In the era of AI and big data, new market demands are emerging in the storage industry, and QLC NAND SSDs, which play a critical role in AI training. This suggests that NAND flash may be on the verge of a new growth opportunity.

QLC, or Quad-Level Cells, refers to NAND flash where each storage cell can record 4 bits. Each cell maintains a standard size, and compared to traditional SLC, MLC, and TLC technologies, QLC enables higher data density, allowing more data to be stored in the same space.

The Advantages of QLC Flash

QLC flash offers several key advantages:

• Capacity: Compared to mainstream TLC flash, QLC increases storage density by 33%.
• Cost: With the same number of flash chips cut from a wafer, QLC provides about 33% more storage capacity than TLC, leading to a lower cost per unit of storage.
• Total Cost of Ownership (TCO): QLC-based SSDs offer lower TCO compared to traditional HDDs. TCO includes considerations like storage density, reliability, and power consumption. While QLC’s write performance is somewhat inferior to other technologies, its read performance is strong.

However, QLC technology also faces some challenges. Since data is stored across more charge levels, this increases the risk of data vulnerability, requiring more advanced error correction and data protection mechanisms. QLC storage devices typically need stronger error correction algorithms and robust data protection to ensure integrity and reliability.

Despite these challenges, as technology continues to advance, QLC is expected to play an increasingly important role in the storage field, potentially revolutionizing the entire information industry.

QLC NAND is Entering the AI SSD Battlefield

Many believe that QLC SSDs are a complement to TLC SSDs, particularly suited for read-intensive and mixed read/write workloads, such as those found in AI, content delivery networks, and machine learning.

In AI applications, TrendForce points out that SSDs are used in AI inference servers to help adjust and optimize AI models during inference, especially by updating data in real-time to fine-tune model outputs. AI inference mainly supports Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs), and SSDs can store reference documents and knowledge bases for RAG and LLM to generate richer responses.

As more generated information is displayed as videos or images, data storage demands increase, making large-capacity SSDs, such as 16TB+ TLC/QLC models, essential for AI inference.

Currently, companies like Samsung, SK Hynix’s Solidigm, Western Digital, and Kioxia hold key positions in the QLC SSD market.

Industry experts are optimistic about the future of QLC SSDs, emphasizing their benefits for data centers, cloud storage providers, and enterprises handling massive amounts of data. QLC SSDs are seen as a potential game-changer for the NAND flash market.

TrendForce also predicts that QLC will account for 20% of NAND flash shipments by 2024, with this share expected to rise in 2025. Moreover, QLC is likely to expand into consumer-embedded storage such as eMMC and UFS, with some products already adopting QLC as the storage medium.

Industry forecasts suggest that by 2025, large-capacity QLC enterprise SSDs will rise, and smartphones will start using QLC-based UFS storage solutions.

(Photo credit: Samsung)

Would Samsung finally be able to secure major orders from tech giants, even one of the world’s top AI companies? According to a report by The Information, the most successful and lucrative partnership in AI business, which is formed by NVIDIA and its foundry partner TSMC, is showing signs of strain, while Samsung may turn out to benefit from the development.

However, it is worth noting that the orders Samsung might get from NVIDIA may not be the most advanced AI chips. According to the reports by The Information and SamMobile, the U.S. AI chip giant is mulling to team up with Samsung to produce its new GPUs, which are considered less complex to manufacture than its AI accelerators.

The reports also suggest that NVIDIA is trying to secure discounted pricing from Samsung, as it aims for a 20-30% reduction compared to what it pays TSMC.

According to the analysis by SamMobile, NVIDIA’s move is an attempt to reduce its dependence on TSMC for upcoming chips, which is a positive development for Samsung.

Though the struggling semiconductor giant is said to have several clients for 5nm, 7nm and 8nm nodes, the continuous yield issues for 3nm and 4nm makes it unable to attract major customers, according to SamMobile, while the company now hopes to improve its yields and regain clients like Qualcomm and NVIDIA.

According to The Information, Samsung’s opportunities arise while NVIDIA’s Blackwell chips, built with TSMC’s 4nm, reportedly faced delays due to issues discovered in the testing process. The chips are said to have failed in high-voltage environments typical of data centers, which put strain on the decades-long partnership between the two firms.

For now, the issues have been resolved, and Team Green’s Blackwell chips are expected to ramp up starting from Q4 2024. According to a report by Wccftech, citing the projection by analyst Ming-Chi Kuo, Blackwell’s estimated shipments are expected to be around 150,000 to 200,000 units this quarter, and surging to 500,000 to 550,000 units in Q1 2025.

According to Kuo, Microsoft is believed to be the major customer, with its orders for GB200 in Q4 skyrocketing, rising 3 to 4 times from the previous range of 300 to 500 racks (primarily NVL36) to about 1,400 to 1,500 racks (approximately 70% NVL72).

Read more

• [News] NVIDIA’s Blackwell Overcomes Delays, as GB200 Reportedly Sets for December Mass Production

(Photo credit: NVIDIA)

According to a report by the Economic Daily News, U.S. chip giant Marvell Technology has announced that it will raise prices across its entire product line starting January 1 next year, marking the first major price hike in the optical communications sector.

Marvell’s strong financial performance last quarter, fueled by surging demand for AI-related products like ASICs and silicon photonics for data centers, surprised the market. The decision to raise prices is seen by industry insiders as not only a move to capture emerging opportunities but also a reflection of the continued investment in cutting-edge and innovative product development to meet growing market demand.

The Economic Daily News cited a leaked price hike notification letter signed by Marvell’s Senior Vice President of Global Sales, Dean Jarnac. The letter highlighted that the global demand for accelerated computing and AI is driving unprecedented investments across the semiconductor supply chain, including expanding production capacity and establishing diverse manufacturing bases. Marvell is no exception, and it plans to continue investing heavily in innovation to deliver advanced products and technologies to its customers.

