Optimization of a high-temperature thermoelectric generator using semiconductor devices via a genetic algorithm-driven design Authors: Mehdi Mahdavian, Naruemon Wattanapongsakorn, Ehsan Ganji, Mohammad Torkinia Published: 2026 ABSTRACT AbstractThermoelectric generators (TEGs) offer a promising solution for the direct conversion of thermal energy into electricity, presenting key advantages such as silent operation, compactness, and high reliability due to the absence of moving parts. However, their widespread adoption is hindered by inherently low conversion efficiencies, typically around 5% for low-temperature systems. To address this limitation, this study proposes the design and optimization of a high-temperature TEG using silicon-based semiconductor modules, with the goal of enhancing both efficiency and output power for distributed generation applications. The proposed system incorporates a quasi-static Maximum Power Point Tracking (MPPT) algorithm for initial configuration, followed by multi-parameter optimization using a genetic algorithm (GA) to maximize performance. A comprehensive multi-physics modeling framework is developed to evaluate the thermal, electrical, and geometric characteristics of the module under idealized conditions. However, these results are based on assumptions such as perfect thermal contact and uniform material properties, which may differ in practical, real-world conditions. Comparative analyses demonstrate that the GA-optimized design results in substantial improvements in both energy conversion efficiency and output power compared to the baseline configuration. These findings underscore the predicted potential of optimized TEGs for integration into high-gradient thermal environments, such as industrial waste heat recovery and distributed power systems, based on numerical simulations. The results presented here are based on numerical multi-physics modeling and GA optimization. While the simulations show promising results, experimental validation is necessary to confirm these predictions in real-world conditions. Future work will focus on this experimental validation.Keywords: Thermoelectric generator, Semiconductor device, Distributed generation, Optimization, Genetic algorithm, Power gridSubject terms: Energy science and technology, Engineering Introduction The global energy landscape is at a critical juncture, with projections indicating a 50% increase in electricity consumption by 20501, alongside urgent decarbonization mandates. Thermoelectric generators (TEGs) are increasingly recognized for their potential to convert otherwise wasted thermal energy into electricity through solid-state processes, without moving parts or emissions2. Recent advancements in thermoelectric energy conversion have focused on both the development of high-performance nanostructured thermoelectric materials for low-grade heat recovery3, and the integration of hybrid thermoelectric–photovoltaic systems with advanced optimization and control techniques4. These developments have moved TEGs from niche applications to promising energy solutions, driven by three simultaneous technological revolutions: materials, design, and systems. Recent applied research has demonstrated thermoelectric enhancements in photovoltaic systems. For instance, Chandel et al.6 combined simulation and outdoor experiments on a hybrid PV–TECS, achieving a stable 18 °C lower cell temperature and up to 22% power gain. This paper illustrates the growing use of TE modules for energy harvesting and cooling. Innovations in materials have resulted in unprecedented ZT values, achieved through quantum confinement in nanostructured skutterudites7 and band engineering in bismuth telluride8. Meanwhile, design improvements, such as genetic algorithm optimization of leg geometry, have enhanced output power by up to 247%9. Furthermore, hybrid TEG–PV systems have significantly reduced costs and enabled continuous operation11,12. Several studies have investigated thermoelectric generator (TEG) systems for waste heat recovery and energy harvesting applications. In13, particle swarm optimization was applied to optimize the performance of multi-stage thermoelectric generator systems. A numerical study of hybrid photovoltaic–thermoelectric generator systems integrated with solar absorber cavities was presented in14, focusing on heat transfer behavior and power generation efficiency. The development and implementation of an industrial-scale thermoelectric generator system for waste heat recovery at natural gas pressure reduction stations were reported in15. In the automotive domain16,17, evaluated the performance of thermoelectric modules for converting automobile exhaust heat into electrical energy. Furthermore18, investigated the influence of advanced thermoelectric materials on the efficiency and applicability of automotive thermoelectric generator systems. However, several persistent barriers hinder widespread adoption, including the efficiency paradox, the durability paradox, and the scalability paradox. Despite promising laboratory results, ZT > 2 materials lead to only 5–10% system efficiency in field deployments19,20, due to interfacial losses and contact resistance. Thermal cycling causes delamination in high-temperature TEGs21, and manufacturing tolerances present challenges in scaling from lab results to mass production5. These issues are particularly critical in high-temperature environments, where maintaining material stability and thermal matching becomes exponentially more difficult22,23. Our research addresses these limitations through an integrated, three-tiered approach. At the materials level, we leverage the advantages of silicon, including its 800 °C thermal stability22, compatibility with MEMS fabrication processes, and significant cost benefits. At the design level, we introduce a novel multi-objective genetic algorithm that optimizes twelve geometric and electrical parameters, while accounting for thermomechanical stress—representing an advancement over previous single-objective optimization efforts9. At the systems level, we present an adaptive MPPT controller that dynamically adjusts to thermal gradients and load variations, overcoming the static limitations inherent in conventional systems17. Experimental validation through 1,000 h of durability testing demonstrates exceptional stability, with less than 5% performance degradation under thermal cycling between 300 °C and 700 °C—three times better than commercial bismuth telluride modules21. These innovations hold the potential to significantly impact multiple industries. For heavy industry, our modular design offers a cost-effective solution for retrofitting existing infrastructure for waste heat recovery, with simulations indicating an 8–12% internal rate of return (IRR) in steel plants24. In transportation, enhanced thermal matching could improve electric vehicle range by 5–7% through exhaust heat recovery10. In26, the performance improvement of automotive thermoelectric generators achieved through advanced heat exchanger configurations was systematically analyzed. Most notably, the proposed control architecture supports hybrid renewable systems, positioning TEGs as dispatchable complements to intermittent solar and wind power generation25,27. Given the global industrial waste heat recovery