Samsung, SK, and Micron show off 5th-generation HBM technology at ‘major’ NVIDIA event for customers
Busy with ‘tailored’ development to meet various customer needs… The brain battle for ‘timely investment’ is also fierce.
(Seoul = Yonhap News) Reporter Jang Ha-na = While governments around the world, including the United States, are making an all-out effort to grow their own semiconductor industries, the semiconductor industry is competing’ n fierce for market leadership based on the competitiveness of advanced semiconductor technologies such as high bandwidth memory (HBM). .
To survive in the ‘war without gunshots’, they love Nvidia, a ‘big player’ in the artificial intelligence (AI) semiconductor market, and at the same time, a battle of brains is going on to secure higher -conductors. gap technology and avoid failing to make timely investments.
Is HBM competition intensifying? Following SK, Samsung also joins in (CG)
[연합뉴스TV 제공]
◇ “We are the first in the 5th generation HBM”… Competition between SK, Samsung, and Micron intensifies
According to the industry on the 24th, Samsung Electronics stock price increased 5.63% in the stock market on the 20th and closed at 79,300 won, up 3.12% on the 21st. On the 22nd, it rose to 79,900 gained during the day, raising expectations for ‘80,000 electronics’.
This is largely due to NVIDIA CEO Jensen Huang’s statement that “We are currently testing Samsung HBM and have high expectations.”
CEO Hwang, who attracted attention from all over the world by holding the annual developer conference ‘GTC 2024’ in San Jose, USA from the 18th to the 21st (local time), visited the Samsung Electronics booth in during this period he personally signed the HBM3E product on display, saying ‘Approved’, he also left behind.
Signature left by CEO Jensen Huang on Samsung HBM3E
(Seoul = Yonhap News) NVIDIA CEO Jensen Hwang reportedly visited the Samsung Electronics booth at the annual developer conference ‘GTC 2024’ held in San Jose, USA, and left a handwritten signature on HBM3E, a bandwidth memory high next generation (HBM). ). 2024.3.21 [한진만 삼성전자 부사장 SNS 캡처. 재판매 및 DB 금지] photo@yna.co.kr
CEO Hwang is working to cement the throne of AI semiconductors by introducing the next-generation AI chip ‘Blackwell’, which achieves up to 30 times the performance of existing products in ‘GTC 2024’. NVIDIA controls more than 80% of the graphics processing unit (GPU) market, which is essential for productive AI learning and inference.
Since NVIDIA’s new chips are supposed to have multiple high-performance HBMs, memory companies have gotten busy.
Currently, SK Hynix is taking the lead in the HBM market by almost exclusively supplying HBM3 to NVIDIA, while its competitors, Samsung Electronics and Micron, are in hot pursuit, anticipating a change in perception.
All three companies put the 5th generation HBM, HBM3E, at the forefront of their booths at ‘GTC 2024’.
In line with the opening of ‘GTC 2024’, SK Hynix announced that it would seriously mass produce the new 8-layer HBM3E product and supply it to its customer (NVIDIA) first. At the exhibition, it unveiled the HBM3E 12- real product tier, currently in the commercialization stage, and supplied HBM3 to NVIDIA It was also highlighted
SK Hynix “We supply NVIDIA”
(San Jose[미 캘리포니아주]= Yonhap News) Correspondent Kim Tae-jong = The fact that SK Hynix supplies HBM3 to NVIDIA is highlighted in the exhibition hall set up in the San Jose Convention Center in California on the 18th (local time) when a semi company -conductors of the United States The annual developer conference NVIDIA GTC 2024 was held. . 2024.3.19 taejong75@yna.co.kr
Samsung Electronics unveiled the industry’s first 12-layer, 36-gigabyte (GB) HBM3E, and Micron also showcased an 8-layer HBM3E product.
Micron, a latecomer to HBM, announced its performance for the second quarter of fiscal year 2024 (December last year to February this year) on the 20th, saying, “Sales have started to happen from HBM3E,” and also said sales contracts for most of the HBM production in 2025 has already ended.
In the future, the HBM production capacity (capa) of the three memory companies and ensuring a stable product are also considered major variables in the HBM competition.
Kim Rok-ho, a researcher at Hana Securities, said, “(Micron) announced that it has started mass production of HBM3E for NVIDIA, and the product issue is a point to watch.”
◇ “What’s next for HBM”… Samsung unveils ‘Mach 1’, which will change the AI semiconductor landscape
The semiconductor industry is also focusing on developing next-generation memory technology to succeed HBM, including ‘adaptive’ memory solutions that can respond to various customer needs as the AI market expands.
In December last year, Samsung Electronics established a new ‘Memory Product Planning Department’ which serves as a memory management tower, and prepares for the future by taking care of all areas, including technology trend analysis, product planning, commercialization and technical. support.
Samsung Memory Research Center (SMRC) at Samsung Electronics Hwaseong Campus
[삼성전자 제공. 재판매 및 DB 금지] photo@yna.co.kr
Samsung Electronics and SK Hynix are also focusing on expanding the ‘Compute Express Link’ (CXL) ecosystem, a next-generation memory technology.
CXL is a next-generation interface that can efficiently connect different types of products in applications that require high-performance calculations such as AI, machine learning, and big data. It can expand bandwidth and capacity while coexisting with existing main DRAM.
