Advancements in Thin-Film Thermoelectric Generator Design for Agricultural Applications

Abstract

Thin-film thermoelectric generators (TFTEGs) emerge as critical components of self-sustaining agricultural systems because they can utilize temperature gradients to generate plant-transpiration-induced thermovoltage signal quantifiable to plant health status. This study examines the latest developments in TFTEG materials, device structures, manufacturing processes, and their integration into agricultural systems such as plant-wearable, canopy-level and stem-clipped TEGs. Key questions addressed include the ideal materials for TFTEG fabrication, their biocompatibility and eco-stability in agricultural settings, recent design and AI-assisted optimization advancements, and future research directions in non-conventional TEG applications. The analysis consolidates evidence from inorganic, organic, and hybrid thermoelectric materials with respect to performance in terms of flexibility, thermal stability, output power, and biocompatibility. Bibliometric analysis was employed to determine dominant research topics and gaps, especially with respect to sustainability and AI-augmented design. The review emphasizes the latest breakthroughs in structural optimization, flexible substrates, encapsulation strategies, and sensor integration for reliability enhancement in field environments. In addition, applications of AI, including neural network-based conditional Generative Adversarial Network, surrogate modeling, and multi-objective optimization, are discussed in relation to the improvement of thin-film TEG design and simulation processes. This study suggests that TFTEGs exhibit great potential in agricultural monitoring and plant wearable applications but material toxicity, mechanical degradation, and integration with AI are still major obstacles.

1. Introduction

Self-powered technologies have become important in driving sustainable and smart agriculture production. They enable renewables like solar and wind energies to be converted into usable electrical power, minimizing the need for external power grids or batteries, facilitating energy efficiency, sustainability, and enhanced monitoring. Advanced self-powered technologies are capable of harvesting other energies, such as heat, acoustic, and radio frequency (RF), even garnering energies from the forces of the environment like mechanical vibrations [1]. Boosting these technologies can reduce reliance on fossil fuels, immensely aiding diminishing greenhouse gases and excessive use of agrochemicals [1]. Various self-powered technologies, such as thermoelectric generators (TEG), electromagnetic generators (EMG), piezoelectric generators (PZG), and triboelectric nanogenerators (TENG), are continually utilized to promote sustainable agriculture. A stackable, hybridized electromagnetic and triboelectric nanogenerator was designed to harvest energy from wind speeds ranging between 3 m/s to 6 m/s and convert it to electrical energy to power electronic devices and microsensing systems in a smart greenhouse [2]. This innovation operated for extended hours, reducing charging time and the need for external power supplies, fostering energy efficiency [2]. These energy harvesters are also integrated into wearable sensor technology, where a triboelectric energy harvester was incorporated into a wearable sensor for a double-network hydrogel plant. It was seen to convert clean energies from dynamic sources, like sound waves, wind, and mechanical pressure, providing a high power density at 424 mW/m² [3]. On the other hand, piezoelectric generators, which convert mechanical energy to electrical energy from the cantilever beam, were used to power a sensor system and wireless radio transmitter designed to detect vibrations from buildings and have continuous and extended monitoring for structural health in the event of anomalies [4]. While all these self-powered technologies offer unique and varying mechanisms, TEGs are more prominent due to their simplicity, having no moving parts, robustness, and capability to harness otherwise wasted thermal energy and transform it into a working energy [5].

Thermoelectric generators are solid-state devices that convert the temperature differences between the ground and the environment into electrical energy, utilizing the Seebeck effect [6]. The Seebeck effect occurs when a temperature difference is applied across different materials, generating voltage. Moreover, TEGs can continually generate power even at night when they utilize the stored heat from the soil [7], where different soil compositions have been found to release varying temperature gradients [8]. While conventional thermoelectric generators have demonstrated their potential in various agricultural and waste heat recovery applications, their design has been extended to suit flexible and lightweight applications, such as wearable systems. The thin-film thermoelectric generator (TFTEG) design offers flexibility and compactness in wearable and portable devices [9]. TFTEGs are continually developed by integrating materials, such as polymer or metal composites, as their core thermoelectric elements to enhance their electrical performance, flexibility, stretchability, and efficiency [9,10]. A multifunctional thin-film TEG was developed for waste heat recovery, serving as a power source for wearable electronic gadgets, and as a thermal touch sensor for real-time switching and temperature monitoring in exoskeleton applications. It can also be easily integrated with integrated circuits as well [11]. Several advantages were highlighted, including decreased thermal impedance, an increased temperature gradient, and fewer materials used compared to conventional TEGs [11]. Despite the advantages that thin-film TEGs showcase in agricultural applications, these devices still face challenges, such as biocompatibility issues and difficulties when deployed under field conditions due to environmental factors.

Agricultural fields are exposed to a wide range of environmental conditions, including mechanical stress and temperature fluctuations. These factors have a significant influence on the performance of TEGs. In these fields, thermoelectric generators become susceptible to thermal and mechanical degradation, which impacts their performance and longevity. A study has examined the influence of temperature differentials on the overall performance of TEGs and found that the response of TEGs becomes more efficient in high and stable thermal gradients [12]. In contrast, when thin-film thermoelectric generators are exposed to oscillating environmental conditions, including wind and fluctuations in ambient temperature, thin-film TEGs often suffer from mechanical degradation, substrate limitations, and moisture [13]. When thermoelectric generators, which are mostly composed of brittle thermoelectric materials, are subjected to mechanical stress from external sources, it poses a challenge to the durability of TEGs [14]. Moreover, the choice of thermoelectric materials can otherwise pose several biocompatibility issues, making it unsuitable for green applications [15]. Some of the top-performing thermoelectric materials rely on the use of selenium (Se), and lead telluride (PbTe), which are highly toxic, jeopardizing biocompatibility [15]. While addressing these challenges fronted by thin-film thermoelectric generators is essential, the limitations of conventional designs, techniques, and tools used for thin-film TEGs must also be considered for an optimized deployment.

Conventional designs of thin-film thermoelectric generators have utilized the heat flow running vertically to the film surface, where thick films separate the hot and cold sides of the TFTEG [16]. Due to the heat transmission and thermal radiation from the hot side, an abrupt increase in the temperature occurs on the cold side, narrowing the temperature difference (ΔT), resulting in a low output power [16]. In contrast, high density patterning for thin-film thermoelectric generators is crucial to improve thermal properties of the thin-film TEG. However, traditional semiconductor processes are often incompatible with high-density patterning for TFTEG legs, leading to difficulties in improvisation [17]. Therefore, optimized and innovative design strategies, and more biocompatible materials are needed to improve the overall functionality and sustainability of thin-film thermoelectric generators.

Material innovation and the integration of artificial intelligence are important in enhancing the properties of thin-film thermoelectric generators to fit certain applications in wearable technology. Some materials, such as polymers or insulators, are engineered to improve the conductivity, mechanical properties, especially during bending, energy harvesting capabilities, and the Seebeck coefficient of thin-film TEGs [18]. On the other hand, artificial intelligence is applied for spatial feature analysis, material selection, and predictive models for the TEG performance. A multi-objective optimization was utilized to maximize the output power, system efficiency and cost of thermoelectric generators [19]. Moreover, an artificial neural network (ANN) and a conditional generative adversarial network (cGAN) were proposed for fast optimization of TEGs, including accurate modeling and field analysis, wherein these models achieved a 97% prediction accuracy [20]. The integration of artificial intelligence in thermoelectric generators would be a great supplement to the entirety of the performance of thermoelectric generators.

The initial topic of the search focused on generative AI tools in general as an assistive method in TEG design. Searches on the implementation of the topic in actual experimental literature have shown that generative artificial intelligence (GAI) as a design tool has not seen much success in the field of thin film TEG design with zero articles found on a search on Scopus and the limited results showing no significant studies on the topic on ScienceDirect.

Table 1. Literature search keywords strategy for focusing the search to the most related articles.

Website	Keywords and Limiting Term	Search Count
Science Direct	“Generative AI tools” AND “Design” Subject Areas: “Medicine and Dentistry”, “Computer Science”, “Engineering”, “Nursing and Health Professions”, “Biochemistry, Genetics and Molecular Biology”	307
Scopus	“Generative” AND “AI” AND “Tools” AND “Design”	1331
Scopus	“Generative” AND “AI” AND “Tools” AND “Design” AND “Optimization”	93
Science Direct	Generative AI Design; Limitations: Book Chapters, 2024–2025	592
Science Direct	Generative Design AND Thin-film	37
Science Direct	“Thin film” and “Thermoelectric generator”; Limiting term: Book chapters	222
Science Direct	“Thermoelectric Generator” AND “Optimization” AND “Thin film” AND “Simulation” AND “ Modelling” AND “Artificial Intelligence”	155
Science Direct	(“Thermoelectric generator” OR “Teg”) AND (“Generative Artificial Intelligence” OR “Generative AI” or “GAI”) and (“Thin film” OR “Thin-film”)	63
Scopus	(“Thermoelectric generator” OR “Teg”) AND (“Generative Artificial Intelligence” OR “Generative AI” or “GAI”) and (“Thin film” OR “Thin-film”)	0

showcases some keywords used with their limiting conditions as well as the total number of results from the search. Studies within the last 10 years (2015–2025) were prioritized, however the primary range for the studies was from the last 20 years (2005–2025). The following papers were excluded in the selection process: Articles that did not discuss GAI and only declared their usage. Articles that did not discuss the technical and design aspects of GAI and its tools were also not included. Articles that focused primarily on the ethical and legal aspects of GAI were also not considered. Other results not considered include encyclopedia indices, abstract compilations, mini-reviews, editorials, and other similar articles. Scientific articles were prioritized in the discussion of sample studies for topics such as those discussed in Section 6. The limited amount of literature on certain topics such as life cycle assessment and agricultural use cases for thin-film thermoelectric generators led to the use of studies that incorporated general TEGs as a reference point.

A bibliometric analysis was conducted to identify key themes in the existing research space on thin film thermoelectric generators. Eight key clusters were formed with their own identifying themes as follows: cluster 1 represented by the color red discusses general components of TEGs such as biocompatibility, morphology, and performance; components such as single-walled carbon nanotubes, composite films, and conducting polymers, and semiconductors; flexible thermoelectrics and thin-films; cluster 2 is represented by the color green and discusses thermoelectrics and related terms such as thermoelectric generators, thermoelectric equipment, thermoelectricity, and energy harvesting; cluster 3 is represented by the color blue and discusses thermal energy focusing on solar, photovoltaics, and energy conversion, it also covers finite element analysis, the internet of things, and electronic equipment; cluster 4 represented by the color yellow on the other hand focuses on material alloys including semiconductor alloys such as semiconducting gallium and indium, as well as thin films and their relevant terminologies such as substrates; cluster 5 represented by the color violet covers TEG parameters such as the figure of merit and electrical conductivity, as well as geometric features such as superlattices; cluster 6 represented by a lighter blue color encompasses compounds and nanocrystals such as antinomy, sulfur, bismuth telluride, and chalcogenides; cluster 7 represented by the color orange focuses on thin film circuits, devices, and their capacity alongside other microelectronic devices; and lastly, cluster 8 represented by the color brown covers open circuit voltage, max output power, electrode configurations, and thermal evaporation processes.

The largest nodes in the bibliometric network in Figure 1 were more common in cluster 2 (green) where it focused on concepts more directly related to thermoelectricity followed by cluster 4 (yellow) which is focused more on terms more closely related to thin films and their design. The three more relevant clusters are cluster 1 (red) which goes into more detail on thermoelectric performance and polymers, cluster 5 (violet) which tackles the general parameters and geometric structures of TEGs, and cluster 3 (blue) which delves more into the field of photovoltaics and the internet of things. Overall, the key themes identified through bibliometric analysis are as follows: TEG performance and structural parameters; material alloys and their fabrication; general energy conversion and thermoelectric equipment; and microelectronics. Topics such as optimization, artificial intelligence, and sustainability did not exhibit a noticeable presence throughout the analysis and can be identified as possible research gaps in the field of thin film thermoelectronics. The lack of presence of life cycle assessment (LCA) within the bibliometric network also suggests it as a possible research gap.

The primary points of concern of this study revolve around thin film TEGs, their role in the agricultural setting, their manufacturing and presentation, their design and optimization processes, and the general trends on their research and application. Thus, the goal of this study is to answer the following research questions to identify the current state of thermoelectrics in the field of agriculture and its integration with AI design tools:

• What are the ideal materials for use in the fabrication of thin film thermoelectric generators? This is answered in Section 3.
• What is the current state of thermoelectric generators in the field of agriculture and how does its properties such as biocompatibility and eco-stability allow it to positively interact with the environment? This is answered in Section 2 and Section 4.
• What are the key advancements in thermoelectric generator design on topics such as their substrate characteristics, integration with flexible circuits, co-design with other materials, and their integration with artificial intelligence? This is answered in Section 5.
• What are the future directions for further research into the field of thermoelectrics? This is answered in Section 3, Section 4, Section 5 and Section 6.

This study mainly contributes to the following:

• A comparative analysis of different thermoelectric materials used in TEGs alongside their fabrication, deposition, and processing techniques.
• A discussion and comparison of different literature showcasing the applications of TEGs in the agricultural setting.
• A technical analysis of the advancements in AI-integration in the field of thermoelectric generators, their general process flow, and a discussion of the various literature, or lack thereof. A discussion on various literature on advancements in thin-film TEG design was provided.
• A scientific elucidation of the general research trends in the field of thermoelectric generators as well as other points of possible development such as AI-integration, material discovery, and other possible applications for TEGs.

This study is composed of six key sections, namely: Section 2 focusing on the fundamental information of thermoelectric effect and thin-film TEGs with their corresponding agricultural use-cases; Section 3 focusing on the chemistry, properties, fabrication, and processing of thin-film thermoelectric materials; Section 4 focusing on the thin film TEG’s biocompatibility with agricultural environments; Section 5 tackles advancements in thin film TEG design as well as the integration of AI tools in the design process; and lastly, Section 6 discusses the future directions on Thin-film TEG research and development.

2. Thermoelectric Effect and Thin-Film Thermoelectric Generators

The Seebeck effect, also known as the thermoelectric effect, covers the phenome-non wherein a temperature differential creates a voltage [21]. The Seebeck coefficient is the numerical representation of a material’s ability to convert heat into electricity and is identifiable as the ratio between the potential difference and the temperature differential between the two ends [22,23]. It is also a key component in the identification of the figure of merit, which is the primary quantifier for a material’s performance as a thermoelectric material [21,23]. The figure of merit can be maximized by improving the material’s Seebeck coefficient and electrical conductivity and minimizing the thermal conductivity of the material. The base formulas for these properties among other basic properties can be found in Table 2.

TEGs are applied in a wide variety of fields, including but not restricted to: the automotive industry for sensors and waste energy recovery; the industry sector for cooling and power generation; energy sector for self-powered purposes such as the conversion of solar energy into electricity; the medical sector for wearable devices; fire protection through the detection of temperature differentials; and AI and robotics for cooling of hardware and components [5,22,24].

A TEG is composed primarily of an n-type leg and a p-type leg which are electrically connected in series, thermally connected in parallel, and located in between two reservoirs of differing temperatures as can be seen in Figure 2 [22,23]. The power generated by the TEG through the Seebeck effect is generally too weak with a single construct thus a TEG module is generally composed of multiple interconnected TEG units to increase the overall power output [23]. Figure 2 showcases a single TEG module represented within the dotted box, where it’s composed of an n-type and a p-type leg attached to metal connectors meant to serve as interaction points with the hot and cold reservoirs.

A TEG can be expressed in the form of a thermally equivalent circuit in which case the heat transfer through both the hot and cold sides can be calculated as seen in Table 2 where the heat transfer is expressed in three terms that address the joule heating, thermal induction, and the Seebeck effect [22]. Further equations may be extracted from their thermally equivalent circuit and more efficiently optimized to provide a more accurate representation of a given TEG with a sample equation found in Table 2. However further research is needed for a more accurate estimate [22]. A thin film TEG is a TEG composed of a thin material of a certain length as such the specific material, and its deposition methods are important aspects of thin film TEG design [5].

Thin film TEGs, as the name suggests, are TEGs designed as. Some general advantages of thin film TEGs over bulk TEGs are that thin film TEGs have lower thermal conductivity, smaller size and weight, greater flexibility [23]. These advantages do come with their own set of limitations as their energy generation capabilities in self-powering devices are not equivalent to the powering capabilities of general microelectronics [23]. The more lightweight and flexible features of thin film TEGs make them more ideal for use in certain cases despite their limitations in the actual amount of energy they can generate, primarily for use in wearable electronic devices in sectors such as healthcare, smart watches, military health monitoring, and entertainment for wearable gaming equipment [5].

The equations demonstrated in Table 2 are primarily based on the following properties: TEG dimensions, voltage and temperature differentials of the hot and cold sides, Seebeck coefficient, electrical conductivity, and thermal conductivity. The TEG dimensions and temperature differentials can be easily identified through methods such as electron microscopy and the use of digital thermometers respectively. There are specialized equipment available for identifying these properties, however it is also possible to identify these properties through simpler experimental configurations. An example is in the identification of the Seebeck coefficient wherein a basic circuit with the following components: a DC source to power a heating plate for the hot side of the TEG with an optional implementation of a cooling device such as a peltier cooler for the cold side of the TEG, a digital thermometer for the measurement of the cold and hot side temperatures, and a digital multimeter for the measurement of the cold and hot side voltages [25].

Electrical conductivity is generally identified through the use of either two-point or four-point probe techniques which can be simplified through the use of equipment such as the Jandel RM3000 (Jandel Engineering, Leighton Buzzard, UK) [25]. Lastly, thermal conductivity can be found by various methods including calculations based on formulas that are dependent on the material with one study utilizing the specific heat capacity of the material found through a differential scanning calorimeter [25]. The two primary divisions for determination of thermal conductivity are the steady state-method which is based on the application of a constant heat flux to one side of the material until thermal equilibrium is reached; and the transient methods which apply short heat pulses, through method such as laser flash analysis, and is based on time-dependent temperature responses which identify the thermal diffusivity which eventually leads to the thermal conductivity [26]. Once these base properties are identified, the other properties can be identified through the use of different formulas that may vary dependent on various aspects of TEG design including material and shape.

Table 2. Thermoelectric generator properties and their basic formulas.

Property	Formula	Description and Symbols	Reference
Seeback Effect	$S={\displaystyle \frac{∆V}{∆T}}$	where S is the Seebeck coefficient, ∆V is the thermal voltage, and ∆T is the temperature difference between the hot and cold sides of the TEG.	[22,27]
Figure of Merit	$\mathcal{z}T={\displaystyle \frac{\sigma S^{2}T}{\kappa }}$	Where zT is the Figure of Merit, $\sigma$ is the electrical conductivity, T is the absolute temperature; and $\kappa$ is the thermal conductivity.	[23,28,29]
Thermal Efficiency	${\eta }_{TE}={\eta }_c{\displaystyle \frac{\sqrt{1+\mathcal{z}T}-1}{\sqrt{1+\mathcal{z}T}+1-{\eta }_c}}$	${\eta }_{TE}$ is the thermoelectric thermal efficiency, ${\eta }_c$ is the Carnot efficiency; and zT is the Figure of merit.	[22,27,28]
Carnot Efficiency	${\eta }_c={\displaystyle \frac{T_h-T_c}{T_h}}$	Where $T_h$ is the temperature of the hot side, and $T_c$ is the material of the cold side.	[22,27]
Heat transfer through high temperature contact	$Q_h=-{\displaystyle \frac{I^2\rho l}{2A}}+{\displaystyle \frac{\kappa A\left(T_h-T_c\right)}{l}}+SI{T}_h$	Where A is the cross-sectional area, l is the length of the leg, $\rho$ is the resistivity, and I is the current.	[22,27,29]
Heat transfer through low temperature contact	$Q_c={\displaystyle \frac{I^2\rho l}{2A}}+{\displaystyle \frac{\kappa A\left(T_h-T_c\right)}{l}}+SI{T}_c$	No new term introduced	[22,27,29]
Fill Factor	$FF={\displaystyle \frac{A_P-A_n}{A}}$	Where $A_P$ is the cross-sectional area of the p-type leg, $A_n$ is the cross-sectional area of the n-type leg, A is the total device area, And FF should be less than or equal to 1.	[22,28]
Optimized TEG Output power	$P_{out}={\displaystyle \frac{{FFS}^2{(T_h-T_c)}^2}{4\rho l}}$	Where FF is the Fill Factor, and The external load resistance is equal to the internal electrical resistance of the TEG.	[22]

Table 2. Thermoelectric generator properties and their basic formulas.

Property	Formula	Description and Symbols	Reference
Seeback Effect	$S={\displaystyle \frac{∆V}{∆T}}$	where S is the Seebeck coefficient, ∆V is the thermal voltage, and ∆T is the temperature difference between the hot and cold sides of the TEG.	[22,27]
Figure of Merit	$\mathcal{z}T={\displaystyle \frac{\sigma S^{2}T}{\kappa }}$	Where zT is the Figure of Merit, $\sigma$ is the electrical conductivity, T is the absolute temperature; and $\kappa$ is the thermal conductivity.	[23,28,29]
Thermal Efficiency	${\eta }_{TE}={\eta }_c{\displaystyle \frac{\sqrt{1+\mathcal{z}T}-1}{\sqrt{1+\mathcal{z}T}+1-{\eta }_c}}$	${\eta }_{TE}$ is the thermoelectric thermal efficiency, ${\eta }_c$ is the Carnot efficiency; and zT is the Figure of merit.	[22,27,28]
Carnot Efficiency	${\eta }_c={\displaystyle \frac{T_h-T_c}{T_h}}$	Where $T_h$ is the temperature of the hot side, and $T_c$ is the material of the cold side.	[22,27]
Heat transfer through high temperature contact	$Q_h=-{\displaystyle \frac{I^2\rho l}{2A}}+{\displaystyle \frac{\kappa A\left(T_h-T_c\right)}{l}}+SI{T}_h$	Where A is the cross-sectional area, l is the length of the leg, $\rho$ is the resistivity, and I is the current.	[22,27,29]
Heat transfer through low temperature contact	$Q_c={\displaystyle \frac{I^2\rho l}{2A}}+{\displaystyle \frac{\kappa A\left(T_h-T_c\right)}{l}}+SI{T}_c$	No new term introduced	[22,27,29]
Fill Factor	$FF={\displaystyle \frac{A_P-A_n}{A}}$	Where $A_P$ is the cross-sectional area of the p-type leg, $A_n$ is the cross-sectional area of the n-type leg, A is the total device area, And FF should be less than or equal to 1.	[22,28]
Optimized TEG Output power	$P_{out}={\displaystyle \frac{{FFS}^2{(T_h-T_c)}^2}{4\rho l}}$	Where FF is the Fill Factor, and The external load resistance is equal to the internal electrical resistance of the TEG.	[22]

Agricultural use cases of thin-film TEGs include plant-wearables for leaf and stem monitoring, canopy-level installation for microclimate sensing, stem-clipping for xylem sap analysis, soil-embedding for soil nutrient and mineral monitoring, and greenhouse TEG arrays for indoor cultivation environment monitoring (Table 3). Conventionally, TEGs are only used to extract and generate energy in the form of electrical signals because of thermal difference between the environment and the object surface where it is attached with. However, TFTEG technologies are advanced up to using it for extracting bioelectrical signals from living organisms such as plants [3,30]. Flexible thin-film TEG with a composite material of Bi₂Te₃ is attached to the lower leaf surface using soft polymer adhesive. It is connected to low-power transpiration rate sensor and has changeable Bluetooth and LongRange (LoRa) transmitter for intermittent data bursting to transmit the plant data collected to a Cloud server [3,31]. It is then imperative that improving the thermo-electric, mechanical and chemical properties of a TFTEG should be prioritized in a data and plant-centered food system.

