Anthony Cibié | CEA-Leti: How do you prevent optical crosstalk when coupling a 5 µm LED array to a 50 µm pitch multi-core fiber?
17:48 - 19:14
Other snippets from this talk
Summary of the clip:
How do you prevent optical crosstalk when coupling a 5 µm LED array to a 50 µm pitch multi-core fiber?
The question targeted optical crosstalk when fiber bundles sit close to the emitters, since most microLEDs emit in all directions rather than into a guided mode. The current geometry already suppresses this: the array pitch is 50 µm while each LED is 5 µm. That dimensional mismatch limits optical crosstalk between the LED and the fiber.
Beyond that geometric margin, work with optical back end aims to make the LED more directive into the fiber, so emission is better matched to the guided path. That directionality is an active development target to avoid crosstalk problems as the array evolves. The approach combines the LED with backend optics rather than relying solely on the pitch ratio.
On shrinking pitch, the answer clarified that the structures are multi-core fiber: one fiber containing multiple cores, not a bundle. Fibers with much more cores exist, and the limitation is probably not in the fiber itself, the speaker said. The harder constraint is whether the electronics can be fitted into such a small pitch.
In this short video, you can learn:
* Optical crosstalk between a 5 µm microLED and a 50 µm pitch is limited by the large pitch-to-emitter ratio.
* Optical back-end integration is being developed to make LED emission more directive into the fiber.
* The barrier to shrinking pitch is fitting the electronics, not the multi-core fiber itself.
📋 **Clip Abstract** A question on optical crosstalk in fiber-coupled microLED arrays is answered by the current 50 µm pitch versus 5 µm emitter geometry, which limits coupling between neighboring channels. Optical back-end work aims to make the LED more directive into the fiber, while pitch reduction is limited by electronics integration rather than the multi-core fiber.
About the speaker:
* Speaker: Anthony Cibié
* Company: CEA-Leti
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#MicroLEDCrosstalk, #MultiCoreFiber, #OpticalBackEnd, #50MicronPitch, #MicroLEDDisplays, #OpticalInterconnect
This is a highlight of the presentation:
Towards parallel communication using GaN microLEDs for chip to chip communication
More Highlights from the same talk.
02:22 - 03:30
How do you reach 10.5 Tbps per millimeter with LEDs that only switch at 2 Gbps?
How do you reach 10.5 Tbps per millimeter with LEDs that only switch at 2 Gbps?
A microLED is not as fast as a laser in silicon photonics: one microLED can go at two gigabits per second. The idea is to parallelize a lot of them together in matrices of microLEDs. This uses a multi-core fiber, which is a fiber with several cores in it, and matrices of photodetectors. Device modeling and integration are also needed to describe how it works.
One channel will be one microLED, one core fiber, and one photodiode. So it brings new constraints. For the data rate, the target is 10.5 terabits per second per millimeter, and the millimeter refers to the size length on the GPUs. We also want to be energy efficient, with under one picojoule per bit.
We need to have a good signal-to-noise ratio to have good communication, and all of that with an acceptable cost. The bit form data rate target is 10.5 terabits per second per millimeter, which sets the scale for chip-to-chip communication. The approach combines matrices of microLEDs, a multi-core fiber with several cores, and matrices of photodetectors, each channel being one microLED, one core fiber, and one photodiode.
In this short video, you can learn:
* A single microLED reaches about two gigabits per second, far slower than a laser in silicon photonics.
* Parallelization uses matrices of microLEDs, a multi-core fiber with several cores, and matrices of photodetectors, with one channel per microLED-core-photodiode triplet.
* The targeted 10.5 terabits per second per millimeter refers to the size length on the GPUs, alongside under one picojoule per bit, good signal-to-noise ratio, and acceptable cost.
📋 **Clip Abstract** This clip explains how chip-to-chip communication can parallelize microLEDs that individually run at only two gigabits per second. It details matrices of microLEDs, multi-core fiber, photodetector matrices, and the one-microLED-one-core-one-photodiode channel, targeting 10.5 terabits per second per millimeter and under one picojoule per bit with good signal-to-noise ratio and acceptable cost.
