Akira Amari, a Japanese lawmaker and leader of the parliamentary association to promote semiconductor strategy, reveals that Japan is gearing up to allocate a subsidy of up to 900 billion yen for the second-phase expansion of TSMC’s Kumamoto fab. The plan involves transitioning from the 22/28 nm and 12/16 nm processes to the more advanced 7 nm process. Once completed, Japan is anticipated to emerge as the leading semiconductor supply hub globally.

Media reports suggest that the cabinet amendment is expected to allot a total of 1.9 trillion yen for semiconductor subsidies in Japan. Japanese companies are slated to receive 590 billion yen, while TSMC’s second-phase expansion project in Kumamoto is in line for the highest subsidy of 900 billion yen, surpassing the market’s earlier projection of 760 billion yen.

Highlighting the unprecedented nature of this subsidy, Amari underscores the imperative of ensuring companies’ operational profitability. Japan envisions becoming a pivotal player in the semiconductor supply chain. Furthermore, contingent on the development scenario, the government is committed to evaluating subsidy reductions, with a pledge to support various schemes for establishing Japan as a long-term semiconductor hub.

As of now, the construction of TSMC’s first-phase fab in Kumamoto is advancing rapidly, with the total workforce anticipated to surpass a thousand. The team is preparing for a timely production launch in 2024.

Although the Kumamoto fab’s announcement and construction preceded that of the U.S. Arizona fab, set to commence production in 2025, TSMC’s Kumamoto fab is garnering robust support from official Japanese channels and partners including SONY Semiconductor Solutions and Denso. The fab is set to utilize 22/28 nm and 12/16 nm processes, with a total capital expenditure of 8.6 billion USD. The Japanese Ministry of Economy, Trade and Industry(METI) granted approval for a subsidy of 476 billion yen in June 2022, which represents approximately 40% of total capital expenditure is supported by the subsidy.

Explore more

• [News] Japanese Lawmaker Confirms Additional ¥900 Billion Subsidies for TSMC’s Kumamoto 2nd Fab
• [News] TSMC’s Global Expansion: Progress in New US, German, and Japanese Fabs

NVIDIA and Intel are adapting to the latest U.S. chip export restrictions by introducing downgraded AI chips specifically tailored for the Chinese market, UDN News said.

According to insider from the China Star Market, a Chinese media, NVIDIA has developed three downgraded AI chip models for the Chinese market. Intel also plans to release downgraded Gaudi 2 chip with an aim to US restriction.

NVIDIA’s latest downgraded AI chips, including HGX H20, L20 PCle, and L2 PCle, are anticipated to be unveiled after November 16th. Chinese companies are likely to receive samples in the coming days. These three chips, derived from the modification of NVIDIA H100, will align their performance with parameters below the new U.S. regulations. Ongoing communication with NVIDIA suggests mass production is slated for the year-end, said by industry sources.

Besides, Yicai also confirms from multiple NVIDIA supply chain sources. The three AI chip products are designed for cloud training, cloud inference, and edge inference, with specific launch times pending confirmation. Sampling is projected between November and December this year, followed by mass production from December this year to January next year.

On the Intel front, there are rumors of a response plan. As reported by The Paper, Intel is planning to release an improved version of its Gaudi 2 chip. Although the rumor exists, specific details are yet to be disclosed.

Since the U.S. government introduced chip export control to China last year, NVIDIA initially designed downgraded AI chips A800 and H800 for Chinese companies. However, new regulations in October this year by the U.S. Department of Commerce brought A800, H800, L40S, and other chips under control. Failure to secure export permission may necessitate order cancellations for NVIDIA.

(Image: Nvidia)

On October 17, 2023, the U.S. government once again expanded its restrictions on the export of semiconductor devices and products to China. The newly added control conditions now encompass NVIDIA’s L40S, A100, H100H800, as well as general-purpose AI server GPUs tailored for the Chinese market, such as A800 and H800. Additionally, AMD’s MI200 series, MI300 series GPUs, and Intel’s Habana Labs’ Gaudi 2, Gaudi 3 GPUs fall under the regulatory framework.

Recalling the U.S. government’s export restrictions on AI chips issued to IC design firms in September 2022, at that time, only A100, H100, and MI200 series were subjected to control, and the U.S. Department of Commerce granted NVIDIA and AMD a one-year buffer period.

