Ok, the last thermoelectrics dataset I'll be looking at before moving onto modelling/prediction/design phase of work: sysTEm (Systematically Verified Thermoelectric Materials) dataset. It's the highest-quality TE dataset I've worked with so far.
Previous TE datasets had issues with missing values, inconsistent units, questionable zT claims. Overall just LLMs not at the place to do reliable data extraction. The datasets we've looked at so far provide a nice overview of the advancement of the technique over the past few years. This most recent one, sysTEm (Tang et al., 2025), fixes some issues by enforcing a simple rule (hence "systematically verified"). Every material must satisfy:
within 10% tolerance. If reported zT doesn't match the calculated zT, it's flagged or excluded. This eliminates some inconsistencies that plagued previous datasets.
Coverage:
8,650 experimental measurements (1991-2024)
1,437 unique materials
Properties: σ, S, PF, zT, κ_total, κ_lattice, κ_electronic
99.2% completeness for transport properties
Top elements (classic heavy-element strategy):
Sb, Te, Se, Bi, Cu - heavy chalcogenides and pnictogens dominate
O (oxides for high-temp), Sn (emerging systems)
Performance:
Mean zT: 0.42
Maximum: 3.082 (Na₀.₀₃Sn₀.₉₆₅Se at 783K)
841 materials with zT > 1.0 (9.7%)
Temperature effects:
Room temp (300-400K): avg zT = 0.24
Mid-temp (400-700K): avg zT = 0.41
High-temp (700-1000K): avg zT = 0.69 (3× better)
I performed the same trade-off analysis I did with NEMAD, identifying materials that optimally balance Power Factor (electrical performance) vs Thermal Resistivity (thermal isolation).
Out of 7,850 materials, 38 are Pareto optimal (0.5%) These elite materials achieve avg zT= 1.31 vs 0.43 overall
Bi-Ge-Te systems (3 materials)
Highest zT: 2.35 at 650K
Recent 2022 optimization
Ag-Sb-Te + Cd doping (9 materials)
Peak zT: 2.54 at 575K
Systematic composition tuning
Ag₂Se-Te composites (6 materials)
All from 2024—very recent
Emerging successor to Bi₂Te₃
Half-Heuslers (Zr-Hf-Ti-Ni-Sn)
Highest power factors (>60 µW/cmK²)
Filled skutterudites (CoSb₁₂)
Ultra-low thermal conductivity
Rattler atoms scatter phonons
Thermal conductivity breakdown (2,035 materials with separated contributions):
Lattice: 70.5% (1.52 W/mK avg)
Electronic: 29.5% (0.64 W/mK avg)
→ Phonon scattering is the dominant challenge
Record holders:
Highest PF (Pareto): FeNb₀.₉Hf₀.₁Sb = 76.05 µW/cmK²
Lowest κ (Pareto): LaCo₀.₅(Ni₀.₅Fe₀.₅)₀.₅O₃ = 0.07 W/mK
The fundamental trade-off: High PF materials tend to have high κ (metals conduct heat). Low κ materials often sacrifice electrical performance. Pareto optimal materials navigate this through:
Nanostructuring (Ag₂Se composites)
Smart alloying (Cd in AgSbTe₂, Ge in Bi₂Te₃)
Rattler atoms in cage structures
Heavy element combinations
Validation: Every material with calculated zT +/- 10% of recorded.
Provenance: Every point includes DOI, year, source, temperature
Completeness: 99%+ coverage vs 65-75% in other datasets
Recency: 50% from 2020+, capturing latest research
This is by far the best thermoelectric dataset available. Perfect for ML, screening, benchmarking, and understanding design principles.
Citation: Tang LZ, Purdy L, Mohanty T, Ng LWT, Sparks TD. Systematically Verified Experimental Thermoelectric Dataset. ChemRxiv. 2025; doi:10.26434/chemrxiv-2025-4gxmc
🤖 In collaboration with Sonnet
On this page