When light meets bytes: Intel introduces crucial optical technology that will boost AI performance: OCI chiplet can reach up to 4 Tbps and consume nearly 70% less power than rivals
Intel unveiled the industry’s first fully integrated bidirectional optical I/O chiplet at the recent Optical Fiber Communication Conference (OFC) 2024.
Presented by Intel’s Integrated Photonics Solutions group, this optical computer interconnect (OCI) supports 64 channels of 32 Gbps data transmission in both directions over up to 100 meters of fiber.
The technology, which can be paired with CPUs and GPUs — a task that was previously complex — addresses the growing demand from AI infrastructure for higher bandwidth, lower power consumption and greater range.
Meeting the demand for AI
It’s well documented that AI-based applications, including LLMs and generative AI, are driving unprecedented demand for I/O bandwidth, driving the need for longer ranges to support larger CPU/GPU clusters. Electrical I/O, which relies on copper traces, offers high bandwidth density and low power, but is limited to short distances. Intel says its co-packaged optical I/O solution can transmit data over much longer distances with higher efficiency and lower power consumption, essential for AI/ML infrastructure scaling.
The OCI chiplet integrates a silicon photonics integrated circuit (PIC) with on-chip lasers and optical amplifiers, with an electrical IC. It supports 4Tbps bidirectional data transfer, compatible with PCIe Gen5, using 8 wavelengths at 200GHz spacing on a single fiber. It also consumes only 5 pico-Joules (pJ) per bit, compared to 15 pJ/bit for pluggable optical transceiver modules.
Intel’s OCI chiplet is still a prototype, but the company says it is working with select customers to combine OCI with Systems-on-Chips (SoCs) and System-in-Packages (SiPs).
“The ever-increasing movement of data from server to server is putting significant pressure on the capabilities of today’s data center infrastructure, and current solutions are rapidly approaching the practical limits of electrical I/O performance,” said Thomas Liljeberg, senior director of Product Management and Strategy for Intel’s Integrated Photonics Solutions Group.
“However, Intel’s achievement enables customers to seamlessly integrate co-packaged silicon photonics interconnect solutions into next-generation computing systems, extending reach, enabling ML workload acceleration that promises to revolutionize high-performance AI infrastructure.”