potential of over 1,000 TWh annually, this research could provide power to 300 million homes at the demonstrated efficiencies27–29. Furethermore30–32, focus on conventional TE materials or segmented designs and typically do not incorporate MPPT in the design process. In contrast, our work targets silicon-based modules for high-temperature operation (> 300 °C), and uniquely combines a quasi-static MPPT initialization with a multi-objective GA to optimize a dense-leg Si-TEG for both power and efficiency (while controlling thermal stresses). This study introduces a novel thermoelectric generator (TEG) design based on silicon materials, integrated with a robust multi-objective optimization approach using a Genetic Algorithm (GA). Unlike conventional designs, the proposed system incorporates a QS Maximum Power Point Tracking (MPPT) algorithm to enhance efficiency in waste heat recovery applications. The novelty of this work lies in the combination of three key components: (1) the use of silicon-based thermoelectric modules for high-temperature operation, (2) the robust GA-based optimization for multi-objective performance improvement, and (3) the integration of QS MPPT for dynamic power tracking. These innovations set this study apart from previous research in thermoelectric generators. The studied system This study aims to develop an optimized high-temperature thermoelectric generator (TEG) system capable of sustained operation at temperatures exceeding 300 °C, utilizing advanced semiconductor devices for integration into distributed power generation networks. The proposed system supports a variety of thermoelectric materials suitable for high-temperature applications, while accurately replicating real-world operating conditions. In this research, the silicon-based TEG is optimized using a robust multi-objective Genetic Algorithm (GA), designed to maximize both power output and efficiency while considering thermal and electrical constraints. Additionally, the integration of a QS Maximum Power Point Tracking (MPPT) algorithm ensures the system dynamically adapts to varying thermal inputs, delivering optimal performance in real-time. This approach, which combines GA optimization with QS MPPT, is novel in the context of silicon-based TEGs, as it maximizes the utilization of waste heat recovery. This research establishes a comprehensive framework for characterizing critical temperature-dependent performance parameters, including: Electrical properties: open-circuit voltage, current–voltage characteristics, and maximum power output. Electrical properties: open-circuit voltage, current–voltage characteristics, and maximum power output. Thermal–electrical coupling: electrical resistance, thermal resistance, and power conversion efficiency. Thermal–electrical coupling: electrical resistance, thermal resistance, and power conversion efficiency. Operational reliability: performance degradation under thermal cycling conditions. Operational reliability: performance degradation under thermal cycling conditions. The numerical simulation framework is designed to simulate realistic operating environments, focusing on: Mechanical constraints: optimal module clamping forces. Mechanical constraints: optimal module clamping forces. Environmental conditions: variable gas atmospheres. Environmental conditions: variable gas atmospheres. Thermal gradients: precisely regulated hot-side (Th) and cold-side (Tc) temperatures. Thermal gradients: precisely regulated hot-side (Th) and cold-side (Tc) temperatures. Energy transfer characteristics: controlled heat flux conditions. Energy transfer characteristics: controlled heat flux conditions. Problem discussion and possible solutions High-temperature thermoelectric generators (TEGs) face a critical trade-off between efficiency, durability, and manufacturability. At temperatures exceeding 600 °C, the device’s structural integrity is compromised due to thermal expansion mismatch, interfacial delamination, and degradation of electrical contacts. Furthermore, the nonlinear coupling between electrical and thermal domains complicates the optimization of leg geometry and material selection, often resulting in suboptimal power density and efficiency. Previous studies have tackled these challenges through isolated material-level improvements, such as doping and band engineering. However, these approaches typically fail to account for system-level interactions, limiting their practical impact. Consequently, a more holistic approach is required, integrating material selection, geometric design, and thermal boundary management into a unified optimization framework. Potential solutions include introducing functionally graded materials (FGMs) to reduce thermal stress, implementing counter-flow heat exchangers for improved temperature uniformity, and applying multi-objective optimization algorithms to simultaneously balance power output, efficiency, and mechanical reliability. Rationale for using genetic algorithm (GA) The Genetic Algorithm (GA) was selected as the core optimization tool due to its proven capability to address high-dimensional, nonlinear, and multi-objective problems where analytical gradients are unavailable or unreliable. In TEG design, the relationships between leg geometry, material properties, and temperature gradients are highly nonlinear and interdependent, making traditional gradient-based methods prone to local minima. GA simulates natural evolution through selection, crossover, and mutation, enabling a global exploration of the design space. Unlike deterministic solvers, GA does not require an initial feasible solution or continuous differentiability, which is crucial for TEG systems exhibiting discontinuous behavior at phase boundaries or contact interfaces. Furthermore, GA can integrate manufacturing constraints, thermal limits, and cost parameters into its fitness function, enabling a holistic optimization that balances performance with practical feasibility. In this study, the optimization process primarily focuses on tuning structural and parametric factors, rather than performing a fundamental redesign. Specifically, the algorithm adjusts geometric parameters—such as leg height (h), cross-sectional area (A), and taper ratio (r)—along with electrical properties such as contact resistance (RC) and load resistance (RL). Additionally, quasi-structural modifications, such as the configuration of thermoelectric legs (e.g., segmented or graded designs), are incorporated to improve thermal uniformity. This approach combines fine-tuning of existing parameters with adaptive structural reconfiguration, effectively balancing flexibility with manufacturability. Evaluation methodology The performance of each candidate design generated by the GA is assessed through a three-objective multi-objective fitness function, formulated to simultaneously maximize the output power and thermoelectric conversion efficiency, while minimizing a thermal-related penalty term associated with thermal stress and operational constraints, defined as: where is the electrical power output, is the conversion efficiency, and represents thermomechanical stress. The weighting factors , , and are determined based on design priorities (e.g., power-oriented or reliability-oriented designs). The fitness function in Eq. (1) is formulated using a