Samsung to Power Qualcomm’s AI Accelerator for Data Centers

SammyGuru is reader-supported. We have affiliate and sponsored partnerships, so we may earn a commission when you buy through links on our site — at no extra cost to you. Learn more.

Samsung is gearing up to play a role in Qualcomm’s push into the AI data center space. The Korean firm’s FC-BGA (Flip-Chip Ball Grid Array) substrates will power the American chipmaker’s first AI200 accelerator for data centers. The mass production of the substrates is already underway.

Samsung expands Qualcomm partnership into AI data Center chip packaging

Samsung’s components manufacturing arm, Samsung Electro-Mechanics, recently began mass production of FC-BGA substrates at its Busan facility, according to ZDNet Korea. The substrate connects high-integrated semiconductor chips and substrates using flip-chip bumps. It is mainly installed in CPUs and GPUs for PCs, servers, and automobiles.

Last year, Qualcomm unveiled the AI200, its first dedicated AI accelerator for data centers. The accelerator integrates an in-house Oryon CPU and Hexagon NPU with power-efficient LPDDR5 memory. Moreover, it offers a low total cost of ownership (TCO) and optimized performance to support AI workloads. The company is expected to launch the AI200 in the second half of this year.

Samsung Electro-Mechanics already supplies package substrates for Qualcomm’s application processors (APs) used in IT devices. Now, with the expansion into AI accelerators, the two companies are deepening their relationship. The report also says the initial supply volume of FC-BGA for the AI200 is not large. However, this could increase in the future as Qualcomm scales up production.

“It appears that the supply of FC-BGA for AI accelerators was smoothly concluded, given Samsung Electro-Mechanics’ long-standing cooperation with Qualcomm,” said an official from the semiconductor industry. “Qualcomm plans to release the AI200 this year and the AI250 next year, and Samsung Electro-Mechanics can benefit from the diversification of its revenue sources as a result.”

The FC-BGA for the AI200 bears low to mid-10s internal layers. It builds these layers using copper circuits and ABF insulating material. Ultra-high-performance AI accelerators can use more than 20 layers.

Written by

Binay Konwar

Binay Konwar started his blogging journey in 2014 and has since written plenty of tech articles. At present, he is working as a News Writer at SammyGuru, covering everything about Samsung. He holds a Master's degree in Mathematics, but his real passion lies in tech and writing. In his free time, he enjoys playing chess and watching movies.

XMore from Binay Konwar →