In contrast, the recent regulations not only cover all mainstream AI server GPUs but also eliminate the buffer period for these chip companies. In essence, companies or institutions in countries not permitted for export can only acquire AI server chips with performance potentially inferior to NVIDIA L40S or AMD MI200 series for the next few years.

Furthermore, stricter control thresholds for lithography equipment have led to the inclusion of ASML’s DUV, the 1980Di, in the control list. This equipment is primarily used in the 28 ~ 7nm process. Previously controlled products were focused on the EUV 3000 series for 7nm and below processes and the DUV 2000 series for 16/14 ~ 5nm processes.

This move indicates that the U.S. government’s desire to control semiconductor process technology has officially extended to mature processes of 28nm.

The expanded U.S. controls on AI chips and semiconductor manufacturing devices not only target China but also countries that might collaborate with Chinese institutions and businesses in AI development.

In this scenario, China is left with only two viable options to establish efficient AI computing resources: (1) designing and mass-producing AI server chips itself or (2) utilizing the computing resources of cloud service providers.

As the U.S. is also discussing the potential inclusion of cloud service providers in semiconductor control policies and currently formulating relevant countermeasures, this path remains unreliable for China. Therefore, the only dependable option is to independently design and manufacture AI server chips.

Read more

• [News] Nvidia says US speeded up new export restrictions on AI chips
• [News] Chinese Chip Equipment Rises Amid U.S. Restrictions

Wistron experienced a slowdown in shipments for product lines like PCs and displays in October, following the prior demand surge. However, their GPU-related AI server products continue to maintain their growth trajectory. Simultaneously, Wiwynn, a subsidiary of Wistron, witnessed a remarkable 20% month-over-month revenue increase due to the rising momentum in AI server-related project shipments, positioning them at the third-highest monthly revenue level in their history for the same period, reported by CTEE.

Both Wistron and Wiwynn hold an optimistic outlook for their AI server products, expecting the growth momentum to extend into the next year. In contrast, they foresee a return to growth trends for non-AI general-purpose servers and cloud data center servers next year, while AI server growth is expected to remain notably strong.

Wistron plays a pivotal role in the AI server supply chain and remains unaffected by high-end GPU shortages and U.S. export restrictions. Shipments in Q4 continue to exhibit consistent month-to-month growth, and the anticipated trend to peak in the second half of the year remains steadfast. Moreover, there are indications of a slight seasonal increase in general-purpose servers in Q4.

In a recent earnings call, Wiwynn maintains an optimistic stance for Q4 and the upcoming year. With the evident growth momentum from AI servers, they anticipate that developments in AI-related projects will lead to a continuous improvement in AI server product shipments.

Furthermore, Wiwynn’s third-largest customer business and AI server revenue both exceeded 10% in the third quarter, marking a significant milestone for the company. Back in October, Wiwynn had set up  a server plant in Malaysia to meet the surging demand for AI servers.

According to TrendForce’s anticipation, in 2023, the shipment of AI servers (including those equipped with GPUs, FPGAs, ASICs, etc.) is expected to exceed 1.2 million units, with a year-on-year increase of 37.7%, accounting for 9% of the total server shipments. In 2024, it is projected to further grow by more than 38%, with shipments reaching approximately 1.676 million units, and the share of AI servers will exceed 12%.
(Image: Wistron)

Explore more

• [Insights] Quanta, Wiwynn, and Major Manufacturers Scale Up to Meet Rising Demand for AI Servers
• [News] Rumors Suggest All Nvidia B100 Baseboard Manufacturing Orders Are Secured by Wistron

The fusion of AIGC with end-user devices is highlighting the importance of personalized user experiences, cost efficiency, and faster response times in generative AI applications. Major companies like Lenovo and Xiaomi are ramping up their efforts in the development of edge AI, extending the generative AI wave from the cloud to the edge and end-user devices.

On October 24th, Lenovo hosted its 9th Lenovo Tech World 2023, announcing deepening collaborations with companies like Microsoft, NVIDIA, Intel, AMD, and Qualcomm in the areas of smart devices, infrastructure, and solutions. At the event, Lenovo also unveiled its first AI-powered PC. This compact AI model, designed for end-user applications, offers features such as photo editing, intelligent video editing, document editing, and auto task-solving based on user thought patterns. 

