Learn the fundamentals of thermal design and how it can be implemented using practical methods and materials to produce acceptable high-power LED applications.
L
ED manufacturers are experts regarding the thermal design inside of the LED package. They can provide the relevant data to describe the thermal performance, and they also have good ideas about how an LED should be soldered to the circuit board. But when it comes to the thermal path between the LED leads and the environment, you are on your own.
Interestingly, LED data sheets often specify such parameters as maximum DC current and thermal resistance from junction to ambient, which depend greatly on the ability of your board and mechanical hardware to conduct heat way from the LED. Yet, these mechanical details are not usually discussed, beyond a suggested solder pad diagram.
To be sure, application notes sometimes provide some good ideas. But they often focus on one or two implementations, without a lot of discussion of the theory. Metal-substrate boards are a favorite technology, because they work well, despite the high cost that will often exclude them from consideration. This article doesn't discuss metal-substrate boards.
What engineers need is a discussion of the fundamentals of thermal design, and how it can be implemented using practical methods and materials, to produce acceptable high-power LED applications, such as that in Figure 1.
LED-specific issues
An LED dissipates heat, which causes the die temperature to rise. Although every semiconductor device has an upper temperature limit, LEDs have special issues with high temperature. The light output reduces significantly at the upper limit, by as much as 50%, which defeats the purpose of driving the LED at high current. The lifetime (usually specified as the time for the light output to degrade by 50%) is significantly reduced by high-temperature operation, from practically infinite to as short as a few months. Also, the wavelength (color) of the light is temperature-dependent, and is sometimes critical. Therefore, it is often a design goal to keep the LED die temperature well below the spec limit.
Low-power LEDs, such as parts in T 1¾ or small SMT packages, are not intended to dissipate much power, and have rather high internal thermal resistance. They have no good thermal path to conduct the heat from the die to the ambient, so their current is limited to the 20-mA range. These devices are usually used for indicators, and higher brightness is not necessary. More recently, white LEDs in T 1¾ packages have been used in flashlights and other lighting products, probably due to their wide availability and low cost, although it is not a very effective package for that application.
A high-power LED (loosely defined as an LED that dissipates more than 0.25W) has the die attached to a thermally conductive material, usually a piece of metal, that can conduct the heat out of the package, and produce a minimal internal temperature rise.
Many smaller high-power LEDs rely on a leadframe, often using several pins, to conduct the heat to solder pads, as well as provide the electrical connections. The solder pads are then designed to spread the heat into an area much larger than the LED itself, so it can be transferred into the air. This method is limited, because the copper pads are very thin, and thus can't conduct the heat over large distances. An example is the OSRAM LW G6SP, a white LED that can dissipate about 0.5W.
A larger high-power LED needs a better thermal path. The die in such a device is mounted on a metal slug that is exposed on the bottom of the package. The slug is then soldered to a circuit board. The dissipation may be several watts, and no amount of circuit board copper can transfer that heat to the air. The heat must flow from the LED, through the board, and into a metal heat sink that is bonded to the bottom side of the board.
So, how do we transfer heat from the top of the board to the bottom? FR4 epoxy-glass material will conduct some heat. It has a typical thermal conductivity of 0.4 W/m-K. If the LED copper pad measures 0.3 in. square (7.61 mm), and the board is 0.031 inch thick (787 µm), then the thermal resistance will be (787 µm)/((0.00761 m)2 * 0.4 W/m-K) = 34 K/W. This is way too high, and we can do better.
Here is a better solution:
Create a top-side pad that is somewhat bigger than the LED, and a matching pad on the bottom. Then fill the pads with thermal vias, which are plated throughholes specifically intended to conduct heat. The reason the pads need to be large is to spread the heat to a wider area, so it will conduct well through the board and, especially, the bond to the heat sink.
Vias are the copper-plated holes that normally connect traces on the top and bottom of the board, or to internal layers. In power circuits, they can also be placed just to conduct heat vertically, in which case they are called thermal vias. If you completely fill the LED pad with vias, you can get about 3% of the area filled with copper. This may not seem like much, but because copper is about 1000 times better than FR4 as a thermal conductor, the board will conduct heat very well.
