1. Color Point Stability of LEDs
Models for
Color Point Stability of LEDs
An Introduction to Differential Chromaticity Analysis
Presented at
Strategies in Light 2017

2. Models for Color Point Stability of LEDs
Strategies in Light 2017 Invited Presentation
Models for Color Point Stability of LEDs
An Introduction to Differential Chromaticity Analysis
Eric Bretschneider
EB Designs & Technology

3. DOE Solid-State Lighting Program Manager
Color Point Stability of LEDs
A Universal Problem of LEDs and Solid-State Lighting
“A light source that shifts in color too much over time is just as useless as one whose lumen output drops below an acceptable threshold*”
James Brodrick
DOE Solid-State Lighting Program Manager *

4. Importance of Projecting Color Shift
Lumen maintenance estimation via LM80/TM21 is the currently accepted metric for projecting LED lifetime
Excessive color shift is a failure metric, but . . .
A lack of predictive standards or methods means
Determining color shift requires excessive amounts of time (test to failure)
Du’v’<0.007 in 6,000 hours is not adequate ! ! !
Lumen maintenance estimation via LM80/TM21 is the currently accepted metric for projecting LED lifetime
Excessive color shift is a failure metric, but . . .
A lack of predictive standards or methods means
Determining color shift requires excessive amounts of time (test to failure)
Lumen maintenance estimation via LM80/TM21 is the currently accepted metric for projecting LED lifetime
Excessive color shift is a failure metric, but . . .
A lack of predictive standards or methods means
Lumen maintenance estimation via LM80/TM21 is the currently accepted metric for projecting LED lifetime
Excessive color shift is a failure metric, but . . .

5. Importance of Projecting Color Shift
L70(6khr) = 21,211 hrs
Du’v’(6khr) =
LXX(10khr) = 83.3%
Du’v’(10khr) =
LXX(12khr) = 78.8%
Du’v’(12khr) =
LXX(8khr) = 87.2%
Du’v’(8khr) =
Color shift failure before lumen maintenance failure

6. Importance of Projecting Color Shift
L70(6khr) = 842,725 hrs
Du’v’(6khr) =
LXX(14khr) = 92.51%
Du’v’(14khr) =
Color shift failure before L90 ! ! !

7. Today’s Lessons Why does Du’v’ remain constant then start to change?
Is it possible to predict Du’v’?
Do you need to know details of package construction and materials to attempt a prediction?
Reminder: the CIE 1976 u’v’ chromaticity diagram is the most perceptually uniform chromaticity coordinates - (u’, v’) chromaticity coordinates are preferred for quantifying differences in color

8. LED Color Point Stability
Which LED has the best color point stability?

9. Terminology
Relative chromaticity (du’, dv’): Chromaticity relative to chromaticity at time = 0
Incubation: period of time during which chromaticity shift (Du’v’) is essentially constant
Recovery: a temporary decrease in chromaticity shift (Du’v’)
Emergence: period of time during which chromaticity shift (Du’v’) changes monotonically with time
CSn: Time for an LED to exhibit a chromaticity shift (Du’v’) of x n
CS4  Time to Du’v’ = 0.004
CS7  Time to Du’v’ = 0.007

10. Du’v’ – a misunderstood metric
Du’v’ is a scalar metric
Chromaticity change is a vector (2 dimensional metric)
Constant Du’v’ does not imply static chromaticity
The incubation period is an artifact of the industry standard color shift metric

11. Understanding Incubation
During incubation chromaticity is changing
After emergence modeling Du’v’ will give adequate results
Predicting emergence requires new models

12. Differential Chromaticity Analysis
Differential Chromaticity (du’*, dv’*) is the rate of change in relative chromaticity over a given time interval
dut’* = (du’t+Dt – du’t)/Dt
dvt’* = (dv’t+Dt – dv’t)/Dt
Differential Chromaticity (du’*, dv’*) is the rate of change in relative chromaticity over a given time interval
dut’* = (du’t+Dt – du’t)/Dt
dvt’* = (dv’t+Dt – dv’t)/Dt
First Principle of Differential Chromaticity: Differential chromaticity is a linear function of time
Second Principle of Differential Chromaticity: Keep all data for t ≥ 2,000 hours
Differential Chromaticity (du’*, dv’*) is the rate of change in relative chromaticity over a given time interval
dut’* = (du’t+Dt – du’t)/Dt
dvt’* = (dv’t+Dt – dv’t)/Dt
First Principle of Differential Chromaticity: Differential chromaticity is a linear function of time

