"Until just 2 years ago, artificial intelligence (AI) was centered on the cloud. Even now, the cloud accounts for a large share of the AI market. But the changes in our daily lives ultimately happen around "Edge AI." Over the past year and 6 months, I have felt that more customers have begun adopting more Edge AI."
Amichai Ron, executive vice president at Texas Instruments Incorporated (TI), visited Korea on the 21st on the previous month and said this in an interview with ChosunBiz. He oversees embedded processing and digital light processing (DLP) optical semiconductor products.
U.S. semiconductor company TI is considered a leader in analog chips that measure continuous signals such as voltage and temperature, and in embedded semiconductors that control in-device functions such as sensors and communications. TI's roughly 80,000 products are used by more than 100,000 corporations. First-quarter revenue this year was $4.825 billion (about 7.16 trillion won), and operating profit was $1.808 billion (about 2.68 trillion won).
◇ Rolling out an "embedded intelligence" strategy as the Edge AI market expands
Ron said the fierce battle for AI leadership that raged on the cloud is gradually moving inside devices. As performance gains through Edge AI (technology that runs AI functions directly on devices where data is generated) are beginning to be felt in everyday life, the related market is growing rapidly. TI has set a strategy to deliver "embedded intelligence" products—devices that judge and act on their own—to the market in a timely manner to produce results.
Ron said, "Ten to fifteen years ago, the key question was how to connect more devices to the cloud; now, how intelligently each device operates has become more important," adding, "How efficiently a battery is used, and how precisely and safely a device makes decisions, have become the central issues." He added, "Using Edge AI enables machines to make more intelligent decisions and improves user experience," and "Edge AI is being integrated into an increasing number of machines—from computers and smartphones to washing machines, refrigerators, and even electric toothbrushes."
In cloud-based AI such as ChatGPT and Gemini, memory and compute capacity act as "bottlenecks" that hinder performance gains. Ron said that although different in kind, bottlenecks also appear in Edge AI. "Customers often have deep knowledge in their domains but lack experience in implementing AI," he said. "This knowledge gap is acting as a bottleneck to the spread of Edge AI."
To address this, TI provides training and development tools for customers adopting Edge AI. In addition to Code Composer Studio, an integrated development environment (IDE) for embedded processors, TI recently launched Edge AI Studio. Ron said, "It's a program that lets you implement desired functions without expertise in AI algorithms," adding, "TI's role extends to helping customers deploy Edge AI on actual devices."
◇ "Choosing products with high performance per watt is crucial to realize Edge AI"
Ron said selecting chips with high performance per watt (performance relative to power) is important for spreading Edge AI. That is because squeezing maximum performance needed for AI implementation from limited device power translates into better convenience. "Chips that consume a lot of power also generate a lot of heat, adding cooling expense," Ron said. "AI computation that uses device resources at scale ends up undermining product competitiveness."
TI is focusing on developing products that meet customers' low-power requirements while delivering high AI compute capabilities. The Neural Processing Unit (NPU) TinyEngine and the TDA5 AI chip for autonomous vehicles are representative.
Ron said, "TI's microcontrollers (MCUs—integrated circuits that miniaturize by integrating a central processing unit and related modules onto one chip) with the TinyEngine NPU integrated can reduce latency per AI inference by up to 90 times and energy use by more than 120 times compared with existing solutions," adding, "Designed for power efficiency, the TDA5 chip can be applied across various autonomous driving levels, which is its strength."
The TDA5 chip delivers up to 1,200 TOPS (trillion operations per second; 1 TOPS is 1 trillion operations per second). It supports 24 TOPS per watt. Automakers often have to consume more than 150 watts (W) to achieve 400 TOPS of compute, but the TDA5 can run 400 TOPS at under 50 W. The following is a Q&A with Ron.
- What is the purpose of your visit to Korea?
"Korea is a market I visit often because it has many innovative customers. I visited to hear what technologies customers need and what support they need from TI to make better products."
- What are customers most asking for in the Edge AI market?
"It varies by industry. There are already hundreds of Edge AI use cases. The application areas differ, but the impact of adopting Edge AI is clear. Take solar power systems, for example. In this field, it's critical how quickly and accurately you detect anomalies or errors. That's because the longer two wires in a solar installation are in contact—even by 1 millisecond (ms) increments—the higher the risk of fire can become. Applying Edge AI can raise accuracy from the existing 85%–90% range to as high as 99%.
Security systems are another example. You need an alarm when a person enters, but you don't need to respond when a single cat comes in. Using Edge AI lets you distinguish between a person entering and a cat entering. It creates an environment where devices can respond appropriately to the situation."
- Recent Middle East wars and tariffs have also made supply chain stability more important.
"It's very important to deliver good products to customers at the promised time. TI has its own manufacturing capabilities and has continued to invest over the past five years to maintain and expand capacity. With production bases in Texas, Utah, Malaysia, and the Philippines, we can supply products stably. This has allowed us to respond to the surge in demand for AI data centers.
Regulations such as tariffs expose competitors to the same environment, so it's hard to say any particular corporations are advantaged or disadvantaged. However, the regulatory environment is becoming more complex over time. TI aims to adapt to this complexity while maintaining competitiveness."
- What value do you want to deliver to customers with embedded intelligence?
"It can be summed up as "helping customers build better systems." Customers decide what a better system is. TI helps customers build the systems they want through a wide range of products, high quality, development tools, and training."