Technological Evolution of PCBs in the Era of Artificial Intelligence

The technological shift from traditional through-hole mounting to high-density interconnects, coupled with the explosive growth of artificial intelligence, is fundamentally reshaping the PCB industry’s technological trajectory, product structure, and value distribution.

AI PCB

Technological Requirements Upgrade of AI Computing Hardware for PCBs

Demand for High Layer Count and High-Density Interconnects

Traditional server motherboards typically employ 12–16 layers, whereas current mainstream AI training servers (such as the NVIDIA DGX H100 series) require PCB layer counts of 20–30 layers. Particularly for GPU substrates, interconnect densities exceeding 5,000 BGA solder points are necessary, with trace width/spacing compressed from the conventional 4/4 mil to 2/2 mil or even 1.5/1.5 mil. This design demand directly drives the adoption of the mSAP (Modified Semi-Additive Process), as traditional subtractive processes can no longer meet precision requirements.

Signal Integrity Challenges and Solutions

At 112 Gbps PAM4 transmission rates, insertion loss must be controlled within -0.6 dB/inch. Through simulation analysis, we have found that the dissipation factor (Df) needs to be reduced from 0.02 for conventional FR-4 to below 0.005. The current leading industry solution involves using a hydrocarbon resin/ceramic filler composite system (such as Rogers RO4835™), which maintains a stable Dk value of 3.5±0.05 and exhibits good dielectric properties even at 77 GHz.

Innovations in Thermal Management Technology

Taking the NVIDIA H100 as an example, the single-chip peak power consumption reaches 700W, rendering traditional thermal design solutions entirely inadequate. Our developed embedded copper block + thermal via array technology can reduce thermal resistance to 0.8°C/W. In terms of substrate material selection, high Tg (≥170°C) and high thermal conductivity (≥0.8 W/m·K) have become basic requirements, with some high-end applications already adopting hybrid structures of metal substrates and organic materials.

Technological Breakthroughs and Localization Progress of Key Materials

Advances in Yüksek Frekans ve Yüksek Hızlı Copper-Clad Laminates

Shengyi Technology’s S7439 series has been certified by major OEMs, achieving a Df value of 0.0058 at 10 GHz, approaching internationally leading standards. Sinoma Science & Technology’s development of Low Dk electronic glass fabric (Dk=4.2) breaks the technological monopoly of Nittobo, with mass production expected by 2025.

Specialty Chemical Materials

In solder resist inks, Taiyo Ink’s SR-7200G series supports laser direct imaging with resolutions up to 20 μm. For plating additives, MacDermid Enthone’s Circuposit 8800 series enables uniform plating with 1:1 aspect ratios, addressing the issue of uniform copper plating in through-holes for high-layer-count PCBs.

AI PCB

Technical Bottlenecks and Breakthroughs in Manufacturing Processes

Lazer Delme Teknolojisi

For microvia processing below 0.1 mm, CO2 lasers are approaching physical limits. We have introduced UV laser processing systems combined with beam shaping technology to enhance processing precision to 35 μm. Han’s Laser’s UV laser drilling machines, using a 355 nm wavelength, achieve a minimum hole diameter of 50 μm with a positional accuracy of ±15 μm.

Innovations in Lamination Processes

For ultra-high-layer boards exceeding 30 layers, we have developed a lamination process involving segmented heating and pressure application. By precisely controlling resin flow, the interlayer fill rate is increased to over 95%, while interlayer alignment accuracy is maintained within ±25 μm.

Upgrades in Inspection Technology

A comprehensive solution combining Otomatik Optik Muayene (AOI) and electrical testing is adopted. Keysight’s PathWave ADS software supports 3D electromagnetic field simulation, enabling early identification of signal integrity issues. For in-circuit testing, Teradyne’s TestStation architecture supports bit error rate testing for 112 Gbps interfaces.

Industrial Chain Restructuring and Business Model Transformation

Reshaping of Supply Chain Relationships

The AI server PCB supply chain is divided into three tiers: GPU board sets are led by chip manufacturers (e.g., NVIDIA’s designated supply chain); CPU motherboards follow the traditional server supply chain; and module manufacturers independently procure accessory modules. This differentiation requires PCB manufacturers to possess differentiated customer engagement capabilities.

Increased Concentration Due to Higher Technical Barriers

Capital investment for 18-layer or more PCBs is 3–5 times that of traditional products, with R&D cycles extending to 12–18 months. This has led to market share concentration among leading companies, with the top three manufacturers accounting for over 60% of the domestic AI server PCB market in 2024.

Changes in Value Distribution

In the Bill of Materials (BOM) cost of AI servers, the proportion of PCBs has increased from 2–3% in traditional servers to 6–8%. Particularly for GPU substrates, due to their high technical complexity, gross margins can reach 35–40%, significantly higher than the 15–20% for traditional products.

AI PCB

Future Technological Development Trends

Integration of Advanced Packaging and PCBs

The Chiplet architecture requires PCBs to assume some interposer functions, driving Substrate-Like PCB (SLP) technology toward trace width/spacing of 10/10 μm. Shennan Circuits’ developed eSLP technology has achieved 8/8 μm process capability and is undergoing sample validation with major chip manufacturers.

Silicon Photonics Co-Packaging Technology

For optical modules above 1.6T, Co-Packaged Optics (CPO) has become an inevitable choice. This requires PCBs to integrate photonic waveguides, and we are developing hybrid substrate technology based on silicon dioxide waveguides, expected to achieve engineering applications by 2026.

Sustainability Requirements

The EU’s CE-RED directive imposes new environmental requirements on PCBs, including halogen-free materials and lead-free processes. Our developed bio-based epoxy resin system reduces carbon footprint by 40% and has obtained UL certification.

Recommendations for Technical Teams

Transformation of Talent Structure

A shift from traditional process engineers to “material-process-system” composite talents is necessary. In our team, the proportion of engineers with materials science backgrounds has increased from 10% a decade ago to 35% today.

Focus of R&D Investment

It is recommended to allocate 60% of R&D resources to high-layer-count HDI, 30% to advanced packaging, and 10% to sustainable development technologies. Particular emphasis should be placed on early collaboration with chip manufacturers and participation in front-end design.

Patent Layout Strategy

Focus on patent layouts in three directions: high-speed materials, thermal dissipation structures, and high-density interconnects. Among our core patents applied for in recent years, those involving special thermal dissipation structures account for 40%, which will become a future technological barrier

Artificial intelligence is elevating PCBs from auxiliary components to core parts of computing systems. This change in status requires us to redefine product development processes with a system-level mindset, transitioning from pure manufacturing service providers to technical solution providers. Future industry competition will be a comprehensive contest of material systems, process capabilities, and system design prowess.