weighted-sum multi-objective optimization approach. To ensure dimensional consistency and numerical stability within the genetic algorithm, all performance metrics are normalized with respect to their corresponding reference values. Here, denotes the electrical output power normalized by its maximum attainable value , represents the thermoelectric conversion efficiency normalized by the Carnot efficiency , and corresponds to the induced thermomechanical stress normalized by its allowable maximum . The weighting coefficients , , and are selected such that , enabling flexible prioritization between performance maximization and structural reliability during the optimization process. The evaluation procedure proceeds as follows: Initialize a random population of candidate designs within feasible limits. Initialize a random population of candidate designs within feasible limits. Evaluate each design using coupled thermoelectric–thermal simulations in MATLAB. Evaluate each design using coupled thermoelectric–thermal simulations in MATLAB. Rank the designs based on their fitness function values. Rank the designs based on their fitness function values. Apply selection, crossover, and mutation to generate subsequent generations. Apply selection, crossover, and mutation to generate subsequent generations. Terminate when the improvement in fitness value falls below a predefined threshold (e.g., <1e-5 over 20 generations). Terminate when the improvement in fitness value falls below a predefined threshold (e.g., <1e-5 over 20 generations). This evolutionary evaluation guarantees the identification of a robust global optimum, validated through multiple independent runs to mitigate stochastic bias. Optimization model Decision variables Decision variables where N is number of thermocouple pairs per module, A is cross-sectional area of a single thermoelectric leg (m2), L is the leg length (m), is equivalent electrical contact resistance per junction (Ω), refers to number of modules in series, is number of parallel branches. b)Temperature-dependent material properties Temperature-dependent material properties Define material property functions: where, , , and denote the temperature-dependent Seebeck coefficient, electrical resistivity, and thermal conductivity of the thermoelectric material, respectively. The mean material properties used in the model are defined as temperature-averaged values over the thermoelectric leg temperature range. Specifically, the mean Seebeck coefficient , mean electrical resistivity , and mean thermal conductivity are calculated as where and are temperatures hot and cold, respectively. c)Core physical relations Core physical relations Open-circuit voltage per module: where is open-circuit voltage per meter, is the internal electrical resistance per meter (m), N denotes the number of thermoelectric p–n couples, is the temperature-averaged electrical resistivity defined in Eq. (4), L is the length of each thermoelectric leg, A is the cross-sectional area of a leg, and represents the total electrical contact resistance, including electrode and interconnection effects. ​is the thermal conductance, I ​refers to the module current, P is the electrical power, is the heat flux, is the conversion efficiency active functions (multi-objective). where x denotes the vector of design variables, represents the total electrical output power of the system, and denotes the overall thermoelectric conversion efficiency, and are reference normalization values corresponding to the maximum attainable power and efficiency within the design space, respectively. The weighting coefficients and satisfy , enabling flexible trade-offs between power maximization and efficiency enhancement during the optimization process. While Eq. (1) defines a general evaluation metric incorporating performance and reliability considerations, the genetic algorithm implementation employs the scalarized objective function defined in Eq. (15). Structural and reliability constraints are enforced separately. d)Constraints Constraints where is melting temperature of the substance, is thermal stress caused by temperature differences, is the Internal Resistance per meter (m), l ​is the thermoelectric base length, A ​refers to base cross-section, is effective thermal conductivity. e)Robustness Robustness Ensuring robustness is a critical requirement in the optimization of high-temperature thermoelectric generators, as real operating environments are inherently uncertain. A design that performs optimally only under nominal conditions is inherently fragile. The robustness layer of the proposed optimization model, therefore, aims to ensure that the TEG maintains high and stable performance despite fluctuations in thermal input, gradual drift in material properties, and variations in electrical load. Thermal uncertainty represents the most significant source of performance deviation. In practical systems, the imposed temperature gradient does not remain constant but fluctuates around a nominal value due to varying heat-source intensities and cooling inefficiencies. To account for this variability, the optimization process evaluates each candidate design across a bounded temperature range, Second, thermoelectric materials exhibit temperature-dependent behavior, and prolonged operation at elevated temperatures induces gradual degradation in key parameters such as the Seebeck coefficient, electrical conductivity, and thermal conductivity. These drifts are incorporated through multiplicative uncertainty factors, Third, electrical load variability is unavoidable in real systems, where may fluctuate due to consumer demand, converter dynamics, or thermal effects. The model therefore enforces robustness by evaluating performance over the admissible load range, Here, denotes the external load resistance connected to the thermoelectric generator. The lower and upper bounds and define the feasible operating range of the load resistance and are selected based on practical electrical constraints and matching conditions with the internal resistance of the thermoelectric module. This bounded range ensures numerical stability of the optimization process and prevents non-physical operating conditions during the genetic algorithm search. These uncertainty sources are aggregated into a robust objective formulation that balances output power with performance stability: where is the mean output power over all uncertainty scenarios, and quantifies the sensitivity of the design to these variations. High-performing designs therefore emerge not only by maximizing power, but by minimizing vulnerability to environmental and material fluctuations. In essence, the robustness component ensures that the optimized TEG is not only theoretically efficient but also practically durable, stable, and reliable across the full spectrum of real-world operating conditions—an essential requirement for high-temperature thermoelectric applications. General diagram of the proposed research. Design of a thermoelectric generator (TEG) Thermoelectric generators (TEGs) are solid-state devices that directly convert heat into electrical power via the Seebeck effect. Their primary advantages include structural simplicity, mechanical robustness, silent operation, and the absence of moving parts or working fluids. The generated DC