news

Thermal Resistance Network and CFD Simulation: Engineering Methodology for Quantitative Design of Heat Sinks

First, from a one-dimensional thermal resistance network to a three-dimensional temperature field

The starting point of the heat sink design is often a diagram of the thermal resistance network. The path of heat from the chip junction to the ambient air is decomposed into: junction to shell (Rhtjc, chip package internal resistance), shell to heat sink (Rhtcs, TIM thermal resistance), heat sink to environment (Rhtsa, convection + radiation). Among them, Rhtsa can be decomposed into the diffusion thermal resistance of the heat sink substrate (Rhtspread), the one-dimensional thermal conductivity thermal resistance of the fins (Rhtfin) and the convection thermal resistance (Rhtconv). Series circuit model: total thermal resistance = Rhtjc + Rhtcs + Rhtspread + Rhtfin + Rhtconv.

This centralized parameter method is fast and effective in initial estimates, but the biggest drawback is that it assumes a uniform temperature distribution, when in fact there is a violent 2D/3D thermal diffusion effect on the substrate below the chip. For advanced chips with local heat flux up to 200 W/cm ², the diffusion thermal resistance may dominate, and even lead to the formation of "hot spots" in the substrate, causing the local temperature to be much higher than the average temperature. CFD simulation must be relied on at this time.

The core equations of computational fluid dynamics simulation

CFD solves three coupled partial differential equations:

1. continuity equation
   (Conservation of mass): Would/Would + ∇·(ρ u) = 0
   

2. momentum equation
   （Navier-Stokes）：∂(ρu)/∂t + ∇·(ρuu) = -∇p + ∇·(τ) + ρg
   

3. energy equation
   ：∂(ρh)/∂t + ∇·(ρuh) = ∇·(k∇T) + S_h
   

For heat conduction within a heat sink, the energy equation is reduced to the solid heat conduction equation (with zero convection term). For the air domain, a complete turbulence model (the most common k-ε model or the more advanced SST k-ω model) needs to be solved to accurately capture the velocity and temperature within the near-wall boundary layer - because
80% of the convective heat transfer coefficient depends on the viscous sublayer in the boundary layer with a thickness of only a few tens of microns
。

III. Ansys Icepak: Facing Complex Surfaces and Multiphysics

Icepak is based on the Fluent solver and uses an unstructured mesh (tetrahedral/hexahedral core), which is highly adaptable to curved geometries (e.g. circular fins, special-shaped air ducts). Icepak's unique advantage is that it can be seamlessly coupled with Ansys Mechanical and Maxwell for electro-thermal-structural three-field analysis. For example, heat sinks in high-power RF amplifiers need to consider both the spatially distributed heat source generated by electromagnetic losses (from Maxwell), the change in contact thermal resistance caused by thermal deformation (from Mechanical), and the fatigue life assessment under transient thermal cycles. This coupled simulation is far more accurate than isolated thermal analysis.

As a meshing strategy, Icepak recommends generating prismatic meshes at the solid-fluid interface, at least 3-5 layers, to resolve the boundary layer temperature layer. For typical CPU heatsinks, the meshes are typically 5 million-20 million, and the solution time is about 2-4 hours on a 16-core workstation.

FloTHERM: The king of efficiency focusing on electronic heat dissipation

Simcenter FloTHERM uses Cartesian meshes (orthogonal meshes), which are generated almost instantaneously without user intervention. Although the approximation of the curved geometry produces a step error, this error can be controlled within the engineering acceptable range for flat fin heat sinks commonly found in consumer electronics (

FloTHERM
Command Center
The module has powerful DOE (Design of Experiments) and optimization capabilities. Engineers can define objective functions (minimum thermal resistance or minimum weight), set design variables (fin height, spacing, thickness, fan speed), and let the software automatically iterate hundreds of simulations to find the laws of the vital few frontiers. This process is almost impossible to do manually.

V. Key traps for simulating boundary conditions

Simulation accuracy is highly dependent on the authenticity of the input boundary conditions. Here are three common pitfalls:

1. Heat source assumption error
   : Simplify the chip as a uniform surface heat source, ignoring the multi-hotspot distribution inside it. Advanced practice is to use the power distribution map provided by the chip manufacturer, or calibrate it through thermocouple measurement.
   

2. Natural convection does not activate the gravitational term
   In natural convection cooling, buoyancy is the only driving force. Without activating the gravity term and setting the air density to the Boussinesq approximation, the simulation results incorrectly predict that there is little flow and the temperature is unusually high.
   

3. Radiation is ignored or overestimated
   : When the surface temperature is lower than 100 ° C, radiation usually accounts for only 5-15% of the total heat dissipation, which can be simplified. But if the surface is blackened with high emissivity (emissivity > 0.9), and the air flow rate is extremely low (
   

Six, grid independence verification and convergence criteria

Any CFD simulation must be verified for grid independence before formal analysis. Operation method: Generate three sets of coarse, medium and fine grids (the number of grids differs by at least 2 times), and calculate the temperature of key locations (such as chip junction temperature). Differences between grid and fine grid results

Convergence criteria are usually set as follows: energy residuals drop below 1e-6, momentum residuals drop below 1e-4, and monitor point temperature changes are less than 0.01 ° C for 100 consecutive iterations.

VII. Closed-loop calibration from simulation to testing

Simulation is never the same as physical reality. The most rigorous development process is: thermal simulation design open mold production sample thermal test (using thermal imaging camera and thermocouple) comparison test and simulation deviation calibration simulation parameters (such as air side convection correlation, TIM thickness deviation) correction design secondary proofing. After two rounds of closed loop, the temperature difference between simulation and test can be controlled within ±3 ° C. This calibration database is the core knowledge asset of the enterprise.

VIII. Conclusion

Thermal simulation has revolutionized the design paradigm of heat sinks, moving from "experience-plus-test" to "predictive-driven design." But software is just a tool, and true expertise lies in setting up the right physical model, accurately interpreting the simulation results, and continuously calibrating the model through thermal testing. In the future, with the proliferation of AI-assisted simulation and high-performance computing in the cloud, real-time thermal simulation (Digital Twin) will become possible - each heat sink will have its digital twin, which reflects the operating temperature in real time and predicts the remaining life.

BQUQ is a professional metal heat sink manufacturer, please send us drawings, and our company will quote you within 12 hours.