Samsung secures 3nm GAA new orders amid persisting foundry revenue challenges

Amanda Liang, Taipei; Vyra Wu, DIGITIMES Asia 0

Credit: Samsung

According to Business Korea, there are rumors suggesting that Samsung foundry, including its System LSI division, may experience losses of up to 1 trillion KRW (approximately $780 million USD) in the first half of 2023. Despite these rumors, during the 2Q23 earnings conference, the top executives of Samsung Electronics did not directly address this issue.

In an unusual move, Samsung openly acknowledged that they are currently facing significant challenges with their gross profit margin. Nevertheless, the company started to produce its third 3nm chip design applying gate-all-around (GAA) transistor architecture.

Samsung's foundry profit plunges

Gibong Jeong, executive vice president of Samsung Foundry, has highlighted that the division is currently grappling with a profitability crisis. This is mainly attributed to the decrease in capacity utilization resulting from ongoing factory expansions and the heightened uncertainty surrounding short-term demand. To tackle this challenging situation, Samsung Foundry has placed significant emphasis on cost improvement and overall profitability management throughout the year 2023.

While Samsung has not publicly disclosed specific revenue and profit figures for the foundry business independently, this division's performance is combined within Samsung's Device Solutions (DS) Division, along with other memory businesses like DRAM and 3D NAND. However, examining the DS Division consistent losses of 4.58 trillion and 4.36 trillion KRW in the past two quarters, which roughly translates to an operating deficit of about 3 billion USD per quarter, it becomes evident that the foundry division is facing substantial challenges in avoiding losses.

The global economic downturn and ongoing customer inventory adjustments have resulted in sluggish demand for mobile, PC and other end-user devices, leading to reduced capacity utilization at Samsung foundry. This situation has become a critical factor contributing to the division's current profitability crisis. While Samsung might not have secured AI orders during this economic downturn, unlike what TSMC claims, the company's top executives remain confident that the long-term capacity will be filled by emerging trends such as High-Performance Computing (HPC) and 5G. Moreover, Samsung is to focus attention on the current growth in demand for AI-related technologies, and this has prompted them to observe an increasing interest and active participation from customers in HPC and 5G applications that utilize their GAA process.

Certainly, it has been one year since Samsung started mass production of the 3nm GAA process. While Samsung has not disclosed the specific customers using this process, a research institution recently revealed that Chinese crypto-mining equipment manufacturers are utilizing the 3nm GAA process through reverse engineering. Despite facing financial challenges, Samsung's top executives expressed their satisfaction with achieving mass production of the third 3nm GAA product. Samsung has taken a leading position in the industry by venturing into the demanding path of GAA process yield improvement, which has shown significant improvements in yield rates compared to previous FinFET generations. This is crucial for Samsung's 2nm technology to compete with TSMC and Intel's 2nm GAA processes in the 2024 to 2025 timeframe.

Samsung looks to challenge TSMC

In contrast to TSMC, which has secured nearly all high-end GPU and data processing unit (DPU) orders for AI accelerators, what is Samsung's strategy for AI-related foundry and advanced packaging? According to Samsung foundry's top executives, the key to these AI chips lies in achieving high performance and high energy efficiency. This is precisely what Samsung's GAA process, which has been in mass production since June 2022, aims to achieve for AI chips at an advanced node. The executives further pointed out that current AI chips are often integrated with high-bandwidth memory (HBM) in a single package.

Samsung's strategic layout currently involves having both GAA process technology and HBM memory, and they are rapidly expanding their advanced packaging capacity.

Various South Korean media outlets have previously reported that Samsung foundry is expected to share a portion of the orders for high-end AI GPUs, including those from Nvidia and AMD, which are traditionally handled by TSMC. NVIDIA was even mentioned to be adopting Samsung's HBM3 memory. AMD has been collaborating with Samsung since late 2022, using their Programmable In-Memory Computing (PIM) solution called HBM-PIM, which optimizes AI engines through DRAM within the memory, achieving parallel processing and minimizing data movement. As for the collaboration between Samsung and AMD, whether it will lead to gaining a share of TSMC's orders remains to be seen and is still subject to observation.