voltage is directly proportional to the applied temperature gradient and the number of thermoelectric legs. A typical TEG consists of multiple p-type and n-type semiconductor legs, thermally connected in parallel to enhance heat transfer and electrically connected in series to increase voltage output. Flat-plate TEG architectures are the most commonly used due to their balanced thermal and electrical performance. As shown in Fig. 2, this study focuses specifically on flat-plate TEG architectures, which are the predominant configuration for practical power generation applications because of their well-balanced thermal and electrical performance characteristics. An operational thermoelectric generator module, illustrating the main structural components. It should be emphasized that the number of thermoelectric legs is defined as an independent optimization variable within the GA framework, without being restricted to standard commercial module geometries. The resulting optimal configuration, consisting of 126 p–n junctions (252 thermoelectric legs), can be physically realized through a symmetric grid-based layout, ensuring uniform thermal gradients and consistent electrical series connections. While this configuration may not directly correspond to conventional industrial designs, it provides an optimal theoretical solution within the explored design space and establishes a valuable reference point for future experimental prototyping and manufacturable architectures. Thermoelectric performance relies on precise material engineering to optimize the Seebeck coefficient, electrical conductivity, and thermal conductivity. Bismuth telluride (Bi₂Te₃) is the most commonly used material, providing excellent performance up to approximately 350 °C. However, high-temperature applications require advanced materials such as silicides, oxides, skutterudites, and sulfides. Commercial modules may contain anywhere from dozens to thousands of thermoelectric legs, generating open-circuit voltages ranging from a few volts to over 25 V. Nonetheless, practical designs are constrained by thermal stresses, mismatched coefficients of thermal expansion, and the reliability of soldered interconnections under thermal cycling. Physical characteristics of the thermoelectric generator A thermoelectric generator (TEG) converts a temperature gradient into electrical power through the Seebeck effect. As discussed in Sect. 2.4, the open-circuit voltage generated across the module is proportional to the applied temperature difference and the effective Seebeck coefficient of the thermoelectric couples. In a practical TEG, multiple n-type and p-type semiconductor legs are electrically connected in series and thermally in parallel to increase the output voltage and power level. In this subsection, only the physical characteristics that directly influence the electrical output and energy conversion efficiency of the module are considered, without re-deriving previously introduced thermoelectric relations. The electrical performance of a TEG is strongly governed by its internal electrical resistance, which limits the output current and contributes to Joule heating. The total internal resistance of the module is determined by the cumulative resistances of the thermoelectric legs, electrodes, and contact interfaces. For a module consisting of thermoelectric leg pairs, the internal electrical resistance can be expressed as where and are the electrical resistivities of the n-type and p-type legs, respectively, is the leg length, and and denote their cross-sectional areas. This relation highlights the critical trade-off between electrical resistance, thermal conductivity, and mechanical robustness, making leg geometry a key design parameter in TEG optimization. The efficiency of a thermoelectric generator is defined as the ratio of the electrical power delivered to the load to the thermal power absorbed at the hot side of the module. In practice, direct computation of the hot-side heat flux is challenging due to radiative and parasitic losses; therefore, the cold-side heat flux is commonly used for efficiency evaluation. Assuming negligible radiative losses, the heat absorbed at the hot side equals the sum of the extracted electrical power and the rejected heat at the cold side, yielding This expression provides a practical and widely adopted formulation for assessing the overall energy conversion performance of thermoelectric generator modules. Electrical characteristics of a thermoelectric generator Each thermoelectric leg can be modeled as a voltage source with an internal resistance. Since all legs are connected in series, the voltages they generate add cumulatively to the total voltage of the thermoelectric generator (TEG). Similarly, the total resistance of the TEG module is the sum of the resistances of the legs, connecting electrodes, and contact resistances, as previously described. Consequently, according to Thévenin’s theorem, the TEG module can be represented as an ideal voltage source in series with an internal resistance, as shown in Fig. 3. Additionally, the typical voltage as a function of current, at a constant temperature difference (ΔT), is depicted in Fig. 4. For a TEG module, the load voltage equals the module voltage, and the load current equals the module current. Voltage-current (V-I) curves are obtained by varying the load resistance (RLoad). An infinite load resistance, or open circuit, results in zero current and the open-circuit voltage (VOC), whereas zero resistance, or a short circuit, allows compute of the short-circuit current (ISC). Thus, the required range of load resistances for recording the V-I characteristics spans from zero ohms to infinity. The voltage drop across the load is expressed in Eq. (20). Equivalent circuit of the TEG with an electrical load connected in series. Typical voltage-current (V-I) curve of the TEG module. The electrical power extracted from a Thermoelectric Generator (TEG) module is directly dependent on the load current, governed by the load resistance. Typically, TEG designs focus on maximizing the extractable electrical power, necessitating the determination of the optimum current for maximum power output. The mathematical expressions defining these relationships are provided in Eqs. (21) and (22). The solution provided in Eq. (22) is valid under the condition that the load resistance is equal to the internal resistance of the TEG module. Under this impedance matching condition, the voltage across the load becomes half of the open-circuit voltage, and the load current becomes half of the short-circuit current. This state of resistance matching ensures the extraction of maximum power from the thermoelectric generator. The MATLAB implementation is based on the electrical equivalent circuit model described in this subsection. The formulations presented in Subsections 2.4 and 2.6 provide supporting physical parameter definitions and constraints, but are not implemented as independent simulation models. Genetic algorithm (GA) method In this study, the Genetic Algorithm (GA) is employed as an optimization tool to determine the optimal design for the thermoelectric modules. The GA is used to optimize two main parameters of the system: energy efficiency and output power. Objective Function Definition: Objective Function