Industry News
Fiber-to-Chip Leap: Unleashing CPO & Glass Waveguides
Views : 502
Author : JIUZHOU
Update time : 2025-07-28 09:34:52
Data center networks are undergoing unprecedented changes. The rise of artificial intelligence and the large-scale deployment of its clusters have pushed this process to a climax.
The demand for high-performance computing to support AI training and reasoning tasks is the core driving force behind this change. Industry forecasts say that in the next five years, shipments of AI-optimized graphics processing units will reach millions. By 2028, this number will grow to an enormous scale.
Building such a large AI network requires a large number of high-speed optical transceivers. Analysts expect the yearly demand to reach millions of units. These devices must operate at high speeds of 1.6Tbps and 3.2Tbps.
El análisis muestra que cada acelerador de IA puede tener más de 10 transceptores en el futuro. This means the demand for fiber connections will increase significantly compared to now.
Energy consumption issues have become prominent. In current data centers, a standard pluggable optical transceiver consumes about 20 watts of power. Experts expect the power consumption of the next generation of high-speed transceivers to nearly double. Based on the current shipment estimates, the power required to drive these transceivers alone will reach about 200MW in 2024.
The need for optical connections is growing rapidly. Transceivers are also using more power.
In the future, their yearly power consumption could reach 2GW. This is about the same as the power output of a large nuclear power plant. This does not include the additional power consumption of driving host-side electronic devices, signal retimers and other links.
For a data center using a million-level AI accelerator, new technologies can save about 150 megawatts of power. This greatly lowers the cost of building power generation facilities. It also reduces operating costs. Depending on energy prices in different regions, annual electricity savings can easily exceed 100 million euros.
In China, the "East Data West Computing" project is advancing. This has increased the need for high-bandwidth, low-power optical connections in supercomputing and intelligent computing centers. These technologies are expected to be key in improving the energy efficiency of local computing units. Faced with this exponentially growing energy consumption pressure, technological innovation has become the key to breaking the deadlock.
CPO: A weapon to break through the power consumption bottleneck
CPO is the core technology that is most likely to break through the above energy consumption bottleneck in the short term. The main idea is to take the electro-optical conversion module out of the pluggable transceiver. We will then integrate it directly into the package substrate of the computing unit.
This "light into chip" method reduces the distance for electrical signals. This creates a more energy-efficient optical link. CPO can cut power use by over 50% compared to traditional pluggable solutions. In some cases, it can reduce it by up to 75%.
Its energy-saving benefit comes from reducing high-loss copper channels. It also simplifies or removes digital signal processors. These processors are used to fix long-distance transmission losses of electrical signals.
CPO can greatly improve signal latency. This is because it reduces the pre-processing and post-processing steps needed to send signals through CPO modules. CPO offers fast, low-power, and low-latency optical connections. These features are key to creating advanced AI networks.
Another energy-saving solution worth paying attention to is the linear drive pluggable optical module. LPO cuts power use and delays by eliminating the DSP chip. It keeps the front panel's plug-in design and works with existing systems.
CPO has benefits in signal quality and speed. However, LPO is cheaper and works well for short distances.
Its high cost-effectiveness, low power use, and quick setup may slow down the full use of CPO for a while. When the link rate goes up to 200G per channel or more, the power savings of LPO will decrease. At the same time, it will become much harder to maintain high signal quality. The industry thinks that as technology grows and speeds up, CPO will be a better long-term choice in the future.
Glass substrate: a key material to enable the next generation of CPO
Glass materials are expected to play a vital role in the next generation of CPO technology. To place the electro-optical conversion core near the computing chip, we need new packaging substrate technology. This substrate must support a larger package size. It also needs to connect well with silicon photonic chips.
Traditional semiconductor packaging mainly relies on organic substrates. The thermal expansion coefficient of organic materials is much higher than that of silicon chips. This difference limits the maximum size of the package. As the industry continues to pursue larger packages, organic substrates face reliability challenges and manufacturing difficulties, resulting in rising costs.
