Baidu first announced its ambition to develop in-house chips in 2018, and in 2019 it declared that the chip had entered risk production, with volume p

Baidu’s New AI Chips Have Entered Volume Production

submited by

Style Pass

2021-08-19 07:00:06

Baidu first announced its ambition to develop in-house chips in 2018, and in 2019 it declared that the chip had entered risk production, with volume production scheduled in 2020. Kunlun, as the processor was named, was intended as an AI accelerator in cloud and edge computing, and adopted Samsung’s 14nm process as well as its 2.5D Interposer-Cube package solution.

In March this year, Baidu revealed that its chip unit had completed a new round of fundraising, with the company’s valuation estimated at around US$2 billion, followed by its decision to split off its chip unit into a separate entity.

Now, Baidu has officially revealed that its 2nd-generation AI chip, Kunlun 2, has entered volume production on 7nm process node. However, it remains unclear who manufactured the chips. So far, only Samsung and TSMC have mastered this particular process node, with SMIC’s 7nm process only entering risk production this year. Baidu has also claimed that its newest chips are compatible with multiple Chinese homegrown chips and operating systems, notably Kylin – an OS developed by China’s National University of Defense Technology, and FeiTeng – a processor also developed by the National University of Defense Technology for supercomputing. Naturally, Kunlun 2 is also compatible with Baidu’s own deep learning framework, PaddlePaddle.

So far, no further details have been given regarding the chips, only that they will be used in various scenarios, including autopilot. The Kunlun family of processors can be traced back to Baidu’s earliest effort to design chips in 2017. Back then, Baidu cooperated with Xilinx and unveiled its FPGA-based, 256-core XPU.