• 2 min read
Intel targets 240 mm packages for giant AI chips
Intel says its HLFF packaging can support 240 × 240 mm packages—24 times the photomask area—for multi-chip AI systems.
Image: iXBT
Intel says it has solved several packaging problems that could enable 240 × 240 mm chip packages—structures it describes as the largest ever attempted in the industry. The company is not talking about a single giant monolithic die, but about chiplet-based systems that combine multiple compute and memory components in one package.
At Intel’s facilities in Rio Rancho, New Mexico, advanced packaging technology is expanding the practical limit imposed by the photomask. Intel says the approach can increase package size to eight times today’s standard industry figures, rising to more than 12 times by 2028.
“We moved from the era of one large chip to a system of chips, almost like a silicon mosaic, where specialized elements are connected in one massive package.”
Intel’s 24× photomask package
The company’s next step is a class of highly large form factor (HLFF) packages. Measuring 240 × 240 mm, they would provide a 24-fold increase over the photomask size, according to Intel—exceeding anything previously seen in the industry.
The packaging process is designed to place more dies on a single substrate, overcoming limits that normally constrain package dimensions. Intel Foundry developed a void-free encapsulation process and tested it on packages containing up to seven dies. The company says this directly addresses one of the main manufacturing problems identified by the HLFF project.
The challenge of substrate warping
Intel also worked on deformation across the unusually large substrate area. Mechanical strength has been one of the obstacles to building such packages; the source points to Nvidia’s decision to abandon the Rubin Ultra design with two GPUs on one substrate as an example of the difficulty.
Recommended reading
Aston Martin F1 keeps humans at the center of AI
These oversized packages are primarily intended for the AI market. Their purpose is to let a single substrate carry several compute chiplets alongside numerous memory chips, creating a much larger integrated system without requiring one enormous monolithic processor. Intel has not provided pricing or a release date for HLFF packages.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via iXBT
