In a study published in the journal NPG Asia Materials, researchers utilized Bayesian optimization (BO) to enhance the thermoelectric properties of multicomponent III–V materials. They focused on improving the figure of merit (ZT) and explored a five-dimensional space, including film composition, dopant concentration, and deposition temperatures.
The above cycle is repeated: (i) thin-film deposition, (ii) thermoelectric property measurements, and (iii) BO recommendations. Image Credit: https://www.nature.com/articles/s41427-024-00536-w
After six optimization cycles, ZT exhibited a remarkable threefold improvement compared to initial trials. The analysis highlighted the effectiveness of high composition and low substrate temperatures in boosting ZT. These findings advance thermoelectric device development and underscore the efficacy of BO for multicomponent materials.
Related Work
Past work has seen significant interest in thermoelectric generators due to the growing demand for energy-harvesting technologies. Flexible thermoelectric devices, crucial for IoT applications like remote sensing, hold promise. The ZT quantifies a device's heat-to-electricity conversion performance, influenced by parameters like conductivity and thermal conductivity. III-V compound semiconductors show potential for thermoelectric thin films, but optimizing their performance requires considerable effort due to numerous combinations.
Recent advances include machine learning techniques like BO, which is effective for multicomponent thermoelectric thin films. Despite limited reports on their development, polycrystalline III-V compound thin films are gaining attention for their potential in utilizing phonon scattering.
Thin Film Deposition
Tin (Sn)-doped indium (In₁₋ₓ), gallium (Gaₓ), arsenic (As₁₋ᵧ), and antimony (Sbᵧ) films, each 500 nm thick, were deposited onto silicon dioxide (SiO₂) glass substrates through vacuum evaporation with Knudsen cells. BO was employed to explore optimal cell temperatures for Ga, As, Sb, Sn, and substrate temperature.
The In-cell temperature varied based on TGa-cell to maintain consistent film thickness at 500 nm. BO implementation utilized a Gaussian process regression (GPR) algorithm with expected improvement (EI) as the acquisition function, generating randomly created training data for the first five cycles.
Sample evaluation included scanning electron microscope (SEM) examination, where energy dispersive X-ray spectroscopy (EDX) spectra obtained determined In₁₋ₓGaₓAs₁₋ᵧSbᵧ compositions. Researchers determined the Sn concentration through secondary ion mass spectrometry (SIMS) measurements. They obtained out-of-plane XRD patterns using a diffractometer with a Ge monochromator and Cu-Kα radiation source. Hall effect measurements were performed using the van der Pauw method, averaging n and μ over 10 measurements per sample.
Researchers measured electrical conductivity (σ) and Seebeck coefficient (S) using a zero-field electron mobility (ZEM)-3 system, utilizing Ag paste to fix the sample on a ceramic stage. Cross-plane thermal conductivity (κ) was determined using a PicoTherm (PicoTR).
BO Optimization Study
The experimental approach utilized in this study involved a cyclic process, where BO was employed to iteratively enhance the thermoelectric properties of Sn-doped In₁₋ₓGaₓAs₁₋ᵧSbᵧ films. This iterative cycle comprised thin-film deposition, thermoelectric property measurements, and BO recommendations, facilitating continuous improvement in material performance.
Researchers investigated the impact of substrate temperature (Tsub) on the crystallinity and composition of In₁₋ₓGaₓAs₁₋ᵧSbᵧ films without Sn doping. SEM images revealed surface morphology and grain size variations corresponding to different Tsub values. Additionally, EDX analysis confirmed uniform composition across the samples, with a consistent ratio of Into Ga and As to Sb.
The out-of-plane XRD patterns indicated polycrystalline structure, with peak shifts attributed to compositional changes induced by varying Tsub. These findings underscore the complex influence of both cell and substrate temperatures on film composition and crystallinity.
Further electrical and thermoelectric properties analysis revealed significant dependencies on substrate and Sn-cell temperatures. Carrier concentration (n) exhibited sensitivity to Tsub variations, while mobility (μ) remained relatively stable. The conductivity type shifted between p-type and n-type behavior, correlating with changes in In₁₋ₓGaₓAs₁₋ᵧSbᵧ composition and Sn doping levels. This intricate interplay highlights the importance of precise control over growth conditions for optimizing thermoelectric performance.
This study delves into the influence of composition on thermal conductivity (κ) at a fixed substrate temperature. Compositional alloying resulted in decreased κ, attributed to enhanced alloy scattering, with significant reductions observed with increased In and Sb content. These findings emphasize the potential for fine-tuning material composition to achieve desired thermoelectric properties, particularly in mitigating thermal conductivity.
The BO process facilitated the rapid enhancement of thermoelectric performance, with ZT values approximately tripling within six optimization cycles. Analysis revealed vital factors contributing to improved ZT, including high composition, low substrate temperature, and controlled Sn doping levels. These insights underscore the effectiveness of machine learning-driven optimization in navigating the intricate parameter space of multicomponent thermoelectric materials.
This study demonstrated the utility of BO coupled with machine learning for optimizing the thermoelectric properties of Sn-doped In₁₋ₓGaₓAs₁₋ᵧSbᵧ films. Systematically exploring growth conditions and material parameters led researchers to significantly improve thermoelectric performance, enhance efficiency, and expand the applicability of III–V compound semiconductor thermoelectric materials.
Conclusion
In summary, this study showcased the efficacy of BO and machine learning in enhancing the thermoelectric properties of Sn-doped In₁₋ₓGaₓAs₁₋ᵧSbᵧ films. By systematically exploring growth conditions and material parameters, researchers achieved notable improvements in thermoelectric performance. The iterative cyclic process facilitated continuous material enhancement by incorporating thin-film deposition, property measurements, and BO recommendations.
The study emphasized the intricate interplay among substrate temperature, composition, and Sn doping levels, stressing the need for precise control over growth conditions for optimizing thermoelectric properties. Additionally, it highlighted the potential of fine-tuning material composition to achieve desired thermoelectric properties, particularly in reducing thermal conductivity.
Overall, this research offered valuable insights into optimizing multicomponent thermoelectric materials, with implications for enhanced efficiency and broader technological applications.
Machine Learning Boosts Earthquake Prediction Accuracy in Los Angeles