Electronics Cooling Lessons from "Hot Air Rises & Heat Sinks" by Tony Kordyban
1. The temperature rise of air passing through an instrument enclosure is a poor indicator of component temperatures; you need to measure individual component temperatures.
2. Know maximum allowable component temperatures in reference to other more identifiable temperature landmarks, e.g., glowing metal, melting solder, boiling water, flesh-burn point, environmental air temperature extremes, freezing water, etc.
3. Environmental test chambers may not simulate the actual environment in an instrument, especially when the test chamber is heated/cooled by forced convection (blowers) and the actual design relies on natural convection.
4. Thermally conductive epoxies/greases are poor conductors when compared to metals (such as copper heat spreaders and aluminum sinks). Their use is justified but they rarely have the same impact on reducing total thermal resistance as a good heat sink with adequate air flow. Chips with diamond heat spreaders have extremely high thermal conductivity and allow heat sinks to be placed without epoxies or grease, but they are expensive.
5. CFD is very useful for designing board layout of components, especially if natural convection is the cooling mode.
6. A heat sink is not a sink for heat; it really only transfers heat from one location to another very efficiently by virtue of high conductivity and a large surface area for convection. Common heat sink design failures include a) fins or pins that are spaced too closely to allow adequate air flow, b) fins aligned perpendicular to the prevailed air flow, c) locating hot heat sinks in places where people may touch them.
7. Typically, less than 10% of all components placed on a board are critical in terms of cooling them to below their maximum operating temperature. Know the maximum limits of the critical components and keep those temperatures satisfied with some derating under the worst-case environmental conditions. Operating limits are usually based upon reliability and not functionality - a component that operates hot but is below the limit may still drift in performance.
8. Allowing for fan cooling from the very beginning of a design phase is prudent since board designs may change and component temperatures may be hotter than expected; fan retrofits are problematic.
9. Hand calculation of natural convection is difficult and very approximate in complicated geometries; this is where CFD can be very useful and cost-effective.
10. Electronic component heat fluxes are always going to increase despite lowering operating voltages. At some point, cooling by natural air convection gives way to forced air convection, and eventually, cooling by forced air convection will give way to liquid or phase-change convection.
11. Fans curves are important in determining the actual flow rate that a fan will deliver, however, determining the system pressure is often difficult. Rule of Thumb: use a fan that is rated at twice the flow that you really need.
12. Setting thermal design goals for an electronic instrument is good practice.
13. The use of a manufacturer's to determine component junction temperature is flawed in at least two ways: a) it is not apparent where you should measure or locate Ta since it is not constant in an actual instrument enclosure, and b) most manufacturers make this measurement on an isolated component in "still air", which is hardly representative of an actual system. As a result, this spec is nearly worthless.
14. Side-by-side redundant fan designs can be useful, but if one fan fails there could be the problem of short-circuiting air flow.
15. Do not forget the following basic notions of convection heat transfer: a) heat flows from hot to cold and no matter how large a fan you put in an enclosure, you will never make the component cooler than the local air temperature; b) the faster the air flow, the higher the convection coefficient and heat transfer rate; and c) the more surface area exposed to the air flow, the higher the heat transfer rate.
16. Some rules concerning thermocouple installation and voltage effects: a) do not mount TCs directly onto electrically live conductors; b) use a meter that has good electrical isolation between the TC measuring section and the rest of the meter; and c) check for immediate effects on the temperature reading by switching the power on/off to the instrument.
17. CFD predictions are only as accurate as the input data and boundary conditions. In determining component power dissipation, be sure to recognize frequency dependence and consult the circuit designer.
18. Pin-fin heat sinks are over-rated; avoid them unless your particular application requires such a geometry. Even though pin fin heat sinks work equally well (or poorly) with any air flow direction, there are some misconceptions: a) they do not have more surface area than an extruded fin heat sink of identical fin thickness; b) the pin spacing is often too small to allow adequate air flow through the pin array; and c) in natural convection or with low forced convection velocities, the flow past the pin produces vortices and turbulence (i.e., pressure drop) that reduces total air throughput and heat transfer rate. For most electronic component sinks under natural convection, about ¼" spacing between fins is about right. Pin fins do have a genuine advantage in integrated blower-sink packages which utilize impinging flow since the flow can spread in a radial direction.
19. CFD software can become a valuable piece of test equipment by using "what-if" type scenarios to investigate new ideas or troubleshoot unexplainable experimental results.
20. More thermocouple advice: it really doesn't matter if you crimp, weld, or solder TC junctions - just so that you don't use too much solder and that the resulting bead is small enough so that the entire junction is at the same temperature during your measurement. Rule-of-Thumb: use about 30-gage wire with beads about the size of the tip of a BIC pen for electronic component measurements.
21. Take advantage of natural convection, even in forced air instrument designs. Do not have a fan pulling air in a direction opposite to the natural rise of hot air. Opposing mixed flow can give worse cooling than natural convection alone.
22. Component package power limits depend upon a number of things, such as: a) natural convection or forced convection cooling; b) local air temperature; c) copper content in PCB; d) neighboring components; e) operating temperature limits of component; and f) package size and composition.
23. Temperature comparisons in terms of percent are meaningless and confusing since they depend upon which temperature scale is used and whether the comparison is based on a temperature difference or absolute temperature.
24. Convective heat transfer depends upon ambient temperature and fluid velocity. Natural air convection at 50C may be more, or less, effective for cooling components than force air convection at 70C, depending upon the type of component and its placement on a circuit board.
25. Experimentally determining the steady-state temperature of a massive component (e.g., a transformer) may take a long time, sometimes on the order of 2-3 hours. The lumped capacitance method (LCM) of transient conduction can be useful in estimating the thermal time constant of a component if material masses and properties are known. It takes about four time constants for a system to attain 98% of its steady-state temperature rise (or decay).
26. A complex electronic assembly may have more than one operating temperature limit, and the "limiting limit" may change under different environment conditions.
27. Conjugate heat transfer - the operating temperature of a component depends upon convection and conduction into the PCB. How much enters the PCB depends upon the amount of copper in the board, where the component is located on the board (edge vs. middle), and the proximity and type of neighboring components.
28. The manufacturer's specification is often the only data available to estimate junction temperature, even though it is recognized that a single resistance cannot possibly represent all the heat paths from a component junction(s). In reality, there should be resistances specified between the junction and i) top of case, ii) sides of case, iii) bottom of case (or board), and iv) leads. In addition, there are even heat paths and resistances between each of these four latter locations. A single can only aid in estimating the junction temperature, which should be done in the most conservative manner.
29. Thermoelectric coolers have there place in situations where component temperatures must be precisely controlled or must be below ambient air temperature. In all other cases the use of thermoelectric coolers aggravate the heat dissipation problem by introducing additional joule heating which must be removed from the system. The amount of joule heating is typically 2 to 4 times the amount of heat pumped by the device, so the additional heat load is enormous.
30. Thermal design and testing of electronic instruments is like a house of cards - the entire foundation is built on some rather uncertain concepts and data: a) controlling temperature is thought to improve component reliability, but when or if this applies is controversial; b) the maximum operating junction temperature for a particular component is really a best-guess, rather than one based upon documented reliability tests; c) components vary considerably from lot to lot, so specs should be quoted with some uncertainty value; d) the problem (see #28) yields significant uncertainty in extrapolating junction temperatures from case temperatures; and e) thermocouples are easy to use but yield about 2C uncertainty in temperature measurement.