The rise of Co-Packaged Optics (CPO)
Over the past decade, the capacity of data center Ethernet switches has surged from 0.64 Tbps to 25.6 Tbps, driven by the adoption of 64x 400 Gbps or 32x 800 Gbps pluggable optical transceiver modules. However, these high-speed modules, within their current form factors, pose significant challenges. Issues include the required densities of electrical and optical connectors, as well as escalating power consumption.
To achieve next-generation optical engines supporting 800 Gbps and beyond per module, the communication rate must double to at least 100 Gbps per lane. This increase introduces substantial signal integrity issues across the switch socket, motherboard, and edge connector, leading to heightened power dissipation at SerDes interfaces. In future Ethernet switching, these signal integrity problems may cause I/O power consumption to exceed that of the switch core. Additionally, the integration density of standard pluggable modules is limited by the QSFP/OSFP form factor, necessitating advanced thermal management solutions not yet widely available.
Co-Packaged Optics (CPO) presents a promising solution to these challenges. Unlike traditional pluggable models, CPO integrates optical modules directly onto the switch ASIC substrate, reducing electrical reach and effectively addressing signal integrity issues. This approach has gained traction among major data centers. However, optimizing the packaging strategy for CPO remains a topic of ongoing industry discussion and development. IDTechEx’s latest report, “Co-Packaged Optics 2025-2035: Technologies, Market, and Forecasts,” explores these advancements in CPO technology and packaging techniques enabling its adoption.
Key trend of optical transceiver in high-end data center. Source: IDTechEx
The importance of advanced semiconductor packaging technologies for Co-Packaged Optics (CPO)
The integration of CPO in data centers aims to boost I/O bandwidth and reduce energy consumption. The way photonic integrated circuits (PICs) are combined with electronic integrated circuits (EICs) and switch ICs can significantly influence the areal and edge bandwidth density, as well as packaging parasitics. These factors directly affect the transceiver’s I/O bandwidth and energy efficiency, meaning improper integration can negate the advantages of silicon photonics.
For CPO, the integration of photonic and electronic components can be achieved through various methods, each with unique advantages and challenges. The most advanced and still in the R&D phase is the 3D monolithic integration embeds photonic components within an existing electronic process node with minimal alterations, co-locating active photonics and driving electronics within the same die. This reduces parasitics and simplifies packaging by eliminating the need for interface pads and bumps. However, monolithic integration typically uses older CMOS nodes, resulting in suboptimal photonic performance and higher energy consumption. Despite these limitations, it offers minimal impedance mismatch and simplified packaging.
Conversely, 2D integration places the PIC and EIC side by side on a PCB, connected by wire bonds or flip-chip. This method is straightforward and cost-effective, but introduces significant parasitic inductance, limiting aggregate I/O due to single-edge connections. While 2D integration is easy to package, the reliance on wirebonds constrains the transceiver bandwidth and increases energy consumption, making it less efficient for high-performance applications.
3D hybrid integration offers a more advanced solution by placing the EIC on top of the PIC, via various advanced semiconductor packaging technologies, including Through-Si-Via (TSV), high density fan-out, Cu-Cu hybrid bonding, active photonic interposer, etc, significantly reducing parasitics. The use of advanced semiconductor packaging technologies in 3D integration allows for dense pitch capabilities, enhancing performance. However, thermal dissipation remains a challenge, as the heat generated by the EIC can impact the PIC, necessitating advanced thermal management solutions. Despite these thermal challenges, 3D hybrid integration achieves higher performance due to minimized packaging parasitics.
2.5D integration serves as a compromise, with both the EIC and PIC flip-chipped onto a passive interposer with TSVs. This approach maintains manageable parasitics and dense pitch capabilities similar to 3D integration but adds complexity with the need for interposer traces. While 2.5D integration balances performance, cost, and fabrication turnaround, it incurs higher parasitics than 3D hybrid integration.
In summary, each integration method presents trade-offs between performance, complexity, and cost, guiding the choice based on specific application requirements and constraints.
Co-Packaged Optics (CPO) Market trajectory
According to IDTechEx, the Co-Packaged Optics (CPO) market is projected to exceed $1.2 billion by 2035, growing at a robust CAGR of 28.9% from 2025 to 2035. CPO network switches are expected to dominate revenue generation, driven by each switch potentially incorporating up to 16 CPO PICs. Optical interconnects for AI system will constitute approximately 20% of the market, with each AI accelerator typically utilizing one optical interconnect PIC to meet increasing demands for high-speed data processing and communication in advanced computing applications.
Total CPO market growth 2025 vs 2035. Source: IDTechEx
IDTechEx’s latest report, titled “Co-Packaged Optics (CPO) 2025-2035: Technologies, Market, and Forecasts,” offers an extensive exploration into the latest advancements within co-packaged optics technology. The report delves deep into key technical innovations and packaging trends, providing a comprehensive analysis of the entire value chain. It thoroughly evaluates the activities of major industry players and delivers detailed market forecasts, projecting how the adoption of CPO will reshape the landscape of future data center architecture.
Source: idtechex.com