2003网站太阳集团(中国)官方网站

Enterprise News（July 10）

Intel temporarily suspends shipments for having detected bugs in its CPU.

TrendForce reported on July 10 that Intel had detected bugs in the 4th Generation Intel® Xeon® Scalable Processors (codenamed Sapphire Rapids), which began mass production in 1Q23.

Intel used two types of underlying designs on Sapphire Rapids: the XCC, created by four computing modules for a single chip, and the single-chip MCC. The former is intended for chips with 36 to 60 cores, and the latter for those with 32 cores. MCC is the Sapphire Rapids version with the highest yield. After identifying the issue, Intel has halted shipments of the dual- and quad-socket models. Most affected clients are in the corporate sector, making Intel proactively suspend shipments of SPR MCC SKUs. Other models, such as the 20/24C and 36C and higher, remain unaffected.

Overall, the DDR5 adoption rate is still constrained by clients’ extension of the product cycles of older models and delays in introducing new models. TrendForce estimates that the Server DDR5 adoption rate will be approximately 13.4% this year. And it is expected DDR5’s adoption rate will not surpass DDR4’s until the end of 3Q24.

Comment:

The impact of this situation is limited. As Enterprise Server OEMs have generally extended the support lifespan of older platforms, the adoption of SPR MCCs has significantly slowed down in 2023. The latest model-specific CPU issues mentioned above may also deepen corporate clients’ concerns and reduce their willingness to adopt SPR. Furthermore, in terms of scale, the suspended shipments in 2Q23 only account for less than 1% of the total number of Intel DP server CPUs, resulting in a very limited impact on the shipment performance of Intel server CPUs.

Enterprise News（July 12）

Intel announces to abandon its NUC business.

An Intel spokesperson wrote in a statement that Intel had decided to stop direct investment in the Next Unit of Computing (NUC) business and adjust its strategy to enable its ecosystem partners to continue NUC innovation and growth. This decision will not affect the rest of Intel’s Client Computing Group (CCG) or Network and Edge Computing (NEX) businesses.

Furthermore, Intel left the door open, stating it was cooperating with its partners and clients to ensure a smooth transition and the fulfillment of all current promises, including the continual support for the NUC products currently in market. This means Intel can still work with its partners to bring NUC or NUC-like products to market. It just will no longer design PCs.

In fact, Intel initially launched its NUC for sale in 2012. It is a compact yet highly-integrated computing device that delivers powerful performance and flexible deployment options. Meanwhile, the NUC usually uses Intel processors and other internal components, such as GPUs, RAMs, disks, and network cards.

Comment:

Intel did not mention any specific reason to stop the NUC in its statement, but many guess it is due to the impact of its financial difficulty. This April, Intel decided to close its Data Center Solutions Group responsible for designing and planning the server business. Furthermore, the NUC is not very practical for a broad consumer base. To purchase the Intel NUC, you would have to find a dealer, many of whom operate business-to-business, and you would have to get ready to spend upwards of US$1,000 for a nearly bare-bones unit. Therefore, it does not seem hard to understand that the NUC is being phased out. Although Intel is also investing in some non-processor product lines, the company’s future focus remains on its IDM 2.0 strategy.

Enterprise News（July 12）

Samsung’s 4nm yield increases to over 75%.

July 12 -- Samsung Electronics recently improved its 4nm process yield to over 75%, sparking speculation that it is expanding its semiconductor foundry clients. A researcher of Hi Investment & Securities noted in a report that Samsung Electronics had managed to improve its 4nm process yield, increasing the chances that Qualcomm and Nvidia outsource capacity to Samsung Foundry.

Previously, Samsung Foundry had experienced delays in product launches and slow improvement in the yield of processes below 10nm, causing major clients to turn to TSMC. TSMC’s capital expenditures and capacity last year turned out to be 3.4 times and 3.3 times that of Samsung Foundry. Regarding advanced processes below 7nm, TSMC’s market share achieved 90%, further expanding the gap between the two companies. As Samsung Electronics reached over 75% yield of the 4nm process and over 60% yield of the 3nm process this year, coupled with the impact of TSMC’s price increases, the industry believes Samsung promises to win back the customers poached by TSMC.

Comment:

The delay in improving its yield made Samsung’s major customers turn to TSMC. To break TSMC’s dominance in the market, improving and stabilizing the yield of its advanced processes may be Samsung’s only option. Market demand being at a low point is somewhat of an opportunity for the company to increase its revenue, with lower operating rates allowing it to focus more resources on advanced processes. In addition to the 4nm process, Samsung also improved its 3nm process yield to more than 60%, boosting its confidence in winning back the orders going to TSMC. TSMC’s continual price increases in foundry services have made many chipmakers reconsider diversity in production outsourcing, which has increased Samsung’s opportunities.

