Glass Substrate, Catch Up At Full Speed

In AI data centers, thousands of chips need to continuously exchange data. From the inside of chips to between packages, and then between servers and cabinets, data transmission speeds keep increasing, making interconnection technology a key focus of the industry.

In the past few years, the industry has formed a relatively clear judgment: as transmission rates continue to rise, copper interconnection faces growing pressure on bandwidth and power consumption, optical interconnection will become an important development direction, and co-packaged optics (CPO) is expected to further shorten the distance between optical devices and chips.

However, outside this technical route, another seemingly less noticeable technology may be catching up rapidly.

This is the glass substrate.

Compared with CPO, glass substrates do not have such strong technical topic appeal, but their industrialization tasks are relatively specific: how to manufacture micro through-holes on glass, how to fill copper, how to control the warpage of large-size substrates, and how to integrate these processes into a stable production flow.

If these problems can be solved step by step, glass substrates will have the opportunity to take the lead in entering the next-generation AI chip packaging system. At least from the perspective of technological advancement, it may achieve substantial breakthroughs earlier than CPO.

Two Distinct Sets of Technical Challenges

Both CPO and glass substrates face engineering challenges, but their difficulties are not the same.

CPO does not lack basic technologies. Key components such as lasers, modulators, waveguides, and fiber optic connectors already exist. What is truly complex is how to integrate these components into a system suitable for large-scale deployment.

For example, how to configure the light source, what connection method to use for optical fibers, how to dissipate heat from the optical engine, whether it is convenient to repair or replace after a failure, and how devices from different manufacturers can work together. These issues will affect the overall design of CPO, and are also related to product reliability, cost, and actual deployment efficiency.

For system solutions that are not yet fully mature, the industrial chain often needs repeated verification and coordination. Progress in a single technology does not mean that the entire system can immediately enter the large-scale commercial stage.

The situation of glass substrates is different.

Its basic structure and application objectives are relatively clear, and the main problem the industry needs to solve is manufacturing. How to process a large number of micro through-holes on glass, how to avoid cracks, how to achieve reliable copper filling, and how to control the deformation of the substrate during processing and heating, all have relatively clear technical directions.

This does not mean that mass production of glass substrates is easier. In fact, problems in any link may affect the yield of the entire substrate. However, compared with the system architecture and industrial chain coordination problems that CPO needs to face, some technical difficulties of glass substrates are easier to break down into specific process tasks, and can be gradually promoted through continuous testing and manufacturing verification.

One requires the industrial chain to continuously refine the system solution, and the other requires the manufacturing process to be gradually mature. The industrialization pace of the two may therefore differ.

AI Chips Are Getting Larger, Bringing Opportunities for Glass Substrates

The growing attention to glass substrates is backed by the trend of expanding AI chip packaging sizes.

As the performance of AI processors such as GPUs continues to improve, more and more HBMs need to be connected around a single chip. To meet bandwidth and capacity requirements, the packaging design must not only consider the processor itself, but also reserve space for surrounding memories, interconnection structures and other components.

This puts higher requirements on packaging substrates.

At present, traditional packaging substrates mainly use organic material systems. As the packaging area increases, warpage control of the substrate becomes more difficult. During heat treatment and subsequent packaging processes, material deformation may affect line accuracy, chip assembly, and solder joint reliability.

This is where the advantages of glass lie.

Compared with traditional organic substrates, glass has better dimensional stability and flatness, and its thermal expansion properties can also be designed according to specific material systems. At the same time, the glass dielectric has good high-frequency electrical properties, which is conducive to high-speed signal transmission; panel-based processing provides another possibility for manufacturing large-size substrates.

These characteristics are practically attractive for packaging solutions that need to carry both large processors and multiple sets of HBMs at the same time.

However, glass does not come without costs. Its brittleness, the difficulty of micro through-hole processing, and subsequent metallization processes may all increase manufacturing complexity. Whether glass substrates can truly replace some traditional organic substrates ultimately depends on whether manufacturing cost, yield and reliability meet product requirements.

Therefore, the focus of the industry is no longer just whether glass has advantages, but whether these advantages can be transformed into mass-producible products.

Through-Holes, Warpage and Copper Filling, Several Barriers Before Mass Production

The manufacturing of glass substrates is not as simple as cutting glass to the required size and then making circuits on the surface. To make it a real packaging substrate, the first problem to solve is the TGV (Through Glass Via) problem.

These through-holes are responsible for connecting the circuits on both sides of the glass, with the aperture usually on the order of tens of micrometers, and the specific size depends on the product design. During processing, it is necessary to control the aperture and position accuracy, and avoid cracks or other defects in the glass.

One process route is to use laser to modify the local part of the glass, and then remove the modified area through chemical etching to form the required through-holes. This method can be used for micro-structure processing, but the processing quality, efficiency and subsequent metallization still need further optimization.

The importance of TGV is not only whether it can penetrate the glass. The size, position and quality of the through-holes will also affect copper filling, circuit design and the reliability of the final package. Even if breakthroughs are made in through-hole processing, it is only one link in the glass substrate manufacturing chain.

Another problem is warpage.

The fact that glass has good flatness does not mean that large-size glass will not deform under all processing conditions. As the substrate area expands and the number of circuits and metal structures increases, thermo-mechanical differences between materials may still cause warpage.

