AI chip packaging is transitioning from silicon and organic materials to glass, ceramics, and M8/M9-level PCBs, but these new materials are generally hard and brittle, making them difficult to process.Author and source: Huatai Securities research report
AI chips are growing larger, pushing packaging materials and processing equipment into the spotlight.
On June 8, Yang Yunxiao and others from Huatai Securities’ Machinery Team wrote in their research report: “The rapidly growing demand for AI computing chips, combined with shortages in traditional packaging materials, is accelerating the shift toward new materials such as glass, ceramics, and M8/M9-grade PCBs for advanced packaging.”
The key lies in the second half: glass, ceramics, and M8/M9 PCBs are all difficult to process. Traditional mechanical drilling often causes chipping and cracking; wet etching has limited efficiency and control over morphology; and conventional lasers tend to induce thermal damage. This is precisely where ultrafast lasers deliver their value.
Traditional mechanical drilling, wet etching, and conventional laser processing yield poor results, while ultrafast lasers, with their cold processing characteristics, have become the solution for precision machining.
The institution’s estimation framework includes potential applications such as glass interposers, glass substrates, M9 materials, and optical module substrates, projecting a long-term market space exceeding 100 billion yuan. However, this market potential is not linearly realized and depends on whether CoPoS, CoWoP, glass substrates, and M9 PCB technologies can enter mass production.
The larger the AI chip, the sooner the packaging materials hit their limits.
The pressure on advanced packaging originates first from the chip itself.
During NVIDIA's product iterations, the number of chips integrated and HBM configurations have continuously increased. Data shows that the GP100 features 4 HBM modules with a capacity of 16 GB and 5 integrated chips, while the GB100 has reached 8 HBM modules with a capacity of 192 GB and 10 integrated chips.
As the chip size increases, the package area also increases, making heat dissipation, warping, and signal transmission more challenging.
The current mainstream CoWoS routes are categorized into three types: S, R, and L. CoWoS-S offers the best performance but at the highest cost, with a maximum packaging size of approximately 2,700 square millimeters; CoWoS-R is more cost-effective, but controlling warpage in large-scale packaging remains challenging; CoWoS-L strikes a balance between cost and performance, yet still faces limitations in maximum size, thermal dissipation, and reliability.
The next steps primarily depend on CoPoS and CoWoP.
CoPoS stands for "Cubic Packaging on Substrate." It involves cutting encapsulation substrates from circular wafers into square panels to improve area utilization, while replacing silicon or organic interposers with glass.
CoWoP takes a more direct approach: eliminating the ABF substrate and directly bonding the silicon interposer to the PCB. This shortens the interconnect path but imposes higher requirements on the PCB, including narrower line widths and spacings, void-free via filling, and low-thermal-expansion materials.
According to TrendForce and China Television News, TSMC plans to establish a CoPoS mass production line at its Chiayi facility, with equipment delivery for the pilot line set to begin in February 2026. At TSMC’s annual shareholders’ meeting on June 4, Chairman and CEO Dr. C.C. Wei mentioned that pilot production lines for CoPoS and glass substrates have already been completed, with large-scale mass production expected within two to three years. According to IT Home, NVIDIA aims to achieve CoWoP mass production on Rubin Ultra.
Material shortages are also pushing the industry forward.
High-performance ABF substrates for CoWoS, with core materials including ABF film and T-Glass specialty low-dielectric glass fabric. T-Glass is supplied almost exclusively by Nittobo of Japan, and current production capacity is fully utilized.
Meanwhile, NVIDIA’s next-generation Rubin high-end GPUs require advanced ABF substrates for packaging, necessitating early inventory buildup, which further exacerbates supply constraints.
This is one of the reasons why glass-based materials have been brought back to the forefront.
Glass, ceramic, and M9 are not the same layer material and cannot be simply substituted for one another.
The market often compares glass, ceramics, and M9 materials together, but they don't fully compete at the same level.
Huatai Securities stated directly: "Glass substrates are primarily used in interposers and substrates for advanced packaging and do not conflict with ceramic materials or M9 used in PCB applications."
The traditional CoWoS structure, from top to bottom, consists of the chip, interposer, substrate, and PCB layer. Glass is primarily used for the interposer and substrate. Ceramic substrates and M9 materials are more commonly used for PCB-related materials or high-power thermal dissipation scenarios.
The advantages of glass include dimensional stability, a flat surface, low high-frequency loss, and the ability to create TGV (through-glass vias). Data shows that glass has a dielectric constant ranging from 3.5 to 10, a CTE adjustable between 2.7 and 12.4 ppm, and surface flatness achievable below 4 nm.
The upgrade path for PCB is M8, M9, and even M10.