Recently, as interest in on-device AI increases, demand for ‘customized’ memory semiconductors such as low latency DRAM (LLW) supporting smartphones and augmented reality (XR) devices is also growing .
SK Hynix CEO Kwak No-jeong said at ‘CES 2024’ earlier this year, “Capacity and power efficiency may be important to some customers, while other customers may prefer bandwidth and information processing functions, ” and “SK Hynix is committed to responding to these demands.” “We are preparing our own dedicated memory platform,” he said.
SK Hynix also unveiled its next generation graphics DRAM, ‘GDDR7’, for the first time at ‘GTC 2024’. The maximum bandwidth is 128GB per second, which is twice that of the previous generation, and power efficiency is improved by 40%.
SK Hynix explained, “GDDR7 will process large amounts of data faster and play a vital role in advancing AI technology.”
Next generation SK Hynix DRAM ‘GDDR7’ graphics
[SK하이닉스 제공. 재판매 및 DB 금지]
As the global AI market is rapidly moving from learning to inference, Samsung Electronics surprisingly announced the development of ‘Mach 1’, a solution to solve the memory bottleneck, at the general meeting of shareholders on the 20th.
Gyeonghyeon Kyeong-hyun, head of Samsung Electronics’ Device Solutions (DS) division, said, “The Mach 1 AI (assembly) chip is currently being developed with the goal of reducing memory throughput by one-eighth and achieving eight work the power efficiency. “It will be the start of an innovation,” he said, “We are preparing to enable Large Language Model (LLM) compilation even when using low power (LP) memory instead of HBM.”
The industry interprets that Samsung Electronics, which lost leadership in the HBM market to SK Hynix, is accelerating the development of next-generation HBM and has adopted a ‘two-track’ strategy to completely change the AI landscape of the semiconductor market by introducing Mach 1, which will more or less replace HBM.
According to market research companies Omdia and KB Securities, the collective AI chip market is expected to reach $140.3 billion in 2030, a 24-fold increase compared to 2023 ($6 billion).
Kim Dong-won, a researcher at KB Securities, said, “In the AI learning process, the use of GPUs that can use computing resources flexibly is inevitable, but in conclusion, customized AI chips with better collection functions for all industries and company more effectively.”
CEO Kyeong-Hyeon Gyeong-Hyeon receives the ‘Samsung Highway’ road sign
(Seoul = Yonhap News) On the 8th (local time), Kyung Kye-hyun, head of Samsung Electronics’ Device Solutions (DS) division, visited the construction site of a foundry factory in Taylor, Texas, USA, and called it ‘ Samsung Highway’ to Williamson County Commissioner Bill Gravel, Getting road signs. 2023.1.15 [경계현 삼성전자 사장 인스타그램 캡처. 재판매 및 DB 금지] photo@yna.co.kr
◇ Samsung seems to be announcing additional investment plans in the US… SK begins construction of the Yongin cluster in March next year
As the competition for leadership in the high-tech semiconductor market is at its peak, major companies continue to invest.
In addition to research and development to bring even more advanced semiconductors to market, it is necessary to have production facilities and processes to supply customized products in a timely manner to large users such as NVIDIA.
For this, investment is essential. It is no exaggeration to say that this is a ‘rag war’ between semiconductor companies.
Samsung Electronics is currently investing more than $17 billion to build a foundry (manufacture of semiconductor loads) in Taylor, Texas, USA.
Although the US government recently announced that it would provide $19.5 billion (about 26 trillion won) in subsidies to its domestic company Intel, it is expected that $6 billion in subsidies will also be provided to Samsung Electronics . In line with this, there is also talk of the possibility that Samsung Electronics will announce additional investment plans focusing on AI semiconductors.
Samsung Electronics plans to invest 20 trillion won in the Giheung R&D complex by 2030 and double its semiconductor research center in size and quality.
Recently, as Samsung Display paid its first dividend since launch, Samsung Electronics also received dividends worth 5.6 trillion won, ensuring ‘real ammunition’ for investment in semiconductors.
Minister Ahn Deok-geun visits SK Hynix Yongin Semiconductor Cluster
(Seoul = Yonhap News) Minister of Trade, Industry and Energy Ahn Deok-geun visits the SK Hynix Yongin semiconductor cluster in Yongin-si, Gyeonggi-do on the 21st and tours the semiconductor production base construction site progressive memory. 2024.3.21 [산업통상자원부 제공. 재판매 및 DB 금지] photo@yna.co.kr
SK Hynix is investing more than 120 trillion won in Yongin, Gyeonggi Province to create the Yongin Semiconductor Cluster, a new memory production base. Construction will begin in March next year and a total of four fabs (semiconductor production factories) will be built by 2046.
Trade, Industry and Energy Minister Ahn Deok-geun recently visited the site to check on the progress and said, “Since the semiconductor gap depends on speed, all ministries will work together to ensure that our companies are not behind in the cluster speed. War.”
hanajjang@yna.co.kr
Report via KakaoTalk okjebo
< 更多代(c) 周年物利,
Unauthorized reproduction/redistribution, learning and use of AI prohibited>
2024/03/24 07:02 Sent
#반도체 #머니전쟁 #③ #Aiming #super #gap.. #Battle #leadership #advanced #semiconductors #HBM
