DELIVERCHIP（HK）LIMITED — 30 December: This Week’s News

Highlights of the Week

1. NXP signs radar partnership agreement with South Korean start-up bitsensing

2. Lisa Su: AMD will not merge with Intel

3. NVIDIA launches generative AI supercomputer priced at just US$249

4. Reports suggest it is highly unlikely that Samsung will supply HBM3E chips to NVIDIA this year

NXP Signs Radar Partnership Agreement with South Korean Start-up Bitsensing

According to reports, South Korean start-up Bitsensing has established a partnership with Dutch computer chip manufacturer NXP to sell radar systems for automotive applications.

Bitsensing CEO Jae-Eun Lee stated in an email that the agreement will focus on combining NXP’s radar chips with Bitsensing’s radar hardware and software, though financial terms were not disclosed.

Jae-Eun Lee noted that radar is typically used in safety applications and offers advantages over lidar and cameras in adverse weather conditions. The company’s radar technology boasts a longer detection range than that of its competitors and is capable of operating in both horizontal and vertical planes; the company has raised US$52 million since its establishment in 2018.

It is understood that bitsensing’s core business is 4D imaging radar. Previously, when discussing the prospects for imaging radar, Jae-Eun Lee remarked, “The current performance of imaging radar cannot yet rival the ultra-high resolution of existing lidar systems, but it holds enormous potential. This means that, as imaging radar technology continues to evolve, it is expected to achieve performance levels capable of replacing lidar, and its rate of development is likely to surpass that of lidar itself.”

Lisa Su: A merger between AMD and Intel is impossible

There had previously been market rumours that, in response to the serious crisis facing Intel, the US government believed it should facilitate a merger between Intel and AMD if necessary. However, AMD CEO Lisa Su has dismissed this possibility.

In an interview with *Time* magazine, Lisa Su discussed Intel’s change of CEO, noting that the role is indeed a “daunting” one and that she holds Mr Kisinger in high regard. At the same time, she denied speculation regarding a potential merger between Intel and AMD, stating that the US government harbours no such intentions.

Analysts generally believe that Kissinger’s departure may pave the way for Intel’s future reorganisation and break-up. Many experts believe that Intel may sell its manufacturing business in the future to focus on chip design.

Citi analyst Chris Danely pointed out that the foundry business is the main cause of massive losses and the key reason why the board’s patience has run out.

NVIDIA launches generative AI supercomputer priced at just $249

On Tuesday (17th), NVIDIA launched a generative artificial intelligence (AI) supercomputer called the ‘Jetson Orin Nano Super’. NVIDIA CEO Jensen Huang stated that this developer kit achieves higher performance and a lower price through software upgrades, costing just $249. Compact enough to fit in the palm of your hand, it can boost generative AI performance by up to 1.7 times.

The company stated that this product is suitable for commercial AI developers, enthusiasts and students. Compared to its predecessor, it offers 1.7 times the generative AI inference performance, a 70 per cent increase in overall performance, and can execute up to 67 trillion INT8 operations per second (67 INT8 TOPS), whilst memory bandwidth has also increased by 50 per cent to 102 GB/s.

In a video recorded to promote the product, Jensen Huang stated that whether creating Retrieval-Augmented Generation (RAG)-based large language model (LLM) chatbots, building visual AI agents, or deploying AI-powered robots, the Jetson Orin Nano Super is the ideal choice, as this kit delivers powerful computing performance at a lower cost.

As the product supports low-power designs and can operate at power consumption levels ranging from 7 watts to 25 watts, it is suitable for a wide range of edge-to-cloud computing scenarios, whether in smart cities, smart agriculture or robotics development.

Huang also noted that the Jetson Orin Nano Super is ideal for those wishing to develop their skills in the fields of generative AI, robotics or computer vision. Amid the trend in the AI sector towards a shift from task-specific models to foundational models, it also provides developers with an easy-to-use platform to turn their ideas into reality.

Reports suggest it is highly unlikely that Samsung will supply HBM3E chips to Nvidia this year

Samsung’s HBM3E process has yet to pass Nvidia’s qualification tests, with supply expected to commence in 2025.

Sources familiar with the matter state that Samsung has failed to gain entry into NVIDIA’s supply chain, with reports suggesting the company has not passed the qualification tests and HBM standards set by NVIDIA. The South Korean giant acknowledged this development in an investor report, stating that it remains optimistic about potential breakthroughs; however, a new report claims that it appears ‘highly unlikely’ that Samsung will supply HBM3E to NVIDIA this year, although the outlook for 2025 appears positive.

Given that it is now a foregone conclusion that Samsung will not supply its HBM products to NVIDIA in 2024, it is no exaggeration to say that Samsung’s fierce rival, SK Hynix, has taken the lead in the race for HBM supremacy. It is reported that the main reason for Samsung’s failure to pass NVIDIA’s sample testing was the extremely high standards set by SK Hynix, primarily due to the use of advanced methods such as ‘MR-MUF’. This made it difficult for Samsung to gain NVIDIA’s trust and, ultimately, to secure orders.

Samsung, which once reigned supreme in the HBM and NAND markets, appears to be losing its dominance, as competitors such as SK Hynix and Micron are already supplying components to industry giants, whilst Samsung remains conspicuously absent. Not only that, but even in the NAND sector, SK Hynix has become the ‘world’s first’ manufacturer to demonstrate NAND solutions with 321 layers or more, whilst Samsung has once again been left behind.

However, it is not all doom and gloom for Samsung, as the company plans to begin supplying HBM3E to giants such as NVIDIA as early as the first quarter of 2025. Furthermore, with the arrival of the next-generation HBM4 process, the company is expected to gain a slight edge due to its in-house semiconductor and memory production lines. However, given that SK Hynix has already partnered with TSMC to develop HBM4, the South Korean giant may need to act swiftly to capture the industry’s attention.

The news in this article is sourced from the internet.