➀ Samsung Electronics' 4nm advanced process yield has approached 80% and recently received ASIC foundry orders from Chinese companies.

➁ The new management team has adjusted the strategy, no longer chasing the forefront competition, and focusing on winning customers with high yield.

➂ The popularity of the AI development platform DeepSeek has ignited the enthusiasm of Chinese technology companies to develop AI ASICs, with Samsung's price and capacity advantages becoming a key driving force for orders.

➀ Alchip预计2024年营收将超过10亿美元，实现重大里程碑；

➁ 公司专注于先进技术，包括2nm测试车间的发布和3DIC设计流程的准备工作；

➂ Alchip通过灵活且强大的3DIC设计流程应对先进封装的挑战，优化功率传输、互连和热特性。

➀ Arm is reported to be planning to sell proprietary chips, with Meta being one of its first customers.

➁ The move is driven by the need for hyperscalers to use ASICs in datacentres to avoid high costs associated with Nvidia GPUs.

➂ Arm's recent trial with Qualcomm highlighted its CEO's strategy document, which included plans for Arm to build its own chips.

➀ EnSilica has been awarded £10.38 million over three years from the UK Space Agency's C-LEO programme.

➁ The funding will support the development of a chip family for future satellite broadband user terminals.

➂ The terminals will connect with various satellite constellations and utilize advanced semiconductor technology.

➀ The rise of ASICs in the capital market is challenging the dominance of GPUs in AI computing;

➁ ASICs and GPUs are both semiconductor chips used for computing, with ASICs being dedicated to specific tasks;

➂ The performance and efficiency of ASICs are highly matched to the task algorithms, making them more reliable and energy-efficient than general-purpose chips like GPUs;

➃ NVIDIA's GPUs have a strong market position due to their performance, ecosystem, and integration capabilities, but the rise of ASICs presents opportunities for diversification in AI computing.

➀ China currently lacks a company that can fully match Broadcom in the AI chip ASIC field;

➁ Broadcom, as a global leader in semiconductor and infrastructure software solutions, has made significant achievements in AI;

➂ Baidu, Apple, and AMD have grown through self-developed technology, while Broadcom has focused on acquisition to become a tech giant.

➀ Marvell's market capitalization surpassed $100 billion for the first time after releasing its Q3 fiscal year 2025 financial results;

➁ Marvell's ASIC business is a key driver of its growth, with over 2,000 custom chips designed over 25 years;

➂ Marvell's custom AI chip services are becoming a major revenue source, with significant partnerships and a goal of capturing 20% of the $40 billion custom chip market in data centers;

➃ Marvell's interconnect business is growing rapidly, with advancements in PAM4 SerDes technology and optical interconnect solutions;

➄ Marvell's 3nm optical DSP and 5nm PCIe retimer products are set to drive future growth;

➅ Marvell's 51.2T Ethernet switch, Teralynx 10, is designed for low power and high performance in data center environments;

➆ Marvell's focus on AI and data center technologies is expected to continue driving growth and market capitalization.

➀ Computing power is an important indicator of a computer's information processing capability, with AI computing power focusing on AI applications, commonly measured in TOPS and TFLOPS, and provided by dedicated chips such as GPU, ASIC, and FPGA for algorithm model training and inference.

➁ AI chip accuracy is a way to measure computing power level, with FP16 and FP32 used in model training, and FP16 and INT8 used in model inference.

➂ AI chips typically use GPU and ASIC architectures. GPUs are the key components in AI computing due to their advantages in computation and parallel task processing.

➃ Tensor Core, an enhanced AI computing core compared to the parallel computation performance of Cuda Core, is more focused on the deep learning field and accelerates AI deep learning training and inference tasks through optimized matrix operations.

➄ TPUs, a type of ASIC designed for machine learning, stand out in high energy efficiency in machine learning tasks compared to CPUs and GPUs.