There are two golden rules to thermal design: start simple and start early. The heat-flow path from the junction to the ambient, usually air in the local environment, determines a component’s temperature. Controlling temperature is therefore a system design challenge. Designers should employ a top-down approach (see the table) that increases confidence in the thermal design as the design progresses.
Heat-transfer correlations abound for natural and forced convection inside pipes and ducts, airflow over flat plates, and other scenarios. However, we advocate using rules of thumb, because correlations give a false impression of accuracy and are more difficult, and hence error-prone, to use.1
The heat-transfer coefficient for a 0.2-m long plate varies from 5 Wm-2K when horizontal in natural convection to 15 Wm-2K for forced convection at 3 ms-1, often the upper limit due to flowgenerated noise. To add a contribution for thermal radiation, we recommend using 10 Wm-2K for natural convection and 10 to 20 Wm-2K for forced convection.
First, calculate the temperature of the air in the enclosure if it’s a sealed system or the airflow needed to provide an air temperature rise of 10°C to 15°C for a forced convection system. The resulting internal air temperature is then used to calculate the printed-circuit-board (PCB) temperature. Finally, a package thermal metric like junction-to-board resistance (θJB) is used to calculate component temperatures.
For simple systems, the total thermal resistance can be considered a series resistance sum from component to board, board to internal air, and internal air to ambient. Systems that are more sophisticated need to be treated as a network of resistances, requiring enough experience to know what assumptions to make about the resistance network itself, and then to calculate the resistance values.
Heat transfer is always three-dimensional, so hand calculations and rules of thumb are of limited applicability. In practice, boards aren’t uniform in temperature due to the distribution of the heat sources and non-uniformity of the flow. However, the main disadvantage is that these approaches provide little insight into how the system can be improved.
COMPUTATIONAL FLUID DYNAMICS
We suggest a top-down approach, building a simple computational fluid dynamics (CFD) model of the entire system at the earliest phase of the design to:
• Provide insight into the way the system behaves using 3D flow and temperature visualization
• Explore different cooling strategies early in the design cycle
• Serve as the platform for incorporating further detail as the design evolves
• Build confidence in the design as the fidelity of the thermal model increases.
The duration of this phase is often very short, sometimes requiring just a few days. The CFD tool must support very rapid turnaround, from geometry creation to results in a couple of hours. CFD tools employ a computational mesh to split the model into many small control volumes (cells) over which the basic equations for flow and heat transfer are solved. Each cell provides predictions of temperature, air speed, and pressure.
To be useful in conceptual design, the meshing has to be 100% reliable with no effort required on the part of users to control mesh quality and density. This tends to rule out general-purpose CFD tools that use automatic (i.e., algorithmic) meshing techniques in favor of EC-specific (electronics cooling) software.
The focus at this stage in the design is to investigate the basic mechanisms involved in removing heat from the system. For aircooled electronics, the objective is to predict the system-level airflow. So what does the model need to include?
The first step is to create a simple box for the enclosure, with a 2D bounding- box representation of any vents. The array of small holes should be accounted for using a porosity and loss coefficient. CFD tools specific to EC should provide this capability by allowing the diameter, pitch, and layout of the holes to be defined.
Internal electromagnetic compatibility (EMC) screens should also be included. On top of that, it’s a good idea to include buoyancy in forced convection-cooled systems, because natural convection effects can influence the flow behavior. The inclusion of buoyancy should not cause a delay in obtaining simulation results.
Thanks to their low cost, axial fans are preferred for forced cooling. A 2D representation of an axial fan is typically adequate at this stage, if care is taken to measure the correct open area for the fan. At minimum, use a linear fan curve to represent the fan’s performance.
In general, it’s considered better to draw air through the box, exhausting hot air to the surroundings, since this produces a more uniform flow through the enclosure. While it reduces the risk of dead spots, the fan will run hotter, affecting longevity.
It may be tempting to specify the cheapest, smallest fan to cool the system. This usually isn’t a good idea, though, because fans are noisy when operating close to their maximum flow rate and their reliability is compromised. Instead, specify a fan that can deliver two or more times the airflow needed to cool the system and de-rate it by reducing its speed.
