Reliability and Performance of Pseudomorphic Ultraviolet Light Emitting Diodes on Bulk

Aluminum Nitride Substrates

James R. Grandusky

1

, Yongjie Cui

1

, Mark C. Mendrick

1

, Shawn Gibb

1

Crystal IS, 70 Cohoes Avenue, Green Island, NY 12183, U.S.A.

ABSTRACT

devices at much higher current densities and input powers than devices on sapphire substrates.

This is due to the better thermal properties and lower dislocation density of devices on AlN

substrates. Devices with active area of 0.001 cm

2

emitting at ~265 nm have been measured for

their reliability and change in power output over time at input currents of 20 mA (20 A/cm

2

), 100

mA (100A/cm

2

) and 150 mA (150 A/cm

2

). When operating at currents of 20 mA over 3500

hours of consecutive operation has been demonstrated with typical decay of ~27% over the 3500

hours. Extrapolating the decay with a linear fit gives a L50 (time to 50% of initial power) of

disinfection to water, air and surfaces without the use of chemicals. This requires a light source

that is emitting in the ultraviolet-C (UVC) range (<300 nm) for efficient disinfection due to a

peak in DNA absorption at ~ 265 nm

suffer the disadvantages of short lifetimes, slow start up, and the use of toxic mercury which can

be a hazard if the bulb is broken and leads to problems with disposal. Light emitting diodes

(LEDs) have the capability of high efficiencies, fast start up which can be synchronized with

water flow, long lifetimes, variable wavelengths, and no toxic materials.

Al

x

Ga

1-x

N with x>0.6 can be obtained resulting in device layers with low dislocation densities,

low resistivities, and atomically smooth surfaces

3

.

Mater. Res. Soc. Symp. Proc. Vol. 1195 © 2010 Materials Research Society 1195-B03-04

Figure 1. Performance of pseudomorphic UVC LEDs on bulk AlN substrates when driven to

high currents for 4 different devices. The Y axis (output power) is normalized to 20 mA and

an n-type Al

type GaN contact layer. Devices were processed using standard LED processing with a mesa

diameter of 350 µm. The devices are flip chip mounted to an AlN submount and packaged in

TO-8 and surface-mounted design (SMD) packages.

For testing of the diodes, the packaged devices were mounted on a pin fin anodized Al heat

sink (40 mm x 40 mm) with forced convection cooling which had an experimentally validated

theoretical thermal resistance of 2.5

o

C/W. With an applied forward bias, the output power was

measured in a calibrated integrating sphere. For lifetime measurements a photodiode was used

equilibrium and the output power was stable. The output power was then normalized to the

power at 20 mA and plotted versus forward current as shown in figure 1. In each device, the

increase in power at currents up to 150 mA is linear with no roll off or thermal effects. The

forward voltage of the devices is quite high and typically 20 V at 150 mA forward current

resulting in an input power of 3 W. The forward voltage is expected to become lower as the n

and p contact metallization is improved.

currents greater than 150 mA without letting the device reach thermal equilibrium (quasi-cw).

figure 2 shows the results with an output power of 1.3 mW at 400 mA of current with a peak

wavelength of 258 nm.

pseudomorphic LEDs is the reliability. It is expected that the low dislocation density in the

substrates will lead to increased reliability and increased lifetimes over devices fabricated on

sapphire substrates. Lifetime measurements were carried out at input currents of 20 mA, 100

mA, and 150 mA. For measurements at 20 mA no heat sinking was used and the devices were

suspended in air. However for higher currents, such as 100 and 150 mA, power inputs of up to 3

W necessitated the use of a heat sink.

The first mechanism was temperature related and due to the device reaching thermal equilibrium.

This was reversible, i.e. if the device was turned off and allowed to cool the power would

fact that there is no external heat sinking of the devices. For 100 mA very little heating is

considerable rise in junction temperature leading to severe power decay in the first several

minutes of operation. The second decay mechanism is burn in, shown in figure 3b. It is unclear

exactly what the mechanism of this decay is, but it could be related to impurity diffusion in the

material

4

or further alloying of the contacts. When run in pulsed mode with low duty cycle the

decay is much slower indicating a stronger dependence on junction temperature as opposed to

current density.

currents as expected, but it is unclear if this is thermally related or current density related. This

decay is shown in figure 4. As the decay is relatively slow, it becomes important to be able to

predict the lifetime of devices based on as little experimental data as possible. In order to do

(2) Logarithmic

(3) One channel exponential decay

(4) Two channel exponential decay

with R

2

values of 0.98 while the log fit shows a very good fit for the 100 mA data with a R

2

Figure 5. Fitted L50 as a function of time used in fit for 4 different models at a) 20 mA and b)

100 mA.

increase for the fits using the square root and exponential decay functions. However for the log

fit the value of L50 predicted after a short time is similar to the predicted value after a long time.

It appears that fitting the data to a log fit after only a few days worth of data gives the best

method of estimating the L50 with limited data; however we are continuing to collect data and

analyze the fitting functions to determine the best way to predict the L50.