Author: Tronserve admin
Wednesday 4th August 2021 02:18 PM
S. Korea Doubles Down Development of Ai Chips to Cement Chip Leadership
South Korean IT industry has doubled down development of neural processing unit (NPU), or the memory behind the next-generation chip powering smart applications spanning from homes to roads and industrial sites.
Korea has the lead in the AI chip race as it is home to the world’s top two DRAM market leaders, Samsung Electronics and SK Hynix, and has 5G network in full swing.
AI chips often called neural processing units (NPUs) rely mostly on non-memory chips to process and send data in parallel as the human brain works.
AI semiconductor is essential in the AI era, when a huge amount of big data flows in real time at high speed. As with DRAM chips, the competitiveness in AI chips hinges on the technology to make them as small as possible.
Samsung Electronics laid out its 2030 semiconductor vision. When AI semiconductor technology becomes further sophisticated, memory and processors will ultimately be integrated into one. If Samsung takes the lead in AI semiconductor, it will create tremendous synergy combined with its global market leadership in memory technology, experts forecast.
Samsung Electronics aims to become a leader in system semiconductor by 2030 with a total of 133 trillion won ($108.7 billion) investment, which breaks down to 73 trillion won for developing system chips such as NPU and GPU and 60 trillion won for production facilities. It will also increase its current 200 top engineering brains to 2,000 by 2030.
Korea’s ICT policy is now headed by Science and ICT Minister Choi Ki-young, who has been researching a low-power memory semiconductor system for over 40 years. He is an expert in next-generation AI semiconductor technology. In December 2019, when Choi was picked as the new minister, the government announced a 10-year plan to foster the AI semiconductor industry. Of its 1 trillion won budget for the project, 247.5 billion won will be invested to help advance semiconductor designing. Four consortiums that include SK Telecom and dozens of private companies, universities and labs are planning to commercialize 10 AI chips by the target year.
[Photo provided by ETRI]
Some results are already in. State-run Electronics and Telecommunications Research Institute (ETRI) and SK Telecom recently developed a coin-sized AI chip that can crunch 40 trillion floating-point operations per second with a max power consumption of just 15W, a remarkable progress from AlphaGo that won games against Korea’s top Go player Lee Se-dol in 2016. The AlphaGo brain then performed 30 trillion integer operations per second with huge power consumption due to backup servers.
A typical smartphone processes 100 gigabyte data per second or perform 100 billion operations per second, according to Kwon Young-soo, Group Leader of the Processor Research Group in ETRI.
Both Google and Huawei developed AI chips that can perform 100 trillion operations per second, but they consume 300W at peak, which means the technology barrier is still a matter of fact before full commercial application.
The AI semiconductor market is expected to grow above the memory semiconductor market in 10 years. Market research firm Gartner predicts that the AI semiconductor market will grow to up to $34.3 billion in 2023, and Allied Market Research says that its global market value will reach $37.8 billion in 2025.