Power Electronics

Designing LED Lighting Systems For Optimal Light Output

An LED lighting system can be optimized for efficacy, footprint, lifetime and cost by varying the LED's current and controlling its temperature with a heat sink. To get accurate results, however, the dynamic nature of LEDs requires modeling their behavior

Find a downloadable version of this story in pdf format at the end of the story.

With the rising costs of energy and increasing governmental requirements for high efficiency lighting, the high brightness LED market is expanding rapidly. LEDs promise high efficiency, but the cost can be high, so how can a designer get more light out of an LED to reduce the number of LEDs required in an array? The most direct way is to increase the current, but that in turn increases temperature, which can degrade the light output, decrease efficiency and reduce lifetime. Thus, heat sinking is necessary. A proper LED lighting design will take into account all these factors and also the LED driver and LED string sizing. New tools have been developed to ease the LED lighting design process and allow users to make tradeoffs between high efficiency, small footprint and low cost.


The first goal a lighting designer needs to set is the light output of the system. This is typically specified using luminous flux, in units of lumens, which is a measure of visible light from a given source. Traditionally, the designer would go to an LED datasheet and look at the specifications for luminous flux and use those parameters to choose the number of LEDs required for the system. But as anyone who has studied an LED datasheet knows, LED behavior is dynamic based on the current being used to drive the LEDs and also the LED temperature.

Typically, the luminous flux is specified at a constant 25°C temperature using a short burst of current in the lab. But in reality, LEDs run hot and the temperature is considerably above the ambient. The best production LEDs on the market today are only about 25% to 30% efficient, which may be considered surprising given the energy efficiency hoopla surrounding them. In fact, this is great compared to the efficiency of a tungsten filament bulb, which may run around 2.2% for a 100W bulb giving off 1500 lumens. But the question is where does that 70% power loss in an LED go? Unlike a tungsten bulb which emits a significant amount of infrared radiation to give off heat, LEDs must get rid of heat through conduction. And that means heat sinks and temperature control are a must. What specific parameters should a designer be concerned with which vary with the LED current and temperature? The important ones include luminous flux, Vf (LED forward voltage drop) and luminous efficacy (luminous flux divided by the power consumed in units of lumens/watt) which is a measure of the efficiency of the LED.

The luminous flux of LEDs goes up with LED current, which can be useful if a designer wants to reduce the number of LEDs in the array to lower the cost (Fig. 1A). In fact, LEDs can often be driven with up to 2x or 3x the nominal current (check the datasheet for maximum current) to get more light output. But the tradeoff is high temperature which increases with increasing current for a fixed heat sink size (Fig. 1B). Higher temperatures mean decreased lifetime and reliability for the LEDs. This also lowers the light output of the LED, perhaps significantly (Fig. 1C). To lower the temperature, a larger heat sink can be used, but this will increase the cost and footprint of the design.

In contrast to the luminous flux, the luminous efficacy goes down with increasing current (Fig. 1D). This drop in efficacy may cause the loss of a governmental efficiency standard approval and certainly make the product less appealing from an energy conservation standpoint. In addition, the forward voltage of the LEDs increases with current and decreases with temperature which may affect the driver design. For example, in a series string of LEDs, the total forward voltage must be kept below the minimum input voltage for a simple buck design, otherwise a boost or buck-boost topology may be required. Thus, we see that the lighting designer must make compromises between cost, footprint, reliability and efficiency when designing an LED system. It's not as simple as just raising the current.

Continue on next page


Let's take a look at several scenarios using a high efficacy LED for a design targeting 2500 lumens. This can be done using conventional manual methods, or by using a design tool like National Semiconductor's WEBENCH LED Architect. This new tool draws from an extensive library of LEDs, drivers and heat sinks to design complete systems.

