Download as PDF, PPTX
![FeBx: it looks like a real superconductor
Susceptibility
shows diamagnetism
Bc2(T)= Bc2(0)[1-(T/Tc)2]/ [1+(T/Tc)2]
gives Bc2(0) = 2 T
-> Type II BCS superconductor
Partial R drop
~ 10 K?
Superconducting phase was detected in 2 spread wafers
APL Materials 1, 042101 (2013)](https://image.slidesharecdn.com/ichiro-180812201024/75/Combinatorial-Experimentation-and-Machine-Learning-for-Materials-Discovery-10-2048.jpg)
More Related Content
Similar to Combinatorial Experimentation and Machine Learning for Materials Discovery
Combinatorial Experimentation and Machine Learning for Materials Discovery
- 1. No impurity Ti(3 Å) Ti (6 Å) Ti (9 Å) Cu (3 Å) Cu (6Å) Cu (9 Å) 5 Å 10 Å 15 Å 20 Å 25 Å 35Å 45 Å 55 Å ti(Å) ts(Å) Permanent magnet library Ferroelectric library Superconductor library Combinatorial Experimentation and Machine Learning for Materials Discovery Ichiro Takeuchi University of Maryland
- 2. - Combinatorial searchof superconductors - Active learning for directing high-throughput experiments Outline Supported by NIST, ONR, and AFOSR University of Maryland V. Stanev X.-H. Zhang H. Yu S. Lee Y. Liang J.P. Paglione NIST A. G. Kusne B. DeCost J. Hattrick-Simpers Duke Univ. S. Curtarolo C. Oses SLAC A. Mehta
- 3. Composition Spreads of TernaryMetallic Alloy Systems Co-sputtering scheme Ni Mn Al 3” spread wafer Al Ni Mn Phase diagram Composition is mapped using an electron probe (WDS) Review article: Green et al., JAP 113, 231101 (2013)
- 4. Rapid mapping ofmagnetic properties: scanning SQUID SQUID assembly inside vacuum Room temperature samples are measured 0 13 25 38 50 63 75 80 60 40 20 0 col row -2.50e+007 0.00e+000 2.50e+007 rho1_25 Mn Ni Ga Raw data Nature Materials 2, 180 (2003)
- 5. Rapid mapping ofmagnetic properties: scanning SQUID 0 13 25 38 50 63 75 80 60 40 20 0 col row -2.50e+007 0.00e+000 2.50e+007 rho1_25 Mn Ni Ga GaNi 0 1 2 3 4 5 6 7 8 9 10 Mn 50 100 150 200 250 M (emu/cc) 0 13 25 38 50 63 75 80 60 40 20 0 col row -2.50e+007 0.00e+000 2.50e+007 rho1_25 Mn Ni Ga Raw data Nature Materials 2, 180 (2003)
- 6. Rapid mapping ofmagnetic properties: constructing functional phase diagram GaNi 0 1 2 3 4 5 6 7 8 9 10 Mn 50 100 150 200 250 M (emu/cc) Nature Materials 2, 180 (2003)
- 7. Rapid mapping ofmagnetic properties: comparison with phase diagram Nature Materials 2, 180 (2003) C. Wedel and K. Itagaki, Journal of Phase Equilibria 22, 324 (2001)
- 8. Targeting superconductors predictedby theory Prediction: FeB4 is a superconductor with Tc ~ 15‐20 K
- 9. Ch 11 Ch 3 Ch13 Middle region: FeB2 – FeB4 more Bmore Fe temperature resistance 4.2 K 300 K Fe-B composition spread: FeBx(x =2-4), 16 spots on one 1 cm2 chip
- 10. FeBx: it looks like a real superconductor Susceptibility shows diamagnetism Bc2(T)= Bc2(0)[1-(T/Tc)2]/ [1+(T/Tc)2] givesBc2(0) = 2 T -> Type II BCS superconductor Partial R drop ~ 10 K? Superconducting phase was detected in 2 spread wafers APL Materials 1, 042101 (2013)
- 11. Mining superconducting databases Superconductorsgrouped by 3 golden coordinates “Predictions of new compounds” Based on ~ 600 superconductors (1988)
- 12. Solution: create ourow from literature 29000 entries in MatNavi (2014) Visualization and mining of NIMS SuperCon database 29000 entries (Cuprates: more than 10000) Data mining via supervised machine learning using Magpie descriptors Removing misentries, duplicates, etc. results in 14000 entries Magpie descriptors developed by Wolverton et al