Designers should make everything in the model thermally conducting, including the enclosure. PCBs should be represented as thermally conducting blocks, with an isotropic thermal conductivity of 5 to 10 Wm-1K-1, depending on the expected copper content. The total anticipated heat dissipated for each PCB should be included to ensure the total heat input to the system. The model should take no more than an hour to build and even less to solve.
Despite the approximations, the simulation result will back up earlier hand calculations and provide valuable information about the system-level airflow. Component temperature rises can be estimated using the θJB metric.
Thermal simulation can contribute to physical partitioning decisions, such as helping to determine the impact of splitting functionality between a PCB and mezzanine board or between components.
Exploring potential cooling strategies to identify the best system-level solution before defining the system architecture is a lowcost, high-value activity, providing insight into the way the system behaves thermally. Much of the thermal design work during this phase is speculative, but that’s the point.
The optimum cooling strategy depends on the specifics of the system. However, the thermal engineer’s role is clear: Explore as many options as possible before the mechanical design solidifies so wise choices are made early.
Just because it looks like the system will get too hot doesn’t necessarily mean it needs a fan. There may be better options, such as adding a heatsink to one or more components or using the enclosure as a heatsink by adding a gap pad. The thermal engineer’s job is to split up the allowable temperature rise, leaving enough headroom as a safety margin.
During architectural design, refine the system model wherever there’s a large temperature rise to accurately resolve such jumps, particularly in the PCBs and in any thermally critical components (Fig. 1). As the architectural design progresses, expectations emerge about the components that will be used, their power consumption, and the need for heatsinking.
As a minimum representation, use a thermally conducting block to represent high-power, high-power-density, and temperature-sensitive components. For plastic packages, use a thermal conductivity of 5 Wm-1K-1. For any ceramic components, use a thermal conductivity of 15 Wm-1K-1. Non-critical components with an available power estimate can be treated as footprint heat sources, with the heat directed into the PCB, refining the local board temperature from which the junction temperature is estimated.
Obtaining component positions before layout is finalized is a challenge, but the designer’s best guess is likely to be close to what will be tried first when it comes to placement. A “what-if ” analysis could save time for both the thermal engineer and electrical engineer later on. Further, the discussion may flag key information, such as the need for the board to be shielded.
Placing the fastest components close together, or even in the same location on opposite sides of the PCB, minimizes timing issues. Unfortunately, these faster components often dissipate lots of power. Thus, a lack of thermal design before layout closure can be disastrous. Minor concessions, such as arranging a row of hot components across the oncoming airflow, can produce a much better overall solution.
It’s generally better to align the components’ main axis with the predominant airflow direction, particularly in densely populated systems. That’s because it increases the total airflow over the board and reduces flow separation on top of the components, delivering more effective cooling.
One of the biggest challenges in thermal design is getting useful estimates of component power consumption. Most products are designed for continuous operation, so the time-averaged power dissipated by the component during normal use is required, not the nominal or maximum rated power for the part.
Dialogue between thermal, mechanical, and electrical engineers to access thermally relevant design data and feed back results of design changes is a characteristic exhibited by companies with mature thermal-design processes. Component power estimates should be checked and re-checked as the design evolves, since they can change significantly from their original values.
Heatsinks increase the effective area between the component and the air, yet it’s important not to get carried away. The optimum is the lowest-cost heatsink that reduces the component temperature below the maximum design value by some safety margin. Using more fins provides more surface area, but they also reduce the area available for airflow. This increases the flow resistance, causing more flow to go around, rather than through, the heatsink (Fig. 2).
Avoid anything fancy at this stage. If the airflow over the component is mainly in one direction, use an extruded heatsink with fins parallel to the flow. If not, maybe there is latitude for improving the airflow distribution through the enclosure. If airflow is at an angle or impinging onto the component, though, use a pin-fin heatsink.
