on the standardization of thermal characterization of leds andrás poppe mentor graphics micred...

On the Standardization of Thermal Characterization of LEDs

András PoppeMentor Graphics MicReD Division, Budapest, Hungary

andBudapest University of Technology & Economics,

Department of Electron Devices, Hungary

Clemens LasancePhilips Research, Eindhoven, The Netherlands

MicReD Division

2MicReD Division A. Poppe and C. Lasance: On the Standardization of Thermal Characterization of LEDs 18/03/2009

Red

0,00

0,50

1,00

1,50

2,00

2,50

3,00

3,50

4,00

550 600 650 700

wavelength, nm

rel.

int.

3 °C

21 °C

34 °C

50 °C

Why to deal with thermal issues?► Reliability is connected to thermal issues

life time (failure mechanisms are thermally assisted) mechanical stress

► Optical properties strongly depend on temperature spectra emitted flux / efficiency / efficacy

Courtesy of J. Schanda


Main drive for thermal characterization

► Fair comparison with competitors' data► In an ideal world: provide customer support

from a manufacturer point of view

from a customer point of viewAre temperatures within specs?

Starting point:

Tj = Rth j-ref P + Tref► Tref from (often unspecified) measurements (own responsibility)► P from estimated power dissipation (own responsibility)► Rthj-ref from Component Data Sheets (other’s responsibility)

If Tj calculated > Tspecification Redesign!

► In case of lighting: specified lumens at operating temperature


The main reason for standardization

Exactly because of the many uncertainties in the simple equation, everything can be quoted by vendors.

Standardization = define strict rules. It is a consequence of living in a world that is based on money, not on honesty.

Differences and similarities: LED vs IC► Similarities

process of how to establish thermal standards

► Differences The good news: LEDs are thermally much less complex The bad news: Tj plays a major role not only in thermal, but also in

lighting design

we need multi-domain models


A few words about thermal resistance

► Rearrange the equation: Rth j-ref = (Tj – Tref)/P

► Original definition in the JEDEC JESD51-1 document

► Accurate; the questions are: what is the dissipated power? Issue for LEDs… what is the TX reference temperature Use cold-plate!


A few words about thermal resistance

► Alternate formulation: instead of spatial difference, let us build a temporal difference:

TJ1 = Rth J-X PH1 + TX (1a)

TJ2 = Rth J-X PH2 + TX (1b)

If t1=0 and t2= Rth J-X =TJ/ PH

TJ2 – TJ1 = Rth J-X (PH2 – PH1) (1b-1a)

Let TJ1=TJ(t1) and TJ2=TJ(t2):

Rth J-X =[TJ(t2) – TJ(t1) ] / (PH2 – PH1)

Rth J-X =TJ(t) / PH

If PH2 = 0, then TJ2= TX – see (1b)


Extend the Rth concept to transient

Rth J-X =TJ(t) / PH

► t1 = 0; t2 = changes from 0+ to ► Keep reference temperature constant (though, falls out)

► Record TJ(t) very densely

► With known PH dissipation, series of Rth-s corresponding to

each t instance is obtained:

This is preferred for LEDs (e.g. compliance to CIE 127-2007)

RthJX(t) = TJ(t) / PH

is called Zth curve

heating curve

cooling curve


How to standardize Rth j-ref?► Questions one may ask:

How is Tj defined? • Can it be measured in practice? IR / electrical test method• If not, how to derive it from another measurement?

What is Tref? • Can it be unambiguously defined and easily measured in practice?

cold-plate setups

How is P defined? • Corrected for non-thermal contributions?

Popt as per CIE-127:2007

What is the physical meaning of Rth?

Is Rth really temperature-independent?

What is the variance in the published data per manufacturer?


"Lumen maintenance" / life time

~5 years ago

By now

Vendors usually mean degradation of ΦV/ΦV0 but not stated explicitly

► Nowadays life-time of LEDs is defined according to the degradation of their light output in terms of ΦV total luminous flux ("lumens")


"Lumen maintenance" / life time► Nowadays life-time of LEDs is defined according to the

degradation of their light output in terms of ΦV total luminous flux ("lumens")


Life time► L70: degradation of the relative light output by 30%

It is already considered as a failure

► B50: when 50% of the population already fails

► Life-time metrics Lxx and Bxx are often used (to be standardized – by whom?)


