News
In response to the high demand for AI chips, TSMC has once again initiated a large-scale expansion in northern, central, and southern Taiwan. This expansion includes the first tool-in and the construction of 2-nanometer fabs and advanced packaging plants.
Industry sources cited by the Liberty Time News have suggested that investing in eight to ten facilities may be necessary for the next-generation 1-nanometer technology in Taiwan.
Hot Demand for AI Chips Sparks Ongoing Installation and Production
As per the report citing industry sources, last year, TSMC’s pace of building factories in Taiwan slowed down due to the downturn in the semiconductor industry. However, with the industry’s recovery this year and the high demand for AI chips, expansion efforts in the north, central, and southern regions have become more active.
In April, first tool-in will be commenced at the Baoshan 2-nanometer plant (Fab 20) in the Hsinchu Science Park. Additionally, the 2-nanometer plant (Fab 22) in Kaohsiung is under construction, with plans for a third plant underway. TSMC’s 2-nanometer process is scheduled for mass production in 2025.
AI chips, emphasizing computational power, universally adopt advanced processes. The surge in demand for AI chips is driving the need for advanced packaging technologies like CoWoS.
Currently facing overwhelming demand, TSMC is expanding its advanced packaging facilities in Zhunan, Central and Southern Taiwan Science Park, with plans to start tool-in and join production lines this year. The advanced packaging facility in Chiayi is also commencing construction this year.
Generation of One-Nanometer Fabrication Plants: 8-10 Facilities Needed
According to the report from Liberty Time News citing sources, Intel’s construction of fab typically takes about five to six years. On the other hand, TSMC’s construction and first tool-in in Taiwan are estimated to be completed within two years.
With a monthly installation capacity of 200 units per month for a two-nanometer plant, TSMC reportedly plans to advance to 1.4-nanometer and 1-nanometer advanced processes after the two-nanometer.
Read more
(Photo credit: TSMC)
News
On March 4th, Chinese assembly and testing company, JCET Group (JCET), announced its intention to acquire 80% equity of SanDisk Semiconductor Shanghai Co., Ltd. (SDSS), a subsidiary of Western Digital Corporation (WDC), for USD 624 million in cash. This strategic move aims to expand JCET’s market share in the storage.
The assembly and testing market competition is intensifying, especially with Moore’s Law slowing down in recent years. While advanced process technology still requires time, the demand for computing is skyrocketing.
As a result, advanced packaging is highly anticipated. Over the past two years, several testing and packaging giants have continuously enhanced their market competitiveness through mergers, expansion, and innovative technologies.
JCET Ventures into Storage with CNY 4.5 Billion Investment
According to the announcement, JCET has signed a “Equity Acquisition Agreement” with SANDISK CHINA LIMITED, the parent company of SANDISK CHINA LIMITED – SSDS.
Per the agreement, JCET Group intends to acquire 80% of the equity of SSDS held by the seller in cash. The transaction price, after full communication and negotiation between the parties, is approximately USD 624 million. After the completion of this transaction, JCET Group will hold 80% of the equity of SSDS, while SANDISK CHINA LIMITED will retain 20% ownership in SSDS.
According to the information available, SSDS was established in August 2006 and primarily engages in the research and development, packaging, and testing of advanced NAND Flash memory. Its product range includes iNAND flash modules, SD cards, MicroSD memory cards, etc., widely used in mobile communications, industrial IoT, automotive, smart home, consumer electronics, and other fields.
The selling party’s parent company, WDC, is a global leader in storage solutions. According to TrendForce’s data, in the third quarter of 2023, Western Digital held a market share of 16.9% in global NAND Flash revenue rankings, securing the third position worldwide.
Per the announcement, WDC has had a long-standing partnership with JCET since 2003, making it one of JCET’s key clients. JCET stated that following the completion of this transaction, the selling party (SANDISK CHINA LIMITED) and its parent company (Western Digital) will continue to be the primary or sole customers of SSDS for a certain period, ensuring a level of stability in its operational performance.
In its announcement of the acquisition of SSDS, JCET explicitly states that this decision stems from an in-depth analysis and forward-looking assessment of the global storage market.
As per the World Semiconductor Trade Statistics (WSTS), storage chips have emerged as the second-largest semiconductor sub-market, capturing approximately 28% of market share, trailing only behind logic chips. Looking ahead, it is projected that by 2024, the storage chip market will reach a staggering USD 130 billion in size.