The higher the number, the lower the signal transmission loss, the faster the speed, and the higher the stability. The core of M8/M9-grade PCBs consists of a lower dielectric constant (Dk), lower dissipation factor (Df), a high-modulus low-expansion fiberglass system, and ultra-low-profile copper foil.
The challenge with M9 lies in its harder materials. It incorporates quartz fabric, Q-glass, as a reinforcing material. High-purity quartz fiberglass has a Mohs hardness exceeding 7, higher than the 5 to 6 range of traditional E-glass fiberglass.
Ceramic emphasizes heat dissipation and thermal expansion matching.
Data shows that ABF materials have a thermal conductivity of 0.8 to 1.2 W/mK, while ceramic substrates can reach up to 200 W/mK. Aluminum nitride has a thermal conductivity of 170 to 230 W/mK, silicon nitride ranges from 60 to 90 W/mK, and sintered silicon carbide is approximately 100 to 250 W/mK.
This explains why three types of materials may be simultaneously of interest: glass for interposers and substrates, M8/M9 for high-speed interconnects, and ceramics for high-power heat dissipation.
Why "scalpel": Ultrafast lasers minimize thermal damage to the lowest possible level
The key point of ultrafast lasers is not "higher power," but "shorter duration."
Huatai Securities defines ultrafast lasers as those with pulse durations in the picosecond (10⁻¹² s) to femtosecond (10⁻¹⁵ s) range.
In such a short time, energy arrives quickly and departs just as fast—the material doesn’t have time to dissipate the heat, yet removal is already complete. This mechanism is called "cold ablation."
This is different from traditional long-pulse lasers.
Traditional lasers more closely "burn" the material, resulting in a large heat-affected zone that can cause chipping, microcracks, melting, and carbonization.
Ultrafast lasers more closely "peel" materials. They act directly on the electron level of the material surface through nonlinear effects such as multiphoton absorption, reducing heat diffusion.
The report states: "Ultrafast lasers are not merely an incremental parameter upgrade over traditional lasers, but a fundamental transformation in the processing mechanism."
In practical processing, this difference results in three outcomes: a smaller heat-affected zone, adjustable hole wall taper, and greater flexibility in shaping.
Analysts noted that laser drilling can process micro-holes of any shape, a feat difficult to achieve with mechanical drilling.
The three processes that best reflect demand are: TGV, M9 microvias, and ceramic etching.
The first one is a glass substrate with TGV.
The key processes for glass substrates include six steps: TGV laser modification, via etching, AOI optical inspection, seed layer deposition, electroplating to fill vias, and polishing.
Among these, "TGV laser modification is the first and most critical process."
The reason is simple: glass lacks the ability to undergo plastic deformation. Excessive or uneven energy can easily cause microcracks, thermal stress concentration, or internal defects. When the aperture is below 30 micrometers, control over the heat-affected zone becomes even more stringent.
Ultrashort pulsed ultrafast lasers can enable non-thermal modification inside glass, reducing thermal stress cracking and edge chipping.
The second process is M9-level PCB microvia drilling.
The M9 is designed for ultra-high-speed transmission environments exceeding 1.6 Tbps. To reduce signal loss, it incorporates high-purity quartz fiber, which also increases manufacturing complexity.
Data shows that the lifespan of traditional mechanical drill bits drops to one-fifth that of conventional materials, and hole diameter accuracy and positional tolerance also deteriorate.
The issue with CO₂ lasers is that they produce an excessively large heat-affected zone, which can lead to resin carbonization and fiberglass tearing. Nanosecond UV lasers have low removal efficiency for high-hardness quartz fiberglass and yield suboptimal hole wall quality.
The value of ultrafast lasers lies in their ability to precisely remove high-hardness fiberglass without carbonizing the resin or short-circuiting the copper layers.
The third step is precision processing of ceramic substrates.
Ceramic materials such as aluminum nitride, silicon nitride, and silicon carbide exhibit high hardness and strong thermal conductivity. Data shows that these materials can achieve a Mohs hardness of 7 to 9 and a thermal conductivity of over 200 W/mK.
Mechanical drilling causes rapid drill bit wear, rough hole walls, and severe edge chipping. The energy from conventional lasers is quickly dispersed, resulting in a heat-affected zone and microcracks.
Ultrafast lasers can improve this issue, but they are not without limits.
Analysts also caution that, in high-throughput processing, thermal accumulation from high-frequency pulses may still induce microcracks, requiring careful control of repetition rate and energy density. The industry is further exploring composite solutions such as high-energy ultrafast lasers, mechanical pre-drilling combined with ultrafast laser finishing, magnetic field-assisted laser processing, and cryogenic-assisted laser processing.
