[Peace&Chips] 'Focus on Neural Networks Following Graphics'... AI-Optimized Semiconductors Flood the Market

Custom-Designed AI Semiconductors for Learning and Inference
Overcoming High Power Consumption Limits of GPUs
Emergence of Various Types Including NPU, IPU, and LPU

Editor's NoteSemiconductors, often called the rice of modern industry. Although it's a term we hear every day, it’s hard to explain when asked. Peace & Chips will make the complex concepts and overall trends of the semiconductor industry easy to digest. Just bring your spoon.

If we were to pick the biggest topic in the semiconductor industry this year, it would undoubtedly be 'Artificial Intelligence (AI)'. With the rise of generative AI demand following the emergence of ChatGPT, AI semiconductors have attracted significant attention. Just last week, the US AI semiconductor company Groq announced that it will produce products at Samsung Electronics’ foundry (semiconductor contract manufacturing) plant in Taylor, Texas, drawing considerable attention.

In the market, AI semiconductors sometimes refer broadly to system semiconductors and memory semiconductors that enable AI technologies or services, but the original meaning is 'semiconductors specialized for AI computation.' It refers to system semiconductors designed to suit the characteristics of AI, which involves learning and reasoning from large-scale data.

The Graphics Processing Unit (GPU), the star of the AI boom, does not perfectly fit the concept of AI semiconductors. GPUs are widely used for AI because they can process simple calculations in large volumes, but as the name suggests, they were originally created for graphic processing tasks. It is said that GPUs have limitations in meeting the diverse technical demands arising from AI development.

The industry is focusing on Neural Processing Units (NPU). NPUs are system semiconductors optimized for AI tasks such as deep learning. They are characterized by lower power consumption and higher performance relative to the cost compared to GPUs. Assuming the same task is processed, NPUs are said to be more efficient than GPUs.

What is noteworthy is that various AI semiconductors are gradually emerging centered around NPUs. There are already several products such as Intelligence Processing Units (IPU, semiconductors that increase data processing speed by integrating memory around the processor), Language Processing Units (LPU, semiconductors specialized for AI inference), and Data Processing Units (DPU, semiconductors that enhance data center processing speed). Although the detailed technologies differ, they share the common goal of maximizing AI applications.

The industry explains, "In the past, AI was image-centric, but recently, with the development of language-based AI technologies, AI semiconductors are evolving to support this direction." Not only fabless companies (semiconductor design) but also giant big tech companies like Google have entered related businesses, bringing various technologies and product news.

Kim Hyung-jun, head of the Next-Generation Intelligent Semiconductor Business Group, said, "Although GPUs are widely used in data centers, their power consumption is very high, so in the future, AI semiconductors will be developed to reduce power consumption while increasing computation speed," adding, "As companies try various approaches, the number of products and types may increase."

The industry’s expectations for the AI semiconductor market are high. Market research firm Gartner predicted that the AI semiconductor market size, which was $44.4 billion last year, will grow to $86.1 billion by 2026. Samsung Electronics’ foundry seems eager not to miss this business opportunity and has recently started actively collaborating with domestic and international AI semiconductor companies such as DeepX, Rebellion, and FuriosaAI.

This article is from [Peace & Chips], published weekly by Asia Economy. By clicking subscribe, you can receive articles for free.

☞Subscribe

This content was produced with the assistance of AI translation services.