Real-world numbers on production HB LEDs: 32% efficiency

Skeptic If you look at Philips data sheet for the K2, DS60, the junction to case is 5.5 C/W. Using the data sheet specification, typical Vf @ 350mA is 3.3V- so we will use that. 3.3V @ 330mA works out to 1.155 Watts. 1.155 Watts * 5.5 C/W = 6.3525ºC rise. So, if you had a 37ºC, 37ºC - 6.3525ºC = 30.6475ºC So, you'd need to keep the heatsink below 30C, which is quite practical. If you use your 700mA, the Vf typical is 3.5V, so you have 2.45 Watts. 5.5C/W thermal resistance * 2.45 = 13.475ºC rise. 37ºC - 13.475ºC = 23.525ºC for the heatsink temperature, which is more difficult to accomplish.

lm/W data at a die temperature of 37ºC have no practical interest at all. For example for a Luxeon K2 at nominal crt. of 700 mA you should keep the heat sink at 15ºC! If we have to compare light sources, please do it at normal die working temperature, 85ºC would be reasonable.

The meters were Fluke 189: Voltage DC Accuracy* ± (0.025%+5) Max. Resolution 1 µV Maximum 1000 V Current DC Accuracy* ± (0.15%+2) Max. Resolution 0.01 µA Maximum 10 A Note: * Accuracies are best accuracies for each function Factory Calibrated Note * Accuracies are best accuracies for each function Going to Fluke's extended accuracy sheet for this meter, and looking up the ranges: DC mA 400.00 mA 0.01 mA 0.15% + 2 DC V 50.000V 0.001V 0.03% + 3 25C Ambient LEDs were left to stabilize until the digits stopped moving, roughly an hour. LED were driven from a 100% fully Analog circuit, an array of eleven as I recall, that was summed together, then divided to find the average numbers- which should be obvious for those that look at the data. Since things sat before each individual measurement until the numbers stopped changing and then some, for each reading, the meter's accuracies for measuring watts should be pretty decent. For the lumen measurement, even spectral absorption for the fixture and the array was measured and corrected. The sphere's diameter was well over 10x the size of the array and fixture diameter. Volume wise, this is a much greater ratio. I've seen times when rounding things off can and does add considerable error, and having to locate more accurate and calibrated equipment to determine why there are bumps in the plotted data- and then realizing it was due to errors introduced by rounding off. So I just let Excel have it's way. Anyhow, lots of great ideas and discussion. I see some of the fluorescent crowd has joined in, welcome aboard! Now there is the usable light to consider. Often times you will find fluorescent bulbs in a fixture with a reflector, and many other light technologies, as often, you are trying to direct the light out of the fixture. There are plenty of metal, metalized, and white painted reflectors that will have a reflectance that ranges from 40-80% (with 80% being rare). As such, it is common to see 60% of say, a CFL bulb's light hit that surface. Guess what? You have to take a 20-60% hit, due to light absorption. Then there are the various diffusion mechanisms in the fixture, that can have serious losses. Then with many of these diffusion techniques, you have light that is kicked back and recycled off the reflector (which folks try to save money on also). One can take some high end T8 bulbs with a 90 lm/W rating, and put them in a ceiling fixture, and start scratching their head when the actual measurement comes in at 20 lm/W. Furthermore, as in the high end T8 bulb case, you also have ballast losses! Then you may have a situation where people are trying to save energy, and only have the bulbs on when they are present. So, they go in for 10 minutes, leave for an hour, and come back in, so on and etc. Many bulbs will start sputtering their cathodes and redeposit the filament materials on the tube walls and reduce the light output. There are some ballast/bulb combinations that will have less of an issue with this. Often, this is why CFL bulbs fail so early and don't often hit anything close to their stated lifetimes. (The surge that travels through incandescent bulbs at turn on...well, save energy, and turn the bulb off and on 10 times a day (also not allowing for a full halogen cycle for halogen bulbs also reduces their lifetime) also shortens their life. There are a few incandescent that have been in operation over 50 years- turns out that these are bulbs that are never turned off) And a wise building manager might also include lamp replacement costs- including labor, consider heating and cooling costs, or even consider the additional energy fluorescent bulbs use in freezer/coolers- as well as the additional energy the refrigeration units use due to the light source heat. A lighting engineer might even consider the average light output during the lifetime of the light source. The lm/W of a light source is just the beginning, there is the drive electronics, lamp fixturing, environment, life cycle costs, average light over the lifetime, and more you could come up with. An example might be the poor power factor of most CFLs on the market, or the CRI (Color Rendering Index) (which can be used to the benefit- such as you might replace sodium vapor lamps and some Metal HID lamps and find that real people only need half the light for high CRI sources (and even at that, still be able to see things much better with only half the light- often it has to be experienced to grasp the idea). Lots of things to think about, if you are trying to get to the "real world"- and there are plenty more still. And then there is the low bar that Energy Star sets...which is another box of monkeys all together! (at least it is a start with good initial intentions.)

Sorry, I need to add to what I said. In the chart the measures V and I have no specified accuracy and we do not know if they lost accuracy by being written down at a resolution before doing the math with them. 5mA is especially bad in this case. The meter used could have measured that 5mA quite accurately but if it was on a 'Scale' that reads x.xxx amps like is listed then it is much more likely to have a HUGE percentage error making the result far less useful. Meters tend to have 2 errors involved, a percentage gain or scaling error and an offset. It is the offset that can be an extremely large error versus the 5mA. An offset of 1 count is common and that alone is +/- 20% in this case. AND when dealing with Power P=V*I is simple math if both V and I are constants but we know that is unlikely. Thus true power is the integral of V*I over time and will differ from a meter's V times a meter's I even with RMS measurements. The only real answer for Power requires instant measurement and math in either high bandwidth analog or high sample rate digital. This is an area where many things fail to get accurate results. Now if those tests were run via an accurate current source and thus quite constant. Then V should be fairly constant as well with some odd effects from thermal. And thus V*I is once more simple math with fair accuracy once thermally stable.