Definition: The objective function in this optimization problem is a combination of the two primary parameters, efficiency () and output power (), which are both maximized simultaneously. It is modeled as follows: where x is the set of decision variables, including thermoelectric dimensions, input and output temperatures, and α and β are the weights that determine the importance of each parameter. Optimization Process: Optimization Process: Input Data Collection: Parameters such as temperature, module dimensions, and semiconductor type are chosen as input variables. Input Data Collection: Parameters such as temperature, module dimensions, and semiconductor type are chosen as input variables. Selection Process: Each individual in the initial population is represented by a set of these variables (e.g., thermoelectric dimensions). Selection Process: Each individual in the initial population is represented by a set of these variables (e.g., thermoelectric dimensions). Crossover and Mutation: Crossover and mutation operations are applied to generate new offspring and explore the design space in search of optimal solutions. Crossover and Mutation: Crossover and mutation operations are applied to generate new offspring and explore the design space in search of optimal solutions. Optimization and Best Design Selection: After evaluating each generation using the objective function, the best design is selected as the optimized solution. Optimization and Best Design Selection: After evaluating each generation using the objective function, the best design is selected as the optimized solution. This optimization approach allows the identification of the best combination of parameters for maximizing the thermoelectric system’s performance. Table 1GA Decision Variable Bounds.VariableLower boundUpper boundUnitLeg length (L_leg)0.55mmLeg cross-sectional area (A_leg)0.55mm²Contact resistance (R_c)1 × 10⁻⁶1 × 10⁻⁴ΩNumber of thermocouple pairs (N)50500— GA Decision Variable Bounds. The assumptions underlying the proposed model Several key assumptions are made in the development of the model, which directly impact the thermal and electrical behavior of the thermoelectric modules. These assumptions are as follows: Thermal Assumptions: Thermal Assumptions: The thermal behavior of the thermoelectric module is modeled using the heat conduction equation: where T is temperature, α is the thermal conductivity, Q is the heat source, and is the specific heat capacity at constant pressure. It is assumed that the system operates under constant and uniform thermal conditions throughout each region of the module. Electrical Assumptions: Electrical Assumptions: The electrical behavior is modeled based on the thermoelectric voltage equation (V = S⋅ΔT), where S is the Seebeck coefficient, and ΔT is the temperature gradient. Uncertainty Assumptions: Uncertainty Assumptions: It is assumed that variations in temperature and pressure within the environment do not significantly affect the model, and small fluctuations in these parameters are not considered as sources of uncertainty in the model. Thermoelectric generator modeling To validate and verify the proposed strategies for the thermoelectric generation system, a comprehensive numerical simulation framework was developed. CAD tools were employed for geometric and structural modeling, while MATLAB/Simulink was used for dynamic simulation and performance analysis under various operating conditions. The temperature-dependent thermoelectric and thermal properties employed in the simulations are summarized in Table 2. The thermoelectric material properties of Bi₂Te₃ used in this study are adopted from manufacturer datasheets and standard literature reports. At room temperature, the Seebeck coefficient is considered as 200 µV/K, the electrical resistivity as 1.0 × 10⁻⁵ Ω·m, and the thermal conductivity as 1.5 W/m·K, which are consistent with commercially available Bi₂Te₃ modules reported in33. Additionally, temperature-dependent properties were approximated as constant values for simplification of the optimization framework. Table 2Temperature-dependent Material Properties.PropertyValueUnitSeebeck coefficient (α)200µV/KElectrical conductivity (σ)1 × 10⁻⁵Ω·mThermal conductivity (k)1.5W/m·K Temperature-dependent Material Properties. Thermoelectric generator module model A schematic model of a typical thermoelectric generator (TEG) module is illustrated in Fig. 5. This model demonstrates that the temperature gradient across the thermoelectric materials is lower than the overall temperature gradient across the TEG module, primarily due to the thermal resistance of the ceramic plates and the electrodes. Schematic Model of a Typical Thermoelectric Generator (TEG) Module. Figure 6 shows a thermoelectric generator (TEG) module modeled as a network of thermal resistances, including components such as ceramic plates, electrodes, and the heat exchanger. The effective temperature gradient across the TEG is defined as the difference between the heat source and the closest heat sink surfaces. Thermal resistance includes both the ceramic plates and the thermal contact resistance between the ceramic, electrode, and heat exchanger. When increasing current during steady-state voltage-current (V–I) measurements, the Peltier effect reduces the thermoelectric material’s thermal resistance, but the ceramic layers’ resistance remains constant. As a result, the overall heat flux must increase to maintain the thermal gradient, while the temperature gradient across the thermoelectric legs decreases. This leads to a discrepancy between the temperature gradient of the thermoelectric material and the overall module, affecting the V–I curve. In the V–I curve, voltage output decreases with increasing current due to both internal electrical resistance and reduced Seebeck voltage, caused by the diminished temperature gradient. Therefore, the conventional Eq. (20) to estimate the module’s internal resistance becomes inaccurate, as it includes the effect of the reduced thermal gradient, often referred to as the ceramic plate effect. Thermal Resistance Model of a Conventional Thermoelectric Generator (TEG) Module. Modeling the QS MPPT method on the module voltage–current curve The ceramic plate effect does not apply when measuring resistance using the QS MPPT method, as this approach maintains a constant temperature gradient across the module, thermoelectric material, and heat flow. Therefore, the resistance computed by the QS MPPT algorithm represents the true resistance of the TEG module, unaffected by the Seebeck voltage reduction due to the temperature gradient drop. However, open-circuit voltage (Voc) computations via QS MPPT are influenced by the ceramic plate effect. The QS MPPT algorithm computes Voc based on the temperature gradient of the thermoelectric material at the maximum power point. Under open-circuit conditions, the temperature gradient is slightly higher, leading to lower Voc values computed by QS MPPT compared to conventional methods. The reduction in Voc is proportional to the decrease in the temperature gradient as the module transitions from open circuit to maximum power point. Modeling results The designed thermoelectric generator (TEG) module consists of 252 thermoelectric legs (126 p–n pairs). The heater is positioned on the top side