Table 3. Comparisons of applications of thin-film TEGs in agriculture.

Use Case in Agriculture	Application Area	Mode of Integration	Mass per Unit Area (g cm−2)	Advantages	Limitations	References
Plant-wearable TEGs	Leaf and stem monitoring	Laminated to leaf surface or stem	Low (0.002–0.01 g cm−2)	Low-mass, non-invasive, monitors plant physiology, battery-free	Limited by small temperature gradient, sensitive to moisture and adhesion issues	[3,6]
Canopy-level TEGs	Microclimate sensing	Suspended or arrayed across canopy levels	Low–Moderate (0.003–0.02 g cm−2)	Enables distributed, energy-autonomous sensing, scalable in deployment	Complexity in installation, shading effects on plants	[3,31]
Stem-clipped TEGs	Xylem sap analysis	Clipped to thicker stems or fruit peduncles	Very Low (0.001–0.005 g cm−2)	Close contact with vascular flow, real-time monitoring of plant water transport	Limited stem area for mounting, potential mechanical stress on plant	[3]
Soil-embedded TEGs	Soil nutrient and mineral monitoring	Buried in topsoil or exposed to surface air near the soil	Moderate (0.005–0.03 g cm−2)	Applicable to smart irrigation, passive energy from natural thermal gradients	Soil compaction, moisture interference, possible corrosion	[30]
Greenhouse TEG arrays	Indoor cultivation environment monitoring	Mounted on panels or heat sources	High (0.01–0.05 g cm−2)	Utilizes controlled gradients, consistent in daily cycles, clean integration	Seasonal variation, need for optimized placement	[2,30]

Table 3. Comparisons of applications of thin-film TEGs in agriculture.

Use Case in Agriculture	Application Area	Mode of Integration	Mass per Unit Area (g cm−2)	Advantages	Limitations	References
Plant-wearable TEGs	Leaf and stem monitoring	Laminated to leaf surface or stem	Low (0.002–0.01 g cm−2)	Low-mass, non-invasive, monitors plant physiology, battery-free	Limited by small temperature gradient, sensitive to moisture and adhesion issues	[3,6]
Canopy-level TEGs	Microclimate sensing	Suspended or arrayed across canopy levels	Low–Moderate (0.003–0.02 g cm−2)	Enables distributed, energy-autonomous sensing, scalable in deployment	Complexity in installation, shading effects on plants	[3,31]
Stem-clipped TEGs	Xylem sap analysis	Clipped to thicker stems or fruit peduncles	Very Low (0.001–0.005 g cm−2)	Close contact with vascular flow, real-time monitoring of plant water transport	Limited stem area for mounting, potential mechanical stress on plant	[3]
Soil-embedded TEGs	Soil nutrient and mineral monitoring	Buried in topsoil or exposed to surface air near the soil	Moderate (0.005–0.03 g cm−2)	Applicable to smart irrigation, passive energy from natural thermal gradients	Soil compaction, moisture interference, possible corrosion	[30]
Greenhouse TEG arrays	Indoor cultivation environment monitoring	Mounted on panels or heat sources	High (0.01–0.05 g cm−2)	Utilizes controlled gradients, consistent in daily cycles, clean integration	Seasonal variation, need for optimized placement	[2,30]

3. Chemistry of Thin-Film TEGs

3.1. Material Classification of Thin-Film TEGs

The performance of thin-film TEGs is fundamentally governed by the intrinsic properties of the materials used. Often used for wearables and miniature energy harvesting applications, thin-film TEGs require materials with a good thermoelectric performance, possessing a large Seebeck coefficient (S), a high electrical conductivity (σ), and poor thermal conductivity (κ) [32,33]. In this section, thermoelectric materials are classified based on three major categories: inorganic, organic, and hybrid inorganic-organic thin film materials, where each class presents unique electrical, thermal, and mechanical characteristics that influence the design and application of thin-film TEGs, particularly in low temperature, low-power, and flexible electronics.

3.1.1. Inorganic Thin-Films

Inorganic TE materials are among the most common materials used in thermoelectric research due to their high-power factors and well understood transport mechanisms. These materials generally possess excellent electrical conductivity and Seebeck coefficients. However, these materials in most cases suffer from poor mechanical flexibility and brittleness [34].

• Silicon-Based Materials

Banking on its relative abundance and reasonably inexpensive, silicon and its derivatives as thin-film materials have garnered popularity due to their integrated systems applications, and compatibility with complementary metal oxide semiconductors (CMOS) and microelectromechanical systems (MEMS) fabrication [35]. Although bulk silicon is a poor thermoelectric material, nanostructuring reduces lattice thermal conductivity by enhancing phonon scattering, while suitable doping mechanisms preserves electrical conductivity, enhancing the Seebeck coefficient. A p-type Si_0.4Ge_0.6 and n-type Si_0.85Ge_0.15 thin film TEG was synthesized on polyimide utilizing silver induced layer exchange and magnetron sputtering [35]. The resulting device could generate high power factors, at low process temperatures and at a 30 K temperature gradient. Other recent efforts to overcome silicon’s poor thermoelectric performance utilized nanocrystalline composites utilizing p-type amorphous silicon (a-Si) dispersed with ultrafine nanocrystals utilizing boron implantation and annealing. Power factor (PF) greater than 1 mW m⁻¹ K⁻² due to carrier mobility enhancement from bimodal crystal distribution has also been demonstrated [36]. In a similar study, a planar microTEG using doped polysilicon, was able to achieve 12.3 μW/cm² at a temperature gradient of 31.5 K [37].

• Chalcogenide Materials

Chalcogenide materials are composed of at least one element from group sixteen of the periodic table. Chalcogenide thin films are most often composed of tellurium, such as bismuth telluride Bi₂Te₃, lead telluride, tin telluride, among others. These alloys, particularly Bi₂Te₃, are recognized as state-of-the-art for room temperature TE applications. The nanostructuring of Bi₂Te₃ consists of closely packed layers of cations and anions, with an anisotropic transport dependent on the direction of heat and electron flow [38]. Bi₂Te₃ superlattices have achieved record ZT values by combining quantum confinement effects with enhanced phonon scattering at interfaces. A recent study presented a promising flexible thermoelectric device with excellent thermoelectric performance, without compromising flexibility and stability [39]. The device features 162 pairs of thin-film legs with high room-temperature performance, using p-Bi_0.5Sb_1.5Te₃ and n-Bi₂Te_2.7Se_0.3, with figure of merit (ZT) values of 1.39 and 1.44 respectively. The use of titanium contact layers enhances power generation and supports the outstanding flexibility of the device. The produced TEG was capable of outputting power of 225.4 μW without the need for a heat sink or booster. Similarly, Ref. [40] fabricated thin film f-TEGs utilizing dispenser printing of a p-type Bi_0.45Sb_1.55Te₃ Se_0.034 and n-type Bi₂Te_2.7Se_0.3, achieving a PF of 23.2 and 19.1 μW/cmK² respectively, which is among the highest among printed Bi-Te systems on PI substrates. The modular TEG system fabricated in the study could power an LED and charging a NiMH battery. Ref. [10] introduced a wavy-structured thin-film TEG leveraging the high performance of sputtered (Bi,Sb)₂Te₃ films. The 36-leg device with optimized geometry achieved a maximum output power of 1.2 mW under a 50 K gradient. In another study by [41], the authors proposed a Tubular Thin Film Thermoelectric Generator (TTTEG), which is composed of flexible TEGs rolled into a tube, able to maximize temperature gradients when partially immersed in heat sources. The TEG utilized p-type Bi₂Te₃ and n-type Sb₂Te₃ fabricated using magnetron sputtering into a PI substrate. The TEG operated effectively in temperatures of 35 to 80 °C, with a maximum power output of 306.8 nW at a temperature gradient of 20 K.

• Carbon Nanomaterials

Carbon nanotubes (CNTs), graphene, and reduced graphene oxides are being studied to a greater extent especially in their use for hybridization [42]. Nanostructuring with carbon nanomaterials typically enhances electrical conductivity, and mechanical robustness, while offering tunable thermal transport. A prime example of this is the synthesis of a Bacterial Cellulose Carbon Nanotube (BC/CNT) composite [43], capable of generating an output power of 14.5 nW, with a temperature gradient of 20 K. A particular highlight of this study was sustainability and biocompatibility of the film, lacking toxic solvents, and being able to enzymatically decompose. Another study compares the performance of amine functionalized multiwalled carbon nanotubes (MWCNT-NH₂), and metal oxide nanoparticles (CuO, NiO, and Fe₂O₃) [44]. The screen-printed TEG, specifically that of the NiO metal oxides, was able to produce a maximum power output of 1.44 nW and a PF of 0.48 nW. Alongside these materials, graphitic carbon nitride (g-C₃N₄) has garnered interest for its tunable semiconducting properties and ease of synthesis, making it a promising candidate for integration into future thermoelectric hybrids [45].

• Other Inorganic TF Materials

Other inorganic material systems include transition metal oxides, half-heusler compounds, and skutterudites, which offer alternative mechanisms for thermoelectric and chemical stability at high temperatures. These materials, however, look promising only at high operating temperatures and temperature gradients (>100 °C) with their integration in thin-film form being far more complex and less studied than chalcogenides [33].

3.1.2. Organic Thin-Films

Organic thermoelectric materials primarily consist of conducting polymers which offer compelling properties, such as mechanical flexibility, low-temperature and low-cost fabrication, biocompatibility, and solution processability, making them ideal for wearables, plant attachable sensors, or deployable field modules [46]. However, their relatively low power output and electrical conductivity remain major challenges for supporting higher energy devices. To improve viability, future research should emphasize balancing electrical performance with sustainability and cost-effectiveness

• Conducting Polymers

Aside from the unique advantages presented by organic thin films, polymer-based thermoelectric materials are known for their typically low thermal conductivity due to their highly disordered structures, making them attractive as TE materials. However, the generally lower electrical conductivity, and Seebeck coefficients, compared to their inorganic counterparts, make conducting polymers less popular in the current landscape [47]. Materials such as poly(3,4-ethylenedioxythiophene):poly(styrenesulfonate), (PEDOT:PSS), polyaniline (PANI), and poly(3-hexylthiophene) (P3HT), and polypyrrole (PPy) are among the most studied conductive polymers due to their ability to conduct electricity through delocalized π-electrons throughout the polymer. Chemical modifications such as doping and functionalization further enhance the thermoelectric properties of these materials. Beyond these commonly explored polymers, several other polymers are important contributors to advancing the thermoelectric field. Polyacetylene (PA), poly(p-phenylene vinylene) PPV derivatives, poly(2,7-carbazolenevinylene), and poly(3-methylthiophene) (PMeT) were early and emerging materials which highlighted the diversity of conjugated polymer backbones, with tunable thermoelectric performance through doping, side-chain engineering, and molecular ordering [48]. However, the integration of these materials into TEG devices, or systematic thin-films remain limited. Among the various conductive polymers, PEDOT:PSS is the most widely used and studied in the TE landscape due to its low thermal conductivity, moderate electrical conductivity, and solution processability [49]. PEDOT: PSS is composed of two ionomers, sulfonated polystyrene (PSS) and PEDOT, which are often pre- and post-treated to boost TE performance [50]. A TEG fabric was synthesized utilizing PEDOT:PSS-coated polyester fabric strips connected with silver wires [51]. TEG could generate a low power output of 12.29 nW at a temperature gradient of 75.2 K. In a more recent iteration of the study, TEG was able to generate a power output of 212.6 nW of power at the optimal temperature gradient of 74.3 [52]. This generation of fabric utilized n-type constantan wires which were able to double the Seebeck coefficient of the material.

