A Startup Wants to Put Light Inside AI Chips

A semiconductor startup is attempting to replace short electrical data links with light generated inside the silicon chip itself. On November 10, 2026, Staera Semiconductor, formerly New Silicon Corporation, emerged from stealth with US$4.6 million in pre-seed financing for its Lightwire optical interconnect. The engineering concept targets one of the hardest problems in AI hardware: transferring vast amounts of data between compute dies, memory and surrounding systems without excessive electricity consumption.

Staera says its architecture integrates silicon CMOS circuitry, gallium-nitride (GaN) light-emitting diodes and silicon-nitride optical waveguides on commercial wafers. Unlike optical modules attached outside a package, its proposed links can be built into an interposer, the bridge that connects multiple dies. Commercial availability is targeted for 2030, not claimed for today.

Video: MediaTek’s official 2026 demonstration of microLED optical transceivers. It illustrates a related industry approach, not Staera hardware or independent verification of Lightwire.

The promised energy savings

Staera lists a design objective below 0.2 picojoules per bit (pJ/bit), compared with its stated 5 pJ/bit reference for co-packaged optics. Using those figures, the proposed link energy would be at least 25 times lower, or at most 4% of that reference. The comparison is a company-defined target, not a controlled third-party benchmark with identical link distance, modulation, throughput and power-accounting boundaries. Actual systems must also power transmitters, receivers, control logic and cooling.

The company claims 4 to 8 gigabits per second per link and more than 600 links per millimeter of chip edge. Link density is not the same thing as measured aggregate throughput: simultaneously driving hundreds of channels raises engineering questions involving optical crosstalk, electrical routing, thermal operation, error rates and production yield. Staera also describes a high-bandwidth-memory design goal above 24 terabytes per second. That is a proposed architecture, not a demonstrated commercial HBM product.

Why the manufacturing method matters

Silicon is effective for electronic logic and waveguides but does not efficiently emit light on its own. Staera proposes incorporating GaN emitters directly into the manufacturing flow. The company reports that its earlier GaN-on-silicon integration work achieved 98% to 99% process yields. That historical figure must not be interpreted as a verified yield for complete mass-produced Lightwire systems.

Chipmakers currently employ a spectrum of approaches, including high-speed copper links, pluggable optical transceivers and co-packaged optics. Readers can compare the networking scale involved in our 800G versus 1.6T optics analysis and examine microLED data-center interconnect research. The critical measurement is usually energy per useful delivered bit at a specified distance and error rate, rather than an isolated component result.

Related technology video repeated for technical context. The video is attributed to MediaTek and should not be described as a Lightwire demonstration.

Funding is real; commercial validation is still ahead

The US$4.6 million pre-seed round was led by Openspace Capital, with participation from SGInnovate, Black Kite Capital, 28 Holdings, In Group and Decimus Technologies. The financing supports a move from research toward customer development. It does not demonstrate a booked order, production ramp or independently certified efficiency improvement.

A useful way to judge commercial progress is to look for third-party measurement of pJ/bit at matched distance, full package power, link bit-error rate, sustained bandwidth, yield per wafer and qualification by named customers. Our broader future semiconductor materials comparison explains why photonics is most likely to complement rather than replace electronic processors. Separately, recent photonic-chip research shows how varied light-based hardware applications remain.

Sources and verification

Primary source: Staera announcement, November 10, 2026; company technical background: Lightwire introduction and integration approach. Independent coverage: Optics & Photonics News. The energy comparison is editorial arithmetic using Staera-stated design targets; no independent field performance is implied.

Editorial note: Lightwire efficiency, density and HBM figures are company claims and forward-looking engineering goals, not independently verified deployed AI data-center results. The linked MediaTek video demonstrates a different manufacturer’s related optical-transceiver research.

We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiatives, please scan the QR code below

Leave a Reply