Specially made glass substrates can expand in heat like silicon chips. They are much better than regular organic substrates. This gives glass substrates excellent thermal stability, effectively reducing mechanical stress and risk of damage when the temperature changes.
Its excellent mechanical strength and flatness also provide a solid foundation for high-reliability chip packaging. In addition, glass substrates can support higher interconnect density and finer lines, thereby improving electrical performance and reducing parasitic effects.
These characteristics make glass a promising high-reliability, high-precision substrate choice in the field of advanced semiconductor packaging. The semiconductor packaging industry is investing in research and development. They want to make advanced glass substrates the next core technology.
Glass waveguide: a bridge between integrated light and electricity
In addition to excellent thermal and mechanical properties, glass can also form embedded optical waveguide structures through special processes. This process swaps ions in the glass with ions in the solution.
This helps control the refractive index in a specific area. As a result, it creates a channel that can guide and limit optical signals. Its characteristics can be precisely designed for a variety of optical scenarios.
Optical waveguides built into glass are like optical fibers. They can effectively send light in and out of optical fibers or silicon photonic chips. This makes glass an ideal optical interconnect media platform for CPO applications.
Combining electrical and optical interconnects on the same substrate is important. This helps solve the problem of interconnect density in large AI clusters.
Currently, the number of optical channels is limited by the physical size of the optical fiber. Glass waveguide structures can hold more connections. This greatly increases the number of input and output channels in a small area. This is much better than the direct fiber-to-chip connection method.
This high-density integration solves space issues. It also improves the performance and scalability of AI clusters. Glass waveguides are compact. This allows more optical channels to fit in the same space.
As a result, the system's data transmission capacity and efficiency improve greatly. This is critical for the next-generation AI infrastructure that needs to process massive amounts of data.
By using glass waveguides, we can create a full optical subsystem on the same substrate. This allows photonic chips to communicate directly through optical waveguides.
This integration method can reduce signal loss. It also offers higher bandwidth density and better structural durability. This is true for high-density, multi-interconnection systems when compared to discrete optical fibers. These advantages establish the core position of glass waveguides in high-performance optical interconnection systems.
Outlook: Reshaping AI interconnection architecture
Using CPO technology in new data centers and AI supercomputers will improve chip performance. This technology uses glass substrates and waveguides. It will significantly increase the "light escape" bandwidth of the chips. This will pave the way for building switches with rates of 102Tbps and above and higher port density.
Network architects are facing an excellent opportunity to rethink and redesign network topologies. The increase in bandwidth and simpler architecture will improve network performance. This will lead to better efficiency and processes.
Optoelectronic co-packaging technology, combined with breakthroughs in glass substrates and waveguides, is expected to reshape the interconnection architecture of AI from multiple dimensions. It can significantly reduce energy consumption, improve sustainability, and make large-scale AI computing more environmentally friendly and economically feasible.
CPO also makes the system more efficient and scalable. This helps it handle larger and more complex computing tasks easily.
CPO technology solves the problem of interconnection density. This speeds up data transmission rates. It also ensures faster and more reliable communication between AI components. Additionally, it effectively reduces system bottlenecks.
In the future, AI interconnection architecture will use more direct optical links. This may even remove some electrical switching links. This change will greatly increase the bandwidth for AI tasks. This will also improve how we process large data sets.
Glass materials, with their excellent optical transmission properties and scalable manufacturing potential, are the ideal carrier to realize this vision. Glass-based optical interconnection technology will become an indispensable infrastructure for building the next generation of high-performance computing and advanced AI systems.
The demand for high-performance computing to support AI training and reasoning tasks is the core driving force behind this change. Industry forecasts say that in the next five years, shipments of AI-optimized graphics processing units will reach millions. By 2028, this number will grow to an enormous scale.
