Gregoire Denis | Avicena Tech: What physical trade-offs limit the bandwidth and internal quantum efficiency of micro LEDs in high-speed optical interconnects?
12:44 - 14:55
Other snippets from this talk
Summary of the clip:
What physical trade-offs limit the bandwidth and internal quantum efficiency of micro LEDs in high-speed optical interconnects?
For micro LEDs, the limit is mostly the recombination bandwidth, not the RC, so the drive conditions dominate. High bandwidth comes from running at high current densities and leveraging Auger recombinations. But this trades against internal quantum efficiency through the balance of radiative and non-radiative recombinations. You cannot push indefinitely: going faster raises the drive current, and you become limited, failing to meet the picosecond permit, so the current must be reduced.
There is a secondary effect: IQE is non-linear and reaches a plateau. To raise the high level to get higher bandwidth, you also raise the lower level, and then there is no more linearity. The extinction ratio, the ratio between the two levels, decreases, reducing gains and reducing OMA. Higher drive also means higher risk in reliability. So there is room for new epitaxy engineering to increase bandwidth while limiting IQE losses.
Those IQE losses will not be zero. The speaker notes years of optimizing this trade-off, but room remains. Semi-polar approaches may help: you do not necessarily need the same EQE and IQE as for display; these are trade-offs. The major opportunity is improving bandwidth at low current density. New epitaxy engineering and new solutions are needed to raise bandwidth while limiting IQE penalties.
In this short video, you can learn:
* Micro LEDs are mostly limited by recombination bandwidth rather than RC.
* High current densities leverage Auger recombinations for bandwidth but reduce IQE and extinction ratio, lowering OMA.
* Improving bandwidth at low current density through epitaxy engineering and semi-polar approaches is a major opportunity.
📋 **Clip Abstract** This clip examines the trade-offs limiting micro LED bandwidth and internal quantum efficiency in optical links. It covers Auger-driven high-current-density operation, non-linear IQE and extinction ratio losses, reliability risk, and the promise of epitaxy engineering, including semi-polar approaches, to improve bandwidth at low current density.
About the speaker:
* Speaker: Gregoire Denis
* Company: Avicena Tech
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#MicroLEDBandwidth, #AugerRecombination, #InternalQuantumEfficiency, #ExtinctionRatio, #OpticalInterconnects, #EpitaxyEngineering
This is a highlight of the presentation:
Wide, Slow, and Efficient: Redefining Optical Link Budgets with MicroLEDs
More Highlights from the same talk.
03:17 - 04:57
Why are micro LEDs emerging as a viable alternative to lasers and copper for short-reach optical interconnects?
Why are micro LEDs emerging as a viable alternative to lasers and copper for short-reach optical interconnects?
There is no single interconnect today, and there won't be a single one moving into this opportunity to remove the copper wall. Multiple types of interconnects will exist. Micro LED can be one of them if we make it right. Regarding reliability, you are not using lasers, and you can add redundancies with micro LEDs, it almost comes for free. High bandwidth densities have been shown by Hail from Mojo; the micro LED is not the bottleneck.
High energy efficiency comes because you don't have the threshold of the lasers. You have a simple architecture, NLZs, which makes no laser, no modulator, no DSP, no SerDes. You are simply fading the chain, so you can get an advantage on cost as well as in latency. That simplicity is the core of the efficiency claim.
We know that GaN is very reliable with temperature. Here we put higher to 200 Celsius. That should be demonstrated on system, but what we already demonstrated in the lab is a just link modulating a GaN LEDs or above 200 Celsius, your eye is still open. So from a modulation standpoint, it is actually reliable there. That temperature robustness supports the reliability case for micro LED interconnects.
In this short video, you can learn:
* Micro LEDs can add redundancy almost for free because no lasers are used, improving reliability for optical interconnects.
* A simple NLZ architecture eliminates laser, modulator, DSP, and SerDes, giving cost and latency advantages.
* GaN LEDs have been lab-demonstrated to modulate above 200 Celsius with an open eye, showing temperature reliability.
📋 **Clip Abstract** Micro LEDs are emerging as a viable alternative to lasers and copper for short-reach optical interconnects because they avoid laser thresholds, enable almost-free redundancy, and offer high bandwidth densities. A simple NLZ architecture removes laser, modulator, DSP, and SerDes, while lab results show GaN LEDs modulating above 200 Celsius with an open eye.
