Silicon photonics First fully integrated optical chip set with 4 TBit/s

From Hendrik Härter | Translated by AI 2 min Reading Time

Related Vendors

Intel has unveiled a prototype of an Optical-Compute-Interconnect (OCI) chip. The OCI chip allows for transmission of up to 4 Tbit/s across 64 channels. The development is particularly interesting for the increasing computing power of AI applications.

Optical Chip: Intel has unveiled a prototype chip based on silicon photonics. With it, the company wants to meet the increasing computational load. Data rates of up to 4 Tbit/s are possible over 64 channels.(Image: Intel)
Optical Chip: Intel has unveiled a prototype chip based on silicon photonics. With it, the company wants to meet the increasing computational load. Data rates of up to 4 Tbit/s are possible over 64 channels.
(Image: Intel)

With the advent and increasing spread of artificial intelligence (AI) and machine learning (ML), the demands on data centers are significantly increasing. Faster data processing can help distribute the load. Intel has now introduced an Optical Compute Interconnect (OCI) chip that meets the requirements for high bandwidth, low power consumption and long-range. This is supposed to enable new computing architectures, but also better scalability for AI applications.

The OCI chip introduced by Intel is designed to transmit 64 channels of 32 Gbps data over a 100 m fiber optic distance. The fully integrated OCI chip uses Intel-developed silicon photonics technology and incorporates a silicon photonics integrated circuit (PIC). This includes on-chip lasers and optical amplifiers and combines them with an electrical IC.

Data rates of up to 4 TBit/s

The OCI implementation presented by Intel supports a bidirectional data transfer of 4 Tbps. In detail, these are 64 channels with 32 GT/s each over fifth generation PCIe (PCIe Gen5). DWDM (Dense Wavelength Division Multiplexing) is used to transmit the data. The hybrid version of optical and electrical components is remarkably energy efficient and uses only 5 pico-joules (pJ) per bit compared to pluggable optical transceiver modules with about 15 pJ/Bit. However, five trillionths of a joule per bit at four trillion bits per second results in a power consumption of 20 watts.

New interface for upcoming computer systems

Integrated photonics has been driven by Intel for over 25 years. Intel was the first company to develop and deliver silicon photonics-based chips, mainly for large cloud service providers. In 2023, Intel showed a processor with four OCI chips, but they only reached half the bandwidth.

Thomas Liljeberg, Senior Director of Product Management and Strategy at Intel's Integrated Photonics Solutions Group: "Our development allows users to seamlessly integrate a so-called Co-Package Silicon Photonics Interconnect into next-generation computer systems. Our OCI chip increases bandwidth, reduces power consumption, and increases range, enabling the acceleration of ML workloads and will fundamentally change high-performance AI infrastructure."

Key role of the next chip generation

The integration of OCI chiplets enables data centers to meet the constantly increasing demands for bandwidth and energy efficiency. HPC applications benefit from the high data transfer rates and low latency. Despite the current prototype phase and the upcoming widespread introduction, Intel's developments and cooperation with select customers show that OCI technology has a promising future and will play a key role in the next generation of computing and data processing architectures. (heh)

Subscribe to the newsletter now

Don't Miss out on Our Best Content

By clicking on „Subscribe to Newsletter“ I agree to the processing and use of my data according to the consent form (please expand for details) and accept the Terms of Use. For more information, please see our Privacy Policy. The consent declaration relates, among other things, to the sending of editorial newsletters by email and to data matching for marketing purposes with selected advertising partners (e.g., LinkedIn, Google, Meta)

Unfold for details of your consent