WEBENCH LED Architect enables designers to perform real-time comparisons and optimize complex lighting systems for performance, size and cost in minutes using graphical visualizations of the critical parameters. To start a design, the user enters the desired light output in lumens and is presented with a listing of suitable LED and heat sink solutions in both table and chart form. The user can tune the design with the unique WEBENCH Optimizer Dial, prioritizing size, efficacy and cost trade-offs. After selecting the LED and heat sink, different driver options are shown allowing the user to choose between driver topologies and LED string configurations. Graphical charts are utilized to visualize the compromises between footprint, efficiency and price.

After choosing the driver, all the components for the system are calculated along with the schematic and operating values such as duty cycle, currents and power dissipation. Electrical simulation is available for analyzing transient behavior. The designer can fine tune the bill of materials (BOM) by choosing different components from a library of over 20,000 passive components or by entering custom component values if desired. Lastly, the designer can order components for prototyping, share the complete system with others, or easily print a complete project report including schematics, BOM and performance characteristics.

Using the tool, the variables will be the heat sink thermal resistance (θSA in °C/W), the LED current, the LED operating temperature and the number of LEDs. The heat sink areas are calculated based on typical extruded aluminum profiles but other solutions, including high thermal conductivity board material, may be used. The scenarios are shown in table form in Fig. 2A and in graphical form in Fig. 2B. The first example is the smallest footprint case. With a small heat sink area we are limited in cooling capability, thus the LED current will need to be kept moderate and the operating temperature will need to be high. We end up with an array of 13 LEDs requiring a heat sink θSA of about 4°C/W, giving an area of 35cm2. The LED temperature is on the high side at 131°C and the price for the LEDs and heat sink is $47.10. The LED efficacy is medium at 92 lumens/W. With the relatively high temperature, the LED lifetime will be reduced.

The second case is the lowest cost scenario. To do this, we will raise the LED current to the maximum to reduce the number of LEDs required and use a low cost heat sink, with the tradeoff that the operating temperature will go to a maximum. In this case, eight LEDs are required, requiring a heat sink θSA of 2.4°C/watt and area of 81cm2. The operating LED temperature is very high at 140°C. Efficacy is the lowest at 74 lumens/W due to the combination of high LED current and high temperature. However, the cost for the LEDs and heat sink is the lowest of all the scenarios at $30.07. The sacrifice here with the high temperature is reduced LED lifetime.

The third scenario is a balance between small size, low cost and high efficacy. Here the current is relatively high, but the heat sink is allowed to increase in size to an area of 119cm2 with a θSA of 2.1°C/W. This lowers the temperature to 109°C, thus giving an efficacy of 88 lumens/W with somewhat increased lifetime, but the temperature is still on the high side. The cost is moderate at $34.27.

We go to the fourth example to get higher efficacy and lower temperature. This utilizes moderate LED current, but the heat sink is larger with an area of 175cm2 and a θSA of 1.8°C/W. The temperature is reduced to 78°C, thus increasing LED lifetime and yielding an efficacy of 108 lumens/W. The $46.08 cost is on the high side.

Continue on next page

The last scenario targets the highest efficacy. It does this by lowering the LED current to the nominal datasheet value and using a large heat sink. This keeps the temperature at a minimum of 48°C, which will result in the longest lifetime of the group. The results also show the highest efficacy of all the scenarios at128 lumens/W but the required heat sink θSA is very low at 0.7°C/W giving a very large area of 837cm2. With the low LED current, there are 19 LEDs required, so the cost is also very high at $88.75.

Thus, we see that the LED performance can vary over a very large range depending on the current and temperature of the LED. Also, there is no single best solution to the problem. High efficacy and long lifetime are achieved only by sacrificing cost and area. On the other hand, going for low cost and small footprint requires a compromise between efficacy and lifetime, which are the two main selling points for LEDs. The good news is that LED performance is improving at a breakneck pace with better efficacy bins and new LED models appearing often. So it is important to monitor the marketplace on a regular basis to keep up with the latest releases.


In the LED system, after the number of LEDs, the heat sink and the LED current are determined, a suitable driver must be found to power the LEDs. To achieve the high efficiencies that go along with LEDs, this typically means a switching regulator is required. Two interacting issues arise: The topology of the driver must be decided upon and the LED array must be determined.