What I am more concerned about is the comparison between LED light and other sources, not absolute efficiency. Do the new LEDs put out more usable light per watt than new florescent etc. From there it is only engineering them into usable form factors and reducing costs. And for costs do Organic LEDs reach the same area of Efficiency as they show a trend for lower net costs. As to number of digits and tolerances. There is a fallacy involved that error percentage and digits are related when in fact they are not. Reducing digits from math increases errors as the accumulated errors are CENTERED about the calculated numbers. But yes beyond a few extra digits it means little. I find 2 extra digits keeps accumulated math errors reasonable. So calculate with lots of digits and reduce final answers to more realistic ones. 50 +/- 1% is 49.5 to 50.5, 50.2 +/- 1% is 49.698 to 50.702 or 49.7 to 50.7 even though that 0.2 is within the 1% tolerance that amount is still significant as the distribution curve is centered on it.

Doug, You have provided a lot of useful information and that is sorely needed in the LED industry. However as general lighting is viewed by many as the holy grail of LED applications, I wanted to qualify that 32% efficiency number and make sure people are comparing apples to apples. Since you did not provide the CCT of your LEDs, I simply made a guess. You measured them to be 102 lm at 1W. Cree only offers that luminous flux bin in cool white. The CCT is important because it is an indicator of the spectrum of the light. Like you said earlier, different color lights will have different conversion factors into lumens, since the eye has a response curve peaked at 555nm. So to translate your number to warm white LEDs, I first applied the simple ratio of efficiency drop from 114 lm to 93.9 lm. This accounts for lumen difference only. There is a further factor not accounted for, that at the lower lumen and warmer color, the lumen to power conversion is different than that for the higher CCT spectrum. I agree with you 1% is not worth quibbling over. Just the engineer in me wanted to point out the fine print. However 20% is nothing to sneeze at. Btw, Cree holds +- 7% tolerance on flux and power measurements. My experience is the LEDs are all clustered at min flux -7%, and the higher bins are not always available. So for all intents and purposes and with all due respect, warm white LED efficiency is in the low 20s percent. Carol

32% is still misleading, as the LEDs are at 330mA. Manufacturers usually specify performance at 350mA. That's 1 % difference. The other factor you left out is the color of the LEDs. The 114 min flux Cree LEDs are 5,000 to 10,000 CCT. That's bluer than cold white fluorescents. The warm white LEDs from the 2600 to 3700 CCT bin maxes out at 93.9 lm. That's 20% lower, which brings the efficiency to 25%. This still does not take into account the conversion loss from radiant energy to lumens which reflects the human eye sensitivity. 3700K is slightly warmer in color than cold white fluorescents. Incandescent color is at 2700K. At that CCT, the 93.9 lm LEDs are way out at the tail end of the lot distribution. So a "hero" production LED for warm white color would max out in the mid 20s. A more commonly available warm white LED would be in the low 20s conversion efficiency.

32% is still misleading, as the LEDs are at 330mA. Manufacturers usually specify performance at 350mA. That's 1 % difference. The other factor you left out is the color of the LEDs. The 114 min flux Cree LEDs are 5,000 to 10,000 CCT. That's bluer than cold white fluorescents. The warm white LEDs from the 2600 to 3700 CCT bin maxes out at 93.9 lm. That's 20% lower, which brings the efficiency to 25%. This still does not take into account the conversion loss from radiant energy to lumens which reflects the human eye sensitivity. 3700K is slightly warmer in color than cold white fluorescents. Incandescent color is at 2700K. At that CCT, the 93.9 lm LEDs are way out at the tail end of the lot distribution. So a "hero" production LED for warm white color would max out in the mid 20s. A more commonly available warm white LED would be in the low 20s conversion efficiency.

On the max 100% luminous efficiency numbers, one has to consider the spectrum?which greatly alters that number. The luminous flux is the part of the power that is perceived as light by the human eye, and the figure 683 lumens/watt is based upon the sensitivity of the eye at 555 nm, the peak efficiency of the photopic (daylight) vision curve. The luminous efficacy is at that single wavelength or frequency-just for 555nm only. However the visible light spectrum needs to be considered. Wise and thoughtful comment but the basis is utterly flawed. At some single wavelengths the maximum efficiency is only 10 lumen/watt. When calculating the efficiency of a YAG White LED, you must add up the eye's efficiency for the entire visible spectrum. For Ce:YAG White, there are sight variations in the emission spectrum, as well as the color temperature of the White, and the maximum efficiency can vary from 230 l/W all the way up to a maximum of 330 lm/W for each LEDs visible emission spectrum. The curve for the human eye also changes from the Photopic, to mezotopic, to Scotopic- depending on the ambient lighting conditions. This is as in the different conditions you are utilizing rods or cones, or a combination of both. A little more on these can be found here: 4colorvision.com/pdf/18abnormalities.pdf If you visit here, you can see the lumens per watt graphed at different frequencies, and the daylight vs. nightime vision response: hyperphysics.phy-astr.gsu.edu/hbase/vision/bright.html#c2 More detailed lm/W graph for the human eye photopic: www.molalla.net/~leeper/human_~1.jpg Both: www.molalla.net/~leeper/humaney.jpg So, for you, look up the lm/W of the human eye at 400nm, and you should start to grasp the concept of having to sum or integrate the spectral output vs. the human eye and find that the 683 lm/W idea is fundamentally flawed. Sharp thinking though, Kudos!