MicroLED: The Optical Technology All Three Memory Makers Are Watching

MicroLED: The Optical Technology All Three Memory Makers Are Watching

When Memory Meets Optics

The most expensive word in this AI cycle is bottleneck. As the datacenter bottleneck moved from compute to data movement, the memory companies holding HBM saw their stocks rerate first. As models grew, what ran short was not chip speed but the ability to feed data into chips.

The same logic carried straight over to optics. As light took over the distances copper could not cover, the entire optical value chain from transceivers to CPO got rerated. Memory and optics, the two protagonists of AI infrastructure investing, have been attacking the same bottleneck from opposite ends.

Yet the company standing where these two currents cross is surprisingly little known.

Avicena, an optical interconnect startup in Sunnyvale, California, counts Samsung Electronics, SK hynix, and Micron all on its cap table.

This article covers why memory companies are paying attention to MicroLED, how this technology differs from conventional optical communications, what is moving on the ground right now, and what investors can take away from it.

Table of Contents

  1. The Cap Table All Three Memory Makers Climbed Onto
  2. LightBundle: Optical Communications Without the Laser
  3. Where Wide and Slow Meets HBM
  4. The Companies Following Avicena: Marvell, MediaTek, Credo, OSRAM
  5. Comparing Against the Laser Camp: Where the Moat Is, and Where the Weaknesses Are
  6. Investment Insight: Don’t Pick the Physics, Pick the Structure

1. The Cap Table All Three Memory Makers Climbed Onto

To follow MicroLED optical communications, there is only one company you need to watch: Avicena. It is the clear leader in this technology, and it is no exaggeration to say that every company now trying to catch up is referencing its architecture.

Start with the timeline.

In 2019, engineers from the optical communications industry founded Avicena. Founder and first CEO Bardia Pezeshki built his career on laser-based optical communications.

August 2022: Series A of $25 million. Samsung Catalyst Fund, Micron Ventures, Cerberus Capital Management, and Clear Ventures participated.

October 2022: acquired a MicroLED fab and engineering team from Nanosys, giving the company the means to produce its own prototypes.

March 2023: signed a joint development agreement with ams OSRAM for volume manufacturing of GaN MicroLED arrays.

April 2025: announced a collaboration with TSMC. LightBundle is the name of Avicena’s MicroLED interconnect platform.

May 2025: Series B of $65 million. Tiger Global led, joined by SK hynix, Maverick Silicon, Prosperity7 Ventures (an Aramco affiliate), VentureTech Alliance (TSMC’s VC arm), Hitachi Ventures, and Lam Research.

September 2025: Marco Chisari joined Avicena as CEO.

March 2026: launched the LightBundle eKit, the industry’s first MicroLED optical communications evaluation platform.

The comment SK hynix left when it invested contains the exact problem this company is trying to solve. High-speed electrical interconnects face a fundamental limit on reach, and laser-based optical links consume too much power.

MicroLED-based interconnects are therefore a promising low-power link candidate for the future memory fabric.

2. LightBundle: Optical Communications Without the Laser

Today’s datacenter optics are all laser-based. Avicena went in the exact opposite direction.

Instead of raising per-channel speed, it lowered it and multiplied the channel count into the hundreds. The company’s own phrase for this is “Wide and Slow.”

The structure works like this. On the transmit side, hundreds of blue GaN MicroLEDs are packed into an array. Each LED carries one independent data lane. The array couples into an imaging-grade multicore fiber bundle, and on the receive side, a matching array of silicon photodetectors (PDs) picks it up.

This architecture produces four points of differentiation.

First, there is no laser.

In laser-based optical links, most reliability, manufacturing, and cost problems trace back to the laser and to laser-fiber alignment.

Second, the power is in a different order of magnitude.

A laser does not operate below its lasing threshold. LEDs have no threshold, so drive current can be lowered as far as the receiver’s SNR allows.

Third, it removes the SerDes.

LightBundle takes the chip’s internal parallel data and maps it nearly 1:1 onto optical lanes.

Fourth, very few components need to be invented.

MicroLED arrays are something the display industry already mass-produces, and silicon PD arrays are something the CIS industry already mass-produces.

According to current employees at SK hynix, Samsung, and Micron in my industry network, all three memory makers are now very seriously evaluating MicroLED for memory optical interconnect.