➤ South Korea's largest mobile carrier launched its first artificial intelligence chip in late 2020.
➤ The chip is being used in carrier services and media content delivery, among others.
➤ The company is looking to work with semiconductor specialist SK Hynix on its next chip.
Vice president and head of T3K Innovation CO at SK Telecom, Lee Jong-min.
Not content with being the first country in the world to launch commercial 5G services in 2018, the South Korean government recently made 600 billion won, or US$540 million, available to local companies for the development of data-processing chips to support industrial automation.
SK Telecom Co. Ltd., the country's largest mobile carrier, launched its first such chip in November 2020.
S&P Global Market Intelligence spoke to the Head of one of SK Telecom's R&D units, T3K Innovation CO, Lee Jong-min, about how the company's entry into the AI chip space will enable it to develop advanced 5G use cases that can be exported around the world.
This interview has been edited for length and clarity.
S&P Global Market Intelligence: Many telecom carriers have adopted artificial intelligence as a way of managing 5G networks, but stopped short of launching their own AI chips. Why did SK Telecom decide to do both?
Lee Jong-min: We believe the convergence between mobile networks and AI will become the most important driver in the Fourth Industrial Revolution (the automation of manufacturing and industrial practices).
We expect AI to have the biggest impact on edge computing, which has become an area of focus for many telecommunications companies. The edge enables the provision of high-quality, ultralow-latency AI services regardless of a user's device type and specification.
Also, as the largest mobile operator in Korea, we can offer new and innovative services by converging our in-house AI technologies, mobile networks and massive amounts of data collected through our network infrastructure.
How have you been combining 5G and AI?
On one hand, for instance, we are offering "5G-AI Machine Vision," which is a solution that connects factories to industrial robots via 5G network. The network enables zero- or low-latency communications between production lines and supervision or analytical software.
On the other hand, our AI-based operational supporting system, named TANGO, controls our network quality and prevents network errors on a real-time basis. TANGO allows us to operate 5G networks in an efficient and stable manner despite network traffic hikes and irregularities caused by COVID-19.
Moreover, hardware-wise, we will combine our AI chip SAPEON X220 with our 5G mobile edge cloud infrastructure.
What are the application scenarios of the AI chip?
We will first apply SAPEON X220 to our Siri-like AI assistant service, "NUGU," as well as a variety of services of our affiliates such as media content selection and media resolution upgrading service, "Supernova," and cloud-based video surveillance service, "T View."
We plan to expand overseas as well. We will begin using our AI chip on Cast.era, a joint venture between SK Telecom and Sinclair Broadcast Group Inc., the owner of a number of TV stations in the U.S.
How will the AI chip help with Cast.era and other media content delivery?
The chip equips Cast.era with the capacity to develop software for broadcasting and over-the-top services. Cast.era will enable Sinclair and other media platforms to offer personalized content, where AI shows what the user wants to see from movies to TV shows. Cast.era will also help analyze data collected from the users' behavior.
For other instance, AI chip enables real-time media resolution upgrade, content picking and recommendation to the television service of our media subsidiary, SK Broadband. Our broadcast partner Munhwa Broadcasting Corp. (MBC) (KR) also uses SK Telecom's service.
You mentioned earlier that SK Telecom aims to make AI affordable. How are you going to achieve that?
With an integrated ecosystem where we develop and apply the technologies in a circle, the cost will be well controlled. We also work with government and partner companies to reduce cost.
We are leading the project assigned by the Ministry of Science and ICT of South Korea to develop AI semiconductors.
With project partners like SK hynix Inc. and Korea Electronics Technology Institute, we are currently developing an AI chip that is competitive both in terms of cost and performance when compared to graphics processing units, or GPUs, for inference in data centers.
Lastly, where do you think your offering sits within the wider chip market?
GPUs take up the lion's share of the AI chip market, where there is an established ecosystem in place for it. The biggest challenge for us is to create a well-established industrial ecosystem based on our own chips.
To distinguish ourselves, we try to stress the point that we offer both hardware and software support.
For example, we have cloud infrastructure, automated AI modeling technology, and Lightning Memory-Mapped Database, which is a technology that can be used to immediately store and search data collected via 5G network for machine learning.
The AI semiconductor market is in its early stages and is expected to expand fast, and therefore provides many opportunities to our businesses.