Energy Efficiency in Co-Packaged Optics

はじめに

As data rates continue to surge past 800G and into multi-terabit speeds, energy efficiency is becoming a critical concern for network operators, hyperscalers, and AI computing environments. A recent study by Resolute Photonics highlights the dramatic differences in energy consumption per bit across different optical interconnect architectures. Traditional Front Plate Pluggable (FPP) Optics are increasingly challenged to meet the demands for higher bandwidth and energy efficiency. This has led to the development of Near Packaged Optics (NPO) and Co-Packaged Optics (CPO), which offer promising solutions to these challenges.

 

Transition from FPP to NPO and CPO

FPP optics involve connecting switch Application-Specific Integrated Circuits (ASICs) to front-panel pluggable optical modules via electrical traces on the printed circuit board (PCB). As data rates increase, these longer electrical paths result in higher power consumption and signal integrity issues.

NPO addresses some of these challenges by placing optical modules closer to the switch ASICs, reducing the length of electrical traces and thereby improving signal integrity and reducing power consumption. However, NPO still relies on separate optical modules, which can limit the potential for further integration and efficiency gains.

CPO takes integration a step further by placing optical engines near, or even within, the same package as the switch ASICs. This approach minimizes the electrical path length between the ASIC and the optical components, significantly reducing power consumption and improving performance. CPO represents a more integrated solution compared to NPO, offering greater potential for energy efficiency and performance improvements.

The energy efficiency of these interconnect technologies varies significantly. A clear trend emerges when analysing energy consumption per bit (picojoules per bit – pJ/bit) across these interconnect technologies

 

Energy Efficiency Comparison

• FPP Optics: In FPP architecture, modules are inserted into switch faceplates, requiring long electrical traces to the chip. the longer electrical traces between the switch ASIC and the front-panel pluggable optics necessitate high-power SerDes (Serializer/Deserializer) to maintain signal integrity. This configuration can result in substantial power consumption, especially as data rates increase. The average power consumption for FPP Optics is about 20 pJ/bit.
• NPO: By positioning optical modules closer to the switch ASICs, NPO reduces the length of electrical traces, leading to improved signal integrity and moderately lower power consumption compared to FPP. However, since NPO still relies on separate optical modules, the potential for energy efficiency improvements is limited compared to more integrated approaches. NPO is seen as a steppingstone between FPP and full CPO adoption.
• CPO: CPO architecture achieves significant power savings by integrating optical engines directly with the switch ASICs. This close integration eliminates electrical traces, reducing the need for power-hungry SerDes. Early implementations of CPO have demonstrated significant power consumption reductions down to less than 5 pJ per bit, which is up to 4 times the energy efficiency over pluggable optics. This is essential for future 1.6Tbps, 3.2Tbps, and beyond networking architecture.

As data center architectures push towards 51.2TB switching speeds, the industry faces mounting power consumption challenges. Traditional pluggable optics continue to increase power demands, making energy efficiency a critical concern. A recent study comparing 4x 800G transceivers to a SiPh CPO chiplet highlights the significant power savings enabled by Co-Packaged Optics (CPO). A pie chart breakdown of power consumption at 51.2TB speeds further reinforces why CPO is the future of high-speed networking.

At 51.2TB, different system components contribute to overall power consumption. The major categories include:

• ASIC Power – The primary computational load of the switch.
• CPU, Timing & Miscellaneous – Supporting processing functions.
• Optics – Power required for signal transmission.
• Power Delivery – The infrastructure needed to distribute power.
• Fan Power – Cooling system energy demands.

CPO drastically reduces the optics power footprint, contributing to an overall system-wide power reduction of 25%-30%.

 

Implications for Hyperscalers and AI Clusters

Improved energy efficiency in optical interconnects has significant implications for hyperscale data centers and AI clusters. Enhanced energy efficiency allows for higher bandwidth densities, enabling data centers to scale their operations to meet increasing data demands without proportionally increasing power consumption and heat generation. Reduced power consumption leads to lower operational costs, including savings on energy bills and cooling infrastructure.

 

SENKO’s Role in Enabling Energy-Efficient Optical Interconnects

SENKO Advanced Components has been instrumental in developing innovative solutions to enhance the energy efficiency of optical interconnects. The company focuses on creating advanced optical connectors and components that facilitate efficient data transmission while minimizing power loss. SENKO’s commitment to research and development in optical technologies contributes to the broader adoption of energy-efficient interconnect solutions in data centers and high-performance computing environments.

 

結論

The evolution from FPP to NPO and CPO represents a significant advancement in optical interconnect technology, with CPO offering substantial improvements in energy efficiency. These developments are crucial for hyperscalers and AI clusters aiming to enhance performance, scalability, and sustainability in an era of exponential data growth.