3.1.3. Hybrid Thin-Films

Hybrid thin films have emerged as a promising strategy to address the potential shortcomings of conventional inorganic and organic TE materials. These materials combine the high electrical conductivity and Seebeck effect of inorganic materials with the low thermal conductivity, mechanical flexibility, and processability of organic materials, forming an enhanced composite or hybrid material, optimizing electrical conductivity, and thermal insulation properties. Typically, such materials are fabricated by embedding inorganic nanostructures within polymer matrices, resulting in optimized electrical and thermal properties. For instance, Ref. [53] developed a flexible n-type TE film by incorporating Cu-doped Bi₂Se₃ nanoplates into a polyvinylidene fluoride (PVDF) matrix. The resulting composite exhibited a PF of 103 μW/cm·K² under a 15 K temperature gradient. A similar study fabricated a lightweight, flexible, and environmentally friendly paper-based TEG using a honeycomb structure inspired by origami and kirigami techniques [54]. The device, which utilized printed Bi₂Te₃ coated with bacterial cellulose nanofibers, achieved a power output of 0.596 μW under a 55 K temperature gradient.

Another work highlighted graphitic carbon nitride (g-C₃N₄) as a promising additive for thermoelectrics [45]. Demonstrated a Graphitic carbon nitride composite with PEDOT:PSS films, which increased conductivity more than 60-fold over a 70 K temperature gradient. While this study evaluated thermoelectric properties of composite films rather than a complete TEG device, it underscores the potential of g- C₃N₄ as a low-cost, scalable filler for future hybrid thermoelectrics. More recently, a high-performance flexible TE composite film by embedding polyaniline (PANI)-coated Ag₂Se nanowires into a PVDF matrix have been reported [55]. This TEG generated a power output of 0.835 μW with a 30 K temperature gradient. These studies collectively highlight the potential of hybrid TE materials in developing efficient, flexible, and scalable energy-harvesting devices. Further advancement in hybrid TE materials is exemplified [9] where a spray-printed PEDOT:PSS/Bi₀.₅Sb₁.₅Te₃ composite film was fabricated. The composite demonstrated robust flexibility, retaining its electrical conductivity and Seebeck coefficient within 10% after 1000 bending cycles.

By integrating inorganic nanostructures into organic matrices, hybrid thin films have shown strong potential for agricultural thermoelectrics, as these composites can be printed to or coated onto flexible substrates, maintaining performance in dynamic environments. However, most devices still demonstrate low power outputs less than 1 μW under ambient environmental conditions < 50 K. Moreover, long-term durability under outdoor exposure is rarely tested. Addressing these operational gaps and challenges should be essential for the deployment of hybrid TEGs in agriculture.

3.2. Thin Film TEG Performance Evaluation and Practical Considerations

3.2.1. Economic and Scalability Considerations

While advances in thin-film thermoelectric materials have focused on optimizing electrical and thermal transport properties, large-scale deployment in agriculture and low-power electronics require consideration of cost and scalability. Conventional chalcogenide thin film TEGs such as Bi₂Te₃ and Sb₂Te₃, require scarce elements such as tellurium, driving production costs and availability. Transition metal oxides, half-Heusler compounds and skutterudites on the other hand rely on more energy-intensive synthesis and fabrication, raising costs. In contrast, organic thermoelectrics, carbon nanomaterials, and other composites benefit from solution-processable, and printing-based deposition, offering low fabrication costs, alongside the low material cost. Hybrid films, which leverage low-cost polymer matrices, combined with optimized inorganic fillers, may present a balance between material performance and economic feasibility, though uniform dispersion and reproducibility remain scale-up challenges.

3.2.2. Optical Activity and Light Material Interactions in Thin-Film TEG Materials

While thermoelectric performance remains the primary focus of thin-film TEG research, optical activity and light material interactions are also relevant, particularly in agricultural settings where the materials are exposed to sunlight and environmental light conditions. Inorganic chalcogenide materials such as Bi₂Te₃ and Sb₂Te₃ exhibit strong absorption in the visible to near-infrared spectrum, enhancing localized heating and temperature gradients [56]. Silicon based films, with their tunable bandgaps, and established optical engineering techniques, allow for partial transparency or selective absorption, optimized for agricultural use. Carbon nanomaterials, such as CNTs on the other hand have been found to exhibit strong light absorption in the broad light spectrum between 250 and 2500 nm [57]. Organic polymers such as PEDOT:PSS and PANI typically exhibit varied absorption bands depending on their doping level and morphology. PEDOT:PSS can be engineered to be highly transparent, while maintaining good electrical conductivity and thermoelectric performance [58]. The optical activity for PANI, and polythiophene derivatives typically correspond to the conjugated π-electron backbones, influencing absorption in the UV-visible region [59]. Hybrid films that embed inorganic structures within organic matrices capitalize on the strengths of both materials, displaying complementary absorption and scattering behaviors. The inclusion of carbon nanomaterials such as graphene and CNTs can introduce additional pathways for scattering and absorption [5]. While optical characterization is less frequently reported than electrical and thermal properties, emerging research indicates that optical activity can complement thermoelectric efficiency by modulating local temperature gradients and contributing to device design considerations such as transparency and aesthetics.

3.2.3. Performance Trends and Evaluation

Due to the nature of this review, thin-film thermoelectric generators were profiled on the basis of their compatibility with typical agricultural environmental conditions such as: Low grade heat utilization and low operating temperatures appropriate for agricultural settings such as greenhouses, compost heaps, and irrigation systems; Structural adaptability, concerning flexibility and stability, for the device to conform to irregular surfaces like soil, leaves, or greenhouse structures; And application relevance such as powering field sensors, or use as temperature or electrical measurement devices. While TE materials are often evaluated based on intrinsic properties such as Seebeck coefficient, electric and thermal conductivity, and the figure of merit (ZT), TEGs present the system-level integration of these materials into functional devices. As such, this review presents TEGs in Figure 3 and

Table 4. Comparison of various TEG materials based on primary material classifications.

Material Classification	p/n Materials	Substrate	Number of Pairs	Treatment	Operating Temperature (K)	Temperature Gradient (ΔT)	Output Power (μW)	PF (μW/cm·K2)	Fabrication Method	Tuning/Doping	Morphology and Thickness	Stability	Flexibility and Robustness	Ref
Silicon-Based	Si0.4Ge0.6/Si0.85Ge0.15	Polyimide (PI)	2 p-type, 1 n-type	SiO2	Room Temperature	30 K	0.45	560/390	Ag-induced Layer Exchange, Magnetron Sputtering	As-doped Ag for n-type	∼1000 nm	High	High	[35]
	Amorphous Silicon	None	None	dispersion with ultrafine crystals by annealing	Room Temperature	None	None	1000	Low pressure chemical vapor deposition	Boron ion implantation	None	None	None	[36]
	Si	None	None	None	313 K	31.5 K	12.3	None	deep reactive ion etching	None	None	None	None	[37]
Chalcogenide	Bi0.5Sb1.5Te3/ Bi2Te2.7Se0.3	Polyimide (PI)	162	None	Room Temperature	5 K	225.4	108.7	Masked Magnetron Sputtering, Flip-Chip Bonding Technique	None	2.5 μm	High	High	[39]
	Bi0.45Sb1.55Te3 Se0.034/ Bi2Te2.7Se0.3	Polyimide (PI)	8	Sintering treatment	Room Temperature	33 K	68	23.2/19.1	Dispenser Printing	None	100 μm	High	High	[40]
	Bi0.5Sb1.5Te3/Bi2Te3	Polyimide (PI)	15	None	Room Temperature	11	10.6	37.8	Magnetron Sputtering	None	1.4 mm	High	High	[10]
	Bi2Te3/Sb2Te3	Polyimide (PI)	16	None	353 K	20 K	0.3068	4.7	RF Magnetron Sputtering	None	1.5 μm	none	High	[41]
Conducting Polymer	PEDOT:PSS-coated polyester/Silver wires	polyester	None	None	300 K	75.2 K	12.29 nW	0.045	PEDOT:PSS coating of polyester fabric	None	520 μm	High	High	[51]
	PEDOT:PSS-coated cotton/Constantan	cotton	5 strips	None	300 K	74.3 K	0.2126	0.057	PEDOT:PSS coating of commercial cotton fabric	None	none	High	High	[52]
Carbon Nanocomposites	Bacterial Cellulose/Carbon Nanotube	Bacterial Cellulose	6	None	Room Temperature	20 K	0.0145	20	Bacterial Cellulose inoculation with K. xylinus	n-doping with polyethyleneimine (PEI), or tetramethyl-ammonium hydroxuide (TMAOH)	10 μm	High	High	[42]
	MWCNT-NH2/NiO	PET	15	None	Room Temperature	100 K	1.44 nW	0.48 nW	Screen Printing	Amine-Functionalization	49 μm	High	High	[44]
	MWCNT-NH2/CuO		15	None	Room Temperature	100 K	1.06 nW	none	Screen Printing	Amine-Functionalization	54 μm	High
	MWCNT-NH2/Fe2O3		15	None	Room Temperature	100 K	0.32 nW	none	Screen Printing	Amine-Functionalization	52 μm	High
Hybrid	PEDOT:PSS/Cu intercalated Bi2Se3:PVDF	Glass	None	None	Room Temperature	15 K	None	103	Drop-Casting	Cu-doping	10 μm	High	High	[53]
	Bi2Te3/Bacterial Cellulose Nanofiber Coating	paper	96	heat treatment	310 K	55 K	0.596	25.5	Printing	None	500 μm	HIgh	High	[54]
	PEDOT:PSS/Exfoliated g-C3N4 composites	glass	None	Ultrasonic homogenization	Room Temperature	70 K	none	419.7	Oxidative chemical polymerizaiton	none	none	High	None	[45]
	PEDOT:PSS/Bi0.5Sb1.5Te3 composite	Kapton Sheet	40	Post treatment processes	25	2–50	0.23–378 pW	0.23	Spray Printing	DMSO doping/NaOH dedoping	227 nm	High	High	[9]
	PANI/Ag2Se	PVDF	6	None	300 K	30	0.835	196.6	Solution Mixing and Drop Casting	Camphosulphonic acid (CSA), dodecyl benzene sulfonic acid (DBSA)	10 μm	High	High	[55]

which are assessed based on operating temperature, temperature gradient, output power, power factor, fabrication method, morphology, stability, and flexibility, noting characteristic material parameters, features, and substrate, relevant to agricultural use.

3.3. Fabrication and Processing Techniques

Beyond the choice of materials, an equally important parameter to consider is the design and deposition of thin films. Vital to the thermoelectric properties of a thermoelectric generator, such as the Seebeck coefficient, electrical conductivity, and thermal conductivity, the fabrication technique influences factors such as film thickness, crystallinity, grain morphology, and interfacial connectivity [60]. Inorganic, organic, and hybrid thin-film TEG fabrication can be broadly classified by physical and chemical deposition techniques (Figure 4, Table 5).