Jarnac explained that the price increases, effective January next year, are necessary to support expanded investment levels. He assured customers that Marvell will minimize the price hike’s impact as much as possible and urged them to place orders based on delivery schedules and provide accurate demand forecasts.

Marvell’s stellar quarterly results were driven by the booming AI demand, which boosted its data center business far beyond market expectations. Key growth drivers included optical products like 800G PAM and 400ZR data center interconnect (DCI) solutions.

“Marvell’s second quarter revenue grew 10% sequentially, above the mid-point of guidance driven by strong demand from AI. We saw strong growth from our electro-optics products, and our custom AI programs began to ramp,” said Matt Murphy, Marvell’s Chairman and CEO, in an August press release. “Next quarter, we expect our combined enterprise networking and carrier end markets to return to growth, while our data center end market growth accelerates. As a result, for the third quarter of fiscal 2025, we expect all our end markets to grow sequentially, with consolidated revenue forecasted to grow 14% sequentially at the mid-point, accompanied by a significant increase in operating leverage.”

(Photo credit: Marvell)

Rumors have been circulating for a while that China’s tech giants are capable of manufacturing chips in advanced nodes even without the need for extreme ultraviolet (EUV) lithography machines by ASML, and that local foundry SMIC has reportedly produced 5nm chips for Huawei. Now there seems to be a new member joining the ranks of the elite. According to the reports by MyDrivers and Wccftech, Xiaomi has successfully taped out its first 3nm SoC.

Though more details are yet to be confirmed, Wccftech indicates that it is possible that Xiaomi will launch the 3nm chipset sometime next year.

Xiaomi, according to the reports, has been developing its own custom chipsets for years. The first product is believed to be Surge S1, which was released with the Mi 5c smartphone in 2017. The report by MyDrivers suggests that the Surge S1, built on 28nm node, is a 64-bit octa-core processor.

Following the Surge S1, Xiaomi went on to develop several chips, reportedly including the Surge C1, Surge G1, and Surge T1, according to MyDrivers.

The information of Xiaomi’s recent success in its 3nm chip tape-out is reportedly disclosed by Tang Jianguo, Chief Economist of Beijing Municipal Bureau of Economy and Information Technology, at Beijing Satellite TV, MyDrivers notes. Citing Tang’s remarks, Wccftech indicates that the achievement is described as a historic milestone for China.

Wccftech further notes that as companies like Huawei have been prohibited from doing business with TSMC or Samsung due to U.S. trade sanctions, if Xiaomi does have reached the tape-out stage for its 3nm chipset, it could enable other Chinese companies, including Huawei, to leverage this technology in their devices.

However, more details are yet to be confirmed, as there are no updates on whether the SoC will use TSMC’s N3E process or the more advanced N3P node, according to Wccftech. Additionally, details about the chipset’s CPU cluster, GPU, or whether it will feature ARM designs or a custom architecture remain unknown. Therefore, it would be better to approach this rumor with caution.

Read more

• [News] Xiaomi Reportedly to Launch In-House Processor in 2025, Manufactured with TSMC’s 4nm
• [News] SMIC Reportedly to Manufacture 5nm Chips for Huawei Without EUV Machinery

(Photo credit: Xiaomi)

TSMC has passed the test of the market with flying colors as it reported record high profit in the third quarter at the earnings call. By confirming that the AI demand is “real,” TSMC Chairman C.C. Wei stated that the foundry giant is expected to enjoy healthy growth over the next five years. But which node would be the one most clients show strong interest in?

According to the reports by the Economic Daily News and MoneyDJ, customer inquiries for 2nm are even higher than those for 3nm, while A16 is highly attractive for AI server applications.

TSMC’s 3nm has already shown robust momentum this year, as its shipments accounted for 20% of total wafer revenue in the third quarter, rising from 9% and 15% in the first and second quarter, respectively.

According to an industrial source cited by MoneyDJ, TSMC started the mass production of 3nm in 2022, while the 2nm is expected to enter volume production in 2025, indicating that the generation cycle for a node has been expanded to three years.

Thus, supported by TSMC’s major clients, the contribution from 3nm will continue to rise next year and remain a key revenue driver in 2026, while the 2nm process is expected to replicate or even surpass the success of 3nm, MoneyDJ notes. According to previous market speculations, tech giants such as Apple, NVIDIA and AMD are believed to be the first batch of TSMC’s 2nm customers.

Citing C.C.Wei, the Economic Daily News notes that the high-performance computing (HPC) applications demand more powerful processors, which accelerates the development of chiplet designs. However, the trend does not seem to impact the adoption of 2nm, and clients are showing even stronger interests for the node compared with 3nm.

And TSMC does plan to expand its 2nm capacity thanks to the strong demand, as the schedule of mass producing 2nm in 2025 remains on track.

According to a previous report from MoneyDJ, TSMC’s 2nm fabs in Hsinchu’s Baoshan and Kaohsiung will achieve a monthly capacity of approximately 30,000 to 35,000 wafers, respectively. By 2027, their combined capacity is set to exceed 100,000 wafers, marking the mainstream transition to the next generation of processes.

As for TSMC’s angstrom-level A16 process, it is creating a buzz even before mass production in 2026. Citing C.C.Wei’s remarks, the report by the Economic Daily News notes that the A16 is highly attractive for AI server applications, and TSMC is actively preparing the related production capacity to meet customer demand.

Read more

• [News] TSMC’s C.C. Wei Declares No Interest in Acquiring Intel’s Fab, Emphasizes ‘AI is Real’
• [News] TSMC 3Q24 EPS Hits Record High, Reaches NT$12.54