Optical transceivers, optical DSPs (oDSPs), and switch ASICs are the core components of data center optical interconnects. The emergence of LPO (Linear-drive Pluggable Optics) and CPO (Co-packaged Optics) is driving the industry toward lower power consumption and higher density.

I. Core Component Functionality Analysis

1. Optical Transceivers: The Bridge for Electro-Optical Conversion

Optical transceivers are critical devices that convert electrical signals to optical signals and vice versa, widely used in data centers and telecom networks. Key functions include:

• Electro-Optical Conversion: Laser diodes modulate electrical signals into optical signals at the transmitter; photodetectors convert optical signals back to electrical signals at the receiver.
• Rate Adaptation: Support multi-rate standards (e.g., QSFP-DD, OSFP) from 100G to 1.6T, covering transmission distances from 50m to 2km.
• Signal Processing: Traditional transceivers rely on DSP chips for equalization, error correction, and dispersion compensation (e.g., DSPs account for ~50% of power consumption in 400G transceivers).

2. Optical DSP (oDSP): The “Brain” of Optical Transceivers

oDSPs are the most valuable electrical chips in transceivers (20%-30% of BOM cost). Key roles include:

• Modulation & Demodulation: In data center scenarios, PAM4 oDSPs use 4-level modulation to boost single-channel rates (e.g., 50G/100G) and compensate for signal distortion. In telecom long-haul scenarios, Coherent oDSPs use QPSK for high-sensitivity transmission.
• Signal Regeneration: Digital signal processing (e.g., FEC) restores degraded signals to ensure long-distance reliability.
• Power Consumption Challenge: In 800G transceivers, oDSPs consume ~6-8W, becoming the primary power source.

3. Switch ASICs: The Central Hub for Data Forwarding

Switch ASICs are the core of switches, responsible for high-speed data frame routing and forwarding. Functions include:

• Port Interconnection: Support multi-port high-speed connections (e.g., 112G SerDes) for low-latency data exchange between servers and storage.
• Protocol Processing: Integrate PAM4 modulation, CDR (Clock and Data Recovery), and flow control to ensure signal integrity.
• Collaborative Optimization: In LPO solutions, ASICs assume partial signal compensation (e.g., linear equalization, clock recovery) previously handled by oDSPs.

II. Technological Breakthroughs and Impacts of LPO & CPO

1. LPO: Cost and Efficiency Gains in Pluggable Architectures

Technical Features:

• DSP-Free Design: Replaces DSPs with high-linearity Driver/TIA chips, eliminating CDR and complex digital processing to reduce power, cost, and latency in 800G LPO modules.
• Compatibility: Retains pluggable formats (QSFP-DD/OSFP) with hot-swappable maintenance, suitable for short-reach (<2km) AI clusters and cost-sensitive scenarios.
• Standardization Progress: Based on OIF CEI-112G-Linear-PAM4, supports partial 800G commercialization, though 224G SerDes requires further validation.

Industry Impact:

• Power Revolution: A rack with 100 400G LPO modules saves >¥2,000 annually in electricity costs (PUE 1.5), with reduced cooling expenses.
• Supply Chain Shift: Reduces reliance on DSP vendors like Marvell/Broadcom, driving localization of Driver/TIA chips.
• Scenario Limitations: Relies on switch ASICs for signal compensation, limiting competitiveness in heterogeneous networks.

2. CPO: Performance Leap in Co-Packaged Architectures

Technical Path:

• Near-Packaging Evolution: From NPO (co-board optical engine) to CPO (co-packaged optical engine), reducing signal transmission distance from 10cm to millimeters, cutting power by 30%-50%.
• Integration Morphologies: Includes Type A (2.5D), Type B (Chiplet), and Type C (3D) packaging, advancing deep integration of silicon photonics with switch ASICs.
• Silicon Photonics Core: CPO relies on silicon photonics for high-density optical device integration, with silicon photonics expected to capture 60% market share by 2030.

