Thermal inverse design for resistive micro-heaters

Khoi Phuong Dao, Juejun Hu

TL;DR

This paper introduces a novel "inverse design" method for creating resistive micro-heaters with precise temperature profiles. By directly calculating the required material properties (doping distribution) instead of iteratively simulating designs, this approach significantly speeds up the design process and enables complex, custom temperature patterns for advanced microelectronic and photonic devices.

What Problem Does This Paper Address?

Designing micro-heaters with specific temperature distributions has traditionally been a trial-and-error process. Engineers often rely on finite element method (FEM) simulations, adjusting heater geometries iteratively to achieve a desired temperature map. This is computationally expensive, limiting the complexity of designs and the number of iterations possible, especially for intricate temperature profiles. Existing inverse design methods for heating devices also suffer from this computational bottleneck, requiring full device simulations in each step. This paper aims to overcome this by developing a computationally efficient and automated inverse design framework for resistive micro-heaters.

Methods

The core of the proposed method is an inverse design scheme that directly infers the heater's electrical resistivity distribution (specifically, the doping profile in silicon) to achieve a target steady-state temperature profile. Instead of simulating the entire heater, the approach models each small "unit cell" of the heater independently.

  1. Unit Cell Modeling: Each unit cell's thermal behavior is characterized by two key parameters: an out-of-plane thermal conductance ($C_1$) and an in-plane thermal conductance ($C_2$). These parameters are derived from localized FEM simulations of small groups of unit cells, not the entire heater.
  2. Heat Flow Equations: The heat flow into and out of each unit cell is described by simplified equations based on temperature differences (for out-of-plane flux) and temperature gradients (for in-plane flux).
  3. Inverse Problem Formulation: Given a target temperature profile $T(x, y)$, the method deterministically solves for the heat generation rate $g(x, y)$ within each unit cell using the derived thermal parameters. This heat generation rate is directly linked to the electrical resistance and thus the doping profile (filament width $W(x, y)$) of the unit cell via Joule heating.
  4. Iterative Optimization (for Boundary Conditions): Since the exact temperature gradient at the heater's boundaries is often unknown, an iterative optimization loop is employed. This loop defines three zones (target, transition, peripheral) and uses the simulated response in the peripheral zone to update the target temperature for the transition zone in the next iteration, eventually converging to a design. This iterative part still requires full heater simulations, but it's significantly reduced in number (e.g., 5 iterations) and is primarily for handling boundary conditions, not for determining the core design.

Datasets

Not discussed in the paper. The paper uses simulated data generated through FEM simulations to derive constitutive parameters and validate the inverse design approach.

Key Insights and Ideas

  • Decoupling Thermal and Electrical Design: By focusing on the doping profile to control resistivity, the method effectively decouples the thermal and electrical design aspects, which is particularly beneficial for photonic devices where optical properties are also critical.
  • Computational Efficiency through Unit Cell Modeling: The key innovation is using localized unit cell thermal properties to bypass extensive full-heater simulations during the core inverse design step. This drastically reduces computational cost.
  • Automated Design Generation: The approach is fully automated, requiring minimal human intervention once the target temperature profile is defined.
  • Robustness to Initial Guesses: The iterative optimization for boundary conditions proved to be robust, converging to similar designs regardless of the initial temperature profile guess in the peripheral zones.
  • Assessing Design Feasibility: The method can also serve as a tool to assess whether a desired temperature profile is physically achievable given the device constraints and boundary conditions.

Key Results

  • Uniform Temperature Achieved: The inverse design method successfully created a doped silicon micro-heater that achieved a uniform temperature of 699.7 K with an RMS variation of only 0.18 K across a 30x30 µm² target zone, a significant improvement over a reference heater with a 102 K variation.
  • Complex Temperature Profiles: The method was demonstrated to generate non-uniform temperature profiles, including a linear gradient along a diagonal and a custom MIT logo shape, with close agreement to the design targets.
  • Scalability: The approach allows for scaling designs to different target zone sizes (e.g., 10x10 µm² to 40x40 µm²) by reusing derived boundary conditions from previous designs, further reducing simulation needs.
  • Application to Non-Planar Structures: The method was extended to a doped Si heater with a ridge waveguide structure, demonstrating its applicability beyond simple planar geometries.
  • Manageable Iterations: The iterative optimization loop required a maximum of only 5 iterations in the examples, making the overall process computationally manageable.

Limitations and Future Work

  • Mesh Size Limitation: The accuracy of the inverse design is currently limited by the mesh size of the FEM simulations used to derive unit cell properties. This can lead to deviations between analytically derived filament widths and the discretized grid. Future work could explore higher-resolution meshing or alternative methods to mitigate this.
  • Physical Viability of Targets: While the method can assess feasibility, some deviations from design targets were observed, indicating that certain complex temperature profiles might be physically unattainable under given boundary conditions.
  • Generalization to Other Materials: While the concept is adaptable, the paper focuses on doped silicon. Future work could explore its application to a wider range of resistive heating materials and structures.
  • Irregular Geometries: The paper notes that the approach can be extended to irregular geometries by calibrating thermal properties for parallelogram-shaped unit cells and compiling them into a lookup table. This is a promising avenue for future research.

Summary updated

Abstract

This paper proposes an inverse design scheme for resistive heaters. By adjusting the spatial distribution of a binary electrical resistivity map, the scheme enables objective-driven optimization of heaters to achieve pre-defined steady-state temperature profiles. The approach can be fully automated and is computationally efficient since it does not entail extensive iterative simulations of the entire heater structure. The design scheme offers a powerful solution for resistive heater device engineering in applications spanning electronics, photonics, and microelectromechanical systems.
BibTeX
@article{dao2024thermal,
  author = {Khoi Phuong Dao and Juejun Hu},
  title = {{Thermal inverse design for resistive micro-heaters}},
  journal = {Journal of Applied Physics},
  year = {2024},
  doi = {10.1063/5.0235526},
}

Loading related content...

Figure