Intel has developed a small chiplet that uses light instead of electricity to transmit data between processors, potentially improving the bandwidth and efficiency of data centers tasked with intense AI processing.
- The OCI chiplet was recently demonstrated by Intel at the Optical Fiber Communication Conference, integrated with one of its CPUs.
- By using laser light traveling through fiber optic connections, the chiplet enables processors to communicate over distances nearly 100 times longer than traditional copper cables, allowing for more flexible and distributed data center architectures.
Addressing AI’s computational demands: The OCI chiplet is designed to help data centers cope with the growing computational workload required by the AI market:
- AI’s increasing hunger for computational power and energy is challenging even the most powerful data centers, which are using more processors and energy to meet the demand.
- The optical signaling in the OCI chiplet allows processors to handle more powerful AI models while running with reduced latency and improved efficiency.
Modularity and scalability: The OCI chiplet’s design offers modularity and scalability advantages for data centers:
- In addition to the optical connections, the chiplet is fitted with an integrated circuit to process information, an electrical circuit to handle electric signals, and a path to attach an optical connector that acts like a port for the light signals.
- The chiplet can be expanded to move more data at quicker speeds, enabling more scalability for data transmission between processors and making it easier for data centers to add them to meet demand.
Broader industry context: While Intel is not the only company exploring photonic technology for semiconductors, its solution stands out:
- Companies like Global Foundries and Ayar Labs have also boasted the use of photonic integration to aid their AI architecture and products.
- Intel’s OCI chiplet is expected to be compatible with most processors, even those from other vendors like Nvidia and AMD, ensuring interoperability within the ecosystem.
Next steps and future potential: Intel plans to deliver the OCI chiplet in large volumes for integration with processing decks and is running pilot programs with select customers:
- As more data centers integrate this technology and manufacturing scales up, Intel believes the OCI chiplet can deliver on both cost and manufacturability.
- The chiplet is expected to support bandwidth requirements for generations ahead, as AI’s computational demands continue to grow.
Analyzing deeper: While Intel’s OCI chiplet represents a significant innovation in using optical technology to boost AI processing in data centers, questions remain about the economic viability of such solutions at scale, given the higher costs associated with optical technologies compared to conventional ones. As the AI industry continues to evolve rapidly, it will be crucial to monitor how Intel’s solution fares in the market and whether it can deliver on its promises of improved efficiency, scalability, and interoperability. The success of the OCI chiplet may also depend on how quickly and effectively Intel can scale up production and integrate the technology with its customers’ data centers.
Intel Is Using Lasers To Help Meet AI Demands On Data Centers