13. Analysis Method Calculate relative chromaticities
du’t = u’t – u’0
dv’t = v’t – v’0
Calculate differential chromaticities
dut’* = (du’t+Dt – du’t)/Dt
dvt’* = (dv’t+Dt – dv’t)/Dt
Linear fit differential chromaticity data for t ≥ 2,000 hours
Project relative chromaticities
du’t+Dt = du’t + dut’*Dt = du’t + (aut + bu)Dt
dv’t+Dt = dv’t + dvt’*Dt = dv’t + (avt + bv)Dt

14. Example 1 6,000 hour data shows barest hint of emergence
Incubation
6,000 hour data shows barest hint of emergence
DCA input data shows almost no variance
Almost immediate emergence predicted
Reasonable agreement to limit of data set
6,000 hour data shows barest hint of emergence
DCA input data shows almost no variance
Almost immediate emergence predicted
6,000 hour data shows barest hint of emergence
6,000 hour data shows barest hint of emergence
DCA input data shows almost no variance

15. Example 1 DCA input = 2 – 14 khrs
Close agreement over long time intervals
Rolling time window not required for accuracy

16. Example 1 Behavior Rapid initial shift towards yellow/orange
Long term shift towards blue
Blue shift begins ~3khrs
Blue shift firmly established ~5khrs

17. Example 1
Model developed projecting forward from 2,000 hour data point
Projecting from final data point increases accuracy
Exceptional accuracy when “close” to CS7

18. Example 2 6,000 hour data shows apparent stability
DCA predicts (surprising) recovery and then emergence
14khr data mirrors the prediction
8khr DCA prediction closely matches full data
6,000 hour data shows apparent stability
DCA predicts (surprising) recovery and then emergence
14khr data mirrors the prediction
6,000 hour data shows apparent stability
DCA predicts (surprising) recovery and then emergence
6,000 hour data shows apparent stability

19. Example 2 Longer term predictions support emergence at 15 - 16khr
Slight changes in CS7 using data out to khr

20. Example 2 Behavior Initial shift towards blue
Long term shift towards yellow
Yellow shift does not start until ~5khr
Yellow trend firmly established 8 - 9khr

21. Example 2 Reminder: shift change at 5,000 hours
Using 1,000 hour data sometimes leads to optimistic results
6khr  CS7 error ~15%
7khr  CS7 error ~9%
8khr  CS7 error ~3%

22. Example 3 6,000 hour data shows apparent stability
DCA predicts recovery followed by emergence
Full data shows recovery
DCA matches full data
Extrapolating forward still predicts emergence
6,000 hour data shows apparent stability
DCA predicts recovery followed by emergence
Full data shows recovery
DCA matches full data
6,000 hour data shows apparent stability
DCA predicts recovery followed by emergence
6,000 hour data shows apparent stability
DCA predicts recovery followed by emergence
Full data shows recovery
6,000 hour data shows apparent stability

23. Example 3 Behavior Initial shift towards yellow/green
Long term shift towards blue begins >7khr
DCA predicted reverse shift before it occurred

24. Example 3 Reminder: final blue shift begins after 7khr data
Using 1,000 hour data leads to optimistic results
Reasonable convergence rate

25. Example 4 6,000 hour data looks exceptional Du’v’ ≤ 0.0006
2 - 6khr change in Du’v’ =
DCA predicts imminent emergence
6khr DCA pessimistic
7khr DCA accurate
6,000 hour data looks exceptional
Du’v’ ≤
2-6khr change in Du’v’ =
DCA predicts imminent emergence
6khr DCA pessimistic
6,000 hour data looks exceptional
Du’v’ ≤
2-6khr change in Du’v’ =
6,000 hour data looks exceptional
Du’v’ ≤
2-6khr change in Du’v’ =
DCA predicts imminent emergence
6,000 hour data looks exceptional
Du’v’ ≤

26. Example 4 Behavior Initial shift towards blue
Second shift towards yellow
Third shift towards blue
Final blue shift begins >5khr

27. Example 4
Reminder: double (180°) change in direction of chromaticity shift
Rapid convergence achieved despite complex chromaticity behavior

28. Behavior of Examples 1-4
Example data sets exhibited variety of sometimes complex behaviors
Multiple direction changes
Late (>7khr direction change)
Despite radically different chromaticity shifts DCA is able to capture and model behavior with no adjustable parameters

29. LED Performance can now be Predicted (from LM-80 data)
Lifetime Color Shift of LEDs
Examples 1- 4
Predictions of these 4 Color Shifting LEDs
(using DCA algorithms with 8Khr of LM-80 data)

30. Comments on the DCA model
Powerful predictive capabilities
Extremely efficient noise filter – even poor correlations for differential chromaticity parameters yield reasonably accurate estimates
Able to accurately model large time spans of complicated data
Inherently unstable model – it will always give finite values for CS7; estimates converge as more data is included in the analysis
Package format and material agnostic

31. Thank you
Questions/Comments Eric Bretschneider