1. Title Slide
Chasing the Rainbow
Eric Bretschneider

2. Chasing the Rainbow This is not an ad for Skittles®
Chasing rainbows – pursuing things that are unrealistic or unlikely
In order to catch something you need to know where it is going.
So if you want to catch a rainbow you need to know where it is going
Chase /CHās/ to follow rapidly to catch or overtake

3. Rainbows A rainbow is a spectral path in space . . .
. . . a path can be followed
One wonders what sort of treasure we will find if we can catch our rainbow
. . . legends tell us there is a pot of gold at the end of the rainbow.

4. Lighting Parameters Flux S/P Ratio Color Rendering/Quality
Luminous (TM-21-11)
Photon (TM-21-19?)
Radiant (TM-21-19?)
S/P Ratio
Color Rendering/Quality
CRI R9 Rf Rg
Chromaticity
xy or u’v’
Du’v’ (TM-35-19?)
CCT
duv
Photobiology
Photobiological hazard (IEC 62471)
Circadian . . .
These are all dependent variables, they are calculated from the SPD.
They are our pot of gold . . .
These dependent variables are our pot of gold

5. Reliability Standards Predict Parameters
Reliability standard development process
define a parameter
develop/approve a standard to calculate the parameter
2-3 years
collect data
analyze trends/develop reliability metric
3-5 years
2-3 years
You have to collect years of data before you can begin trying to predict behavior.
Once you have the data, it still takes years before you can make a prediction. That’s fine if you don’t mind a slow response/reaction to the demands of an industry.
5-8 years after we develop a standard for a metric we might be able to predict future behavior

6. If we calculate everything from the SPD, why don’t we just predict the SPD?
Lumen maintenance: predict 1 parameter vs time
Chromaticity shift: predict 2 parameters vs time
If white LEDs are (typically) made using an LED and phosphor can’t we just model each emitter?
How bad could it be?
After all, it’s only an LED and a phosphor or two . . .

7. Reality Check for Emitter Based Models
For a reasonably accurate model we should expect at ~20 parameters (or more)
Prior knowledge of LED construction required to prevent “mode hopping”
Short summary: it’s really complicated, even if you know the details
For a single phosphor you need 8+ parameters:
4+ for the emission
4+ for the excitation
Mode hopping refers to one of the modeled peaks suddenly shifting dramatically between two time intervals. You either let the peak jump or at that point in time your model starts diverging from the data.
You can do a better job if you know how the LED is constructed, but a generalized approach should not require prior knowledge.
Just try asking manufacturers to give up the material/component list for all of their LED packages This is very sensitive information!

8. Mode Hopping
Peak parameters (center, FWHM, etc.) can make a step change over time
For a single phosphor you need 8+ parameters:
4+ for the emission
4+ for the excitation
Mode hopping refers to one of the modeled peaks suddenly shifting dramatically between two time intervals. You either let the peak jump or at that point in time your model starts diverging from the data.
You can do a better job if you know how the LED is constructed, but a generalized approach should not require prior knowledge.
Just try asking manufacturers to give up the material/component list for all of their LED packages This is very sensitive information!
The negative peak is the absorption spectrum of the phosphor material.

9. Finding an Alternative Approach
If only we had experience in modeling the spectral output of a light source that changed in a predictable manner over time . . .
Wouldn’t it be great if we had a really well known light source
That changed in a predictable way
And that had been modeled?
Oh, wait we do

10. D-Series Illuminants
D-Series (daylight) illuminants are calculated using a linear combination of 3 eigenspectra
S(l) = S0(l) + M1S1(l) + M2S2(l)
M1 and M2 are functions of CCT. If you know the CCT, you can calculate the their values and then calculate the SPD.
Could a similar approach work for LEDs?
Could a similar approach work for LEDs?
This approach was actually developed when “computer” was a job description, not an electronic device – its surprisingly old

11. Exponential Decay Analog
S(l) = S0(l) x e[-S1(l)t]
We have been modeling LED lumen maintenance as an exponential decay for almost a decade now.
Perhaps we could use an exponential decay analog . . .

12. Results
It’s almost impossible to tell the difference between the actual data and the prediction.
TM-21 and TM-35 are highly specialized reliability models, they predict very specific metrics and no additional information.
Even this rudimentary eigenspectral analysis gives very comparable results on lumen maintenance and chromaticity shift – at the same time
These results are at least as good as predictions from existing standards and those in development

13. What else can we predict?
But look at what else we can predict!
All of these parameters are predicted with a single model/method. Every single parameter listed here and even more can be predicted as a function of time
These are examples of the enormous number of parameters we can predict with the spectral power distribution

14. Eigenspectral Analysis
The first eigenspectrum S0(l) does not contain significant information
It’s the initial spectrum
This is boring. Knowing the spectrum at time = 0 doesn’t tell us anything we don’t already know
This doesn’t tell us anything we don’t already know

15. Eigenspectral Analysis
Peak shift in time
The second eigenspectrum S1(l) is far more interesting
Although this is the rate of change, it gives information on construction/materials
The second eigenspectrum though is far more interesting. This shows a peak that shifts location over time
This shows a loss of emission and is an exceptional match to a commercial phosphor
Without knowing anything about the package, we can identify some of the materials and also have a very good idea of what is changing inside the package.
We even have an idea of what we should focus on to improve performance stability over time
This is a decaying phosphor it is an ~96% match to the emission spectrum of a commercial red phosphor

16. Perspective
The accuracy of even rudimentary eigenspectral models is on par with that of existing (or in process) standards that predict a single parameter or a pair of related parameters
A successful eigenspectral model will likely be the only reliability model needed
In addition to predicting performance, the eigenspectra themselves contain valuable information about the aging mechanisms of LED products/devices
This method is potentially a very powerful tool.

17. Takeaways
The traditional reactive approach to reliability standards will never be able to keep up with a changing industry
New and proactive approaches to developing reliability standards need be considered in rapidly evolving markets like the lighting industry
A willingness to embrace ambitious goals can pay dividends far in excess of what we expected
The industry is changing more rapidly than ever. Reacting to the industry means we will on get further behind as time goes on.
If we want to adapt, then we have to take a more proactive stance. It is our only hope to keep up.
Sometimes being ambitious pays off far better than we ever expected.