Within this expansive market, NAND flash chips stand out, constituting approximately 40% of the global storage market. Moreover, it is projected to maintain a compound annual growth rate (CAGR) of 8% from 2021 to 2027. This indicates that the NAND flash chip market is not only vast but also possesses stable growth potential.
Intense Competition in the Semiconductor Packaging Market: Focus on Advanced Packaging
Major semiconductor packaging companies like ASE, Amkor Technology, Powertech, and Huatian Technology are continuously enhancing their competitive edge through expansions and acquisitions, driving rapid growth in the semiconductor packaging industry.
Regarding ASE, it recently announced that its capital expenditure for this year is expected to increase by 40% to 50% compared to last year, with 65% allocated to packaging, especially for advanced packaging projects.
On February 22nd, ASE Group and semiconductor giant Infineon Technologies jointly announced the finalization of an agreement. ASE Group will invest EUR 62.589 million to acquire Infineon’s backend packaging facilities located in Cavite, Philippines, and Cheonan, South Korea.
In November 2022, ASE initiated construction on its fourth and fifth plants in Penang, Malaysia, with completion slated for 2025. Besides Malaysia, ASE continues its expansion efforts in Taiwan, including Kaohsiung, Zhongli, and Tanzi.
On January 16th, Amkor and semiconductor foundry Globalfoundries held a ribbon-cutting ceremony for the joint construction of the Amkor Porto plant in Portugal. Since February 2023, Globalfoundries has relocated 50 devices from its Dresden facility to Amkor’s new Porto plant.
The initial batch of client products has been certified using Globalfoundries’ equipment. Globalfoundries is transferring some of its 300mm production lines from Dresden to Amkor’s Porto facility, which is IATF16949 certified, to establish Europe’s first large-scale test and packaging plant.
In 2023, the packaging and testing firm, Powertech, expressed optimism about the future development of advanced packaging technologies. Powertech has outlined plans to resume significant capital expenditures in the latter half of 2024, ensuring sustained investments to maintain a competitive edge and prepare for long-term growth.
In March 2023, Huatian Technology made significant strides by investing 2.858 billion yuan in its wholly-owned subsidiary, Huatian Technology (Jiangsu) Co., Ltd., for the construction of the “High-Density, High-Reliability Advanced Packaging Research and Industrialization” project.
Once completed, the project will enable an annual wafer-level integrated circuit packaging capacity of 840,000 pieces for Bumping, 480,000 pieces for WLCSP, and 26,000 pieces for ultra-high-density fan-out UHDFO. The construction period spans 5 years, from June 2023 to June 2028.
Currently, the packaging market’s competitive landscape extends beyond traditional OSATs, with major foundries and storage giants actively joining the fray. Technologies like TSMC’s SoIC, CoWoS, and InFO, Intel’s EMIB, Foveros, and Co-EMIB, as well as Samsung’s I-Cube (2.5D) and X-Cube (3D), are emerging and maturing.
The proliferation of various 2.5D and 3D packaging solutions indicates that competition in advanced packaging will only intensify, promising an exciting future ahead.
Read more
(Photo credit: JCET Group)
News
AMD has encountered resistance from the United States when attempting to sell a custom-designed AI chip to the Chinese market, according to Bloomberg citing sources.
The same source further revealed that AMD sought approval from the U.S. Department of Commerce to sell this AI chip to Chinese customers. The chip’s performance is lower than AMD’s products sold in other regions of China, and its design complies with U.S. export restrictions.
However, U.S. officials informed AMD that the chip’s performance is still too robust, requiring the company to obtain approval from the U.S. Department of Commerce’s Bureau of Industry and Security before sales can proceed. It remains unclear whether AMD is currently seeking a license.
The U.S. government introduced initial export control measures in 2022 and bolstered a series of measures in October 2023, encompassing additional technologies and reforming potential sales that could undermine intermediary countries/regions.
The strict limitations on the sale of NVIDIA’s chips specifically designed for China align with the initial export regulations from 2022. Subsequently, the company developed new customized but lower-capacity products for the Chinese market, aligning with the restrictions imposed by the United States in 2023.
The U.S. ban in 2022 prevented both NVIDIA and AMD from selling their most powerful artificial intelligence chips to China. When the restrictions took effect in 2022, AMD anticipated that they would not substantially affect its operations.
While AMD has not publicly discussed its development of new AI chips for China, NVIDIA, on the other hand, immediately introduced improved products with reduced performance.