Industry Impact:

• Performance Boost: A 1.6T CPO system supports 51.2T total bandwidth with sub-nanosecond latency, meeting AI training clusters’ ultra-high bandwidth demands.
• Ecosystem Challenges: Initially reliant on proprietary designs (e.g., NVIDIA Quantum-X), lacking unified standards, and high O&M costs due to whole-unit replacement for optical engine failures.
• Market Differentiation: CPO targets scale-up networks (e.g., multi-rack AI clusters), while scale-out networks still rely on pluggable modules.

III. Future Trends and Technological Dynamics

1. Multi-Technology Coexistence:

• LPO Dominates Mid-Term: By 2025-2027, LPO will rapidly penetrate AI clusters and mid-sized data centers, with >8M 1.6T LPO ports expected by 2027.
• CPO’s Long-Term Potential: Post-2030, CPO will gradually commercialize in hyperscale data centers, especially for 100T+ scenarios, as silicon photonics matures.
• DSP Persistence: DSP solutions will remain mainstream for long-haul and heterogeneous networks, with power-optimized “Link Optimized-DSP” variants.

2. Synergistic Innovation:

• Silicon Photonics Fusion: Enables LPO (reduces Driver/TIA costs) and CPO (achieves high-density integration), serving as a foundational technology for both.
• Packaging Breakthroughs: 3D packaging and TSV (Through-Silicon Vias) advance CPO to Type C, shrinking size and improving thermal efficiency.
• Standardization: IEEE 802.3 and OIF efforts will accelerate LPO interoperability, while CPO requires open ecosystems to resolve compatibility issues.

3. Supply Chain Restructuring:

• Chip Vendors: Marvell/Broadcom maintain DSP dominance but face LPO competition; NVIDIA/Intel integrate silicon photonics with ASICs via CPO to strengthen system-level competitiveness.
• Transceiver Vendors: Innolight, Eoptolink, etc., actively deploy LPO/CPO but must balance investments with market demand to avoid over-commitment.
• Foundries: TSMC, STMicroelectronics, etc., expand silicon photonics capacity to enable CPO mass production.

IV. Conclusion

The emergence of LPO and CPO marks a pivotal shift from “pluggable-dominated” to “integrated-evolving” optical interconnects. LPO’s low power and ease of deployment make it a mid-term mainstream choice, while CPO’s extreme performance represents the long-term direction. Their competition will drive data center architectures toward greater efficiency and intelligence, offering opportunities and challenges for silicon photonics, advanced packaging, and other underlying technologies. In the future, synergistic development across multiple technological paths will become the norm, with standardization and ecosystem collaboration determining the pace of technology adoption.

In the rapid development of optical communication technology, data centers have an increasingly urgent demand for high-speed, efficient, and low-energy-consumption optical interconnect solutions. As two highly anticipated technical solutions, Co-Packaged Optics (CPO) and Linearly DrivenPluggable Optics (LPO) exhibit their respective characteristics in the field of optical module applications. However, CPO has obvious advantages over LPO in many aspects.

I. Advantages in Technical Principles and Integration

(A) Highly Integrated Architecture

CPO technology encapsulates the network switch chip and optical module within the same slot, achieving a tight integration between the optical engine and switch chip. This highly integrated architecture significantly shortens the distance between the switch chip and optical engine, fundamentally optimizing the electrical signal transmission path. Taking the connection between typical data center switches and optical modules as an example, in traditional methods, signals would undergo attenuation and delay in longer transmission lines. In contrast, CPO technology drastically reduces signal transmission distance, effectively enhancing electrical signal transmission speed, ensuring signal integrity, and minimizing the likelihood of signal distortion and interference.

Conversely, while LPO employs linear drive technology to eliminate the DSP/CDR chip within the optical module, integrating related functions into the switch chip on the equipment side, the optical module and switch chip remain relatively independent, falling far behind CPO in terms of signal transmission integration and optimization.

(B) Efficient Collaborative Operation

CPO technology enables deep collaborative operation between network switch chips and optical modules. Due to their physical proximity, they can achieve more efficient coordination in data processing and transmission. For instance, when handling large-scale data traffic, the switch chip can promptly and accurately send data to the optical module for optical signal conversion and transmission, while the optical module can swiftly respond by converting received optical signals back into electrical signals for the switch chip. This efficient collaborative operation mode enables CPO to excel in handling large volumes of data and high-rate data transmission tasks, effectively boosting the performance of the entire optical communication system.