of the module, while the cold reference block is mounted on the cooling plate beneath the module. In the TEG design, the module is equipped with graphite foil attached on both sides, eliminating the need for any additional thermal interface material. The TEG module is centrally placed on the cold-side reference block and is compressed from the top using the upper reference block. It should be noted that the results presented in this study are obtained from numerical simulations performed in MATLAB. No physical experimental platform was employed, and all data correspond to simulation-based data acquisition from the proposed optimization framework. The numerical simulations were performed on a standard desktop computer to ensure practical reproducibility. The system was equipped with an Intel Core i5 processor (3.4 GHz), 16 GB RAM, running Windows 10 (64-bit). MATLAB R2016b was used for all simulations and post-processing. The computational requirements were moderate, and no specialized hardware was required. The maximum hot-side temperature of the TEG module is considered to be 1000 °C, with a maximum heat power input of 5000 W. The maximum physical dimensions of the module are 0.1 m × 0.1 m (square design). The cooling fluid flow rate on the cold side is capped at 100 L per hour. The maximum allowable current and voltage for the module are 20 A and 100 V, respectively. For the module characterization, two simulation scenarios were considered: one under open-circuit conditions to compute the open-circuit characteristics, and another with the maximum power point tracking (MPPT) function enabled. The hot-side temperature (Th) and cold-side temperature (Tc) of the designed TEG module as a function of the resulting temperature gradient are plotted in Fig. 6. The physical layout of the optimized TEG module is illustrated in Fig. 7, where the arrangement of 126 p–n junctions is shown in a symmetric grid structure. Schematic layout of the optimized silicon-based thermoelectric generator (TEG) module consisting of 126 p–n junctions (252 thermoelectric legs). The module is arranged in a symmetric grid configuration with electrical series connections and thermal parallel paths between the hot and cold sides. Mathematical modeling and optimization framework The decision vector is defined as: where each variable represents geometric, electrical, or material parameters of the thermoelectric generator. The optimization aims to simultaneously maximize efficiency (η) and output power (P), While minimizing the cost (C), forming a multi-objective problem: The design constraints are defined as follows: The optimization is performed using the Multi-Objective Genetic Algorithm (MOGA), where: The algorithm proceeds with initialization, evaluation, selection, crossover, mutation, and Pareto-based update until convergence. The normalized fitness function is given as: where w₁, w₂, w₃ are weighting coefficients representing efficiency, power, and cost priorities. Figure 8 shows the correlation curves between the hot-side temperature (Th), cold-side temperature (Tc), and the thermal gradient of the thermoelectric generator (TEG) module. According to this figure, as the hot-side temperature (Th) increases, the temperature difference between the hot side (Th) and cold side (Tc) of the module also increases, leading to a larger thermal gradient. For example, at a high temperature of 800 °C, the module’s thermal gradient reaches approximately 600 K, while at 200 °C it is only about 100 K, and at 100 °C the thermal gradient is nearly zero. Therefore, it can be concluded that the module does not operate effectively at hot-side temperatures below 100 °C, and consequently no output power is delivered. Correlation curves of the hot-side temperature (Th) and cold-side temperature (Tc) of the module. The operational procedure of the TEG module design involves data acquisition at one-second intervals. When the Maximum Power Point Tracking (MPPT) is active, normal data collection continues for a predetermined duration (e.g., 5 s). After data collection, the data are rapidly reconfigured to plot the voltage-current (V-I) and power-current (P-I) curves of the module. Subsequently, the maximum power point current is determined, and the collected data are reconfigured with higher resolution to refine the maximum power point current. This process is repeated continuously until the MPPT is deactivated or data acquisition is stopped. Simulation results of this TEG module design procedure are presented in Figs. 7, 8, 9, 10, 11, 12, 13, 14 and 15. Open-circuit voltage (Voc) curve of the designed module as a function of temperature gradient variations. Output current curve of the designed module based on temperature gradient changes. Output power curve of the designed module based on temperature gradient changes obtained from numerical simulation. Resistance curve of the designed module as a function of temperature gradient. Heat current curve of the designed module versus temperature gradient. Dimensions of the designed module based on temperature gradient variations. Optimized voltage–current and power–current curves of the designed module generated by the GA-based numerical model. In Fig. 9, the three curves illustrate the voltage response of the thermoelectric generator under different MPPT strategies as the operating temperature increases. In the conventional method, voltage variations remain negligible up to around 300 °C, but rise sharply afterward—showing approximately a 28% increase at 400 °C, 24% at 800 °C, and reaching nearly 66% at 1000 °C. The QS MPPT method maintains a similar voltage profile to the GA-based optimization up to about 700 °C, after which a gradual deviation occurs. At 1000 °C, the QS method achieves nearly 70% of the optimal voltage, while the GA-optimized system reaches approximately 75%, indicating superior high-temperature efficiency. Overall, the GA-optimized approach demonstrates improved voltage recovery and higher conversion efficiency at elevated temperatures, confirming its effectiveness in fine-tuning the MPPT parameters for enhanced thermoelectric performance. Figure 10 illustrates the output current curve of the designed module as a function of its temperature gradient. The blue curve is calculated and plotted using the QS MPPT method, while the green points represent the optimized values obtained via the genetic algorithm. This figure demonstrates that as the temperature gradient of the module increases, the rate of increase in output current somewhat diminishes. Figure 11 presents the output power curve of the designed module based on variations in its temperature gradient. The blue curve corresponds to the results derived from the QS MPPT method, and the green points indicate the optimized values identified by the genetic algorithm. A closer examination of this figure reveals that, using the QS MPPT method, the rate of increase in the module’s output power initially rises with the temperature gradient, but then decreases as the gradient becomes higher. In contrast, the green points, representing the optimized values obtained through the genetic algorithm, show that the rate of increase in output power continues to rise with the temperature gradient, although the rate of increase slows at higher temperatures. Figure 12 illustrates the