Some degradation mechanisms► If LEDs are driven within their specified ratings (safe

operating area), failures should not happen (in principle)

► However, some users wish to know what happens in case of uprating

► Possible degradation and failure modes in this case: Yellowing of phosphor containing encapsulations Lens degradation Delamination of adhesive layers Solder joint failures

What temperature really counts?


Some questions / issues► Do quoted data reflect real-life conditions?

E.g. do quoted temperature values correspond to values that happen during operation?

Is junction temperature really the junction temperature? If reference point temperature is quoted, does it allow fair

comparison?

► Are data measured really in a correct way?

► Are the IF and T?? reference values comparable at different vendors?

► There is no idiot-proof way of translating temperature related data to e.g. "useful lumens" under all operating conditions

► Thermal data are often provided by means of series thermal resistance models which may fail in some cases


Some questions / issues► Usually luminous flux data are provided, which can not be

used for thermal design purposes. Total radiometric flux (emitted optical power) would also be required

► Efficacy data (also called luminous efficiency or "lumen per watts") are provided, but efficiency data (emitted optical power per supplied electrical power) are not published Term efficiency is used ambiguously


Example► Let us assume two WPE-s (wall plug efficiency:

Popt/Pel) T = 50oC, Pel = 10W

"Rthel" = T / Pel = 50/10 = 5 K/W

WPE = 25% Rthreal = T / (Pel – Popt) = T / [Pel (1-WPE)] =

= 50/(100.75) = 6.67 K/W

WPE = 50% Rthreal = T / (Pel – Popt) = T / [Pel (1-WPE)] =

= 50/(100.5) = 10 K/W► By neglecting Popt vendors report much nicer data than reality


How to consider light output?

Popt(T,IF) WPE(T,IF) V(T,IF)

steady-state electrical powering for

photometric/radiometric measurements in thermal steady-state

JEDEC JSD51-1 static test method compliant thermal measurement

DUT LED on cold-plate detector

CIE 127-2007 compliant photometric & radiometric measurement

switching-off

record thermal data, calculate Tj and Rth-real

ANY THERMAL TESTER WHICH MEETS CERTAIN

REQUIREMENTS

ANY PHOTOMETRIC TEST SETUP WITH TEMP.CONTROL


WPE – is not a single number► Depends on IF and T

2008

~2004

Set of such plots must be published for typical operating conditions.

Neither is efficacy a single number; people tend to forget

this.


Series thermal resistance models?!► Usual suggestion of LED

vendors on their data sheets

► In some cases does not work for LED packages where the heat-flow path splits into multiple branches:

Multiple heat-flow paths / need to be modeled, based on simulation?

Assumes a single heat-flow path!


Series thermal resistance models?!Luxeon Rebel extended hs

Theta die-thermal pad

10

11

12

13

14

15

0.1 1 10 100 1000

k board (W/mK)

Th

eta

j-t

p (

K/W

)

20

200

2000

20000

Luxeon Rebel extended hs Theta thermal pad-board

0

50

100

150

200

0.1 1 10 100 1000

k board (W/mK)

Th

eta

tp

-b (

K/W

)

20

200

2000

20000

► Flotherm simulations (conduction mode)

► Element values of the series resistance model show dependence on environmental conditions

► Nonsense in some cases

The underlying assumption of the single path must be checked


Multiple heat-sources► What would be the right Rth?

► In general, any # of LEDs (LED chips) may be present in a package or in a level1 assembly

Force Sense


Problem of multiple LED chips► Individual Rth vs 'ensemble' Rth?