Building such a large AI network requires a large number of high-speed optical transceivers. Analysts expect the yearly demand to reach millions of units. These devices must operate at high speeds of 1.6Tbps and 3.2Tbps.
El análisis muestra que cada acelerador de IA puede tener más de 10 transceptores en el futuro. This means the demand for fiber connections will increase significantly compared to now.
Energy consumption issues have become prominent. In current data centers, a standard pluggable optical transceiver consumes about 20 watts of power. Experts expect the power consumption of the next generation of high-speed transceivers to nearly double. Based on the current shipment estimates, the power required to drive these transceivers alone will reach about 200MW in 2024.
The need for optical connections is growing rapidly. Transceivers are also using more power.
In the future, their yearly power consumption could reach 2GW. This is about the same as the power output of a large nuclear power plant. This does not include the additional power consumption of driving host-side electronic devices, signal retimers and other links.
For a data center using a million-level AI accelerator, new technologies can save about 150 megawatts of power. This greatly lowers the cost of building power generation facilities. It also reduces operating costs. Depending on energy prices in different regions, annual electricity savings can easily exceed 100 million euros.
In China, the "East Data West Computing" project is advancing. This has increased the need for high-bandwidth, low-power optical connections in supercomputing and intelligent computing centers. These technologies are expected to be key in improving the energy efficiency of local computing units. Faced with this exponentially growing energy consumption pressure, technological innovation has become the key to breaking the deadlock.
CPO: A weapon to break through the power consumption bottleneck
CPO is the core technology that is most likely to break through the above energy consumption bottleneck in the short term. The main idea is to take the electro-optical conversion module out of the pluggable transceiver. We will then integrate it directly into the package substrate of the computing unit.
This "light into chip" method reduces the distance for electrical signals. This creates a more energy-efficient optical link. CPO can cut power use by over 50% compared to traditional pluggable solutions. In some cases, it can reduce it by up to 75%.
Its energy-saving benefit comes from reducing high-loss copper channels. It also simplifies or removes digital signal processors. These processors are used to fix long-distance transmission losses of electrical signals.
CPO can greatly improve signal latency. This is because it reduces the pre-processing and post-processing steps needed to send signals through CPO modules. CPO offers fast, low-power, and low-latency optical connections. These features are key to creating advanced AI networks.
Another energy-saving solution worth paying attention to is the linear drive pluggable optical module. LPO cuts power use and delays by eliminating the DSP chip. It keeps the front panel's plug-in design and works with existing systems.
CPO has benefits in signal quality and speed. However, LPO is cheaper and works well for short distances.
Its high cost-effectiveness, low power use, and quick setup may slow down the full use of CPO for a while. When the link rate goes up to 200G per channel or more, the power savings of LPO will decrease. At the same time, it will become much harder to maintain high signal quality. The industry thinks that as technology grows and speeds up, CPO will be a better long-term choice in the future.
Glass substrate: a key material to enable the next generation of CPO
Glass materials are expected to play a vital role in the next generation of CPO technology. To place the electro-optical conversion core near the computing chip, we need new packaging substrate technology. This substrate must support a larger package size. It also needs to connect well with silicon photonic chips.
Traditional semiconductor packaging mainly relies on organic substrates. The thermal expansion coefficient of organic materials is much higher than that of silicon chips. This difference limits the maximum size of the package. As the industry continues to pursue larger packages, organic substrates face reliability challenges and manufacturing difficulties, resulting in rising costs.
Specially made glass substrates can expand in heat like silicon chips. They are much better than regular organic substrates. This gives glass substrates excellent thermal stability, effectively reducing mechanical stress and risk of damage when the temperature changes.
Its excellent mechanical strength and flatness also provide a solid foundation for high-reliability chip packaging. In addition, glass substrates can support higher interconnect density and finer lines, thereby improving electrical performance and reducing parasitic effects.