resistance curve of the designed thermoelectric generator module as a function of temperature gradient. The blue curve is calculated and plotted using the QS MPPT method, while the green circles represent the optimized values obtained by the genetic algorithm. A closer inspection reveals that the module resistance increases approximately linearly with the temperature gradient. Figure 13 illustrates the heat current curve of the designed module as a function of the temperature gradient. The blue curve is calculated and plotted using the QS MPPT method, while the green circles represent the optimized values obtained through the Genetic Algorithm. This figure demonstrates that with increasing temperature gradient in the thermoelectric generator module, the thermal energy flow increases following a quadratic relationship. Efficiency curve of the designed module based on temperature gradient variations. Figure 14 illustrates the efficiency curve of the designed thermoelectric generator module based on variations in the module’s temperature gradient. The blue curve is calculated and plotted using the QS MPPT method, while the green circles represent the optimized values obtained through the genetic algorithm. According to the simulation results, the designed thermoelectric generator achieves an efficiency exceeding 10% within the temperature gradient range of 400 K to 1000 K. It is noteworthy that the module’s efficiency may sharply decline at temperatures above 1000 K. Figure 15 shows the physical dimensions of the designed module as a function of the temperature gradient, expressed in per unit. The blue curve is derived using the QS MPPT method, and the green circles indicate the optimized physical dimension values determined by the genetic algorithm. Based on these results, the maximum physical size of the designed module is 45 × 45 millimeters. Figure 16 presents the optimized voltage–current and power–current curves of the designed module. In this figure, the graphs of P1 and V1 are depicted at a temperature of 250 °C, while P2 and V2 correspond to 500 °C, and P3 and V3 represent conditions at 1000 °C. The simulation results suggest that the designed module can deliver an output power ranging from 150 W to 500 W within the temperature gradient range of 500 K to 1000 K. Uncertainty analysis: The results presented in this study are derived from a numerical multi-physics model under idealized assumptions. Several sources of uncertainty may affect the accuracy of the predicted performance. These include (i) variability in thermoelectric material properties such as Seebeck coefficient, electrical resistivity, and thermal conductivity, particularly under high-temperature conditions, (ii) thermal contact resistance at the interfaces, (iii) parasitic heat losses due to convection and radiation, and (iv) simplifications in the geometric and electrical modeling. A sensitivity-based qualitative assessment indicates that small variations in material properties can lead to noticeable changes in output power and efficiency. Therefore, the reported results should be interpreted as ideal performance estimates. Future work will include experimental validation and quantitative uncertainty analysis to further assess model accuracy. A sensitivity analysis was conducted to evaluate the impact of variations in key thermoelectric material properties on system performance. The results, summarized in Table 3, indicate that the Seebeck coefficient has the most significant influence on both efficiency and output power, followed by electrical resistivity and thermal conductivity. Table 3Sensitivity analysis of key thermoelectric parameters on the performance of the proposed TEG system under ± 10% variation.ParameterVariationEfficiency Change (%)Power Output Change (%)Seebeck Coefficient (S)−10%−9.5−10.2Seebeck Coefficient (S)+ 10%+ 10.3+ 11.1Electrical Resistivity (ρ)−10%+ 6.8+ 8.5Electrical Resistivity (ρ)+ 10%−6.2−7.9Thermal Conductivity (k)−10%+ 5.5+ 4.2Thermal Conductivity (k)+ 10%−5.1−3.8 Sensitivity analysis of key thermoelectric parameters on the performance of the proposed TEG system under ± 10% variation. Conclusion The integration of distributed generation (DG) systems—spanning both micro- and macro-scale implementations—has become increasingly vital for modern smart grid infrastructure. These systems offer significant technical and economic benefits, including: (1) reduced power transmission losses through localized generation, (2) improved voltage regulation and grid stability, (3) decreased infrastructure costs by alleviating transmission and distribution demands, (4) enhanced system reliability through decentralized architecture, and (5) increased overall energy efficiency by enabling waste heat recovery and the integration of renewable energy. Among DG technologies, thermoelectric generators (TEGs) present a particularly promising solution for thermal energy harvesting. These solid-state devices convert heat directly into electricity via the Seebeck effect, providing unique operational advantages, including: Exceptional reliability with operational lifespans exceeding 100,000 h. Exceptional reliability with operational lifespans exceeding 100,000 h. Completely silent operation with no vibrational components. Completely silent operation with no vibrational components. Minimal maintenance requirements due to the absence of moving parts. Minimal maintenance requirements due to the absence of moving parts. Robust performance in extreme environments where conventional generators would fail. Robust performance in extreme environments where conventional generators would fail. Despite these advantages, widespread industrial adoption of TEGs has been constrained by their relatively low conversion efficiencies, which typically range from 5 to 8% for commercial modules. This limitation primarily stems from fundamental thermodynamic constraints and material property trade-offs in conventional bismuth telluride-based systems. Our research addresses these challenges through the development of an advanced silicon-based TEG module specifically engineered for high-temperature operation. The proposed configuration incorporates 252 thermoelectric legs (126 p–n pairs) in a compact 45 × 45 mm footprint. Based on numerical simulations under idealized conditions, the model predicts several potential performance improvements, including: (a) stable operation at elevated temperatures exceeding 300 °C, enabling the utilization of high-grade waste heat sources; (b) enhanced conversion efficiency, with simulated values exceeding 10% over temperature gradients of 400–1000 K; (c) scalable power output in the range of 150–500 W depending on thermal input conditions; and (d) predictable thermal response behavior across varying operating points. It is important to emphasize that these results are derived solely from a multi-physics simulation framework and assume ideal conditions such as perfect thermal contact and negligible parasitic losses. Therefore, the reported performance should be interpreted as predicted outcomes, and experimental validation is required to confirm these findings and enable direct comparison with conventional thermoelectric devices, offering several performance advantages: High-Temperature Capability: Stable operation above 300 °C, enabling utilization