner

• Can we create a machine learning model to predict bulk modulus that is accurate to ~20GPa in ~10 mins?

• Let’s find out!

• Code posted at:– https://gist.github.com/computron

24

Page 25: Combining High-Throughput Computing and Statistical Learning to Develop and Understand New Thermoelectric Compounds

Thank you!• Collaborating research groups

– Jeffrey Snyder– Geoffroy Hautier– Mary Anne White

– Mark Asta– Hong Zhu– Kristin Persson– Gerbrand Ceder

• Primary researchers– TmAgTe2 – Prof. Hong Zhu and Dr. Umut Aydemir– YCuTe2 – Dr. Umut Aydemir and Dr. Jan Pohls

– CuPbSbS3 – Dr. Alireza Faghaninia– MatMiner – Dr. Saurabh Bajaj

• NERSC computing center and staff• Funding: U.S. Department of Energy, Basic Energy Sciences

25Slides (already) posted to http://www.slideshare.net/anubhavster