What assumptions underlie the trillion-dollar opportunity?
The market potential for ultrafast laser equipment primarily stems from four categories of potential applications: CoPoS glass interposers, ABF glass substrates, M9 materials, and optical module substrates.
In the calculation, the unit price of the equipment is assumed to be RMB 6 million. Under different penetration rates, the corresponding existing market sizes are:
- 10% penetration rate: approximately RMB 10.3 billion;
- 30% penetration rate: approximately RMB 31 billion;
- 50% penetration rate: approximately RMB 51.6 billion;
- 80% penetration rate: approximately RMB 82.6 billion;
- 100% penetration rate: approximately RMB 103.3 billion.
Among these, the M9 material contributes the most. Under a 100% penetration rate assumption, the M9 material corresponds to a demand of 11,574 units; ABF glass substrates correspond to 3,704 units; CoPoS glass interposers correspond to 1,757 units; and optical module substrates correspond to 174 units.
This calculation is based on two implicit assumptions.
First, advanced packaging is indeed expanding from CoWoS to CoPoS and CoWoP. TSMC has planned mass production lines for CoPoS and has already built pilot production lines for CoPoS with glass substrates, with large-scale mass production expected within 2 to 3 years.
Second, materials such as glass substrates, M9 PCBs, and ceramics are not laboratory-level pathways—they are ready for mass manufacturing. Equipment orders ultimately come from production lines, not from technological narratives.
LPKF maintains a high-end position, while domestic manufacturers are pursuing complete solutions.
The global ultrafast laser equipment landscape is currently dominated by leading overseas companies in the high-end market, while domestic manufacturers are accelerating their efforts to catch up.
LPKF Germany's advantage lies in its LIDE (Laser-Induced Deep Etching) process, which enables crack-free, high aspect-ratio, and low-thermal-damage processing of glass substrates. Its Vitrion S 5000 is designed for thin glass micromachining and TGV via formation, achieving aspect ratios of up to 1:50.
Domestic manufacturers' paths are more diverse and closer to downstream process validation.
Han's Laser offers specialized equipment for PCB manufacturing, including ultrafast laser drilling systems. Its glass laser drilling machine achieves via diameters as small as 10µm, with a depth-to-diameter ratio of up to 50:1 for mainstream materials.
Hans Laser possesses advanced laser source technologies, including picosecond ultraviolet and picosecond infrared, and offers solutions for micro-processing of brittle materials such as glass, sapphire, and ceramics, while also being compatible with high-end processes like mSAP and TGV.
Dir Laser focuses on glass TGV. Its wafer-level TGV laser micro-via equipment supports various glass materials with a minimum via diameter of ≤5µm and an aspect ratio as high as 1:100.
Inno Laser offers a product portfolio spanning from nanosecond to femtosecond pulses and from infrared to deep ultraviolet wavelengths, with laser sales exceeding 22,000 units, and its ultrafast PCB drilling equipment has secured its first order.
The laser drilling machine from Lianying Laser has a repeatability accuracy of ±1 µm and can process glass up to 20 mm thick. The glass TGV drilling equipment is currently in customer validation.
Delong Laser specializes in precision laser microprocessing equipment and core lasers, with independently developed picosecond and femtosecond solid-state lasers covering applications such as TGV, wafer cutting, and ceramic and glass processing.
Haimax focuses on "laser + automation," developing processes such as laser etching and laser-induced methods, with applications spanning lithium batteries, photovoltaics, consumer electronics, and semiconductors.
The focus of competition is no longer just on the specifications of individual devices. Advanced packaging materials are complex, with narrow process windows, and downstream manufacturers require comprehensive solutions including light sources, motion control, process parameters, inspection, and automation. Domestic manufacturers also have opportunities through localized delivery and strong engineering responsiveness.
The risk lies in the production timeline, not in the concept itself.
The logic behind ultrafast lasers is clear: the harder the material, the smaller the hole, and the less thermal damage that can be tolerated, the more valuable ultrafast lasers become.
However, there are still three types of risks associated with device volume.
First, the development of advanced packaging technologies has fallen short of expectations. If progress on routes such as CoPoS, CoWoS, and glass substrates is slower than anticipated, demand for related equipment will also be delayed.
Second, AI computing power investment has fallen short of expectations. The primary driver for upgrading packaging materials is the demand for high-end AI chips; if capital expenditures downstream slow down, equipment orders will be affected.
Third, risks from international trade friction. Advanced packaging materials and equipment rely on global supply chains; trade restrictions could impact verification, delivery, and the pace of customer expansion.
For the market, ultrafast lasers are not merely a story about "laser equipment," but rather a precise tool essential to the manufacturing process following the transition to advanced packaging materials.