Investigation of the possibility to alleviate the low-contrast problem from the top of the sapphire substrate:

Light emerging from 100um Sapphire substrate a) directly above pixel area: 23.1%:

b) Below: one pixel adjacent from the emitting LED: 6.16%

c) Over a 10x10 pixel area:

d) LED active region thickness reduced to 3um (Weve changed to using log irradiance plot later since we are more interested in the background)

Using spherical convex sapphire microlens with varying radiia) (r=35um) i) Pixel directly above LED, ii) Pixel adjacent i)

b) (r=100um, rays ~ normal to surface) I) Directly above LED,ii) Pixel adjacent i)

c) (r=50um, ~ collimation) i) directly above LED, ii) pixel adjacent i)

iii) (Below) integrated over 10x10 pixels, iv) log scale (closer to eyes intensity dependent response)

d) (r=43um) i) directly above LED, ii) pixel adjacent i)

Effects of coating on top of sapphire substratea) Below: Perfect transmitter, flat sapphire surface. Each 70x70um pixel transmitted 23.6%, 7.3% (adjacent pixel) respectively. Hence the best AR coating stack alone would show no significant improvement to the problems above. Ie. TIR is the main problem hence we should focus on surface texturing.

b) Below: Just moved the detector a few nm into the sapphire. High irradiance shows that TIR is the main problem limiting extraction efficiency as well as cause of interpixel crosstalk. (Flux/Emitted Flux: 0.91 here)

Above: L to R: light directly emitted from LED active region; light emitted after being reflected by bottom mirror; both sets of rays combined

Surface Texturing: Other shapesa) i) Simple pointed cone with cone angle 30 deg, pi/4 fill factor

ii), iii) (Below) slightly improved EE but log plot shows larger spillover

Major noises due to internal reflection of cones. How to optimize this? Lets see if we can shrink the cones further.

b) Results from micro-cones r=3.5um, 10 deg. i) over 10x10 pixel area; ii) over single pixel above LED.

c) R=1.75um, cone angle 10 deg. It is probably impossible to simulate smaller structures with TracePro since it uses ray optics (and edge diffraction effect). The transmittance becomes less efficient as the no. of reflections increase.

d) Below: Cone angle = 64 deg, r=35um (an angle such that most rays incident from neighbour pixel will be reflected away. This is the lowest neighbour background weve got so far (300ish at 0.2). However, The background doesnt seem to drop with distance hence would produce a low contrast image.

Tracepro cant simulate waveguide effects in microstructures ~< lambda hence we only simulated structures at least several microns large. (The software is able to reproduce diffraction effects at material edges, but the calc is limited to edges/interfaces.)Several other possible shapes from Tracepros reptile function. Our computer is not powerful enough to simulate shapes other than the pre-defined reptile shapes below:

Parameter Number (Param@)

Geometry Type1234567891011121314

Conex centery centerheight / depthend radiuscone anglechamfer heightchamfer angle

Spherex centery centerradiusheight / depth

Hip Roofx centery centerheight / depthy widthy anglex widthx angleorient. angle

Ellipsoidx centery centercenter ht/dpthx radiusy radiusz radiusx rotatey rotatez rotate

Logx centery centercenter ht/dpthlengthend1 radiusend2 radiusx rotatey rotatez rotateradius ratio

Enhanced Prismx centery centerheight/depthx widthy widthx0 anglex1 angley0 angley1 angleorientation angley0 peak radiusy1 peak radiusy0 trough radiusy1 trough radius

Flattened Cone x centery centerheight/depthend radius cone angle peak radiustrough radius

Pointed Cone x centery centerheight/depthcone angle peak radiustrough radius

(Chamfer height = 0 in our simulations above)

Sub-dividing each LED pixel into sub-arrays of even smaller pixels. Can this limit the emission angle from each pixel?

Weve sub-divided what was originally one LED pixel into 5x5 9um sub-pixels (w/ 2um gaps in between)

It looks like there are no significant difference between the plots above and Fig. 1. This could be that the original emission angle with 50um single pixel was already small enough such that reducing the pixel to 10um would have no effect. (pixels are ~1um thick so we would not see significant constrains in emission angle unless sub-pixel sizes are reduced to ~1um but this obviously could not be modelled by TracePros ray optics approach. In addition, slicing one pixel into 50x50 smaller pixels would also introduce extra gaps in between, reducing the utilization ratio). + Diffractive effects (see articles on single mode fibres)

Instead, can we have flat mirrors on each of the 5x5 sub pixels for light collimation? Will this be cheaper than building a single parabolic mirror on a pixel with a curved bottom surface?

Rotationally symmetric parabolic mirrors had been built by a company (http://www.infiniled.com/technology/microled). Well see whether rotational symmetric parabolic mirror works better or 2-direction symmetrical parabolic works better. We believe the later would have a higher ultilization ratio and is easier to be fabricated (?).

Effects of Substrate thickness on EE (assume interference effects are negligible at 100um+)TBD

Omni-directional reflectorsSimulationMatlab thin film simulation

TraceProa) Spherical ReflectorsWe simulated a spherical perfect metallic reflector with r=37.36um. The well depth protruded under the flat sapphire surface was 9.5966um, such that the well makes an angle of 42 deg with the flat surface at the edge. (H. Kim) Fig. from H. Kim, showing EE

b) X-y parabolic reflectorc) Chamfer

Can use the Interactive Optimizer option to optimize the shape of the reflector/ cups/ any geometries. Eg: