Though Samsung has denied the rumor that its HBM3e passed NVIDIA’s qualification tests, multiple Taiwanese companies in the supply chain reportedly learned that the product is expected to receive certification soon, and will start shipping in Q3. As memory manufacturers are said to shift at least 20-30% of their production capacity to HBM, tightening supply further, DDR5 prices in Q3 will reportedly be on the rise.

It is reported that some of Samsung’s supply chain partners have recently received information to place orders and reserve capacity as soon as possible, which indicates the memory giant’s HBM may begin shipments smoothly in the second half of the year. The move may also imply that the internal capacity allocation within Samsung will accelerate, shifting the focus of production lines to HBM.

Taiwanese memory supply chain sources reportedly believe that the news of Samsung’s HBM certification is likely to be confirmed at the upcoming Samsung financial report meeting, which will take place on July 31. It is said that memory manufacturers will relocate at least 20-30% of their production capacity, driving DDR5 prices to rise.

TrendForce notes that a recovery in demand for general servers—coupled with an increased production share of HBM by DRAM suppliers—has led suppliers to maintain their stance on hiking prices. As a result, the ASP of DRAM in Q3 is expected to continue rising, with an anticipated increase of 8–13%. Due to high average inventory levels of DDR4 among buyers, purchasing momentum will be focused on DDR5.

On the other hand, regarding NAND prices in Q3, TrendForce reports that while the enterprise sector continues to invest in server infrastructure, the consumer electronics market remains lackluster. This, combined with NAND suppliers aggressively ramping up production in the second half of the year, is likely to curb the blended price hike to a modest 5–10%.

According to TrendForce’s latest analysis, Samsung’s initial plan to pass NVIDIA’s certification in Q2 was delayed, making it falling behind SK hynix and Micron. Simultaneously, some HBM suppliers also faced lower-than-expected production yields, leading to concerns about a shortage of HBM3e 8hi materials for the H200 GPU shipments starting in Q2 2024.

However, Samsung adjusted its 1alpha nm front-end production process and back-end stacking process in the first half of 2024, leading the industry to expect that sample production could be completed in Q3 2024, followed by product certification.

Read more

• [News] Samsung’s HBM3e Reportedly Passed NVIDIA’s Certification Though It Denies the Rumor
• Demand for Servers to Support Second Half of the Year; DRAM Prices Expected to Increase by 8–13% in Q3, Says TrendForce

(Photo credit: Samsung)

As top memory giants and AI chip companies all gear up for the combat of next-gen high bandwidth memory (HBM), JEDEC, the leader in the development of standards for the microelectronics industry, revealed the preliminary specifications of HBM4 last week. According to its press release and a report from Wccftech, HBM4 is poised to deliver substantial memory capacities, with densities up to 32Gb in 16-Hi stacks.

According to JEDEC, HBM4 aims to boost data processing rates while preserving key features such as higher bandwidth, reduced power consumption, and increased capacity per die or stack, which are crucial traits for applications that demand efficient management of large datasets and complex calculations, such as generative AI, high-performance computing, high-end graphics cards, and servers.

According to JEDEC’s preliminary specifications, HBM4 is anticipated to feature a “doubled channel count per stack” compared to HBM3, which indicates a higher utilization area, leading to significantly enhanced performance. It is also worth noting that in order to support device compatibility, the new standard ensures that a single controller can work with both HBM3 and HBM4.

JEDEC notes that HBM4 will specify 24 Gb and 32 Gb layers, offering support for TSV stacks ranging from 4-high to 16-high. The committee has initially agreed on speed bins up to 6.4 Gbps, with ongoing discussions for higher frequencies.

Interestingly enough, JEDEC did not specify how HBM4 integrates memory and logic semiconductors into a single package, which would be one of the major challenges the industry has been eagerly trying to solve.

Earlier in June, NVIDIA announced its next-gen Rubin GPU, targeting to be released in 2026, will feature 8 HBM4, while its Rubin Ultra GPU will come with 12 HBM4 chips.

The roadmaps for memory giants on HBM4 is generally in accordance with NVIDIA’s product pipeline. Samsung, for instance, is said to be developing a large-capacity HBM4 memory with a single stack capacity of 48GB, which is expected to enter production in 2025.

The current HBM market leader, SK hynix, on the other hand, has collaborated with TSMC on the development and production of HBM4, scheduled for mass production in 2026.

Micron has also disclosed its next-generation HBM memory, tentatively named HBM Next. It is expected that HBM Next will offer capacities of 36GB and 64GB, available in various configurations.

Read more

• [News] The HBM4 Battle Begins! Memory Stacking Challenges Remain, Hybrid Bonding as the Key Breakthrough

(Photo credit: SK hynix)

The semiconductor industry, driven by AI, is entering a new upward cycle. According to a forecast report from SEMI, after the trough in 2023, the total sales of equipment in 2024 will hit a new high, with growth momentum continuing into 2025. Among this trend, per a report from Commercial Times, major companies including TSMC, Intel, Samsung, SK hynix, and Micron are all actively preparing, with plans to continue increasing capital expenditure next year in preparation for the AI era.

TSMC and Intel are the most proactive foundries. Intel plans to increase its capital expenditure by 2% in 2024, reaching USD 26.2 billion; TSMC’s capital expenditure for this year is expected to be between USD 28 billion and USD 32 billion.

The same report further cited sources, indicating that TSMC’s capital expenditure this year will reach the upper end of the estimated range. Next year, the upper limit is expected to increase by another USD 5 billion to USD 37 billion, potentially reaching the second-highest level in its history.

It’s reported that customer demand for TSMC’s 2nm process capacity has exceeded expectations. In addition to Apple securing the first batch of TSMC’s 2nm capacity, non-Apple customers are also actively planning for advanced processes. TSMC continues to advance its goal of mass production of the 2nm process by next year.

Another source cited by Commercial Times reveals that TSMC accelerated equipment orders in the second quarter and further increased momentum in the third quarter, primarily to ensure the smooth launch of the 2nm process by mid-next year.

In the HBM sector, Samsung and SK hynix are reportedly raising funds to prepare for significant production expansion in 2025.  A report from Korean media outlet Korea Economic Daily (KED) indicated that Samsung Electronics and SK hynix are considering applying for loans from the Korea Development Bank, with planned loan amounts of KRW 5 trillion (roughly USD 3.6 billion) and KRW 3 trillion (roughly USD 2.2 billion), respectively.

Micron’s capital expenditure plan for the 2024 fiscal year is about USD 8 billion. In the fourth quarter of the 2024 fiscal year, Micron will spend approximately USD 3 billion on fab construction and new wafer fab equipment (WFE). For the 2025 fiscal year, Micron plans to significantly increase its capital expenditure, targeting 30% of its revenue, or about USD 12 billion. Earlier, Micron’s Chief Operating Officer, Manish Bhatia, stated that the scale of the HBM business is expected to expand to several billion dollars in the 2025 fiscal year.

Read more

• [News] TSMC’s Capital Expenditure to Surge Next Year, Driven by Rising 2nm Demand
• [News] Micron Slightly Raises Capital Expenditure for 2024, HBM Expected to Further Drive Revenue Growth in 2025
  

(Photo credit: Micron)

While still working in the final stage of HBM3e qualification with NVIDIA, Samsung Electronics is also advancing in the AI memory market with custom high bandwidth memory (HBM) solutions. According to reports by PassionateGeekz and China Flash Market, the memory giant is collaborating with major clients, such as AMD and Apple, to develop tailored HBM products, which are expected to be commercially available in the era of HBM4.

Citing Choi Jang-seok, head of Samsung’s new business planning team at memory division, the reports note that many customers of Samsung are switching from traditional, general HBM to customized products, as the latter promises better performance, power and area (PPA), while offering greater value than current options.

PassionateGeekz notes that at the Samsung Foundry Forum 2024 earlier this week, Choi further highlighted two forms of customized HBM Samsung has been developing. It is worth noting that Samsung is developing a large-capacity HBM4 memory with a single stack capacity of 48GB, which is expected to enter production in 2025.

On the other hand, Samsung also illustrated the innovation of the 3D stacking of HBM DRAM and customer-specific logic chips. By bypassing the interposer and base die required in the existing 2.5D packaging solution, the HBM chip can be directly integrated into the computing SoC in 3D. Samsung’s custom HBM, therefore, by eliminating intermediaries and substrates, can significantly reduce power and area.

TrendForce also observed that for HBM4, standard processes and capacities have been settled. The three major suppliers are in the development stage, with each buyer initiating custom requests. For future generations of HBM, new directions have been proposed, as HBM may no longer be just arranged next to the SoC main chip but could also stack directly on top of it.

While all the options are still under feasibility discussion and not finalized, TrendForce believes the future HBM industry will shift towards more customized production. Compared to other DRAM products, this approach aims to break away from the framework of commodity DRAM in terms of pricing and design, offering more specialized solutions.

Read more

• [News] Samsung Establishes New HBM Team to Advance HBM3, HBM3e and HBM4 Development
• [News] Overview of Expansion Plans by HBM Giants

(Photo credit: Samsung)

Recently, Intel, AMD, Samsung, LG Innotek, and SKC’s US subsidiary Absolics have all highly focused on glass substrate technology for advanced packaging. Due to its excellent performance, glass substrate technology has become a rising star in the field of advanced packaging.

In September 2023, Intel announced the so-called “next-generation advanced packaging glass substrate technology,” claiming it could revolutionize the entire chip packaging field. Glass substrate refers to the replacement of organic materials in organic packaging with glass, rather than replacing the entire substrate. Therefore, Intel will not mount chips on pure glass; instead, the core material of the substrate will be made of glass.

Intel stated that glass substrates could lay the foundation for achieving an astounding one trillion transistors on a single package within the next decade. Based on its promising prospect, rumors surface recently that Intel plans to mass-produce glass substrates as early as 2026. Intel has invested approximately a decade in glass substrate technology and currently has a fully integrated glass research line in Arizona, USA. The company stated that the production line costs over USD 1 billion and requires collaboration with equipment and material partners to establish a complete ecosystem. Currently, only a few companies in the industry can afford such an investment, and Intel seems to be the only company so far to successfully develop glass substrate.

Apart from Intel, SKC’s US subsidiary Absolics, AMD, and Samsung also see the broad development prospect of glass substrate.

In 2022, SKC’s US subsidiary Absolics invested around KRW 300 billion to establish the first factory dedicated to producing glass substrate in Covington, Georgia, USA. Recently, the company announced that the factory has been completed and has begun mass production of prototype products. Industry analysts believe this marks a critical moment for the global glass substrate market.

Samsung has formed an alliance composed of Samsung Electro-Mechanics, Samsung Electronics, and Samsung Display to develop glass substrate, aiming to start large-scale mass production in 2026 and commercialize the technology faster than Intel. It’s reported that Samsung Electro-Mechanics plans to install all necessary equipment on a pilot production line by September this year and commence operations in the fourth quarter.

AMD plans to launch glass substrate between 2025 and 2026 and to collaborate with global component companies to maintain its leading position. According to Korean media reports, AMD is conducting performance evaluation tests on glass substrate samples from several major global semiconductor substrate companies, intending to introduce this advanced substrate technology into semiconductor manufacturing.

Currently, with the emergence of new companies like SCHMID and the participation of laser equipment suppliers, display manufacturers, and chemical suppliers, the industry is gradually forming some new supply chains around glass core substrate, and create a diversified ecosystem.

Read more

• [News] Samsung Bullish on Glass Substrate Applications, Setting Up Production Line for Mass Production Starting in 2026

(Photo credit: Intel)