Samsung Focusing on Neural Processing Units
Global semiconductor companies are increasing their investments in M&A and R&D in order to get the upper hand in the AI chip market, which is expected to grow from US$18.4 billion to US$117.9 billion for 10 years to come.
According to market research firm IC Insights, this year’s M&A volume in the global semiconductor industry recently rose to US$72 billion after Nvidia acquired ARM for US$40 billion and Intel decided to sell its NAND flash memory business to SK hynix for US$9 billion. In July, Analog Devices acquired Maxim Integrated for over US$20 billion. AMD’s acquisition of Xilinx is on the horizon.
Those big deals share the keyword of AI chip. Those companies are trying to take the lead by creating inter-segment synergy. For example, AMD’s share in the data center server CPU market stands at 2 percent or so although it is comparable to Intel in the PC CPU market. The main products of Xilinx include AI accelerators for data processing acceleration at datacenters. ARM, in the meantime, is specialized in AI-related chip design and Nvidia is expected to enter the autonomous driving and IoT industries through the takeover.
South Korean semiconductor companies are working on AI chip development and performance enhancement, too. Samsung Electronics’ R&D arms located in Austin and San Jose are concentrating on neural processing units (NPUs), which are optimized for deep learning. At present, NPUs are used for image and voice recognition and so on as a part of smartphone application processors. According to Samsung Electronics, the applications of NPUs will be expanded to cover self-driving and many more.
SK hynix’s acquisition of Intel’s enterprise SSD business is to enter the AI chip market as well. Solid-state drives (SSDs) are expected to be in increasing demand as the AI functions of data centers are being increasingly highlighted.