3.3.1. Physical Deposition Techniques

Most inorganic thermoelectric films are fabricated utilizing physical deposition techniques which offer precise control over film thickness, microstructure, and composition. Among these, physical vapor deposition (PVD) methods such as sputtering techniques, including radio frequency (RF) and magnetron sputtering, are techniques frequently used to deposit films of Bi₂Te₃ and other related alloys [35,61]. These methods enable the formation of dense and uniform films with well-defined microstructures due to the broad tunability of the process parameters. [39,41]. Beyond sputtering, thermal and flash evaporation methods also allow direct evaporation of bulk materials in vacuum onto substrates, able to fabricate high quality Bi₂Te₃ films with uniform thickness and crystallinity [11]. Meanwhile, pulsed laser deposition (PLD) had also demonstrated the ability to produce nanostructured films with confined grain sizes and low lattice thermal conductivity, enhancing thermoelectric performance seen in Bi₂Te₃ thin films [62]. For even greater structural control, molecular beam epitaxy (MBE) allows for atomic-layer control during deposition, capable of growing high quality epitaxial layers with engineered superlattice structures resulting in outstanding thermoelectric properties. Although MBE is limited by low throughput and high cost, it is indispensable for applications where high crystallinity and precise doping control are essential. In recent years, printing techniques such as inkjet printing, screen printing, and direct writing have gained traction as TE deposition methods for both organic and inorganic thin films due to their compatibility with low-cost, large-scale fabrication, and high throughput manufacturing [44,54,63]. Organic thermoelectric films such as PEDOT:PSS, are in most cases, compatible with physical deposition techniques that possess solution processability. Spin coating is frequently employed to produce uniform thin films with a controlled thickness, critical for ensuring a homogenous network of the polymer [64]. Drop-casting is another technique to form thicker films, up to several microns in thickness, which can yield thermoelectric devices that can adapt to mechanical requirements, especially in flexible devices [53,55,64]. Ultrasonic spray deposition (USD) or ultrasonic spray coating bridges the gap between vacuum and solution-based deposition techniques. By atomizing precursor solutions into uniformly sized droplets through high frequency ultrasonic vibrations, USD offers better control over droplet size and distribution for the fabrication of fine-tuned film morphology and porosity in flexible and wearable thin-film TE devices [65,66]

3.3.2. Chemical Deposition Techniques

Chemical deposition techniques can offer distinct advantages over physical methods, notably in enabling low processing temperatures, higher compatibility with flexible or temperature-sensitive substrates, and potential solution-processability. Chemical vapor deposition (CVD), including variants such as metal-organic CVD (MOCVD), and low-pressure CVD (LPCVD), are among the most widely used chemical deposition methods. These techniques have been successfully applied to deposit Cu-based sulfides and chalcogenide thin films, where recent advances in single-source precursors (SSPs) have enabled the deposition of films with improved stoichiometry, conformality, and reduced organic contamination, even on complex substrates [36,67]. For applications that require ultrathin films and precise thickness control, atomic layer deposition (ALD) along with plasma enhanced ALD, has emerged as a powerful technique by self-limiting surface reactions, allowing angstrom-scale control over film growth, even on high-aspect-ratio structures. This method is valued for fabricating high-quality films at lower temperatures, especially for flexible electronics. Additionally, molecular layer deposition (MLD) extends the principles of ALD to hybrid organic-inorganic materials, enabling the fabrication of nanolaminates and interface-engineered architectures that enhance phonon scattering and carrier mobility [68]. Electrodeposition is a low-cost and scalable approach that has been widely used for thermoelectric films such as Bi₂Te₃ and Sb₂Te₃ [69]. By applying galvanostatic or potentiostatic control of an electrochemical cell, the process can produce films with a controllable composition and morphology. Further variations such as pulsed electrodeposition can enhance film uniformity and grain structure [70]. Although organic TE films are often associated with solution-based physical techniques, chemical deposition methods have been recognized for their ability to achieve controlled doping and precise molecular ordering. Vacuum thermal evaporation and in-situ polymerization have been recognized as organic chemical deposition methods, capable of the deposition of organically doped semiconductors with tunable doping levels and structural properties [71].

Table 5. Comparison of advantages and disadvantages of the various fabrication techniques of TEG.

Technique	Category	Advantages	Disadvantages	Ref.
Sputtering	Physical	Uniformity, adhesion, control over thickness, scalability	Requires high vacuum, expensive equipment, low throughput	[35,41,61]
Evaporation		Simple setup, fast deposition, good crystallinity, low cost	Poor Coverage, limited uniformity	[11]
PLD/MBE		Nanostructuring, high control and purity, good interface	Small deposition area, low throughput, high cost	[62]
Spin Coating		Uniformity, ease of control, low-cost and fast	Flat substrates only, solvent waste, Film cracking in thick layers	[64]
Drop-Casting		Simplicity, low-cost, thick film deposition, suitability	Poor thickness control, non-homogeneous morphology	[63]
Printing		Digital, low-waste, scalable, low cost	ink-formulation challenges, resolution-dependence, requires post processing	[53,55,64]
USD/USSC		Uniformity, controlled porosity, waste reduction, scalable, versatile	Liquid delivery challenges, optimization complexity, material limitations,	[65,66]
CVD	Chemical	Conformational coating, good compositional control	High-temp processing, toxic precursors	[36,67]
ALD/MLD		High thickness control, excellent uniformity and conformality	slow growth rate, high precursor cost	[68]
Electrodeposition		Low-cost and scalable, compositional and morphological control	Substrate-dependent, film stress and adhesion issues	[69,70]
In-situ Polymerization		Precise molecular ordering, controlled doping	limited choices for monomers, long processing time	[71]

Table 5. Comparison of advantages and disadvantages of the various fabrication techniques of TEG.

Technique	Category	Advantages	Disadvantages	Ref.
Sputtering	Physical	Uniformity, adhesion, control over thickness, scalability	Requires high vacuum, expensive equipment, low throughput	[35,41,61]
Evaporation		Simple setup, fast deposition, good crystallinity, low cost	Poor Coverage, limited uniformity	[11]
PLD/MBE		Nanostructuring, high control and purity, good interface	Small deposition area, low throughput, high cost	[62]
Spin Coating		Uniformity, ease of control, low-cost and fast	Flat substrates only, solvent waste, Film cracking in thick layers	[64]
Drop-Casting		Simplicity, low-cost, thick film deposition, suitability	Poor thickness control, non-homogeneous morphology	[63]
Printing		Digital, low-waste, scalable, low cost	ink-formulation challenges, resolution-dependence, requires post processing	[53,55,64]
USD/USSC		Uniformity, controlled porosity, waste reduction, scalable, versatile	Liquid delivery challenges, optimization complexity, material limitations,	[65,66]
CVD	Chemical	Conformational coating, good compositional control	High-temp processing, toxic precursors	[36,67]
ALD/MLD		High thickness control, excellent uniformity and conformality	slow growth rate, high precursor cost	[68]
Electrodeposition		Low-cost and scalable, compositional and morphological control	Substrate-dependent, film stress and adhesion issues	[69,70]
In-situ Polymerization		Precise molecular ordering, controlled doping	limited choices for monomers, long processing time	[71]

4. Biocompatibility and Eco-Stability of Thin-Film TEGs in Agricultural Environments

4.1. Interaction with Plant Tissues and Soil

Although literature on thin-film thermoelectric materials designed for agriculture applications is limited, some of both organic and inorganic materials have been used, usually as sensors, for agriculture applications. Despite the outstanding characteristics of inorganic TE materials, most discussed in this section lacks information in terms of application and compatibility in agricultural settings. Nevertheless, their concepts may be adapted for agricultural applications. SWNT-based nanosensor for mechanical damage detection embedded on leaves of several plant species of different genus namely spinach, strawberry blite, lettuce, sorrel, arugula and Arabidopsis, proved non-destructive [70]. A blend of chitosan and PANI nanocomposites carrying biopesticides loaded on tomato plants has shown biocompatibility in terms of physiological aspects and growth when tested in vivo and in greenhouse conditions [71]. Also more recently, PEDOT:PSS integration on plant sensors for in vivo monitoring of growth and stress of crops such as rice, wheat and tomato demonstrated biocompatibility and posed minimal destruction [72,73]. Most of the inorganic TE materials are potentially toxic. Elemental lead and aluminum are toxic to the environment and humans. Aluminum, although abundant in the earth’s crust, when present in toxic levels makes soil acidic and interferes with nutrient absorption by plants. Though toxicity heavily depends on concentration and duration of exposure, a low concentration of lead, for instance, has been found to interfere with plants’ photosynthetic and metabolic processes by causing adverse effects in the cellular level. It is also dangerous to marine life when leached in the soil and carried to bodies of water [74]. A life cycle assessment of non-oxide bismuth telluride alloys and oxide-based lanthanum-doped strontium titanate and calcium cobaltite was conducted to find out the environmental impacts of the TE materials. Considering the acidification potential, eutrophication potential, terrestrial toxicity and human toxicity, the TE materials have negligible impact (<2%). TE materials namely tellurium, and antimony have a slight impact reaching a maximum of 8%, and 20% respectively. Cobalt oxide, on the other hand, has a considerably high freshwater sediment toxicity of around 30% [31]. Bismuth chalcogenides, gold and silver are also known to affect microbial communities. Selective antagonism of a certain organism was not demonstrated but it has been demonstrated to possess antimicrobial properties against gram negative and gram-positive bacteria. The effect was explained to be due to the charge differences between the nanomaterials and the microorganisms’ cell wall and the production of reactive oxygen species (ROS) by the nanomaterials in the presence of light [65]. The same ROS by Pb in response to external factors impose adverse effects on plant cells [74]. Hence, conducting polymers are being explored for its more environmentally friendly properties.

4.2. Environment-Responsive Degradation and Encapsulation

The open agricultural field is exposed to harsh conditions such as fluctuating temperature, relative humidity, precipitation, wind, light and other uncontrolled factors. For instance, temperature fluctuation results in a minimum and unstable temperature gradient challenging the efficiency of a TEG to generate power [6]. Simultaneously, humidity and precipitation may hasten degradation and affect the performance of a device. This also challenges the capacity of wearable devices to adhere and stay on plant surfaces. Apart from all the environmental factors, the plant characteristics and morphology such as rate of growth, waxy cuticles and presence of trichomes affect the adhesion ability of wearable devices [31]. These factors influence the performance and degradation of a device during deployment which make it an important consideration in choosing materials and design.

Chalcogenide-based thin-film can be designed into self-healing, recyclable and Lego-like reconfigurable TEG. A flowable liquid metal wiring with polyimine substrate and wire encapsulant makes it self-healable due to the bond exchange interactions of the polyimine. While it is self-healing, the materials may also be recycled by depolymerizing with amine monomers, and all the materials can be re-built into comparatively functional TEG devices. These characteristics are relevant when trying to address degradation due to mechanical movement by wind and other external factors. On the other hand, this is an environment-friendly option for inorganic materials that are not as biodegradable as those of nanocellulose-based to help reduce waste and conserve scarce materials [75,76].

Inorganic thin-film materials such as Bi₂Te₃ undergo thermal degradation, oxidation and sublimation. Dip coating Bi₂Te₃ with high-temperature polymers has been demonstrated to overcome these weak points and thus suitable for macro-scale applications [77]. To slow down the rate of TE degradation, coating them with glass, plastic and cellulose-based encapsulants somehow work to preserve their electrical properties. Encapsulants protect TE materials from deterioration due to temperature fluctuation, atmospheric oxygen and moisture permeation while enhancing its properties. Another inorganic TE material, CNT, demonstrated delayed oxidation from atmospheric oxygen with the use of glass encapsulant [78]. Further, coating them with a flexible substrate and layering them with nanoscale material significantly enhances their flexibility and thus prevents micro-cracks [43,79].

Nanocellulose substrates show superiority in terms of flexibility and biodegradability. However, its practical application is somehow limited due to its large number of hydroxyl groups making it susceptible to moisture degradation. Hydrophobic materials such as copper iodide were used to coat the nanocellulose making it moisture-resistant while retaining its flexibility [80].

