Advanced packaging amplifies substrate technology barriers, potentially strengthening the medium-term advantage of Japanese supply-chain players such as Ibiden

Overview

This report summarizes a seminar hosted by JPMorgan with Toshihiko Nishio of SBR Technology on future technology trends in FC-BGA (ABF) substrates. The discussion focused on organic cores, glass cores, HDI cores, ceramic cores, and advanced packaging routes such as CoWoS, CoPoS, CoWoP, and EMIB-T. The core conclusion is that AI chips and HBM are continuously driving larger packaging sizes, and substrate warpage, flatness, soldering reliability, low CTE, high rigidity, and mass-production process capability are becoming the main competitive barriers in the electronic components industry.

Core views

The report believes that as CoWoS and CoPoS adopt larger reticle sizes, differences in thermal expansion coefficients among packaging materials will amplify warpage and connection-failure risks, increasing the strategic value of high-rigidity, high-flatness substrates and glass-core substrates. Ibiden's relative competitive advantage in large-size, high-difficulty substrates may strengthen in the medium term; Ibiden and other Japanese companies may also become key suppliers for EMIB-T. At the same time, EMIB-T, CoWoP, and glass cores have not yet fully crossed the mass-production threshold, and competition among technical routes still depends on mass-production yield, customer qualification, supply-chain buildout, and end-platform timing.

Analysis framework

The report uses a technical-route breakdown and supply-chain mapping approach: it first compares the structural differences among CoWoS-S, CoWoS-L, EMIB-T, CoPoS, and CoWoP, then analyzes the impact of larger reticle sizes on warpage, soldering, flatness, and thermal-expansion-coefficient matching, and finally maps these technical bottlenecks to supply-chain segments such as ABF substrates, glass cloth, drilling tools, glass cores, and assembly foundries.

Methodology notes

• Technical route comparisonAdvanced packaging architecture comparison

  Structural differences among CoWoS, CoPoS, CoWoP, and EMIB-T

  By evaluating dimensions such as whether a large interposer is used, whether ABF substrates are relied upon, whether panel processes are adopted, and whether TSVs are added in the silicon bridge, the analysis determines the requirements each route places on substrates, assembly, and yield.

• Manufacturing bottleneck analysisWarpage and connection reliability framework

  Flatness, CTE, and soldering risks after reticle size expansion

  The larger the package size, the more likely differences in thermal expansion coefficients among materials are to cause warpage, which in turn affects soldering connection reliability; therefore, low CTE, high rigidity, and flatness become key indicators.

• Supply-chain competitiveness assessmentKey material and process capability mapping

  Mass-production capability in substrates, glass cloth, microvia drilling, and glass cores

  Technical requirements are mapped to the capabilities of suppliers such as Ibiden, Nittobo, Union Tool, Absolics, and Samsung Electro-Mechanics to identify potential beneficiaries and bottleneck segments.

• Industrialization stage assessmentMass-production maturity assessment

  Differences among development, pilot production, small-scale production, and large-scale mass production

  The report distinguishes among technologies with mass-production track records, technologies under evaluation, and technologies lacking infrastructure, emphasizing that customer qualification, capacity, yield, and the ability to serve external customers determine commercialization timelines.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Ibiden

  Potential beneficiary supplier of high-end ABF substrates and glass-core development

  Strengths

  The report believes larger reticle sizes require high-rigidity and high-flatness substrates, which may enhance Ibiden's relative competitive advantage in the medium term; TSMC's cooperation with Ibiden and Innolux to develop CoPoS glass-core substrates is also attracting attention.

  Weaknesses

  Glass cores are still in the development stage, and no clear timeline for large-scale mass-production launch has been announced; large-size substrates are still constrained by challenges in warpage, flatness, and connection reliability.

  Comparison

  Compared with most substrate makers still in the development stage, Ibiden's position in high-end ABF substrates receives greater attention in the report, but Absolics is more advanced in small-scale production of glass cores.

  Risks

  If the CoWoS-L or CoPoS routes progress less than expected, or if CoWoP reduces reliance on Japanese substrate makers, demand elasticity may weaken.

