Alexander Loesing | ALLOS Semiconductors: How does a wide-and-slow microLED array with many channels beat a fast-and-narrow laser on system-level reliability?
07:57 - 09:38
Other snippets from this talk
Summary of the clip:
How does a wide-and-slow microLED array with many channels beat a fast-and-narrow laser on system-level reliability?
The opto industry pushes a lot of data through a single channel, with lasers that are super, super fast. A microLED approach will never reach that speed. Instead, it uses lots of channels and a much slower rate. Because LEDs are less expensive than lasers, and after one million things to consider, the slow-and-wide path can transmit the same data. It is not about chasing laser frequency. It is about choosing the right point on the curve for each application.
With an array of ten thousand channels, if 10% go down, nobody cares because enough backup exists. A laser goes down: pull the plug, replace the pluggable. But with deeper integration, if the laser is on the chip and goes down, the chip is down, then the board, then the tray, then the rack, then the row. Not good for the data center business. That reliability is built into microLEDs and multiple channels, thinking on a system level.
The approach is not trying to chase the lasers, because we will never reach them. It is not only chasing frequency. Instead, think on a system level, where you can have the right trade-off points for your system. This is data from customers: two examples of devices show significantly different efficiencies. That difference matters for the system. The wide-and-slow array is a different system-level optimization.
In this short video, you can learn:
* Fast-and-narrow lasers push data through one very fast channel, while wide-and-slow microLEDs use many slower channels to transmit the same data.
* With ten thousand channels, 10% failures are tolerable, but a laser failure needs a pluggable replacement, and deeper integration cascades chip to board to tray to rack to row.
* Customer data on two devices shows significantly different efficiencies, supporting system-level trade-off points instead of chasing laser frequency.
📋 **Clip Abstract** The opto industry's fast-and-narrow lasers cannot be matched in speed by microLEDs, so a wide-and-slow array uses many slower channels to transmit the same data. That array redundancy tolerates 10% channel failures, avoids pluggable laser replacement and cascading data-center failures, and lets system-level trade-off points, not laser frequency, guide device selection.
About the speaker:
* Speaker: Alexander Loesing
* Company: ALLOS Semiconductors
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#WideAndSlow, #FastAndNarrow, #MicroLEDArrays, #ChannelRedundancy, #Optoelectronics, #DataCenterOptics
This is a highlight of the presentation:
MicroLEDs for Displays and Optical Interconnects: Different Markets, Shared Manufacturing Challenges
More Highlights from the same talk.
05:45 - 08:02
Why does increasing current density to boost MicroLED bandwidth create a fundamental trade-off with efficiency?
Why does increasing current density to boost MicroLED bandwidth create a fundamental trade-off with efficiency?
The trade-off is framed as a triangle: frequency, or bandwidth, efficiency, and reliability plus scaling. Ramping current density crams more power into the chip, allowing faster switching. But increasing current density lowers internal quantum efficiency, the light extracted and the LED’s intrinsic efficiency. Efficiency sits on the Y-axis, and this drop is physics, not an engineering choice. You gain bandwidth while energy efficiency falls, and there is very little you can do about it.
The rainbow curve captures this trade-off: you can have high bandwidth, high frequency, or high efficiency, but not both. Simplified math and the formula tell the same story. If current density rises, bandwidth goes up and energy efficiency goes down. Everybody follows the current-density path, yet frequency is not the only metric; where the energy goes matters. You can tweak things, but you remain constrained to different points on that curve.
For manufacturing and adoption, reliability and scaling sit alongside speed and efficiency. If the MicroLED cannot be made reliably or scaled, nobody will buy it. Decades of work show a simple path: ramp current density, cram more power into the chip, switch faster. That path collides with internal quantum efficiency. The practical response is to tweak the design and choose different operating points on the rainbow curve for different applications, accepting the physics of the trade-off.
In this short video, you can learn:
* Increasing current density raises bandwidth or frequency but lowers internal quantum efficiency.
* The rainbow curve describes a physics-bound choice between high bandwidth/high frequency and high efficiency.
* Different applications can occupy different points on that curve after tweaks, but the trade-off remains.
📋 **Clip Abstract** Ramping current density makes MicroLEDs switch faster, but it also reduces internal quantum efficiency, the intrinsic light output. The rainbow curve shows a physics-bound trade-off between high bandwidth and high efficiency, so designers must tweak and select different operating points for different applications.