Conducting polymers, such as PEDOT: PSS and PANI gained attention due to its relatively lower environmental adverse effects. However, its application in the open field is limited due to its loss of good mechanical and electrical conductivity as it interacts with oxygen, moisture and UV light in the environment [81]. The hygroscopic characteristic of PEDOT:PSS was previously addressed by encapsulating it with a blend of functionalized graphene oxide and polyvinylidene fluoride. The carboxyl and hydroxyl groups of graphite oxide are converted into ester groups to make it hydrophobic. This process imparted humid-resistance and stability at negative temperature range to the PEDOT:PSS [82]. Evidence proved how encapsulation imparts negligible degradation of PEDOT:PSS at relative humidity of 0–75% and temperature changes from 4–40 °C making it possible to deploy in an enclosed growth chamber such as greenhouses [31].

Although plastic encapsulants are entirely impermeable, moisture eventually gets in overtime. Cellulose-based encapsulants such as ethyl cellulose and polyvinyl alcohol offer a degree of water resistance, flexibility and low carbon footprint. Its moisture resistance was proven by its consistent electric performance after submersion in water for a minute and its low contact angle in the surface with textile substrate. At the same time, it imparts relative thermal resistance of the material with 10% mass loss at 315 °C [83]. Figure 5 shows a comparison based on the inherent characteristics of TE materials as discussed in the previous sections.

5. Recent Advancements in Thin-Film TEG Design

5.1. Design Advancements in Thin-Film Thermoelectric Generators

Recent advances in thin-film thermoelectric generators have focused on improving their integration and efficiency through the interplay between materials and structural designs. A 3D flexible wavy-structured thin film thermoelectric generator was developed to harness the out-of-planar thin-film TEG temperature differences. Moreover, the said design has allowed the utilization of the vertical heat flow, which significantly improved the device output performance. This wavy-structured TEG has transcended other flexible thermoelectric generators roughly over five times as it exhibited a high normalized output power density due to its wavy architecture, 54.6 mV output voltage and a maximum output power of 697.4 nW [10]. To overcome the growing need for thick films, a planar thin film solar thermoelectric generator was designed and optimized. This planar thin film can be efficiently optimized since its thickness is different from the deposited film. Furthermore, conventional microfabrication techniques can be employed since the patterning of the n- and p-type thermoelectric layers can be easily executed. This design was found to be significant in thin-film thermoelectric generators where the increase of film thickness is nonviable [84,85,86].

5.2. Flexible and Stretchable Substrates for Conformal Applications and Integration with Flexible Printed Circuits and Biodegradable Electronics

Thin-film thermoelectric generators can be manufactured in an extensive array of substrates, including several types of polymers and even metals. Integrating these flexible and stretchable substrates enables adaptability to varying curvatures of surfaces and improves several key properties of the TFTEGs. Polyethylene terephthalate (PET) was utilized as a flexible substrate for CuI films for TFTEG applied in a wearable application. It was seen that even though PETs were not as stable as glass or mica substrates, which are inherently insulators, their lightweightness and durability were proven in a variety of flexible electronics applications [18]. CuI films were found to be well-bonded with PET flexible substrates. However, PET substrates have smaller amounts of p and S values for corresponding CuI films, indicating that PET flexible substrates possess a slightly lower thermoelectric performance compared to glass or mica [18].

In contrast, flexible polymer substrates are continually employed as substrates for thin-film TEGs to at least support inorganic thermoelectric materials. A polyimide substrate was used for thin-film TEG for energy harvesting applications. Chalcogenide films were deposited on a polyimide substrate and were found to provide high flexibility and mechanical strength under stretching [10]. When the flexible substrate was sonicated for two cycles, the deposition temperatures were 613 K, 633 K, and 653 K, respectively, inferring the capability of polyimide substrates to withstand high temperatures during deposition [10]. Polyimide substrates were also utilized for a multi-functional wearable thin-film generator where it displayed a relatively constant thermal conductivity of y of 0.15 W/mK between 300 and 550 K [11].

A flexible Kapton substrate was used to support a PEDOT:PSS/Chalcogenide composite, using a spray printing technique, which enhanced several properties, such as the electrical conductivity and the Seebeck effect [9]. Moreover, the use of Kapton substrate has also allowed for an excellent flexibility and mechanical stability of the thin-film after it has undergone 1000 cyclic bending cycles [9]. A stainless-steel substrate was used for a thin-film TEG due to its excellent anti-corrosion properties. An electrodeposition process was performed but a requirement is deposition on conductive substrates and the choice of stainless steel seemed beneficial. However, a major disadvantage includes the rising of unwanted electrical conduction from the stainless-steel substrate during the operation [87]. Temperature is considered a significant factor that affects the properties of the substrates, including some properties such as electrical conductivity and the Seebeck effect. Nevertheless, having these flexible and stretchable substrates allows for a more lightweight application of thin-film TEGs with enhanced electrical conductivity, temperature, and mechanical resilience.

Traditional bulky circuits limit their applicability in numerous applications, especially in wearable applications. Flexible printed circuits become increasingly important in several applications since they are designed to fit into complex geometries and are more mechanically resilient, which broadens their applicability. Moreover, biodegradable electronics are beneficial as they promote sustainability and reduce environmental threats. A biodegradable, flexible, thin-film thermoelectric generator was developed using nanocellulose substrates [88]. These nanocellulose were re-engineered from cellulose fibers to suit advanced applications in biosensors and different electronic devices. The CuI film was deposited into the nanocellulose substrate using the scalable SILAR method to obtain a lightweight and flexible biodegradable thermoelectric material composite CuI/NC. The thermoelectric material produced was non-toxic, promoting environmental friendliness, sustainability, and biodegradability [88].

5.3. Thin-Film Thermoelectric Generator Co-Design with Sensor Technology

Thin-film thermoelectric generators are increasingly integrated with sensors to drive self-powered systems for remote and continuous monitoring. The integration of TEGs with sensor technology reduces the need for external batteries and enables the creation of self-powered systems. An autonomous soil moisture sensor was powered by a thermoelectric generator, where the sensor utilized nanostructured thermo-sensitive resistors fabricated on the ceramic substrate. The system has been able to garner data in 5 days consecutively without the need for additional energy, which can be highly beneficial in monitoring soil health in agricultural applications [30]. It is also applied in thermal sensors for exoskeleton applications. Using chalcogenide thin films, these sensors can be able to monitor temperature changes in real-time. Moreover, the integration of thin-film thermoelectric generators has helped enhance the Seebeck effect and has low thermal conductivity, which is beneficial for efficient thermoelectric conversion [11]. Also, its integration allowed the maintenance of the temperature gradient, which is crucial for the operability, signal strength, and sensitivity of the device [11]. A multifunctional, flexible, thin-film thermoelectric device was engineered for self-powered light sensing aside from energy harvesting for small-scale applications. The co-design approach has demonstrated a maximum response of 4.89 V/cm² W⁻¹. Additionally, its light intensity responsivity has increased over time as the output voltage increased [89].

5.4. Artificial Intelligence in the Context of Thermoelectric Generators

Artificial Intelligence has been gaining traction in recent years in aiding in the execution of various tasks and processes, with the field of TEG use being one of them [90]. Artificial Intelligence can be further divided into different subsets such as Machine Learning which is focused on training a certain AI model through a given dataset to perform specified actions [84]. Machine learning, which is the more prevalent form of AI used for optimization of TEG performance using different neural networks [91]. Artificial neural networks are composed of what could be considered as digital neurons that take certain inputs and pass them onto other neuron layers in iterative processes that aim to optimize the parameters and eventually identify prominent values such as the thermoelectric figure of merit [91,92]. Neural networks also have other benefits in their use such as the general concept being simple for newcomers to analyze, them being more efficient than traditional data handling methods, and with them being useful in predictive models for thermoelectric performance [20].

There are three prominent types of neural networks, namely Feedforward Neural Networks (FNNs), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs). The feedforward neural network displays a unidirectional flow of information from input data to the predicted outputs after passing through a series of computational layers [91]. The convolutional neural network, as the name suggests, contains convolutional layers which apply convolutional operations to a set of input data which allows it to extract visual and spatial features, pooling and flattening them, and then applying them through several interconnected layers for processing to attain the desired output. making it ideal for processing input data such as images [91]. The last major neural network is the recurrent neural network, which utilizes feedback loops to retain information from previous calculations to assist in creating predictions and is ideal for use in sequential data analysis [91].

Generative Artificial Intelligence (GAI) is another subset of deep learning that in the discussion of design can aid in the construction of models and techniques such as the use of generative adversarial networks may aid in material selection [91,92]. GAI has been used for optimization purposes such as in the form of conditional generative adversarial networks which can be used to assist in geometric optimization and has shown effectiveness in surrogate modelling with limited case studies in the field of thermoelectric generators [20]. However, the lack of literature in its implementation in TEG design suggests that the focus of AI in TEG design is more on the optimization of the TEG parameters rather than on preliminary design and modelling. An overview of artificial intelligence and select subsets under machine learning can be identified in Figure 6 with the more prevalent AI tools falling under the different subsets of deep learning including GAI as well as deep and artificial neural networks [20,91,92].

5.5. Device Layout, Structural Optimization, and Data-Driven Multiphysics Simulation

As previously discussed, optimization is one area of interest for the implementation of AI in the field of TEGs. Beginning the discussion with the base optimization process without AI implementation, one study tackles the optimization of bulk TEGs based on a two-stage segmented configuration focused on extracting solar energy, and analyzing its performance through a 4E (Energy, Exergy, Environment, and Economic) analysis of its performance [93]. The energy analysis focuses more on parameters such as energy efficiency and the overall performance of the system whereas Exergy focuses more on the internal efficiency of the system considering aspects such as the thermodynamic processes occurring within the system [93]. The environment aspect addresses the impact of technology on the environment whereas the economic aspect addresses the economic viability of the TEG [93]. The optimization process outlined in this study first tackles the basic concepts and the equations and formulas that may be extrapolated for use during the optimization process as well as the necessary parameters to be identified (TEG output power and efficiency, entropy generated, carbon dioxide emission during operation, and dollar per watt generated) [93]. This is then followed by the creation of the model to be used. This study utilized Autodesk Inventor Professional 2024 for the creation of the conceptual model to be subjected to Finite Element Analysis in ANSYS 2023 [93]. The usual software used for simulation purposes is both ANSYS and COMSOL [92]. The created model was then validated using the mesh sensitivity analysis features offered by ANSYS [92]. Other aspects to be considered would include the boundary conditions for the model, the specific ones identified in this study being heat flux and transfer at the hot and cold junctions, as well as the establishment of the ground at the n-type leg and a terminal voltage at the p-type leg [93].

The thermal parametric optimization tackles the improvement of the energy system and in this case focuses on the identification of the key parameters related to the previously discussed 4E approach and which specific concentrated solar fluxes and heat transfer coefficients to yield the highest values [93]. Geometric optimization focuses on the different aspects of the TEG geometry (thermoelectric leg height, cross-sectional area, and material composition such as skutterudite) and identify the ideal dimensions to maximize TEG’s overall performance [93]. The iterations were then presented in the form of graphs for ease of analysis.

Other studies tackling parametric optimization involved the optimization of a silicon-base nano thermoelectric generator for maximized performance for waste heat recovery purposes [94]. The numeric model based on steady state calculations was created through COMSOL and the 3D model used for the study was verified through comparing its COMSOL-calculated performance with studies that created similar TEGs and compared their performance [94]. The model conditions specified were the material thermal conductivity, the Seebeck coefficient, the electrical resistance, the contact resistance, the doping concentration, and the temperature applied to the hot and cold sides of the TEG, Whereas the key parameters optimized in this study are the dimensions of the blade thermocouples together with their spacing parameters [94]. It was eventually identified that the ideal parameters involved a thermocouple thickness of 2 μm and a width of 60 nm with an identified specific power density of 16.2 μW/(cm²·K²) aiming for optimal performance while maintaining structural integrity [94].