- 13. Machine learning modeling of superconducting critical temperature (random forest) Models a Models are builtbased on 14000 entries Random forest solves the problem of including non- superconducting materials Stanev, et al. npj Computational Materials 4, 29 (2018)
- 14. Machine learning modeling of superconducting critical temperature Overall prediction accuracyis pretty good Different classes of superconductors can be distinguished Low Tc vs cuprates Cuprates vs low Tc npj Computational Materials 4, 29 (2018) Experimental Tc (log) PredictedTc(log) Low Tc vs Fe-based Cuprates vs Fe-based
- 15. Machine learning modeling of superconducting critical temperature Different chemical properties/trendsare revealed BCS superconductors Curprates ~ √ Expected behavior: phonon frequency falls out as the key descriptor Cuprates: Max Tc scales as # unfilled orbitals (?) Stanev, et al., npj Computational Materials 4, 29 (2018)
- 16. Machine learning modeling of superconducting critical temperature DOS of predictedsuperconductors show unusual structure Combining SuperCon (experimental database), machine learning, ICSD, and AFLOW, we predict possible new superconductors Stanev, et al. npj Computational Materials 4, 29 (2018)
- 17. BCC Fe Hexagonal FeGa FCC FePd Not in Database FCC Fe Not in Database Orthorhombic Fe and Pd Silicides Structural phase mapping using libraries Long et al., Rev.Sci. Instrum. 80, 103902 (2009) Stanev et al., npj Computational Mat (2018) 4:43 Fe Ga Pd 3” spread wafer
- 18.
- 19. Phase mapping based on active learning • Semi‐Supervised Learning: Propagate knowledge from measured/known samples to unmeasured samples • Quantify Certainty of Prediction •Graph‐based phase separation algorithm used Not Measured Coarse Measurement Measured (Active) Sample to Query Certainty of Cluster Label (0‐1) Ni Mn Ge Ni Mn Ge Runs Live at SLAC X‐Ray Beam Line!
- 20.
- 21.
- 22.
- 23.
- 24. Pure VO2 Autonomous mapping of composition‐ temperature phase diagram 100 150 200250 300 350 400 10 2 10 3 10 4 10 5 R() T (K) 0.33 0.96 1.36 1.68 2.27 2.61 2.86 3.44 c-Al 2 O 3 substrate W%Nb% W – substituted VO2 Metal-insulator transition temperature shifts to lower temperature with doping 3%W- VO2 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 %W substitution Temperature(○C) 20 40 60 70 Expected phase diagram Tetragonal Monoclinic Mixed Clustering + active learning for autonomous control Monoclinic Tetragonal
- 25. 27.00 27.75 28.50 10 1 10 4 10 7 Intensity 2 V1-x Wx O2 VO2 Autonomous mapping of composition‐ temperature phase diagram V1-xWxO2 Roomtemperature diffraction Composition spread chip 9 mm wide VO2 V1-xWxO2 DeCost, et al. (2018) Bruker C2 software Anton-Parr temperature stage Both controlled by active learning algorithm https://www.youtube.com/watch?v=GSsAykESATM
- 26. Procedures for activelearning of autonomous phase diagram mapping
- 27. Procedures for activelearning of autonomous phase diagram mapping # of all possible measurements: 12 x 9 = 108 Start measurements at room temperature “Project” to higher temperatures Move to higher temperatures once enough confidence is established Need to measure less and less points at higher temperatures
- 28.
- 29. Evolution of the combinatorial strategy (and a future forecast) Circa 1990 20002010 2017 and beyond Challenges: How to make hundreds of samples fast How to measure large number of samples: speed, number and quantitative accuracy How to quickly analyze large amount of data Machine learning to control and reduce number of samples# of samples: x 100-1000 # of samples: x 0.1 – 0.2 # of samples: x 0.1 – 0.2 Reducing the number of data points
- 30.