Baidu, Samsung Electronics gear-up for Edge AI chip’s mass production in 2020

Spread the love

Beijing, China – Seoul, South Korea: China’s Baidu and Samsung Electronics have completed the development of Baidu’s first Edge AI accelerator chip – KUNLUN. And the new cloud to Edge AI chip will under go mass production early in 2020.

Baidu KUNLUN chip

Baidu KUNLUN chip is built on advanced XPU, which is a home-grown neural processor architecture for Cloud, Edge and AI. It uses Samsung’s 14-nanometer (nm) process technology with its I-Cube (Interposer-Cube) package solution.

It offers 512 gigabytes per seconds (GBps) memory bandwidth and supplies up to 260 Tera operations per second (TOPS) at 150 watts.

In addition, the new chip allows Ernie, a pre-training model for natural language processing, to inference at three times faster than that of conventional GPU/FPGA-accelerating model.

Leveraging the chip’s limit-pushing computing ability and power efficiency, Baidu can effectively support a wide variety of functions.

That includes large-scale AI workloads, such as search ranking, speech recognition, image processing, natural language processing, autonomous driving, and deep learning platforms like PaddlePaddle.

Baidu and Samsung’s foundry cooperation

Through the first foundry cooperation between the two companies, Baidu will provide advanced AI platforms for maximizing AI performance.

Samsung will expand its foundry business into High performance computing (HPC) chips, dedicated for Cloud and Edge computing.

“We are excited to lead the HPC industry together with Samsung Foundry,” said OuYang Jian, Distinguished Architect – Baidu.

“Baidu KUNLUN is a very challenging project since it requires not only high level of reliability and performance at the same time, but is also a compilation of the most advanced technologies in the semiconductor industry,” said Jian.

Jian also thanked Samsung’s process technologies and competent foundry services that helped to meet and surpass their goal to offer superior AI user experience.

“We are excited to start a new foundry service for Baidu using our 14nm process technology,” said Ryan Lee, VP – Foundry Marketing, Samsung Electronics.

Beyond mobile to data center applications

Calling “Baidu KUNLUN” as an important milestone for Samsung Foundry, Lee said his company is expanding its business beyond mobile to data center applications by developing and mass-producing AI chips.

“Samsung will provide comprehensive foundry solutions from design support to cutting-edge manufacturing technologies, such as 5LPE, 4LPE, as well as 2.5D packaging,” informed Lee.

Diverse applications such as AI and HPC requires higher performance and that’s where chip integration technology is becoming highly important.

Samsung’s I-Cube technology

Samsung’s I-Cube technology connects a logic chip and high bandwidth memory (HBM) 2 with an interposer, providing higher density/ bandwidth on minimum size using its own solutions.

Compared to previous technology, Samsung claimed that its solutions offer maximize product performance with more than 50% improved power/signal integrity. It is anticipated that I-Cube technology will mark a new epoch in the heterogeneous computing market.

Samsung is also developing more advanced packaging technologies, such as redistribution layers (RDL) interposer and 4x, 8x HBM integrated package.

“Baidu KUNLUN is a very challenging project. It requires not only high level of reliability and performance at the same time, but is also a compilation of the most advanced technologies in the semiconductor industry,” said Jian.

Jian also thanked Samsung’s process technologies and competent foundry services that helped to meet and surpass their goal to offer superior AI user experience.

“We are excited to start a new foundry service for Baidu using our 14nm process technology,” said Ryan Lee, VP – Foundry Marketing, Samsung Electronics.

Calling Baidu KUNLUN as an important milestone for Samsung Foundry, Lee said the company is expanding its business beyond mobile to datacenter applications by developing and mass-producing AI chips.

“Samsung will provide comprehensive foundry solutions from design support to cutting-edge manufacturing technologies, such as 5LPE, 4LPE, as well as 2.5D packaging,” informed Lee.

Chip integration technology is becoming highly important for high performance. Diverse applications such as AI and HPC require higher performance.

Samsung’s I-Cube technology connects a logic chip and high bandwidth memory (HBM) 2 with an interposer, providing higher density/ bandwidth on minimum size using its own solutions.

Samsung claimed its solutions offer maximize product performance with over 50% improved power/signal integrity. I-Cube technology is anticipated to mark a new epoch in the heterogeneous computing market.

Samsung is also developing more advanced packaging technologies, such as redistribution layers (RDL) interposer and 4x, 8x HBM integrated package.

(Image source – Baidu KUNLUN chip )

Leave a Reply

Your email address will not be published. Required fields are marked *