Solder sucking
There is a problem with filling a solder pad with vias. In the soldering process, the pad is covered with solder paste, and the LED placed on it, and then the board is heated in an oven until the solder melts. But each via tends to act like a capillary tube, and sucks some of the solder down the hole. Since these vias are packed closely throughout the pad, they will suck most of the solder away, leaving a marginal solder joint. This is a common problem with BGA (ball grid array) packages, and several solutions exist.
A simple way to avoid solder-sucking is to not put the vias in the solder. Just put vias around the edge, outside of the solder joint. The problem with this approach is that it requires the thin copper on the top surface to carry the heat to the vias. That adds thermal resistance, but it may be acceptable. In high-density applications using many LEDs, there may not be room for such large pads. Also, the LED anode and cathode connections often limit the areas where the pad may be extended.
The best performance results from copper-capped vias (also called via-in-pad technology). After the vias are drilled and plated with copper, they are filled with resin and then more copper is plated on top, leaving a flat pad with no exposed holes. It is somewhat more costly than standard boards, but not excessively so.
Let's consider an example.
We want to use an OSRAM Golden Dragon white LED (LW W5SM), running at 500 mA. That will dissipate about 1.7W.
We are going to bond this board to a heat sink, which we can assume will not exceed 60°C. The design of the heat sink and what keeps it cool is not our problem here. Our task is to get the heat from the LED to the heat sink.
Although the LED is rated to 125°C, we want to maintain a reasonable light output and lifetime, so we shall not exceed 90°C die temperature. This is a temperature rise of 90 – 60 = 30 K, or a thermal resistance of 30 K / 1.7W = 17.6 K/W. This is the sum of the thermal resistances of the LED package (11 K/W), the board, and the adhesive attaching the board to the heat sink.
There is nothing we can do to reduce the internal thermal resistance in the LED.
The board thermal resistance is improved by spacing the vias close together, or plating thicker copper in the holes, or making the pad wider, or making the board thinner. Some of these have practical limits, which are defined by the people who build the boards.
The thermal resistance of the adhesive can be reduced by making the glue line thinner, or using a product with better thermal conductivity, or making the pad wider.
Note that making the pad wider came up twice. While it is true that a wider pad will conduct heat better, there is a problem with this approach: The heat from the LED must conduct laterally through the pad before it can conduct down through the board. If the top-side copper was thick, this would work great, but thin copper foil has its limitations.
The copper foil on a circuit board conducts heat like any other piece of copper. If we calculate the thermal resistance of a square of copper foil (for heat moving laterally), we will see that it is the same regardless of the size of the square. (This does not account for heat dissipating into the air.) If we assume a thickness t, and a dimension x for the square, then the thermal resistance is (x) / ((t * x) * 350 W/m-K) = 1/(t * 350 W/m-K). Since any copper area can be approximated by squares in series or parallel configurations, this is a useful way to predict temperature rise going through copper planes.
A via is a vertical tube made of copper, formed by plating copper onto the walls of a hole drilled through the board. You can calculate the thermal resistance for a single via as follows. Assume that the copper is plated 1 mil (25 µm) thick. That seems to be the standard for plated-throughholes. The cross sectional area is then p * d * t, where t is the thickness of the plating, and d is the diameter of the hole. Let's assume we are using 15-mil (381 µm) diameter holes, and the board is .031 in. (787 µm) thick. Then the thermal resistance of a via is (787 µm)/ (( p * 381 µm * 25 µm) * 350 W/m-K) = 74 K/W.
We will mount the LED on a 0.3 in. square pad and put the vias on 30-mil centers, which allows 100 vias (Figure 2). So the total thermal resistance is 74 / 100 = 0.74 K/W. Actually, it is perhaps twice this number, because some of the vias are connected to the LED slug by the top copper pad, which adds thermal resistance.
Now let's look at the adhesive. We are going to bond the board to the heat sink using a thermal epoxy such as 3M TC-2810, which has a conductivity of 1 W/m-K. If we assume the epoxy is 5 mils thick (127 µm), and the pad measures 0.3 in. (7.61 mm), then we get a thermal resistance of (127 µm)/ ((.00761 m)2 * 1 W/m-K) = 2.2 K/W.
So the total thermal resistance from the die to the heat sink is 11 + 0.74 + 2.2 = 14 K/W, well within our goal.
