Sony and TSMC hook up for fab-lite

Intel's EMIB packaging technology is reportedly growing to become quite popular in the industry as the demand for memory and AI chips continues to rise. A recent report suggested that Google was interested in Intel's EMIB as TSMC's CoWoS (chip-on-wafer-on-substrate) experiences supply bottlenecks. Packaging was among the first bottlenecks that emerged in the AI race, which kicked off in late 2022 and has persisted to date as manufacturers rush to add capacity and develop new technologies.

As Taiwan's TSMC remains the world's premier contract chip manufacturer, IC designers are also seeking alternatives to its technology. One alternative is Intel's EMIB, and a report from Zdnet Korea suggests that SK hynix is working with the firm to research and develop EMIB packaging technology. These efforts are part of using EMIB as 2.5D packaging, which uses an interposer to connect the main chip die with the packaging substrate that connects the package to the circuit board.

Intel's Aggressive EMIB Packaging Marketing Could Make It A Key Industry Player, Say Sources The report suggests that SK hynix is evaluating using EMIB to connect its high bandwidth memory (HBM) with the logic die of a chip. The publication's source adds that hynix is also looking at the raw materials that might be used in case EMIB is used for volume production.

Packaging yields for EMIB could play a key role in SK hynix's decision to use the technology. Recent commentary from well known technology analyst Ming-Chi Kuo discussed the yields. Kuo remarked that while Intel's EMIB-T's 90% yield was a validation figure, it does not represent production yield. The analyst added that production yield would play a key role in Google's decision to use EMIB for the next-generation TPU AI chips.

Intel's aggressive EMIB marketing and market dynamics could lead to the technology becoming a key part of the AI packaging supply chain, say the sources. During the firm's latest earnings call, Tan touted the benefits of his firm's business model, which allowed it to quickly integrate customer feedback into production.

"We're not just at the CPU. We have advanced packaging and foundry, we can really effectively driving some of the changes more quickly to serve the customer in terms of their different workloads," he said.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved