Samsung Electronics CEO Kyung Kye-hyun in charge of chip business hinted at the second generation of its own “Mach” AI inference chip in an Instagram post Friday.
“Client interest in inference-committed Mach-1 is increasing,” Kyung said.
“Some of the clients want to use Mach in large-scale applications with more than 1 trillion parameters, which justifies the faster-than-expected development of Mach-2. We should get down to preparation.”
The Mach-1 chip was first mentioned at the recent shareholder meeting of Samsung Electronics without much detail except that it will be used for inference purposes and will launch in early next year.
It was the first time a specific name for the company’s own AI chip was disclosed.
Kyung also said in the post that Samsung is close to claiming leadership in high bandwidth memory (HBM) chips.
“We have set up a dedicated team and are ramping up productivity and quality and due to their efforts, HBM leadership is coming to us,” he said.
“For AI applications, high-capacity HBM is competitive, which is why clients want 12H (12-layered) HBM3 and HBM3E.”
Samsung Electronics was the first to announce the development of 12-layered HBM3E early this year, with plans for mass production in the first half of the year. Rivals SK hynix and Micron are currently dedicated to 8-layered HBM3E chips.
BY JIN EUN-SOO [jin.eunsoo@joongang.co.kr]
