Alexander Loesing | Allos Semiconductor: What strain mismatch must be overcome to thin GaN-on-Si and scale bow from 100 to 300 mm?
05:11 - 07:38
Other snippets from this talk
Summary of the clip:
What strain mismatch must be overcome to thin GaN-on-Si and scale bow from 100 to 300 mm?
On a silicon substrate, the epi starts on a 1.5-millimetre-thick wafer; the wafer is then thinned, and the know-how lies in making its physical dimensions compatible with a silicon line. Low bow must hold across the whole value chain. A consistent bow value indicates no residual strain from the sandwich of epi layers and substrate—an achievement that is not obvious.
Two mismatches dominate. The lattice constant differs, so a different atomic pattern is built on top and does not match, causing strain, bad crystal quality and wafer bow. The thermal-expansion coefficient is also hugely mismatched: gallium nitride for LEDs is deposited at 1000 degrees Celsius, then cooled to room temperature, and doing that correctly is difficult.
Curvature comes from stress, and bow is the vertical distance that fabs dislike. As diameter increases from the 100-millimetre standard in the LED industry, holding a 50-micrometre bow requires a very significant reduction in curvature. Moving from 200 to 300 millimetres demands more than a two-times decrease.
In this short video, you can learn:
* A 1.5-millimetre silicon substrate is thinned and dimensioned for a silicon line while low bow is kept consistent across the value chain.
* Lattice-constant and thermal-expansion mismatch cause strain, poor crystal quality and bow, with GaN deposited at 1000 degrees Celsius and cooled to room temperature.
* Scaling from 200 to 300 millimetres requires more than a two-times curvature reduction to hold a 50-micrometre bow.
📋 **Clip Abstract** A silicon substrate begins 1.5 millimetres thick and is thinned for silicon-line compatibility, with low bow maintained across the value chain. Lattice-constant and thermal-expansion mismatch create strain and bow; scaling to 300 millimetres needs more than a two-times curvature reduction to hold a 50-micrometre bow.
About the speaker:
* Speaker: Alexander Loesing
* Company: Allos Semiconductor
* Event: MicroLED Connect 2024
* Location: High Tech Campus, Eindhoven, Netherlands
Join the most event MicroLED, AR/VR display and Optical I/O Connect event of the year
Learn more https://www.microledconnect.com/
#GaNOnSi, #WaferBow, #LatticeMismatch, #ThermalExpansionMismatch, #LEDManufacturing, #CompoundSemiconductors
This is a highlight of the presentation:
More Highlights from the same talk.
01:45 - 04:09
Why do microLEDs force a move to semiconductor fabs, and what wafer specs must they meet?
Why do microLEDs force a move to semiconductor fabs, and what wafer specs must they meet?
At Display Week last year, the industry belief shifted: microLEDs need semiconductor processing lines to reach the yield levels and cost points the industry needs. Conventional LEDs can accept many defects on a large chip, but the same defect in a microLED array for glasses or head-up displays kills the complete product. That defect must be avoided.
The semiconductor industry is big, powerful and mainly digital, with acceptance criteria different from the LED world. These LEDs use different materials, not silicon, so fabs must be convinced to take the wafers. Most prototypes use LEDs from the traditional LED world: small diameters and thick wafers to control strain. Strain remains high, producing large bow that harms bonding yield and raises cost.
The silicon industry is very digital and cannot work with those wafers. You need the wafer size they use: 200 millimeter diameter. The wafers must have the same physical dimensions, the right thickness, edge rounding and overall surface quality. Many details would simply not be accepted in their line. Ideally, bow should be 50 micrometer.
In this short video, you can learn:
* MicroLED arrays cannot tolerate the defect levels accepted on conventional large LEDs because one defect can kill a complete product.
* Moving to semiconductor processing lines is framed as necessary for the yield and cost points microLEDs need.
* The required wafer specs include 200 millimeter diameter, correct thickness, edge rounding, surface quality and bow ideally at 50 micrometer.
