DeepX meets AMD... Physical AI and heterogeneous computing cooperation

DeepX, a domestic AI semiconductor startup, met with senior executives at the headquarters of American semiconductor company AMD to discuss the possibility of cooperation in the fields of next-generation physical artificial intelligence (AI) and heterogeneous computing.

DeepX announced on the 3rd that officials including AMD Global AI Market Executive Vice President Keith Stryer and AMD Korea CEO Jang Jeong-hee visited its headquarters in Pangyo, Seongnam-si, Gyeonggi-do, and met with executives including CEO Kim Nok-won.

This meeting is a follow-up discussion held after AMD CEO Lisa Su and CEO Kim Nok-won met at the 'AMD/Domestic AI Semiconductor Company Roundtable' held on the 7th of last month.

Following the visit of AMD Korea's working-level staff on the 21st of the same month, discussions on technological cooperation began in earnest with the visit of the company's senior vice president, who oversees global AI strategy and public-private cooperation, in person.

At the meeting, DeepX proposed two key cooperation plans targeting the physical AI market.

First, the plan is to combine DeepX's ultra-low-power NPU (neural network processing unit) with the AMD Kria platform, an embedded module.

Accordingly, the idea is to simultaneously implement real-time precise control of the robot and on-device AI recognition and inference functions.

In addition, cooperation was proposed linking AMD's GPU open software platform ROCm with DeepX's dedicated SDK (software development kit).

The plan is to build an efficient heterogeneous computing ecosystem in which the central processing unit (CPU), graphics processing unit (GPU), and NPU share tasks according to their respective computational strengths.

Kim Nok-won, CEO of DeepX, said, "As physical AI spreads to industrial sites such as robots, the importance of heterogeneous computing that maximizes the advantages of each processor is growing. Through cooperation with AMD, we will continue to seek ways to link DeepX's low-power NPU with various computing environments."