► No way to measure Rth-ji-s unless LED array is designed for thermal testability

Force SenseRth-ij

Rth-ensamble


If we have access to individual LEDs:► If each LED can be measured individually: thermal

resistance or impedance matrices (see e.g. T. Treurniet or M.W. Shin – SEMITHERM'06)

► The JEDEC JC15.1 committee goes this way for IC-s


Short pulse measurements► During in-line testing photometric/colorimetric

properties are measured with a short pulse Tj = Tref = constant is assumed

In 10 ms significant junction temperature change may take place

During 10 ms Tj changes almost by 5 oC

1e-6 1e-4 0.01 1 100 100000

5

10

15

20

25

Time [s]

No

rma

lize

d t

em

pe

ratu

re r

ise

[°C

]

T3Ster Master: Zth

dimco_kisf2 - Ch. 0

Question is if this causes big problems or not…

A multi-domain LED model could correlate this with long term data


PR

PlossPD-Popt

Pheat=IF, Tref ΦV(IF, Tj), Tj

Create a 'standardised' LED model?

In research phase at Budapest University of Technology

Input: IF, Tref output: ΦV dream of SSL designers

Lot of work and round-robin tests would be needed


How about existing standards? ► Is JEDEC JESD51-1 "good" for LEDs?

yes, • but the "power dissipated in the device" has to be carefully specified• reference temperature needs to be well established and kept constant

Possible new measurement guidelines:• measure Popt according to CIE 127-2007• measure on a cold-plate, Tref = Tcold-plate

• calculate junction temperature as follows: TJ = Tref + RthPH

► Is CIE 127-2007 "good" enough? yes,

• thermal aspects of total flux measurement are precisely specified, • but scattered around in the document• Arrangement b of Figure 9 allows TEC-based

control of LED package temperature (cold-plate), attach DUT LED to such a cold-plate

Possible measurement guidelines to standardize:• 4 wire connection to DUT LED to comply with

JEDEC JESD51-1 requirements


Standardization activities: JEDEC► JEDEC JC15.1 committee: LED standardization

task group

Initiated by Philips, OSRAM and Mentor Graphics MicReD

White paper published, next versions of the white paper: • Eurosime'09• Electronics Cooling Magazine

Panel discussion at the 14th THERMINIC Workshop in Rome, Italy

• Philips Lighting, OSRAM OS, Lumileds, MicReD, KETI

Next meeting: 20 March 2009 in San Jose, companies interested in LED business invited


Standardization activities: CIE► Two new TC-s established, dealing with thermal

aspects TC-63: Optical Measurement of High-Power LEDs

• To develop a CIE recommendation on methods for the operation of high-power LEDs in DC and in pulse mode, at specified junction temperatures, for optical measurements.

TC-64: High Speed Testing Methods for LEDs • To prepare a technical report on high speed testing methods for

electrical, thermal and optical quantities during the production of LEDs and the conversion of the values to DC operational conditions including the related time dependent functions.

► Light and Lighting Conference with Special Emphasis on LEDs and Solid State Lighting, 27-29 May Budapest, Hungary TC meetings


Some conclusions► Data sheets do not provide sufficient information

e.g. for thermal design, efficiency data are also required, efficacy is of no help

► Agreement on measuring 'real' Rth of single chip LEDs is required one should consider the real heating power not juts the supplied

electrical power – would allow real thermal design this would also provide a fair basis of comparison for LED vendors

► Agreement on the way of obtaining LEDs' junction temperature is needed reported metrics (e.g. luminous flux) possibly should refer to this

► Standard test environments are also part of a like-with-like comparison we believe measuring on cold-plates yields meaningful and

repeatable results


Some conclusions► For multi-chip LED systems Rth/Zth 'matrices' can be

introduced (slef/transfer impedances) as this is becoming accepted for conventional silicon devices (ICs) too This assumes access to the individual LED chips (which is assured in

certain cases like RGB modules) Design for thermal testability of LED lighting solutions could support

this If LED chips are not accessible individually, a commonly accepted

concept of 'ensemble' thermal resistance would be needed

► Easy-to-handle yet technically sound, commonly accepted multi-domain LED models would be greatly appreciated by lighting designers Such models could partly replace conventional data sheets: vendors

provide model parameters instead of data sheet charts and values

on the standardization of thermal characterization of leds andrás poppe mentor graphics micred...

Documents

c21 c34 c50 cwavelength

department of electron

netherlandsmicred divisiona