📋 **Clip Abstract** The clip explains why microLEDs are moving toward semiconductor processing lines: defect tolerance is far lower than in conventional LEDs, where a single defect can kill an entire array. It details the wafer requirements fabs demand, including 200 millimeter diameter, physical dimensions, thickness, edge rounding, surface quality and bow ideally at 50 micrometer.
About the speaker:
* Speaker: Alexander Loesing
* Company: Allos Semiconductor
* Event: MicroLED Connect 2024
* Location: High Tech Campus, Eindhoven, Netherlands
Join the most event MicroLED, AR/VR display and Optical I/O Connect event of the year
Learn more https://www.microledconnect.com/
#MicroLEDDefectYield, #SemiconductorProcessingLines, #200mmWaferDiameter, #WaferBow50Micrometer, #MicroLEDDisplays, #DisplayManufacturing
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
18:31 - 20:44
Can GaN-on-Si red match InGaN, and when will it be production-ready?
Can GaN-on-Si red match InGaN, and when will it be production-ready?
No red LED product is shipping today. Two partner projects are using the material to improve red performance; one presentation illustrates a strain-engineering benefit from the substrate. The approach can make strain more tensile or compressive according to need, and produce a convex, concave or flat wafer for the target.
Strain can also be controlled in individual layers at different growth stages, which is said to help address the lattice-constant problem in red. Results are pending and the work is very early. The personal view is that red still has time to improve because most people currently work on quantum dots.
Asked whether three to five years is realistic, the answer is yes: results are extremely encouraging, and using the same material brings manufacturing-integration advantages, with values becoming feasible for efficiency. Production readiness still requires technology for large wafers and growth of quantum wells and the whole LED stack with good uniformity and stable processes.
In this short video, you can learn:
* No red LEDs are shipping; two partner projects are using the material to improve red performance.
* Strain can be tuned tensile or compressive and applied in individual layers at different growth stages to address red's lattice-constant problem.
* Three to five years is judged realistic for red, but production readiness needs large-wafer technology plus uniform quantum-well and LED-stack growth with stable processes.
📋 **Clip Abstract** No red product is shipping yet; two partner projects are using the material to improve red performance through strain engineering from the substrate. While results are pending and the work is early, three to five years is considered a realistic timeframe, with production readiness depending on large-wafer technology and uniform, stable quantum-well and LED-stack growth.
About the speaker:
* Speaker: Alexander Loesing
* Company: Allos Semiconductor
* Event: MicroLED Connect 2024
* Location: High Tech Campus, Eindhoven, Netherlands
Join the most event MicroLED, AR/VR display and Optical I/O Connect event of the year
Learn more https://www.microledconnect.com/
#RedLEDStrainEngineering, #SubstrateStrainControl, #QuantumWellUniformity, #LargeWaferLEDGrowth, #MicroLEDDisplays, #LEDManufacturing
Other talks by Alexander Loesing
- The Path of 300 mm GaN-on-Si into Silicon Semiconductor Fabs · MicroLED Connect 2024
- Development of GaN uLEDs on 300mm Si wafers · Quantum Dots: Material Innovations and Commercial Applications 2022
More highlights from Alexander Loesing
- ▶ Why does increasing current density to boost MicroLED bandwidth create a fundamental trade-off with efficiency? · MicroLED Connect 2026
- ▶ How does a wide-and-slow microLED array with many channels beat a fast-and-narrow laser on system-level reliability? · MicroLED Connect 2026
- ▶ Why must MicroLED manufacturing transition to silicon lines and 300mm wafers to achieve commercial viability? · MicroLED Connect 2026
- ▶ Why are MicroLEDs, rather than lasers, the key to enabling opto-electrical interposers for GPU-to-memory optical interconnects? · MicroLED Connect 2026
- ▶ Why do microLEDs force a move to semiconductor fabs, and what wafer specs must they meet? · MicroLED Connect 2024
- ▶ Can GaN-on-Si red match InGaN, and when will it be production-ready? · MicroLED Connect 2024