Market News（July 13）

With HBMs five times more expensive than DRAMs, a HBM supremacy war starts.

Samsung and SK Hynix hope their semiconductor businesses can be profitable through capacity increases of high-capacity chips such as HBMs. HBM chips, which vertically interconnect multiple DRAM chips, can dramatically increase data processing speed compared to traditional DRAM products. They are at least five times more expensive than DRAMs.

In the HBM DRAM segment, SK Hynix said in April that it had developed the world’s first 12-layer HBM3 DRAM product with a 24GB memory capacity, the thinnest and largest in the industry. The company has already provided samples to customers and plans to start mass production by the end of the year. Meanwhile, Samsung plans to mass-produce HBM3 chips by the end of the year. The company’s top management said it would spend hundreds of billions of won to double the capacity to produce HBM3 chips at its Cheonan plant by the end of 2024. According to local brokerage KB Securities, HBM3 chips will account for 18% of Samsung’s chip sales revenue by 2024, up from an estimated 6% this year. The president of Samsung’s Device Solutions, which sees to its chip business, said the company would try to control more than half of the HBM market.

Comment:

With the explosive growth of data, the impact of memory walls on computing speed becomes increasingly apparent. Increasing memory bandwidths has always been a crucial focus for memory chips to reduce the impact of memory walls. To a great extent, the memory industry’s value proposition has long been guided by system-level requirements, bringing the system’s performance beyond its current caps. One thing clear is that memory performance enhancement is reaching an inflection point, and more and more people are starting to question whether it is always sustainable to improve system performance through memory-level trade-offs.

As a significant part of the memory market, DRAM technology constantly undergoes upgrades and derivatives. HBM, a high-value-added DRAM product, is basically the only option. To realize AI services, HBM memory chips capable of transfering data at high speeds have become an indispensable configuration for AI, resulting in a surge in market demand.

Some Chinese Industrial News

1.Epiworld announces to mass-produce 8-inch SiC epitaxial wafers.

Epiworld Electronic Technology (Xiamen) Co., Ltd. recently announced it had completed developing the 8-inch SiC epitaxial process with independent intellectual property rights, formally capable of mass-producing Chinese-made 8-inch SiC epitaxial wafers. Epiworld signed several long-term contracts lately, including one for 8-inch SiC epitaxial wafers valued at more than US$192 million.

The quality of the 8-inch SiC epitaxial wafers produced by Epiworld has reached the international advanced level, with thickness non-uniformity less than 3%, concentration non-uniformity less than 6%, and the yield of 2mm*2mm tube cores over 98%. This technological breakthrough marks that China has mastered the commercialized 8-inch SiC epitaxial growth technology, further advancing the localization process of SiC epitaxial materials and greatly enhancing China’s international status in the SiC epitaxial area.

2.Selling a domestically-made chip at a loss of 230,000 yuan

Chinese in-vehicle chip start-up Black Sesame Technologies submitted its listing application to Hong Kong Exchanges recently, planning to list on the Hong Kong Main Board. Black Sesame is one of the only two Chinese high-performance computing chip companies that have realized mass production and vehicle onboarding, second only to Horizon Robotics in mass production pace and shipment volume. According to the financial data disclosed in the prospectus, Black Sesame’s revenue in the past three years totaled only 279 million yuan, while the loss climbed year by year, reaching a total of 5.871 billion yuan. That means with 25,000 chips shipped, the company lost 234,800 yuan on each chip sold.

Black Sesame has two product series, Huashan and Wudang. The Huashan series focuses on autonomous driving. Its A1000 SoC started mass production in 2022, with 25,000 chips shipped by the end of that year. The Wudang series, formally released in April 2023, focuses on cross-domain computing. Its C1200 adopts the 7nm process technology to cover the needs of different domains, such as smart cockpit and smart driving. Samples are expected to be provided within 2023.

3.A Chinese team achieves mass production and preparation of 12-inch 2D semiconductor wafers.

Chinese Science Bulletin published the latest research result online under the title of Modularized Batch Production of 12-inch Transition Metal Dichalcogenides by Local Element Supply. The research, proposing modularized growth technology by localized element supply, realized the highly-efficient batch production of wafers of semiconducting 2D transition metal dichalcogenides, making wafer size scalable from 2 inches to 12 inches, which is compatible with modern semiconductor processes.

2D semiconductors are an emerging semiconductor material with excellent physicochemical properties, represented by single-layer transition metal dichalcogenides. Similar to the development path of traditional semiconductors, wafer materials are the foundation to drive the 2D semiconductor technology toward industrialization. How to achieve large-size, low-cost batch production of 2D semiconductor wafers is an urgent scientific problem.

Regarding the core scientific problem of the size enlargement and batch preparation of 2D semiconductor wafers, the researchers proposed a brand-new modularized growth strategy by local element supply and achieved the batch production of up to 12-inch 2D semiconductors.