About the speaker:
* Speaker: Anthony Cibié
* Company: CEA-Leti
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#MicroLEDMatrix, #MultiCoreFiber, #PhotodiodeMatrix, #SubPicojoulePerBit, #OpticalInterconnect, #ChipToChipCommunication
10:45 - 12:25
What 6 dB bandwidth did LEDs reach at 800 A/cm², and what eye-diagram speed held at 25°C and 100°C?
What 6 dB bandwidth did LEDs reach at 800 A/cm², and what eye-diagram speed held at 25°C and 100°C?
Electrical S21 versus frequency for different current density showed 6 dB bandwidth of 1.5 GHz at 800 A/cm². Separate LEDs' bandwidth over current density was also measured. A unified model for DC and AC LED behavior was developed, combining both in one model. Useful for modular device system studies or budget link: tune parameters, improve bandwidth by 1.5, check whole-link behavior. Also gives driver design team a black-box model matching the LEDs.
For model validation, an earlier generation epi stack was used because the latest was too recent. Eye-diagram measurements were performed at 25°C and 100°C at 400 Mbit/s. Modeling was also performed on those devices at the two temperatures. Experimental curves are shown for the devices, linking measured eye diagrams to the modeling effort.
The unified model supports system-level and driver-level use. In a budget link, changing assumed bandwidth by a factor of 1.5 lets engineers see overall link behavior after tuning parameters. For driver design, it provides a black box that behaves like the LEDs. The model covers DC and AC behavior together, based on S21 versus frequency and bandwidth versus current density data.
In this short video, you can learn:
* 6 dB bandwidth reached 1.5 GHz at 800 A/cm² current density.
* A unified model captures both DC and AC LED behavior for system and driver design.
* Eye-diagram measurements were run at 25°C and 100°C at 400 Mbit/s on a previous-generation epi stack.
📋 **Clip Abstract** A unified DC/AC LED model and S21 measurements show a 6 dB bandwidth of 1.5 GHz at 800 A/cm². Eye-diagram tests at 25°C and 100°C at 400 Mbit/s validate modeling on an earlier-generation epi stack.
About the speaker:
* Speaker: Anthony Cibié
* Company: CEA-Leti
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#6dBBandwidth, #S21Response, #UnifiedDcAcModel, #EyeDiagramMeasurement, #MicroLED, #OpticalInterconnect
06:17 - 07:57
What changes when a microLED shifts from display duty cycles to always-on data modulation at 1000 A/cm²?
What changes when a microLED shifts from display duty cycles to always-on data modulation at 1000 A/cm²?
For optical communication, the drive current density is much higher than for microdisplay, probably around 1,000 ampere per centimeter square. The LED size might also be higher to get enough flux and enough photons per period, so that the signal-to-noise ratio is good. The pitch will also be higher because, although the LED can go lower in size, the electronics and fiber cores must fit in the pitch.
The arrangement of LEDs will match the fiber core adjustments. Operating temperature will be quite high compared with display. The operating mode will also be very different: in display the LEDs are either turned off or on, but here they will be always on, and the LED will be modulated around its on state. The modulation bandwidth will also be much higher for optical communication.
These constraints about using micro-LEDs in optical communication ask questions about how the LED behaves in such a configuration. The discussion frames the shift as a change in current density, size, pitch, arrangement, operating temperature, operating mode and modulation bandwidth. That is why the behaviour of the LED in this configuration is being examined.
In this short video, you can learn:
* Optical-communication micro-LEDs are driven at much higher current density, probably around 1,000 ampere per centimeter square.
* Their size and pitch are increased to provide enough photons for a good signal-to-noise ratio and to fit electronics and fiber cores.
* They remain always on and are modulated around the on state, with higher operating temperature and modulation bandwidth than display.
📋 **Clip Abstract** This clip contrasts micro-LEDs for microdisplay with micro-LEDs for optical communication. It highlights higher current density around 1,000 ampere per centimeter square, larger size and pitch, fiber-matched arrangement, high operating temperature, always-on modulation and higher modulation bandwidth, which raise questions about LED behaviour in that configuration.
About the speaker:
* Speaker: Anthony Cibié
* Company: CEA-Leti
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#HighCurrentDensity, #SignalToNoiseRatio, #AlwaysOnModulation, #ModulationBandwidth, #MicroLEDs, #OpticalCommunication