At this point, the relationship of the LED array voltage to the input voltage range becomes a critical parameter. If the total voltage of the LED array is less than the minimum Vin (plus a bit extra to account for losses across the switch), then a buck topology can be used. This is the simplest topology to implement and it carries the advantages of high efficiency and low input current requirement.

If the total LED voltage is above the maximum input voltage, then a boost topology is called for. This is also a proven topology but has the disadvantage of requiring a high-voltage, high-current FET depending on how much the voltage must be increased. This may result in higher cost and larger footprint.

Lastly, if the LED array voltage is between the maximum and minimum input voltage, then a buck-boost topology is required. This allows the most flexibility with the LED array voltage, but the driver design is the most complicated and expensive to implement. Also, like boost, it has the disadvantage of requiring high current if the input voltage goes below the LED voltage.


The LED array configuration can be arranged to allow for a desired driver topology. If a buck topology is desired, the LED array can be broken down into parallel strings such that the LED string voltages are less than the minimum input voltage. However, if parallel strings are combined on the same single output driver with one current sense resistor, it has the disadvantage that the current in each string may be different due to the variations in LED forward voltage. This may lead to differences in brightness and temperature and eventually variations in LED lifetime between the strings. This can be solved by using a driver with multiple outputs and current sense resistors, or by using multiple single output drivers.

To avoid the current sharing problem, the LEDs can be arranged in series. However, the total LED voltage may be quite high. If the LED voltage exceeds 60V, additional safety features and certifications may be required to meet governmental standards.

Taking the balanced optimization from the previous 2500 lumen example which in a series configuration has nine LEDs for a total string voltage of 28.6V, we now examine several driver scenarios using an input voltage range of 15V-25V. Fig. 3 is a chart of the total system footprint vs the system luminous efficacy including the driver losses. The circle size is proportional to the cost. The lower right of the plot shows the highest efficacy, lowest footprint solutions. These use a single series string of nine LEDs. The total LED voltage of 28.6V is above the 25V maximum Vin, so this requires a boost driver topology.

Continue on next page

The middle of the plot shows buck driver solutions which have the LEDs broken into three separate strings of 9.5V each, so they are below the minimum input voltage of 15V. The chart shows three separate drivers being used, but they could also be combined to use one driver with one current sense resistor to reduce the cost, but at the risk of uneven current sharing. The last group in the upper left uses two strings of 5 LEDs, each of which results in an LED string voltage of 15.9V.

Another scenario, shown in Fig. 4, is to vary the input voltages in order to get different driver topologies using one series string of LEDs. This table zeros in on the driver performance only and does not include the LED and heat sink contributions. In the first case, we target a buck topology so we use an input voltage range of 35V to 40V, which is above the 28.6V LED string voltage. The driver performance, excluding the LEDs, produces an efficiency of 93% with a component area of 3.5 cm2 and a cost of $1.90.

To get a boost topology, we lower the input voltage range to 20V to 25V so it is below the LED voltage. The efficiency in this case is the same at 93%, but the footprint is larger at 6.3 cm2 and the cost about a dollar higher at $3.02.

In the last case, we use an input voltage of 25V to 35V which results in a buck-boost topology since the LED voltage is between the maximum and minimum input voltage. This gives a lower efficiency of 88%, a higher component footprint of 8 cm2 and a cost of $4.04.

Thus, we can see that the LED driver comprises just 5% to 15% of the total system cost and the LED driver efficiency is high at 93% vs 24% for the total system (see Fig. 5 ).

The last step in the driver design process is creating the actual design. Driver design tools automatically generate a bill of materials and allow the user to change passive components and run simulations to verify system performance. Fig. 6 displays the results of an input transient simulation run in the WEBENCH LED (http://www.national.com/analog/led#software) Architect design tool, showing the effects of changing the input voltage on the LED current.

Download the story in pdf format here.

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.