NVIDIA is set to commence pre-orders for its AI chip H20 specially designed for the Chinese market by the end of March this year in response to US export bans, according to sources cited by a report from STAR Market Daily.
Still, AMD is now actively expanding into the AI chip market. In December 2023, it launched the new MI300 series, challenging NVIDIA’s chips. According to sources cited by Bloomberg’s report, this customized product for China is known as the MI309. It is still unclear which Chinese customer wishes to purchase AMD’s AI chips, which could affect the company’s authorization.
Read more
(Photo credit: AMD)
News
As per a report from TechNews, Apple’s pivot into AI, abandoning its “Project Titan” for electric cars, signals a shift towards Generative AI. The report further cites sources indicate that Foxconn may provide AI servers to Apple and is currently in testing phase.
Regarding this matter, Foxconn responded with no comment on individual clients or products.
According to reports from Economic Daily News, Apple has conducted extensive AI feature testing and, given Foxconn’s global leadership in server manufacturing, it has emerged as Apple’s preferred partner for the AI project.
In addition to Apple, during a recent financial conference, Dell’s COO, Jeff Clarke, disclosed that NVIDIA is set to launch a new generation server GPU, “B200,” based on the Blackwell architecture in 2025. Notably, this revelation wasn’t part of NVIDIA’s product roadmap released in October 2023, and the company has not officially mentioned this product.
Currently, the H100 utilizes TSMC’s 4-nanometer process technology, with Foxconn securing approximately 90% of the assembly orders last year. While the fabrication process for the B100 and B200 chips remains unconfirmed, industry expectations cited by the report have pointed to the 3-nanometer process.
Previously, media speculation cited by the report from Commercial Times stated that although the B100 chip boasts computational power at least twice that of the H200 and four times that of the H100, still, B100’s tenure in the market is anticipated to be short-lived, with the B200 emerging as the mainstream product. It is rumored that Foxconn Industrial Internet will handle the manufacturing for the B200.
Foxconn’s Chairman Young Liu previously indicated a strong demand for AI servers, with Foxconn securing new projects continuously.
Foxconn spokesperson James Wu noted that Foxconn Group commands over 40% market share in the server industry, particularly in the mid-to-high-end products related to AI servers. Foxconn closely collaborates with customers and aims to maintain its dominance, anticipating substantial contributions once the entire supply chain stabilizes.
Read more
(Photo credit: Foxconn)
News
The surge in demand for NVIDIA’s AI processors has made High Bandwidth Memory (HBM) a key product that memory giants are eager to develop. However, according to South Korean media DealSite cited by Wccftech on March 4th, the complex architecture of HBM has resulted in low yields, making it difficult to meet NVIDIA’s testing standards and raising concerns about limited production capacity.
The report has further pointed out that HBM manufacturers like Micron and SK Hynix are grappling with low yields. They are engaged in fierce competition to pass NVIDIA’s quality tests for the next-generation AI GPU.
The yield of HBM is closely tied to the complexity of its stacking architecture, which involves multiple memory layers and Through-Silicon Via (TSV) technology for inter-layer connections. These intricate techniques increase the probability of process defects, potentially leading to lower yields compared to simpler memory designs.
Furthermore, if any of the HBM chips are defective, the entire stack is discarded, resulting in inherently low production yields. As per the source cited by Wccftech, it has indicated that the overall yield of HBM currently stands at around 65%, and attempts to improve yield may result in decreased production volume.
Micron announced on February 26th the commencement of mass production of High Bandwidth Memory “HBM3e,” to be used in NVIDIA’s latest AI chip “H200” Tensor Core GPU. The H200 is scheduled for shipment in the second quarter of 2024, replacing the current most powerful H100.
On the other hand, Kim Ki-tae, Vice President of SK Hynix, stated on February 21st in an official blog post that while external uncertainties persist, the memory market is expected to gradually heat up this year. Reasons include the recovery in product demand from global tech giants. Additionally, the application of AI in devices such as PCs or smartphones is expected to increase demand not only for HBM3e but also for products like DDR5 and LPDDR5T.
Kim Ki-tae pointed out that all of their HBM inventory has been sold out this year. Although it’s just the beginning of 2024, the company has already begun preparations for 2025 to maintain its market-leading position.
Per a previous TrendForce press release, the three major original HBM manufacturers held market shares as follows in 2023: SK Hynix and Samsung were both around 46-49%, while Micron stood at roughly 4-6%.
Read more
(Photo credit: SK Hynix)