Figure 7 outlines the general flow for the optimization process starting with the necessary modelling in software such as Autodesk, ANSYS, and COMSOL [20,93,94,95]. The key equations will then need to be identified for the calculations done in the optimization process. The model will then need to be verified to ensure its viability for use through either comparison with other literature or through verification through an experimental procedure. Finally, the optimization process will be implemented where the necessary parameters or geometric dimensions are optimized, presented in graphical form, and analyzed.

Aside from conventional optimization techniques, inverse optimization of a thermoelectric material can also be done in the determination of key parameters [95]. The general flow is still the same beginning as the modelling of the TEG however a dimensionless parameter was set to assist with the sensitivity analysis of TEG performance [95]. An experimental set-up was then created to verify the model’s validity with a maximum perceived error of approximately 4.9% [95]. The inverse optimization model for the TEG had the maximum power generation as the dependent variable while optimizing the values of thermal conductivity, the Seebeck coefficient, and electrical conductivity with the optimization continuing in its iteration until the goal of improving TEG performance has been achieved [95]. The study provided interesting conclusions that for their chosen boundary conditions similar to real-life applications, the figure of merit was not effective as an evaluation method to investigate the performance of the TE as the sensitivity of the physical parameters changed for the material at constant figure of merit [95]. Other key findings identified were the importance of thermal conductivity in attaining maximum power generation within a specific range of figure of merit values with inverse optimization showcasing its usefulness in identifying the weight of specific parameters in optimizing material performance [95].

5.6. Surrogate Modeling of TEG Device Performance

Another application of AI in optimization is using surrogate modelling wherein a simpler model is created to emulate the behavior of more complex models. One such example involves the optimization of an Annular TEG (ATEG) device but also conducted surrogate modelling using artificial neural networks and conditional generative adversarial networks [20]. In preparation for optimization, key equations were identified representing the following aspects of the TEG model: the flow fields for both fluid and solid zones, and the coupled thermoelectric equations representing both thermal and electrical fields [20].

The comprehensive surrogate model is composed of the ANN for predicting the performance of the TEG based on the one-dimensional data provided by the model created in COMSOL and the conditional Generative Adversarial Network (cGAN) model which generated the visualized design based on both the two-dimensional image and the design parameters provided by the ANN analysis [20]. The physical field distribution generated from the cGAN model is returned to the ANN model for additional data processing with the two results re-optimized and revalidated in COMSOL, with this process iterated until the ideal parameters have been achieved [20]. The necessary dataset for training the ANN model was extracted from the numerical model of the ATEG in COMSOL and based on five key input parameters (twist ratio, radius, length, inlet temperature, and inlet velocity) and trained in Matlab’s Neural Network Toolbox, whereas the pressure and heat flow distributions were processed into 256 × 256 × 1 images for the cGAN model [20].

Regarding the cGAN model, it was trained using PyTorch in Python two convolution neural networks were implemented: a generator (G) for image compression into a vector for analysis and eventually generate the corresponding pressure and heat flow distribution and a distributor (D) which is a convolutional PatchGAN classifier which differentiates the fake image created by the generator and the real image created from COMSOL [20]. The time taken for the cGAN model to process a sample size of 200 is 11.5 h and through the analysis of the mean square error and structural similarity of the samples led to the identification of 150 samples for the training of the cGAN model [20]. The overall results of the TEG optimization showcased a prediction accuracy of the ANN model exceeding 97% and computational efficiency of 99.97% and SSIM values for the pressure and heat flux distributions being 0.954 and 0.934 for the cGAN model with a physical field prediction and visualization time of 5 s, with the overall optimization leading to an improvement of 22.07% and 59.11% of the TEG output power and efficiency respectively [20]. Overall, this study outlined the effectiveness, speed, and reliability of using AI tools such as neural and adversarial networks in TEG thermal modelling, parametric design and optimization, although time and effort is needed to prepare and train the necessary datasets for the AI models [20].

Multi-objective optimization was implemented in one study, utilizing a transient model exploring the impact of amplitude, frequency, and heat transfer coefficient, resulting in an investment reduction by up to 50% with only minimal losses of power and efficiency of approximately 1% [19]. Once the boundary conditions have been properly established, the model has been validated through actual experimentation and then subjected through multi-objective optimization based on a cost-function and the selected weight factors: output power, efficiency, and investment [19].

Another example in the implementation of multi-objective optimization is in a combined photovoltaic thermoelectric generator system with phase change material and microchannel heat pipe integrated components optimized through a Non-dominated Sorting Genetic Algorithm II (NSGA II) alongside Multi-Objective Particle Swarm Optimization (MOPSO) with the results weighted through entropy weighting methods and finally run through a Technique for Order Preference by the Ideal Solution (TOPSIS) model [96]. The key parameters that were optimized were as follows: photovoltaic reference efficiency, TEG module quantity, external and internal PCM plate thickness, and the external and internal Phase Change Material’s melting temperature [96].

5.7. AI-Assisted FEM

Finite element modelling (FEM) is an important part of the design process. However, existing literature is limited to how AI can be used as an assistive tool to minimize research effort in the field of thermoelectric generators. Examples do exist in other fields and as this focuses on the modelling aspect and not the concept of thermoelectrics, it is possible to consider the AI-assisted FEM methods utilized in different types of equipment or items. The study by Javadi et al. [97] explored the implementation of an evolutionary polynomial regression-based constitutive model (EPRCM) in comparison to standard FEM models and shown that the proposed model can serve as a possible alternative to conventional methods when used to analyze simple case studies on geometric models. In the field of healthcare, Quantitative Computed Tomography-based Finite Element Analysis (QCT-based FEA) was utilized in a machine learning-based surrogate modeling process for the prediction of hip fracture risk and for the identification of possible fracture location [98]. The dataset was acquired from the Digital Imaging and Communications in Medicine image dataset where CT scans from affected patients were collected, the 3D femurs were then reconstructed through image processing using 3D Slicer which was exported in STL format for the meshing with 4-node tetrahedral elements using HyperMesh, the model was then further manipulated with the use of Bonemat v3.0, the specification of key boundary conditions, the utilization of linear FEA to assist in determining the behavior of the model under certain conditions [98]. A categorical boosting (catboost) algorithm was utilized for the actual prediction for risk of hip fracture which was able to perform effectively with a limited dataset [98].

It can be noted that the degree of complexity in which the previously mentioned femur models can be attributed to the different levels of complexity of the bones of the human body and the finite element analysis of TEGs would likely be a simpler process.

Table 6. Comparisons of optimization techniques applied in TEGs.

Software Used	AI Algorithms/Tools Used	Boundary Conditions/Input Parameters	Parameters to Be Optimized	Optimization Results	Reference
ANSYS 2023 R1; Autodesk Inventor Professional	No AI tools implemented	1. Defined heat flux to hot junction. 2. Defined heat transfer coefficient at cold junction. 3. Ground at n-type semiconductor and terminal voltage determined at p-type semiconductor.	1. Dollar per Watt Value 2. Maximum power output 3. CO2 savings 4. Exergy efficiency 5. Thermoelectric height 6. Thermoelectric cross-sectional area 7. Skutterudite Content	1. Economic Optimal operation attained at a Dollar per Watt Value of 0.16 US$/W at Skutterudite composition of 92.59% and a concentrated solar flux of 40 kW/m2. 2. Optimal power generation of 41.12 W and maximum CO2 savings of 19.5 kg/yr attained at a concentrated solar flux of 95 kW/m2 and heat transfer coefficient of 4 kW/(m2K). 3. Peak exergy efficiency of 9.52% attained at a concentrated solar flux of 20 kW/m2 at a cross-sectional area of 0.1 mm2 and concentrated solar flux of 10 and 20 kW/m2. 4. Optimal leg height and cross-sectional area varies based on ideal parameters to be optimized with a leg height of 7.5 mm ideal for 2.18 W power output.	[93]
COMSOL Multiphysics	No AI tools implemented	Model considerations: 1. Seebeck coefficient, p- and n-type doping concentrations, contact resistance, and temperature applied to both ends specified. 2. Thermal conductivity and electrical resistance to be calculated.	1. Power density 2. Geometric parameters (Length, width, and height of blade thermocouples)	Maximum power density of 16.2 μw/(cm2K2) achieved with a thermocouple thickness of 2 μw and a width of 60 nm.	[95]
Unspecified	No AI tools implemented	1 Constant heat flow at the hot side. 2. Constant temperature on the cold side.	1. Maximum power generation 2. Optimal matching external load. Impact of the following parameters considered: 1. Thermal conductivity 2. Seebeck coefficient 3. Electrical conductivity	A range of parameters for specific figure of merit values was identified with the key result focused on a flowchart of the TE inverse optimization process at the device level. Formulas for identifying maximum power were generated based on thermal conductivity which varied based on the figure of merit.	[96]
COMSOL Multiphysics	1. ANN (Matlab Neural Network toolbox) 2. cGAN Trained using PyTorch in Python	Value ranges and resolutions for the following geometric parameters were identified: 1. Twist ratio = 30–110 with a resolution of 20. 2. Radius = 0.01–0.03 m with a resolution of 0.005 m. 3. Length = 0.017–0.097 m with a resolution of 0.016 m. 4. Inlet temperature 5. Inlet velocity	1. Maximum power output 2. TEG efficiency	1. Prediction accuracy = 97% 2. SSIM of pressure distribution = 0.954 3. SSIM of heat flux distribution = 0.934 4. Net power increase = 22.07% 5. Efficiency increase = 59.11% Computational efficiency achieved = 99.97%	[20]
Unspecified however a multi-objective optimization software was used with response surface methodology.	No AI tools implemented	1. Function of amplitude and frequency as input heat. 2. Heat transfer coefficient as input at cold end. 3. Seebeck effect, Peltier effect, and Thomson effect considered.	1. Maximum output power 2. TEG system efficiency 3. Investment	1. Maximum output power loss of 1%. 2. Efficiency loss of 1%. 3. TEG investment reduction of 30–50%	[19]
MATLAB	NSGA II-MOPSO algorithm alongside other analysis tools such as the TOPSIS method	Sensitivity parameters: 1. PV reference efficiency 2. TEG module quantity 3. Thickness of each of the two PCM plates 4. Melting temperature of each off the two PCM plates.	1. Electrical Efficiency 2. Minimum life cycle cost	1. Average annual efficiency increased by 52.4% 2. Total life cycle costs decreased by 98.4% Optimal configuration identified 1. Electrical efficiency of 25.6% 2. Total life cycle cost of 335.4 CNY Parameters for optimal configuration: 1. PV reference efficiency of 26.4% 2. TEG module quantity of 25.6% 3. Optimal thickness of 21.6 mm and 2.8 mm for external and internal PCM plates respectively. 4. Melting temperatures of 30.8 °C and 11.4 °C for external and internal PCM plates respectively.	[96]
MATLAB/SIMULINK	Generalized Regression Neural Network (GRNN) combined with FDO	1. Hot side temp = 250 K at matched load 2. Cold side temp = 50 K at matched load TEG module parameters: 1. Power output = 24.3 W 2. Open-circuit voltage = 10.8 V 3. Load voltage = 5.4 V 4. TEG internal resistance = 1.2 Ω 5. Load current = 4.5 A	Global Maximum Power Point (GMPP)	Global Maxima tracking efficiency > 99%. Time taken to trace true GMPP = 110.1 ms. GRNN with FDO compared to other methods such as Cuckoo-search algorithm (CSA), Perturb and observe algorithm (PandO), Particle swarm optimization (PSO), and Grasshopper optimization algorithm (GHO) and was found to display the finest results with minimal tracking time.	[99]
ANSYS COMSOL Multiphysics	ANN Genetic Algorithms (GA)	1. Cold side temperature = 300 K 2. Convectional heat flux on all open internal surfaces with heat transfer coefficient of 1 mW/(cm2K). 3. TEG model connected to external load. 4. Inlet and outlet of metal substrate as terminal voltage and ground respectively.	1. Fill Factor 2. Maximum output power density 3. Thermoelectric height 4. Interconnect height 5. Width of n-type leg 6. Width of p-type leg	Values achieved at operating conditions of heat flux density = 300 mW/cm2 and contact resistivity = 10−8 Ωm2 1. FF = 0.11. 2. PDMAX = approx. 3 mW/cm2 3. HTE = 4.81 mm 4. HIC = 0.5 mm 5. Wn = 2.27 mm 6. Wp = 2.25 mm Geometrical Optimization time = 40 s. Neural network optimization is 1000 times more effective than COMSOL simulation optimization.	[100]
COMSOL multiphysics	ANN GA	1. Constant heat flux conditions. 2. Insulation thickness and electrode thickness set to 0.5 mm. 3. TEG device area = 1 cm2 4. Cold side temperature = 293.15 K 5. External temperature = 293.15 K 6. Convectional heat flux on open internal surfaces with heat transfer coefficient = 1 mW/(cm2K) Ranges and resolution of geometric parameters and operating conditions 1. TE leg height with a range of 1–10 mm at a resolution of 0.1 mm. 2. FF with a range of 0.05–0.95 and resolution of 0.01. 3. High temperature n-type and p-type TE leg height ratio both with a range of 0.05–0.95 and a resolution of 0.01. 4. Heat flux with a range of 100–2000 mW/cm2 and a resolution of 1 mW/cm2. 5. Top and bottom side contact resistivities both with a range of 10−9–10−7 Ωm2. with a resolution of 10−9 Ωm2.	Optimization time and performance.	Prediction accuracy of iterative trained ANN increased from 94 to 98%. Average ANN optimization time = 6.3 s Proposed optimization method over 5000 times faster than FEM optimization methods	[101]
COMSOL multiphysics	DNN GA Active Learning (AL)	Boundary conditions: 1. Hot end temperature = 670 K 2. Cold end temperature = 350 K 3. All other boundaries are thermally insulated. 4. Heat convection and heat radiation are neglected. 5. Material properties are assumed homogeneous and isotropic. 6. Thermal and electrical contact resistance at interfaces between any two sections are not considered. 7. External load resistance is considered a thermal insulator. Input parameters: 1. TE leg length = 10 mm 2. P-type and n-type segment length both with a range of 0.1–9.7 mm. 3. Height of aluminum oxide and copper electrode = 1 mm. 4. Radius of external load resistance = 1.5 mm 5. Horizontal bar length = 8 mm. 6. External load resistivity with a range of 10−7–10−2 Ωm. 7. TE area of both legs = 3 × 3 mm2 Five weight factors considered with values of 0, 0.25, 0.5, 0.75, and 1.0.	1. Max power output 2. TEG efficiency	The optimization process resulted in an improvement of 1,91 and 1.5 times in terms of power output and TEG efficiency respectively in comparison to the designs in the initial dataset. Values calculated for each weight factor with a sample power output of 0.1710 W and efficiency of 12.450% at a weight factor of 0 in comparison to actual power of 0.1695 and efficiency of 12.459%.	[102]