These characteristics make glass a promising high-reliability, high-precision substrate choice in the field of advanced semiconductor packaging. The semiconductor packaging industry is investing in research and development. They want to make advanced glass substrates the next core technology.
Glass waveguide: a bridge between integrated light and electricity
In addition to excellent thermal and mechanical properties, glass can also form embedded optical waveguide structures through special processes. This process swaps ions in the glass with ions in the solution.
This helps control the refractive index in a specific area. As a result, it creates a channel that can guide and limit optical signals. Its characteristics can be precisely designed for a variety of optical scenarios.
Optical waveguides built into glass are like optical fibers. They can effectively send light in and out of optical fibers or silicon photonic chips. This makes glass an ideal optical interconnect media platform for CPO applications.
Combining electrical and optical interconnects on the same substrate is important. This helps solve the problem of interconnect density in large AI clusters.
Currently, the number of optical channels is limited by the physical size of the optical fiber. Glass waveguide structures can hold more connections. This greatly increases the number of input and output channels in a small area. This is much better than the direct fiber-to-chip connection method.
This high-density integration solves space issues. It also improves the performance and scalability of AI clusters. Glass waveguides are compact. This allows more optical channels to fit in the same space.
As a result, the system's data transmission capacity and efficiency improve greatly. This is critical for the next-generation AI infrastructure that needs to process massive amounts of data.
By using glass waveguides, we can create a full optical subsystem on the same substrate. This allows photonic chips to communicate directly through optical waveguides.
This integration method can reduce signal loss. It also offers higher bandwidth density and better structural durability. This is true for high-density, multi-interconnection systems when compared to discrete optical fibers. These advantages establish the core position of glass waveguides in high-performance optical interconnection systems.
Outlook: Reshaping AI interconnection architecture
Using CPO technology in new data centers and AI supercomputers will improve chip performance. This technology uses glass substrates and waveguides. It will significantly increase the "light escape" bandwidth of the chips. This will pave the way for building switches with rates of 102Tbps and above and higher port density.
Network architects are facing an excellent opportunity to rethink and redesign network topologies. The increase in bandwidth and simpler architecture will improve network performance. This will lead to better efficiency and processes.
Optoelectronic co-packaging technology, combined with breakthroughs in glass substrates and waveguides, is expected to reshape the interconnection architecture of AI from multiple dimensions. It can significantly reduce energy consumption, improve sustainability, and make large-scale AI computing more environmentally friendly and economically feasible.
CPO also makes the system more efficient and scalable. This helps it handle larger and more complex computing tasks easily.
CPO technology solves the problem of interconnection density. This speeds up data transmission rates. It also ensures faster and more reliable communication between AI components. Additionally, it effectively reduces system bottlenecks.
In the future, AI interconnection architecture will use more direct optical links. This may even remove some electrical switching links. This change will greatly increase the bandwidth for AI tasks. This will also improve how we process large data sets.
Glass materials, with their excellent optical transmission properties and scalable manufacturing potential, are the ideal carrier to realize this vision. Glass-based optical interconnection technology will become an indispensable infrastructure for building the next generation of high-performance computing and advanced AI systems.
Related News
Read More >>
Optical Connectivity: The Engine of AI Scaling
May .08.2026
With the rapid growth of generative AI and large language models, data centers are quickly becoming intelligent computing centers. The growing demand for computing power from AI models is not just about performance.
5G Core & MEC: AI-Driven Growth Through 2030
Apr .17.2026
Global communications infrastructure is currently witnessing a long-awaited "second wave of explosive growth."
6G Outlook: The Future of Next-Gen Wireless Communication
Apr .03.2026
The transition from 5G to 6G is more than just faster internet speeds. 6G will bring new changes to network design, operation, and business models. Wireless communication will become smarter, more efficient, and more energy-efficient.
Ready or Not, Next-Gen Communication Is Here
Mar .20.2026
Standardization efforts for the next generation of mobile communication technologies commenced in 2025. They project commercial deployment to take place around 2030.