Smartphone manufacturers are also significantly extending their efforts into edge AI. Xiaomi recently announced their first use of Qualcomm Snapdragon 8 Gen 3, significantly enhancing their ability to handle LLMs at the end-user level. Xiaomi has also embedded AI LLMs into their HyperOS system to enhance user experiences.

During the 2023 vivo Developer Conference on November 1st, vivo introduced their self-developed Blue Heart model, offering five products with parameters ranging from billions to trillions, covering various core scenarios. Major smartphone manufacturers like Huawei, OPPO, and Honor are also actively engaged in developing LLMs.

Speeding up Practical Use of AI Models in Business

While integrating AI models into end-user devices enhances user experiences and boosts the consumer electronics market, it is equally significant for advancing the practical use of AI models. As reported by Jiwei, Jian Luan, the head of the AI Lab Big Model Team from Xiaomi, explains that large AI models have gain attention because they effectively drive the production of large-scale informational content. This is made possible through users’ extensive data, tasks, and parameter of AI model training. The next step in achieving lightweight models, to ensure effective operation on end-user devices, will be the main focus of industry development.

In fact, generative AI’s combination with smart terminal has several advantages:

1. Personal data will not be uploaded to the cloud, reducing privacy and data security risks.
2. AI models can connect to end-user databases and personal information, potentially transforming general AI LLMs into personalized small models, offering personalized services to individual users.
3. By compressing AI LLMs and optimizing end-user hardware and software, edge AI can reduce operating costs, enhance response times, and increase service efficiency.

Users often used to complain about the lack of intelligence in AI devices, stating that AI systems would reset to a blank state after each interaction. This is a common issue with cloud-based LLMs. Handling such concerns at the end-user device level can simplify the process.

In other words, the expansion of generative AI from the cloud to the edge integrates AI technology with hardware devices like PCs and smartphones. This is becoming a major trend in the commercial application and development of large AI models. It has the potential to enhance or resolve challenges in AI development related to personalization, security and privacy risks, high computing costs, subpar performance, and limited interactivity, thereby accelerating the commercial use of AI models.

Integrated Chips for End-User Devices: CPU+GPU+NPU

The lightweight transformation and localization of AI LLMs rely on advancements in chip technology. Leading manufacturers like Qualcomm, Intel, NVIDIA, AMD, and others have been introducing products in this direction. Qualcomm’s Snapdragon X Elite, the first processor in the Snapdragon X series designed for PCs, integrates a dedicated Neural Processing Unit (NPU) capable of supporting large-scale language models with billions of parameters.

The Snapdragon 8 Gen 3 platform supports over 20 AI LLMs from companies like Microsoft, Meta, OpenAI, Baidu, and others. Intel’s latest Meteor Lake processor integrates an NPU in PC processors for the first time, combining NPU with the processor’s AI capabilities to improve the efficiency of AI functions in PCs. NVIDIA and AMD also plan to launch PC chips based on Arm architecture in 2025 to enter the edge AI market.

Kedar Kondap, Senior Vice President and General Manager of Compute and Gaming Business at Qualcomm, emphasizes the advantages of LLM localization. He envisions highly intelligent PCs that actively understand user thoughts, provide privacy protection, and offer immediate responses. He highlights that addressing these needs at the end-user level provides several advantages compared to solving them in the cloud, such as simplifying complex processes and offering enhanced user experiences.

To meet the increased demand for AI computing when extending LLMs from the cloud to the edge and end-user devices, the integration of CPU+GPU+NPU is expected to be the future of processor development. This underscores the significance of Chiplet technology.

Feng Wu, Chief Engineer of Signal Integrity and Power Integrity at Sanechips/ZTE, explains that by employing Die to Die and Fabric interconnects, it is possible to densely and efficiently connect more computing units, achieving large-scale chip-level hyperscale computing.

Additionally, by connecting the CPU, GPU, and NPU at high speeds in the same system, chip-level heterogeneity enhances data transfer rates, reduces data access power, increases data processing speed, and lowers storage access power to meet the parameter requirements of LLMs.

(Image: Qualcomm)