In comparison, the collaborative operation between LPO’s optical module and switch chip relies on external interfaces and communication protocols, posing certain limitations in terms of data interaction timeliness and efficiency, making it challenging to meet the stringent demands of future ultra-high-speed, large-scale data transmission.

II. Advantages in Performance

(A) Low Power Consumption

With data center energy consumption becoming increasingly prominent, power consumption has become a critical metric for evaluating optical communication technologies. By shortening signal transmission distance and optimizing the integrated architecture, CPO technology significantly reduces energy loss during signal transmission. Relevant research data indicates that optical communication systems utilizing CPO technology can reduce power consumption by 30% to 50% compared to traditional solutions at the same data transmission rate. For large-scale data centers, this translates to substantial annual savings in electricity costs and contributes to achieving green energy-saving goals.

While LPO reduces some power consumption by eliminating the DSP chip, due to its separated design of the optical module and switch chip, it cannot match CPO in overall power consumption control. Especially in scenarios involving long-distance, high-rate data transmission, LPO’s power consumption disadvantage becomes even more apparent.

(B) High Transmission Rate and Stability

Leveraging its unique architecture and collaborative operation mechanism, CPO technology can support higher transmission rates. Currently, CPO technology has achieved high-speed data transmission of 800G and even 1.6T, demonstrating exceptional transmission stability. In complex network environments, CPO can effectively resist external interference, ensuring the accuracy and continuity of data transmission while reducing the bit error rate. For example, in AI data centers, where massive training data requires high-speed, stable transmission, CPO technology can meet this demand, ensuring efficient AI model training.

LPO’s removal of the DSP chip leads to an increased system bit error rate, posing bottlenecks in transmission distance and rate enhancements. It is more suitable for short-distance scenarios with relatively low rate requirements, falling short of CPO in terms of stability for long-distance, ultra-high-speed data transmission.

III. Advantages in Cost and Maintenance

(A) Long-term Cost Advantages

In the long run, CPO technology offers significant cost advantages. Although initial research and deployment costs for CPO are relatively high, as technology matures and scales up for production, costs will gradually decrease. CPO’s highly integrated nature reduces the number of interfaces and external cable connections between optical modules and switch chips, lowering hardware and cabling costs. Additionally, due to CPO’s low power consumption, long-term electricity cost savings are considerable.

While LPO reduces optical module procurement costs by eliminating DSP chip material costs, in terms of overall equipment costs, the increased performance requirements for switch chips on the equipment side, along with cabling costs, diminish its overall cost advantage in the long term.

(B) Maintenance Convenience and Reliability

In terms of maintenance, CPO technology exhibits high reliability. Its high integration reduces external connection points and potential failure points, lowering the probability of faults. In the event of a failure, CPO technology can quickly locate the issue through advanced monitoring and diagnostic techniques, enhancing maintenance efficiency. For instance, after a large data center adopted CPO technology, equipment failure rates dropped by 30%, and maintenance time was reduced by 50%.

While LPO supports hot-swapping, facilitating maintenance of individual optical modules, its relatively loose system architecture necessitates considerations around compatibility and collaborative operation between optical modules and switch chips during overall system maintenance, increasing maintenance difficulty and complexity.

IV. Current Development Status, Challenges, and Feasibility Comparison for Future Phases

(A) LPO Development Status, Challenges, and Future Feasibility

1.Development Status

At the technical level, LPO has made progress in the field of 800G optical modules. Taking 800gbps rates as an example, its power consumption can be reduced to approximately 8W, marking a substantial 40-45% reduction compared to traditional pluggable optical modules. In market applications, some companies have already begun layouts. For instance, at OFC 2023, Arista was the first to share test results for a 51.2T switch equipped with LPO transceivers, and NVIDIA is poised to be the first to deploy 200G per channel LPO transceivers in its AI clusters by 2025. Many industry-leading clients, such as Meta, are actively considering adopting LPO technology.