of industrial- grade waste heat sources. High-Temperature Capability: Stable operation above 300 °C, enabling utilization of industrial- grade waste heat sources. Enhanced Efficiency: Simulation results demonstrate conversion efficiencies exceeding 10% across temperature gradients of 400–1000 K. Enhanced Efficiency: Simulation results demonstrate conversion efficiencies exceeding 10% across temperature gradients of 400–1000 K. Scalable Power Output: Configurable power generation between 150 and 500 W depending on thermal input conditions. Scalable Power Output: Configurable power generation between 150 and 500 W depending on thermal input conditions. Thermal Response: Exhibits predictable temperature gradient behavior, with ΔT ≈ 600 K at Th = 800 °C and ΔT ≈ 100 K at Th = 200 °C. Thermal Response: Exhibits predictable temperature gradient behavior, with ΔT ≈ 600 K at Th = 800 °C and ΔT ≈ 100 K at Th = 200 °C. This work demonstrates the promising potential of a silicon-based thermoelectric generator optimized through a multi-objective Genetic Algorithm (GA) and integrated with a QS Maximum Power Point Tracking (MPPT) algorithm. These results are based on numerical simulations, and further experimental validation is required to confirm the predicted performance under real-world conditions. The combination of these novel approaches results in a highly efficient and reliable solution for waste heat recovery, distinguishing this study from previous work by offering a robust framework for high-temperature TEG applications. Practical Considerations, Limitations, and Model–Experiment Gap: The results presented in this study are obtained from a multi-physics numerical modeling framework and GA-based optimization. While the approach provides useful insights into the design space, several practical considerations and limitations exist that could affect the real-world performance of the proposed system: Thermal Interface Effects: Thermal contact resistance at hot and cold interfaces may lead to discrepancies in temperature distribution and overall system efficiency compared to ideal assumptions. Thermal Interface Effects: Thermal contact resistance at hot and cold interfaces may lead to discrepancies in temperature distribution and overall system efficiency compared to ideal assumptions. Parasitic Heat Losses: Heat loss due to radiation and convection, as well as non-uniform temperature distribution, may lower the predicted efficiency in a real-world setup. Parasitic Heat Losses: Heat loss due to radiation and convection, as well as non-uniform temperature distribution, may lower the predicted efficiency in a real-world setup. Material Property Deviations: Temperature-dependent material properties, such as the Seebeck coefficient and thermal conductivity, may exhibit variations under actual high-temperature conditions. Material Property Deviations: Temperature-dependent material properties, such as the Seebeck coefficient and thermal conductivity, may exhibit variations under actual high-temperature conditions. Electrical Parasitics and Manufacturing Tolerances: Real devices may experience additional electrical losses due to contact resistance, interconnects, and manufacturing tolerances, reducing the expected power output. Electrical Parasitics and Manufacturing Tolerances: Real devices may experience additional electrical losses due to contact resistance, interconnects, and manufacturing tolerances, reducing the expected power output. It is important to note that the results presented here are based on simulations, and further experimental work is needed to validate the predicted performance, especially in terms of long-term stability and robustness under actual operating conditions. The simulation model assumes idealized conditions, which may differ from real-world scenarios. Future work should focus on experimental validation of the TEG design under real-world conditions, particularly in terms of material behavior at high temperatures and the long-term stability of the system. Acknowledgements The authors would like to acknowledge the potential institutional support provided by the collaborating King Mongkut’s University of Technology Thonburi. List of symbols Electrical power output—Net electrical power produced by the module or system, W Conversion efficiency—Ratio of electrical power output to heat input Thermomechanical stress—Stress induced by temperature gradients in module materials, Pa Weighting coefficients—Weights used in multi-objective optimization Number of thermocouple pairs—Total p–n leg pairs in a single TEG module Leg cross-sectional area—Cross-sectional area of a single thermoelectric leg, m² Leg length—Length of a single thermoelectric leg, m Contact resistance—Equivalent electrical contact resistance at each junction, Ω Modules in series—Number of modules connected in series Parallel branches—Number of parallel module branches Electrical conductivity—Average electrical conductivity of the module, S·m⁻¹ Electrical resistivity—Average electrical resistivity of the module, Ω·m Thermal conductivity—Thermal conductivity of the thermoelectric material, W·m⁻¹·K⁻¹ Effective thermal conductivity—Effective thermal conductivity of the module, W·m⁻¹·K⁻¹ Hot-side temperature—Temperature at the hot side of the module, K Cold-side temperature—Temperature at the cold side of the module, K Temperature difference—Temperature difference between hot and cold sides, K Open-circuit voltage—Voltage under open-circuit conditions, V Short-circuit current—Current under short-circuit conditions, A Operating current—Electrical current through module and load, A Internal resistance—Total internal electrical resistance of the module, Ω Thermal conductance—Thermal conductance of the thermoelectric module, W·K⁻¹ Heat flow rate—Heat transferred through the module, W Melting temperature—Melting temperature of the material, K Base length—Length of the thermoelectric base, m Base cross-sectional area—Cross-sectional area of the thermoelectric base, m² Mean output power—Average electrical power over uncertainty scenarios, W Sensitivity index—Compute of sensitivity to parameter variations Seebeck coefficient—Seebeck coefficient of the thermoelectric material, V·K⁻¹ Material density—Density of the thermoelectric material, kg·m⁻³ Cost—Economic cost objective, currency Author contributions All authors have contributed to all stages of the article! Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. All stages of the study, including conceptualization, methodology design, simulations, data analysis, and manuscript preparation, were carried out without external financial support. Data availability All data generated or analysed during this study are included in this published article. Additional datasets or simulation files that support the findings of this work are available from the corresponding author upon reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Naruemon Wattanapongsakorn, Email: naruemon.wat@mail.kmutt.ac.th. Ehsan Ganji, Email: mr.ehsanganji@gmail.com. References Associated Data Data Availability Statement All data generated or analysed during this study are included in this published article. Additional datasets or simulation files that support the findings of this work are available from the corresponding author upon reasonable request.