• Nittobo

  Key supplier of high-end organic-core glass cloth

  Strengths

  Nittobo's T-glass can help control substrate CTE and improve strength and electrical performance; the report says it is almost the sole supplier in high-end products.

  Weaknesses

  Capacity expansion needs to balance demand from customers such as Nvidia and Broadcom against competitor catch-up and future oversupply risks.

  Comparison

  The report says other glass-cloth suppliers such as Taiwan Glass currently find it technologically difficult to keep up with high-end demand.

  Risks

  If competitors catch up technologically or end demand is weaker than expected, gradual capacity expansion may still face supply-demand mismatch.

• Union Tool

  Supplier of microvia drilling tools for hard low-CTE substrates

  Strengths

  Lower CTE and higher-rigidity materials increase drilling difficulty, and the report describes Union Tool as a leading supplier of drilling tools capable of stable microvia processing.

  Weaknesses

  Demand depends on continued upgrades in high-end substrate materials and expansion of advanced packaging.

  Comparison

  The report says other companies currently cannot match its capability in stable microvia processing for hard materials.

  Risks

  If material routes shift or processing methods change, the demand structure for tools may change.

• Intel EMIB-T / Intel Foundry Services

  Advanced packaging route that could potentially replace CoWoS-L

  Strengths

  EMIB-T does not use a large interposer, reducing sensitivity to warpage caused by one-time attachment of the entire module and potentially improving back-end yield; Intel says it can support up to 12x reticle size.

  Weaknesses

  EMIB-T has not yet established mature mass-production technology, Intel Foundry Services lacks reliable mass-production capability for external customers, and substrate-side capabilities are also not fully established.

  Comparison

  Compared with CoWoS-L, EMIB-T has theoretical advantages in connection reliability and warpage control, but CoWoS already has a more mature ecosystem.

  Risks

  If mass-production, substrate, and assembly processes cannot be established in time, onboarding by potential customers such as Broadcom and Google will be limited.

• TSMC CoWoS / CoPoS

  Core platform for AI advanced-packaging capacity and technology routes

  Strengths

  CoWoS is currently an important advanced packaging route; CoPoS improves area utilization through a 310mm square panel and, if successful, could improve production efficiency.

  Weaknesses

  CoWoS-L faces more severe flatness and soldering issues at 5.5x and 9.5x reticle sizes; CoPoS may also be delayed.

  Comparison

  If CoPoS is delayed, TSMC can offset part of the impact by increasing existing CoWoS capacity; EMIB-T, by contrast, attempts to reduce warpage risk through a different structure.

  Risks

  If 9.5x CoWoS-L cannot succeed, packaging-technology limitations may affect realization of the Nvidia Feynman generation.

• Nvidia CoWoP

  A packaging concept that may reduce reliance on traditional ABF substrates in the future

  Strengths

  CoWoP eliminates traditional ABF substrates and mounts the GPU and HBM interposer directly on an SLP-like PCB, which could theoretically simplify the structure, shorten signal paths, improve thermal-design flexibility, and reduce cost.

  Weaknesses

  It requires SLP line width to fall below 10μm, and PCB processing precision, yield, flip-chip mounting, chip-side routing rules, and bump pitch must all change.

  Comparison

  Compared with CoWoS, CoWoP may reduce reliance on Japanese substrate makers and use PCB supply chains in Taiwan and China, but it also brings greater pressure on standardization and development resources.

  Risks

  Because Nvidia is prioritizing support for the Feynman generation, it may lack the resources to simultaneously develop CoWoP as an independent technology, and the report questions its feasibility.

• Absolics

  Leader in small-scale production of glass cores

  Strengths

  The report says Absolics is currently one of the few glass-core companies with mass-production facilities and has advanced to the small-scale production stage.

  Weaknesses

  Small-scale production is not the same as mature large-scale mass production, and industry infrastructure remains insufficient.

  Comparison

  Ibiden, Shinko Electric Industries, and Unimicron are still in the development stage, while Samsung Electro-Mechanics plans to advance mass-production prototypes after 2028.

