How silicon photonics is helping AI models grow

Silicon photonics is poised to play a pivotal role in unlocking the full potential of artificial intelligence (AI). The advent of large language models driving AI capabilities has spurred increased investment and intense competition in the field of silicon photonics. This technology seamlessly integrates silicon-based integrated circuits (ICs) with optical components to efficiently process and transmit vast volumes of data.

Prominent IC designers and manufacturers, as well as leaders in AI systems and telecommunications equipment, have entered this race. Renowned participants include NVIDIA, TSMC, Intel, IBM, Cisco Systems, Huawei, NTT, and imec, the Interuniversity Microelectronics Centre based in Belgium. Some of these entities, like Intel and NTT, have been devoted to advancing silicon photonics for nearly two decades.

Speaking at the Silicon Photonics Global Summit in Taiwan on September 5, Douglas Yu, TSMC’s Vice President overseeing system integration pathfinding, highlighted the potential of silicon photonics, emphasizing its ability to address critical challenges related to energy efficiency and AI computing power. This technology could herald a paradigm shift and mark the beginning of a new era.

Silicon photonics in the semiconductor industry has gained significant attention due to its attributes, including high bandwidth, rapid data transmission, long-distance capabilities, low power consumption, and applicability to advanced networking, cloud computing, data centres, autonomous vehicles, and smart transportation systems. Essentially, it is expected to enhance device and system performance while reducing energy consumption across various high-tech industries.

According to industry projections, the global silicon photonics market is anticipated to reach a value of $7.86 billion by 2030, with a compound annual growth rate (CAGR) of 25.7% from its 2022 valuation of $1.26 billion.

Around September 2022, NVIDIA and TSMC initiated a collaborative research and development project named COUPE (Compact Universal Photonic Engine). This endeavour aims to leverage NVIDIA’s silicon photonic technology to combine multiple AI processors (GPUs) using co-packaged optics technology integration. This approach significantly minimizes signal loss, enabling the creation of a large-scale GPU set.

At the same time, Intel defines silicon photonics as the amalgamation of two pivotal 20th-century inventions, silicon integrated circuits and semiconductor lasers. It facilitates faster data transmission over longer distances compared to conventional electronics, making it ideal for Ethernet switches, routers, and transport networking equipment in large-scale cloud and enterprise data centres.

Intel holds a commanding market share in manufacturing optical networking transceivers, thanks to its scale and integrated solution, which boasts being better than competitors. Intel plans to quadruple its advanced IC packaging capacity by 2025, including establishing a new 3D packaging facility in Malaysia.

Furthermore, NTT, Japan’s national telecom carrier and telecommunications technology developer, aims to produce advanced photonics-electronics convergence devices with significant improvements in energy efficiency, transmission capacity, and latency reduction for mobile and optical networks by 2030. NTT, alongside Intel and Sony, established the Innovative Optical and Wireless Network Global Forum to advance all-photonics network infrastructure.

Huawei and imec initially collaborated on silicon photonics research in 2014, but this partnership was disrupted due to Huawei’s inclusion on the US Commerce Department’s Entity List and restrictions on ASML’s lithography systems to China. Huawei continued its research independently, aiming to create photonic chips to circumvent US chip restrictions.

This technology is seen as highly promising, however it is considered to be in its early stages, with the potential to challenge the dominance of European and American countries in the future.