IN Brief:
- ITRI's silicon photonics engine provides 3.2Tbps aggregate transmission, twice the capacity of its previous 1.6T generation.
- More than 20 partners are participating across chip design, fabrication, testing, and packaging.
- Commercialisation work is running alongside international collaboration on technical specifications for 3.2T optical technology.
Industrial Technology Research Institute has unveiled a 3.2Tbps silicon photonics optical engine developed with support from Taiwan’s Department of Industrial Technology, doubling the aggregate transmission capacity of ITRI’s previous 1.6T generation.
The platform provides a total 3.2Tbps of optical transmission, equivalent to approximately 400GB of data each second. ITRI uses multiple integrated high-speed optical channels, although it has not disclosed the individual lane rate, modulation format, package dimensions, or power consumption in the current announcement.
Those missing details prevent a direct specification comparison with commercial 3.2T transceiver platforms, but the programme extends beyond a laboratory bandwidth demonstration. ITRI has assembled more than 20 partners across chip design, fabrication, testing, and packaging and is working with international companies on specifications for 3.2T optical technology.
The supply network addresses one of the persistent problems in silicon photonics: a working optical circuit is only one part of a manufacturable product. Lasers, modulators, electronic drivers, receivers, fibre coupling, packaging, thermal management, and test all contribute to yield and cost, with several of the most difficult interfaces occurring between different technologies.
Electrical interconnect remains effective over short distances inside servers and packages, but increasing data rates raise insertion loss, equalisation power, and signal-integrity demands. Optical links shift longer or higher-bandwidth connections into the photonic domain, reducing some electrical reach constraints while adding a new set of packaging and optical manufacturing requirements.
ITRI’s 3.2T work arrives as the wider industry moves towards denser optical connectivity around AI systems. Recent market and manufacturing programmes have shifted attention from proving that silicon photonics can carry high-speed signals towards the lasers, connectors, fibres, testing, and production yield needed to manufacture large volumes of optical hardware.
That transition is already visible in Taiwan’s semiconductor ecosystem. Earlier this month, TSMC forecast that silicon photonics could exceed half of the optical-transceiver market during 2027, while highlighting the same manufacturing constraints around lasers, fibre connectivity, and test. ITRI’s programme approaches those constraints by building a domestic supply network around the optical engine itself.
The 3.2Tbps aggregate figure is twice ITRI’s earlier 1.6T generation, but bandwidth alone will not determine commercial adoption. Power per bit, optical loss, thermal behaviour, fibre attachment, package footprint, test time, and the electrical interface to the host silicon will shape where the technology can be used.
Testing becomes particularly difficult as optical engines add more high-speed channels. Electrical performance and optical output have to be characterised together, while coupling loss and assembly variation can affect measurements that would be straightforward on a conventional electrical die. Wafer-level optical probing, package test, and system validation therefore become part of the architecture rather than a final manufacturing check.
Packaging presents a similar constraint. The closer optics move towards switches and accelerators, the shorter the electrical connection becomes, but thermal density and assembly tolerances become more difficult. A high-bandwidth optical engine has to maintain alignment and optical performance beside silicon devices dissipating substantial power.
ITRI’s network of design, fabrication, packaging, and test partners is intended to address those interfaces before the technology reaches commercial production. The programme gives Taiwan’s existing semiconductor manufacturing base a route further into optical interconnect hardware rather than treating photonics as a separate supply chain.
The current announcement establishes aggregate bandwidth and the breadth of the commercialisation programme but leaves important engineering data for later disclosure. Lane architecture, energy efficiency, coupling method, thermal limits, and production package design will be needed before system developers can compare the engine directly with competing 3.2T implementations.
The next milestones are therefore likely to be less about another bandwidth increase and more about product definition. Customer evaluation, detailed electrical and optical specifications, packaging, and repeatable test results will show whether the 3.2Tbps engine can move from an institute-led technology programme into a manufacturable optical component.