  Risks

  Customer qualification and capacity expansion take time, and large-scale adoption of glass cores may be postponed until after 2030.

Key data

• CoWoS-S reticle size upper limit3.3xToshihiko Nishio said 3.3x is the upper limit because a larger interposer increases warpage and reduces mounting soldering reliability.
• TSMC CoWoS-L planned reticle size3.3x -> 5.5x -> 9.5x by 2029TSMC has disclosed plans to expand reticle size, but 9.5x is considered very difficult to achieve.
• Substrate size changes85mm×85mm、110mm×110mm、130mm×140mmThese correspond respectively to 3.3x, 5.5x, and 9.5x reticle size.
• Potential HBM4 bump pitch change65μm -> 36μmA finer pitch requires higher-density routing between HBM and LSI.
• CoPoS panel size310mm square panelCoPoS transfers the CoWoS process from a 300mm wafer to a square panel to improve area utilization.
• Potential CoPoS delayup to around two yearsNishio believes the technology may proceed according to TSMC's plan, or it may be delayed by around two years.
• SLP line width required for CoWoP mass production10μm or lessThe current level is around 15-20μm, and PCB processing precision, yield, and flip-chip mounting capability still need to improve.
• Assessment of glass-core mass-production timingdifficult until 2030The report believes glass-core infrastructure and mass-production factories are still insufficient, making large-scale mass production before 2030 difficult.
• J.P.Morgan global equity research coverage distributionOverweight 53%, Neutral 36%, Underweight 12%From the disclosure table; percentages may not sum to 100% due to rounding.

Impact & implications

In terms of investment implications, AI accelerators and HBM packaging continue to increase substrate-technology difficulty, benefiting in the short to medium term suppliers that already possess capabilities in high-end ABF substrates, low-CTE materials, microvia processing tools, and high-rigidity substrates. Ibiden's relative advantage may strengthen because of larger reticles and high-flatness requirements; Nittobo and Union Tool are positioned respectively in high-end glass cloth and hard-material microvia processing; if glass cores and EMIB-T mature, they could reshape the supply chain, but before 2030 attention is still needed on mass-production facilities, customer qualification, and external customer service capability.

Risks

• CoWoS-L may face difficulties progressing at 5.5x and 9.5x reticle sizes due to warpage, flatness, and soldering reliability issues.
• EMIB-T has not yet proven it can stably mass-produce for external customers, and substrate and assembly processes still need to be established.
• Although CoPoS has production-efficiency potential, commercialization timing may be delayed by up to about two years versus TSMC's plan.
• If successful, CoWoP could reduce reliance on traditional ABF substrates and Japanese substrate makers, but it also faces its own challenges in line width, yield, and mounting processes.
• Glass cores are regarded as a future material, but large-scale mass production before 2030 remains difficult because of insufficient mass-production facilities, customer qualification, and supply-chain infrastructure.
• If key material suppliers such as Nittobo expand capacity while facing competitor catch-up or end-demand volatility, future oversupply risk may emerge.

What to watch

• Progress in TSMC CoWoS-L 5.5x evaluation, and whether soldering issues in 9.5x reticle size are resolved after 1H2027.
• Whether the Nvidia Feynman generation is constrained by packaging technology, and whether implementation assessment related to Rubin Ultra improves.
• Whether Intel EMIB-T can establish reliable mass-production capability for external customers, and progress in Amkor Technology's acceptance of assembly-technology licensing and outsourcing.
• Whether TSMC CoPoS is introduced as planned, or whether a delay of up to about two years emerges.
• Whether CoWoP can achieve sub-10μm SLP line width, acceptable yield, and adjustments to chip-side routing rules.
• Progress in glass-core mass production, pilot production, and customer qualification at Ibiden, Shinko Electric Industries, Unimicron, Samsung Electro-Mechanics, Absolics, and others.
• The pace of Nittobo T-glass capacity expansion, competitors' technology catch-up, and whether Union Tool can maintain its leadership in microvia processing tools for high-rigidity materials.