SEOUL (THE KOREA HERALD/ASIA NEWS NETWORK) - Memory chip giant Samsung Electronics has invested in a Chinese artificial intelligence technology developer as part of its efforts to catch up in the intensifying global race for AI platforms and devices, according to sources Sunday (Oct 22).
In August, Samsung made a significant investment in DeePhi Tech, a Beijing-based start-up with deep learning technologies, a source told The Korea Herald.
"Samsung made quite a large investment in DeePhi Tech, which a group of Korean developers who met the start-up's officials found surprising," the source said.
"Samsung appears to have made considerable investment in the Chinese company, in line with the Chinese government's move to foster home-grown AI tech firms. The investment could have been made out of political consideration, not necessarily over the company's technological prowess."
It is the second investment Samsung has made in foreign AI tech start-ups following a US$30 million (S$40.8 million) investment in the UK-based Graphcore last October.
Samsung is listed as one of the partner businesses on DeePhi Tech's homepage, along with China's fabless chipmaker Mediatek, US fabless Xilinx and cloud service provider Amazon Web Services.
DeePhi Tech is a deep learning solution developer that was founded in 2016 by four co-founders who graduated from Tsinghua University and Stanford University.
The company drew attention from tech giants like Samsung due to its neural network compression technology and neural network hardware architecture.
SK Telecom, a Korean mobile carrier that is also passionate about its emerging AI business, had also offered an investment to the Chinese start-up this summer, but the company turned down the offer, the source said.
DeePhi Tech offers the Deep Neural Network Development Kit, known as DNNDK, which is a deep learning software development kit designed as an integrated framework that aims to simplify and accelerate deep learning applications.
Samsung is reportedly interested in the company's neural network-based AI chipsets for portable devices, which allow instantaneous speech recognition, neural language processing and other recognition tasks on smartphones.
Such AI chipsets, known as neural processing units, or NPUs, are at the height of market attention as a means to widen the use of AI programs, especially in smartphones.
At a seminar held at the Semiconductor Exhibition 2017 on Wednesday, Kim Il San, a principal engineer at Samsung Electronics' system LSI business, said the adoption of NPUs in smartphones will be mainstream in the near future.
"Because it is inefficient to have the current CPUs in smartphones to process deep learning algorithms, adding NPUs will become a mainstream trend," Kim said. "It is still difficult to comment on NPU adoption in the following Samsung Galaxy phones."
Samsung's semiconductor business appears to be ramping up its R&D efforts on AI chipsets lately, after an announcement by Huawei's AI chip Kirin 970-equipped Mate 10 smartphone series.
"Not only the system LSI business, but also Samsung's mobile communications business are increasing research on the development of on-device AI chips at their own level," said Yoo Hoi Joon, a professor at the Korea Advanced Institute of Technology.