About the speaker:
* Speaker: Alexander Loesing
* Company: ALLOS Semiconductors
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#InternalQuantumEfficiency, #RainbowCurve, #CurrentDensity, #PicojoulesPerBit, #OpticalInterconnects, #MicroLEDDisplays
18:17 - 19:27
Why are MicroLEDs, rather than lasers, the key to enabling opto-electrical interposers for GPU-to-memory optical interconnects?
Why are MicroLEDs, rather than lasers, the key to enabling opto-electrical interposers for GPU-to-memory optical interconnects?
Arrays can be built and placed into cables, creating active optical cables, an approach Avicena has preached for many years. That is called a great opportunity. It connects arrays to cable-level optical links. Active optical cables are framed as one route, separate from the scale-in route. That route goes with opto power onto the board and onto the chip level.
The second opportunity is scale-in: moving opto power onto the board and onto the chip level. NVIDIA, at a semi-private conference in Germany hosted by a big chemical company, said it wants not only an electrical interposer but an opto-electrical interposer. The interposer is the substrate piece, under different names, that connects chips on it—for example, the GPU with the memory—by electrical and optical means.
Highly funded companies, and ones with higher valuations when acquired in the US, like Meta and Celestial AI, have been working on this opto-electrical interposer. The goal is to send huge amounts of more data across the chips on the board. That is not possible with lasers, not in a way that people really like. This is seen as a great opportunity.
In this short video, you can learn:
* Arrays can be built and put into cables to create active optical cables, an approach Avicena has preached for many years.
* Scale-in moves opto power onto the board and chip level, where NVIDIA said it wants an opto-electrical interposer, not only an electrical interposer.
* Meta and Celestial AI have worked on opto-electrical interposers to send huge amounts of more data across board chips, which is not possible with lasers in a way people really like.
📋 **Clip Abstract** The clip contrasts cable-level active optical cables with scale-in opto power at board and chip level. It highlights NVIDIA's stated interest in an opto-electrical interposer and notes that Meta and Celestial AI have worked on this because lasers cannot provide the desired data movement across board chips.
About the speaker:
* Speaker: Alexander Loesing
* Company: ALLOS Semiconductors
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#ActiveOpticalCables, #OptoElectricalInterposer, #ScaleInOptoPower, #GPUtoMemory, #OpticalInterconnects, #MicroLEDs
12:45 - 14:11
Why must MicroLED manufacturing transition to silicon lines and 300mm wafers to achieve commercial viability?
Why must MicroLED manufacturing transition to silicon lines and 300mm wafers to achieve commercial viability?
The LED industry's current approach cannot deliver the scale required. It uses the wrong wafer sizes and the wrong mindset. The silicon industry operates completely differently, and being ready for that approach is essential. Without tapping into that scale, a company cannot be in this business or present a scalable option. This is the foundation for commercial viability, before cost and yield enter the discussion.
This work has continued for more than a decade, and with the previous company even longer. That long maturation means the approach is ready to go, including on 300 millimeter. Skepticism remains because the supply chain is not ready. Finally, there is some movement: more people who have 300 millimeter lines are getting engaged. That engagement is important because supply-chain readiness involves more than one wafer type.
The supply chain includes not only the LED wafer but also the driver wafer. That means diameters must match absolutely. If they do not match, things become very expensive. This matching requirement is a manufacturing constraint for scalable production. The industry must deliver on this to tap into scale and present a scalable option, while also addressing cost and high yield.
In this short video, you can learn:
* The LED industry's current wafer sizes and mindset cannot support the scale required for MicroLED commercial viability.
* More than a decade of development has made the approach ready for 300 millimeter, but supply-chain readiness still limits adoption.
* Matching diameters between LED and driver wafers is essential, because mismatched diameters make the process very expensive.
📋 **Clip Abstract** MicroLED commercialization requires silicon-line scale and 300 millimeter capability, which the traditional LED industry approach cannot provide. Supply-chain readiness, including matching LED and driver wafer diameters, remains critical because mismatches drive up cost.
About the speaker:
* Speaker: Alexander Loesing
* Company: ALLOS Semiconductors
* Event: Eindhoven 2026
* Location: High Tech Campus, Eindhoven
#MicroLED, #300mmWafer, #SiliconLineManufacturing, #DriverWaferMatching, #OpticalInterconnects, #DisplayManufacturing