It is understood that Samsung Electro-Mechanics is considering using thicker glass to deal with this problem, and the relevant solution involves a thickness of about 2 mm.

Glass thickness seems to be just a parameter, but in fact it affects the entire subsequent process. The thickness determines the depth of the through-hole, which in turn affects processing time and copper filling difficulty, and also has an impact on the substrate structure and wiring design.

Therefore, determining the glass thickness is not just a matter of material selection, but an important step to connect glass processing, through-hole manufacturing, metallization and packaging design. However, the determination of parameters does not mean that the mass production problems have been solved, and the stability of the entire manufacturing process still needs to be verified subsequently.

In contrast, copper filling is still one of the links that needs to be focused on.

After the through glass vias are formed, copper needs to be filled to establish a reliable electrical connection. The bottom-up filling method is relatively reliable, but the speed may be slow; the method of expanding from the middle to both sides is expected to improve efficiency, but it may bring defects such as internal voids.

Manufacturers need to find a balance between filling quality, production efficiency, cost and reliability.

Relevant materials also mention that the industry is exploring alternative solutions such as adjusting filling specifications and using heat treatment to make copper powder form conductive structures. Whether these routes can be used in actual mass production still depends on subsequent verification, and judgments cannot be made solely based on theoretical efficiency advantages.

From the current technical path, the problems of glass substrates have gradually become specific. Each link, including how to process through-holes, how to control substrate warpage, and how to reliably fill copper, has clear process objectives, but it still takes time to combine them into a stable production process.

Why Is South Korea Gaining Attention?

Another observation clue for the glass substrate industry is the competition between South Korea, Chinese Taiwan and Chinese Mainland.

Chinese Taiwan and Chinese Mainland have a strong display industry foundation, and relevant enterprises have accumulated rich experience in large-size glass processing for a long time. From this perspective, the two regions naturally have potential advantages in developing glass substrates.

However, display panels and semiconductor packaging substrates are not the same manufacturing tasks.

Display panel processing emphasizes large-area manufacturing capabilities, while semiconductor packaging also requires high-density interconnection in a limited space, which puts forward higher requirements for the position accuracy of through-holes, circuit processing and metal filling quality. Being able to process large-size glass does not mean that it can directly meet all manufacturing requirements of semiconductor packaging.

What is really important is whether enterprises can combine glass processing with links such as laser, etching, electroplating and subsequent circuit manufacturing, and stably control quality in repeated production.

According to industry observations, South Korean enterprises are continuously accumulating experience in relevant links, involving substrate manufacturers, enterprises related to laser and etching, and electroplating chemical suppliers. Since TGV requires the coordination of multiple process links, the cooperation between industrial chains has also become part of the competition.

Analysts believe that South Korean enterprises may have a slight lead over Chinese Taiwan and Chinese Mainland in TGV-related manufacturing experience.

However, this judgment still needs more specific data to support. The process routes, product specifications and verification stages adopted by various enterprises are not exactly the same, and the progress of a single process link cannot directly represent the leading level of the entire industrial chain.

For glass substrates, the final competition is not who announces the successful R&D first, but who can integrate through-hole processing, copper filling, warpage control and circuit manufacturing, and achieve repeatable production while meeting product reliability requirements.

This is also why the current technological progress is worthy of attention, but cannot be simply equated with mass production capacity.

Will Glass Substrates Be Implemented Earlier Than CPO?

Back to the original question: why may glass substrates achieve industrialization breakthroughs earlier than CPO?

The key is not which technology is more advanced, but that the industrialization thresholds they need to cross are different.

The long-term value of CPO remains clear. As AI data centers have increasingly higher requirements for bandwidth, power consumption and interconnection density, optical interconnection is expected to play a role in more scenarios. However, from individual devices to system applications, CPO still needs to solve problems such as light source, connection, heat dissipation, testing, maintenance and supply chain coordination.

Glass substrates have a relatively clear advancement route. The packaging area of AI chips continues to expand, and the warpage and size limitations faced by traditional organic substrates are increasingly attracting attention, which provides clear application demand for glass. Next, the industry needs to continuously improve through-hole processing, copper filling and substrate stability, and verify whether these processes can meet mass production requirements.

If relevant manufacturing links can gradually mature, glass substrates will have the opportunity to take the lead in entering the new generation of AI chip packaging.

This does not mean that CPO will be replaced by glass substrates. The two solve different problems, and they may also appear in the same advanced packaging system in the future: the glass substrate is responsible for carrying chips and realizing high-density interconnection, while optical interconnection undertakes the task of higher-speed data transmission.

For the industrial chain, what is worthy of attention is not who will win between glass and light, but whether the progress of the two technical routes can meet expectations, and at what pace they will enter practical applications.

Compared with the picture of light beams transmitting between chips, a piece of glass covered with tiny copper through-holes is not so eye-catching. But today, when AI chip packaging is becoming more and more complex, what determines whether a technology can be implemented is often not how new the concept is, but whether the manufacturing process can truly mature.

CPO represents the future direction of interconnection technology, while glass substrates may take the lead in solving the immediate problems of large AI chip packaging. If this judgment is finally verified, the progress of the next generation of AI chips may start from a seemingly ordinary piece of glass.

This article is from the WeChat Official Account "Semiconductor Industry Observer" (ID: icbank), written by the editorial department, and published with authorization from 36Kr.