FOPLP in AI GPUs: 2027 Mass Production Expected

Derived from TSMC's InFO (Integrated Fan-Out) packaging technology introduced in 2016, the development of FOPLP (Fan-Out Panel Level Packaging) is gaining traction with major players like AMD and NVIDIA exploring its use for chip packaging. This trend has sparked significant interest in FOPLP technology across the market. According to TrendForce, initial applications are focused on mature process and cost-sensitive products such as PMICs (Power Management ICs). Mainstream consumer IC products are expected to adopt FOPLP once the technology matures, possibly by 2027 for AI GPUs.

TrendForce identifies three primary applications for FOPLP technology currently: first, OSATs (Outsourced Semiconductor Assembly and Test) transitioning consumer IC packaging from traditional methods to FOPLP; second, foundries and OSATs shifting AI GPU 2.5D packaging from wafer-level to panel-level; and third, panel manufacturers entering the consumer IC packaging domain. This highlights proactive industry-wide adoption of FOPLP technology across various segments of the supply chain.

Analyzing industry collaborations, significant progress has been made in partnerships such as AMD with ASE and Nepes for PC CPU products, and Qualcomm with Nepes for PMIC products. However, limitations in FOPLP linewidth and pitch compared to FOWLP currently restrict broader applications.

Regarding AI GPU packaging, major players like AMD and NVIDIA are leading discussions with TSMC and SPIL to transition from wafer-level to panel-level 2.5D packaging. This collaboration aims to scale up packaging sizes and reduce unit costs, although technical challenges mean the transition is still in the evaluation phase.

Industry experts clarify that while rectangular substrates are replacing traditional circular silicon interposers during packaging stages, this does not imply a complete shift from circular wafers. The use of rectangular substrates in packaging aims to increase area utilization efficiency, particularly in 2.5D packaging, enhancing component transmission efficiency and reducing costs to accommodate more chip combinations, which is beneficial for large-scale AI chip layout arrangements.

FOPLP technology presents both advantages and challenges, including lower unit costs and larger packaging sizes. However, uncertainties remain in the commercialization of the technology due to ongoing technical and equipment system developments. TrendForce estimates that mass production for consumer ICs and AI GPU applications using FOPLP could occur between the second half of 2024 and 2026, and from 2027 to 2028, respectively.