2.Challenges

LPO requires specific modules to pair with ASIC switch chips, which diminishes the versatility advantage of pluggable modules. Additionally, link performance and responsibility delineation are challenging, testing is complex, and there currently lacks unified electrical and optical standards, making interoperability between modules from different vendors difficult to guarantee. Furthermore, eliminating the DSP chip reduces power consumption and cost but increases the system bit error rate, posing bottlenecks in transmission distance and rate enhancements.

3.Future Feasibility

Despite challenges, LPO has emerged as a viable alternative to CPO in the 100G per channel interconnect field. By 2029, LPO’s penetration rates in 800G, 1.6T, and 3.2T ports are projected to be 3%, 33%, and 15%, respectively. If NVIDIA adopts LPO for NVLink expansion in its next-generation GPUs, it will significantly boost market demand for 1.6T-DR8 LPO modules, potentially adding over 8 million 1.6T LPO ports by 2027. With continuous technological improvements and the gradual establishment of industry standards, LPO has substantial growth potential in short-distance, cost-sensitive scenarios with moderate rate requirements.

(B) CPO Development Status, Challenges, and Future Feasibility

1. Current Status

Technologically, CPO has achieved high-speed data transmission of 800G and even 1.6T, and it excels in reducing power consumption. Broadcom and Cisco’s CPO products have achieved low power consumption of approximately 7pJ/bit, while CPO solutions from IBM, Coherent, and other companies have further reduced power consumption. In terms of industrial layout, Broadcom announced its next-generation switching ASIC product line equipped with CPO as early as 2021, and its product Bailly has been successfully put into production. Cisco showcased a CPO prototype at OFC2023. Intel placed its Silicon Photonics Group under the Data Center and AI (DCAI) group to fully develop optical engines based on silicon photonics. Additionally, numerous companies such as Ranovus, Marvell, Nubis Communications, Lightmatter, IBM, and others have joined the ranks of CPO technology research and development.

2. Challenges

From a technical perspective, the wiring process of connecting optical fibers from ASICs to the front panel of the PCB is challenging, and the overall testing and optimization after system integration are also complex. In terms of reliability, CPO technology integrates multiple optical modules with chips, leading to an increase in failure rates. For instance, Broadcom’s CPO lab product, which integrates 16 optical modules on a single board, exhibits a failure rate that is more than ten times higher, reducing reliability and stability to one-tenth of their original levels. From a market perspective, CPO research and development require significant financial and human resources, and the production process is complex. Cost reduction requires time and scale effect accumulation, making it difficult to compete with optical modules in the short term. Furthermore, several major CSP cloud vendors in North America have limited interest in CPO, and some overseas companies have even scaled down their research and development teams. Additionally, CPO mass production faces numerous technical challenges, with 3.2T as the generational standard. Mass production is expected no earlier than 2027, with Broadcom currently progressing the fastest, and its first-generation CPO engineering test scheduled for 2025.

3. Late-Stage Feasibility

According to the “2025 China CPO Industry Market Development Trends and Industrial Demand Forecast Report” published by Beijing Zhiyan Kexin Consulting Co., Ltd., the commercialization of CPO technology is expected to start with 800G and 1.6T ports, with commercial use beginning in 2024-2025 and subsequent scaled growth in 2026-2027. By 2033, the global CPO market size is projected to reach $2.6 billion. With the continuous maturation of technology, gradual reduction in costs, and increasing market demand for high-speed, low-power optical communication solutions, CPO is expected to occupy an important position in the future field of optical communications, particularly in data center scenarios with extremely high requirements for transmission rates and power consumption. Market news also indicates that NVIDIA may introduce a new CPO switch at the GTC conference in March 2025. The supply chain reveals that this CPO switch is in the trial production stage, and if progress is smooth, mass production can be achieved in August of this year. These developments demonstrate the potential for the future development of CPO technology.

In summary, CPO technology has significant advantages over LPO in terms of technical principles, performance, cost, and maintenance. Although both CPO and LPO face their own challenges during development (with CPO’s commercialization appearing more difficult relative to LPO), in the long run, with continuous technological breakthroughs and the gradual maturation of the market, CPO, with its unique advantages, is more likely to dominate the future field of optical communications and lead industry trends.