CoWoS packaging technology is the first choice of many international arithmetic chip makers and one of the mainstream solutions for high-end performance chip packaging. With the growing demand for Artificial Intelligence (AI) chips, NVIDIA is facing tremendous pressure on its packaging capacity due to the surge in demand. In particular, TSMC's (TSMC) CoWoS packaging technology, its capacity has been very tight in the past year or so. To meet this challenge, NVIDIA had originally planned to introduce FOPLP packaging technology in 2026, but in light of the fast-changing market situation, the company has decided to move up the timeline to 2025 to produce its next-generation GPU, the GB200 chip, using fan-out panel level packaging (FOPLP) technology.
FOPLP technology, known as Fan-Out Panel-Level Packaging, is an extension of Fan-Out Wafer-Level Packaging (FOWLP). Compared with traditional packaging methods, FOPLP technology offers higher area utilization, production efficiency, and cost effectiveness. It enables the integration of multiple chips, passive components, and interconnects in a single package to achieve higher performance and better electrical performance.
FOPLP technology utilizes larger substrate sizes and has an area utilization of more than 95%, much higher than the 85% of FOWLP technology. The high area utilization allows more chips to be processed in a single package, improving packaging efficiency and creating a scale effect. Moreover, FOPLP technology is suitable for a wide range of applications such as sensors, power ICs, RF ICs, RF, connectivity modules, PMICs, etc., especially in the automotive electronics field with a wide range of application prospects. Relying on the precision rewiring layer (RDL) process, FOPLP technology realizes high-speed, high-density interconnections between chips (D2D), which is crucial for AI computing.
figure:CoWoS is shortage and Nvidia shift to FOPLP
The GB200 is NVIDIA's next-generation AI chip based on the Blackwell architecture, combining two GPUs and a Grace CPU to deliver up to 30x performance for LLM inference workloads. NVIDIA's GB200 supply chain has been launched and is currently in the design fine-tuning and testing phase. Shipments are expected to reach approximately 420,000 units this year, while next year they are expected to climb to 1.5 to 2 million units. With the introduction of FOPLP packaging technology, the performance and reliability of the GB200 chip will be further enhanced. By improving the chip's integration and electrical performance, the GB200 will be able to better meet market demand for high-performance AI chips. At the same time, FOPLP packaging technology will also improve the chip's packaging efficiency and reduce production costs, providing strong support for NVIDIA's competition in the AI chip market.
NVIDIA's early introduction of FOPLP packaging technology not only demonstrates its ability to respond quickly to market changes, but also reflects the company's foresight in packaging technology innovation. With the GB200 chip going into production in 2025, NVIDIA will further consolidate its leadership in high-performance computing and drive the development of AI and data center technologies.