About the speaker:
* Speaker: Gregoire Denis
* Company: Avicena Tech
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#GaNLED, #MicroLEDInterconnect, #NLZArchitecture, #HighBandwidthDensity, #OpticalInterconnects, #ShortReachOptics
17:42 - 19:04
Why does blue remain optimal for microLED optical interconnects despite longer-wavelength benefits?
Why does blue remain optimal for microLED optical interconnects despite longer-wavelength benefits?
Paolo noted that longer wavelengths reduce latency and lower fiber dispersion, advantages that apply mainly to longer reach, but only if the full-width half-maximum stays the same. Co-optimizing components, reach, and data rate raises the wavelength question. The theoretical case for longer wavelengths depends on holding FWHM constant while shifting the emission wavelength. That condition is rarely met in practice, which changes the comparison.
MicroLEDs do not simply change wavelength. The FWHM also increases, so the expected gain from longer wavelengths erodes. After balancing all trade-offs across the link, blue probably wins. The overall balance favors blue despite the theoretical latency and dispersion benefits at longer wavelengths. This is the co-optimization lesson: you cannot isolate wavelength from the rest of the component set or from the reach and data rate targets.
Shorter reach may behave differently from longer reach. The trade-offs that apply at long distances do not automatically transfer to short distances. The conclusion is that there will not be a single interconnect. Different reaches will need different solutions. This reach-dependent behavior reinforces that wavelength choice must be evaluated against the specific link budget, not as a universal rule. Blue remains the likely choice for the targets discussed, but the analysis is reach-specific.
In this short video, you can learn:
* Longer wavelengths reduce latency and fiber dispersion only if FWHM stays constant, but microLEDs increase FWHM when wavelength shifts.
* After balancing all trade-offs, blue probably wins for the targeted reach and data rate.
* Shorter reach may differ from longer reach, so there will not be a single interconnect.
📋 **Clip Abstract** Longer wavelengths offer latency and dispersion gains for longer reach, but only if FWHM remains the same. MicroLEDs increase FWHM when wavelength changes, making trade-offs minimal and blue the likely winner, though shorter reach may require different choices.
About the speaker:
* Speaker: Gregoire Denis
* Company: Avicena Tech
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#MicroLED, #BlueWavelength, #FWHM, #FiberDispersion, #OpticalInterconnects, #DataCenterOptics
08:31 - 10:25
How does a one-terabit-per-second micro LED link hold up in a live ASIC-to-ASIC transceiver?
How does a one-terabit-per-second micro LED link hold up in a live ASIC-to-ASIC transceiver?
The light bundle kit couples a fiber bundle directly on top of micro LEDs transferred in the fab, with blue light visible at the interface. A March OFC demonstration ran at 500 gigabits per second, and the fast learning cycle pushed the same kit to one terabit per second. It is now offered to customers and goes live next week at the ECO Conference in Malaga.
Architecturally it is an ASIC-to-ASIC transceiver: LEDs sit on the ASIC on one side, photodetectors sit on the ASIC on the other, joined by bundles up to five meters long with a 90-degree bend. The link carries 335 micro LEDs at three gigabits per second each, and the ASIC side supplies the power needed to prove these links for optical data communication.
Validation runs on open eye monitors and PRBS generators, with live eye diagrams visible while power is consumed, and any link or channel selectable. Co-designing the photodetectors and surrounding components is what produces the very fast cycle of learnings behind the 500-gigabit to one-terabit jump, letting the team iterate on the same platform rather than rebuild it.
In this short video, you can learn:
* The light bundle kit moved from a 500-gigabit-per-second OFC demonstration in March to one terabit per second, and is now shipping to customers.
* Each of the 335 micro LEDs runs at three gigabits per second across bundles up to five meters long with a 90-degree bend.
* Open eye monitors and PRBS generators drive the link, with photodetector co-design compressing the learning cycle.
📋 **Clip Abstract** A fiber bundle is attached directly on top of fab-transferred micro LEDs to form a light bundle kit, first shown at 500 gigabits per second and then pushed to one terabit per second. The resulting ASIC-to-ASIC transceiver pairs LED-on-ASIC and photodetector-on-ASIC dies over up to five-meter bundles, with 335 micro LEDs at three gigabits per second each and live eye monitoring.
About the speaker:
* Speaker: Gregoire Denis
* Company: Avicena Tech
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#MicroLEDLightBundle, #ASICtoASICTransceiver, #OpenEyeMonitoring, #FiberBundleCoupling, #OpticalInterconnect, #DataCenterOptics