Other aspects of the model should also be refined, including fan selection. EC-specific CFD tools will allow you to de-rate the fan and automatically recompute the fan curve for the lower speed. Such tools often have a library of fans that can be dragged into the model, so this should become a “plug-and-play” activity. It can even be automated as a series of runs, swapping the fan for each run.
DETAILED DESIGN PHASE
Once the architectural design is complete, overall system design is set. This leads to two main thrusts to the thermal design. One is refining the model of the enclosure, including any fans and vents. The other is to refine the model of the PCBs and components. This continued refinement of the models serves to provide greater confidence in the validity of the thermal simulation.
Thermal performance varies considerably and is designed to exploit different heat-flow paths, either down to the board or up to a heatsink. Package selection is largely based on cost and electronic performance. But the least expensive package can be the most expensive overall when you include the cost of the thermal solution. Consider thermal performance during package selection.
The block models described earlier are crude, providing “indicative” case temperatures. In conjunction with package selection, the component thermal model should be updated to at least a tworesistor compact thermal model.2 This will capture the package’s thermal behavior as two resistances: junction-to-case and junction- to-board, or preferably a Delphi compact thermal model.3
Alas, the availability of Delphi models from suppliers is limited. Fortunately, one Web-based tool allows system builders to create both two-resistor and Delphi compact thermal models for a wide range of chip packages using defaults based on the package outline.4 Knowledge of the die size further improves model accuracy in the absence of other vendor-supplied data. Detailed models that directly represent the thermally important package internals can also be created.
Heatsink design should be finalized before PCB layout is complete. If a heatsink with a base that’s larger than the component is needed, space on the PCB will need to be allocated for mounting. Heatsinks partially block the flow, so components in the wake behind a heatsink are likely to become hotter as they get less airflow. As a result, the layout could be affected.
Heatsink cost correlates well with weight. Thus, cheap also means light, which is desirable from a reliability perspective. Heavy heatsinks place more stress on the package interconnect and require attachment to the PCB. Some EC-specific CFD tools with robust meshing allow you to automatically optimize the design of a heatsink for a particular duty, minimizing the weight for a specified target temperature.5 If surface area increases minimally, maybe a heatsink made from a thermally conducting plastic will suffice.
Before PCB layout closure, the thermal design should have evolved to where there’s full confidence that the packages can be cooled. Further, for any packages requiring heatsinks, a working design of the heatsink should be available. Usually, there’s some expectation about the required PCB stackup well before routing.
The thermal representation of the PCB can be improved by estimating the percentage coverage of copper for each layer, (e.g., 20% for signal planes and 90% for power and ground planes) and including these discretely in the model. Once the boards are routed, the thermal design can be further refined by importing the trace details from the EDA system. This will account for the local variation of copper in each layer and any thermal and electrical vias.
PROTOTYPING THE SYSTEM
Prior to the first physical prototype, it’s common to perform a single, highly detailed thermal verification simulation (Fig. 3). If thermal issues were considered from early in the design cycle, final verification should be a formality before physical prototyping starts. Early consideration of thermal issues is perhaps the hallmark of a mature thermal design process.
Companies with sophisticated thermal design processes have thermal sign-offs throughout the project and consider thermal before committing to the concept. They also learn from experience. By applying the insights gained into the product’s thermal performance and measurements on prototypes, they can improve their thermal modeling and seek ways to push thermal design higher up the design flow.
1. “Sense and nonsense of heat transfer correlations applied to electronics cooling,” Lasance, C.J.M., Proceedings of the 6th EuroSimE Conference, 18-20 April 2005, pp. 8-16
2. “Two-Resistor Compact Thermal Model Guideline,” www.jedec.org/download/search/JESD15-3.pdf
3. “DELPHI Compact Thermal Model Guideline,” www.jedec.org/download/search/JESD15-4.pdf
5. “Simulation-based design optimization methodologies applied to CFD,” Parry, J., Bornoff, R., Stehouwer, P., Driessen, L., Stinstra, E., Proceedings of 19th SEMI-THERM Symposium, 11-13 March 2003, pp. 8-13