showcases a comparative analysis of selected studies involving the optimization of TEGs. An interesting point to identify is that in the optimization methods integrated with AI use, the focus is more on the efficiency of the proposed AI integration methods rather than the optimized properties themselves. Neural networks were also identified as the preferred mode of AI optimization alongside the use of genetic algorithms. While most studies on optimization were not focused on the optimization of thin film type TEGs, the general process and flow can serve as reference points for future studies.

6. Future Directions and Implications

While the primary use of TEGs is in the field of power generation, studies have been conducted to explore their use for other purposes such as in temperature sensing [27,103]. A temperature sensor was developed which made use of a self-powered temperature probe composed of a TEG and a temperature compensation component (TCC) and the general concept for the sensor’s operation involves the use of the TCC as a temperature reference whereas the TEG generates a voltage that is supplied to modules such as the A/D converter module for signal processing and a power management module for self-powering purposes [103]. The performance of the proposed sensor was improved through the utilization of four TEGs and through the addition of solid paraffin to assist in cold side temperature stabilization, and the developed sensor showcased the ability to detect fire before it reached the flashover state with a maximum absolute error of 1.6 K and a mean absolute error of 0.55 K [103].

AI can also be used to improve the performance of a TEG-based temperature sensor. A three-terminal temperature sensor created by connecting a PbTe TEG and a Bi₂Te₃ TEG in series and utilizing the differing Seebeck coefficients to generate independent equations for the calculation of the temperatures from both the hot and cold sides [104]. Machine learning was used for the optimization of the temperature sensor’s performance using data collected from COMSOL software and the Tensor Flow open-source library. The study resulted in a self-powered temperature sensor with a maximum absolute error of 0.98 °C in a temperature range of 0–70 °C [104]. Integrating the principle of TFTEG design principles with the optimal chemiresistive and electrochemical sensing materials [105] may potentially result in a very practical standalone plant wearable sensor.

Both instances discussed involve the integration of multiple TEG units with the first one primarily for improving the maximum power output and the second instance utilizing the difference in Seebeck coefficient to identify the key temperatures.

The concept of integrating multiple TEG units to maximize power output would necessitate the ability to produce large quantity of TEG units at once thus the economic feasibility model for large scale production of such TEG units should be considered as a possible direction for further research as some of the fabrication methods discussed in the study may be too costly. Roll-to-roll production of TFTEGs may be considered for future studies as it has been used for TFTEG production and in the large-scale production of nanostructured thermoelectric materials at lower production costs [26,106].

Recent TEG systems have also been seen integrated into Internet of Things (IoT) systems for a variety of purposes including as power source for purposes such as waste heat recovery in automotive and for smart home sensing, and a temperature sensing system for fire detection purposes [5,106]. As previously mentioned, however, the maximum power output and temperature sensing capabilities are affected by a variety of factors including the quantity of TEG modules used, the thermoelectric material properties, and the structure of the TEG to be considered. In the case of TFTEGs with a more limited power generation potential than conventional TEGs, the power generated may not be sufficient for the IoT system and also have a limited range for sending desired signal data. This is more pronounced in the consideration of agricultural purposes if the agricultural area to be considered is one with limited data connectivity. Thus, in the designing of TFTEG systems, the system level power-budgeting for Long Range/Narrow-Band Internet of Things (LoRa/NB-IoT) wireless modules should also be considered among other aspects.

Lastly, it is essential to ensure that TFTEGs remain environmentally sustainable particularly if they are intended for use in agricultural purposes. Thus, a life cycle assessment for determining the overall impact of the TFTEG system should be considered during the research and design process. It was previously established in a Scopus database search that there is a data gap in the implementation of the life cycle assessment of TFTEGs and a search on science direct showcases 15 research articles under the search conditions: [“life cycle assessment” AND (“Thin-film” OR “Thin film”) AND “Thermoelectric generator”], of which only 3 articles utilized LCA for analysis. One such topic included an LCA framework for analyzing the sustainability of utilizing thermoelectric systems for heat utilization from geothermal seepage incorporating two main divisions: the life cycle inventory from the extraction of thermoelectric materials such bismuth telluride until the end-of-life management and the life cycle impact assessment including global warming potential, resource depletion, acidification potential, and eutrophication potential [107]. It was identified that thermoelectric systems are generally environmentally sustainable with research efforts in alternative thermoelectric materials, recycling and recovery methods, and different aspects of the system design suggesting possibilities for improvements in long-term sustainability [107]. The two other articles utilized the LCA for a comparative techno-environmental analysis of two materials for use in TEGs and the carbon dioxide emissions of a TEG-integrated photovoltaic system respectively [79,108]. It should be noted that none of the identified studies utilized TFTEGs and that possible microplastic residues after field degradation was not discussed, suggesting possibilities for future research in this direction.

7. Conclusions and Recommendations

The study identified several critical points of the current and future state of thin film thermoelectric generators:

• The complex relationship between the material properties and fabrication strategies primarily governs the application and performance of a thin-film TEG. Inorganic materials deliver superior thermoelectric efficiency but require polishing in flexibility and compatibility with soft substrates. Carbon-based materials, on the other hand, provide enhanced mechanical flexibility, and sustainable processing, but suffer from low power outputs. Hybrid materials bridge these gaps and show promise in agricultural and wearable use-cases. Likewise, fabrication techniques, whether physical or chemical, greatly influence film morphology, interfacial quality, and eventual device performance. In agricultural use-case scenarios, the strategy in aligning the materials and processing methods should encompass the specific tailoring fabrication techniques for low-temperature, flexible substrates, improving power output under ambient and sub-optimal temperature gradients, and establishing durability under real environmental conditions.
• Organic and inorganic thin-film materials were mostly utilized for electrochemical sensing, and crop growth and health monitoring for agricultural applications. Organic materials are generally more biocompatible and conform due to their flexibility. Inorganic TEs, on the other hand, have good power output but suffer from relatively poor mechanical flexibility. It was, however, proven that the choice of materials and fabrication methods may address this drawback. Encapsulating TE devices has been proven to enhance their performance and properties while slowing down their rate of degradation due to different external factors. While inorganic TE materials displayed impressive performance and mechanical flexibility, there is lacking evidence to support their biocompatibility and interaction with plant, soil and microfauna considering their cytotoxicity in their elemental state.
• To overcome common challenges faced by thermoelectric generators, such as miniaturization, ease of integration, and efficiency, solutions are leveraged through the mesh of artificial intelligence, flexible and stretchable substrates, and exploration of design architectures. It was seen that the choice of materials for substrates and design architectures affect the overall performance of thermoelectric generators. Nevertheless, the advantages and improvements presented in current applications can be foundational when implementing these technologies, especially for prolonging the operation of a system in smart agriculture, such as monitoring plant conditions. There have been multiple studies on the implementation of AI in the field of TEGs with most studies focusing on the use of AI in the optimization of TEG parameters through methods such as inverse and multi-objective optimization and with a focus on the use of neural networks. The use of AI tools such as GAI in the initial modelling of the TEG has not been explored however GAI tools have been implemented to assist with optimization using cGANs to generate visual representations and thermal models. The use of AI to assist in other aspects of TEG design such as FEM was explored however there is a lack of literature on the topic to confirm the actual viability in TEG design. Recent studies on the integration of AI in TEG optimization focus more on the improvement brought about in both the optimization process in comparison to traditional methods and the general performance of the TEG itself.
• The field of thin-film thermoelectrics has great potential for development in a variety of research directions. The utilization of TEGs for temperature sensing purposes, the combination of multiple TEGs to improve overall performance, and AI integration are directions for performance improvement of TEGs. There are other directions in terms of the improvement and feasibility analysis of TFTEG performance through the considerations for economic feasibility of widescale production, power supply management considerations for IoT integration, and life cycle assessment for sustainability viability.

The lack of relevant studies on the implementation of thin film thermoelectric generators in agriculture makes it a field for potential research in developing precision agriculture. The use of TFTEGs in the field of healthcare in powering wearable sensing devices suggests a possibility in their implementation in flexible crop sensors. However, further research is required to consider various aspects such as whether they can provide sufficient power generation and if the TFTEGs themselves have minimal negative impact on their implementation in the agricultural setting.