SK hynix revealed that HBM3E not only meets the speed specifications necessary for AI memory, but also achieves the world’s highest level in all aspects such as thermal control and customer convenience. In terms of speed, this product can process up to 1.15TB of data per second. It is equivalent to processing 230 full HD movies (5GB) in 1 second.

2. Samsung will launch more than 300 layers of NAND flash memory next year

According to Korean media reports quoted by Fast Technology, Samsung will launch more than 300 layers of flash memory in 2024, ahead of SK Hynix. A few days ago, SK Hynix announced the world’s first flash memory with more than 300 layers, reaching 321 layers, the type is TLC, and the single Die capacity is 1Tb, but it will not be mass-produced until 2025.

It is not yet clear the specific details of Samsung’s 300-layer flash memory, such as the exact number of layers, capacity density, flash memory type, etc., only know that it is still a double stack, not a triple stack like SK Hynix. This means that each layer of Samsung’s flash memory has more than 150 layers, which is undoubtedly more difficult. According to the roadmap planned by Samsung, by 2030, the flash memory stack will exceed 1,000 layers.

3. It is said that IC design manufacturers are unwilling to expand production for discounts

According to Taiwan Electronics Times quoted by Science and Technology Board Daily, most IC design companies point out that they need to place a certain amount of orders in order to obtain discounts from foundries.

Although the discounts given by suppliers in terms of volume-for-price are increasing, judging from the current market outlook, most IC design companies are unwilling to expand production and increase inventory levels in order to obtain more discounts. pressure.

4. In the second quarter, silicon wafer shipments ended their consecutive decline, but the peak season benefits were still weak

According to the Taiwan Business Times report, the latest quarterly wafer industry analysis report of the International Semiconductor Industry Association (SEMI)’s silicon product manufacturers organization shows that global silicon wafer shipments in the second quarter of 2023 will reach 3.331 billion square inches, a year-on-year decrease of 10.1. %, an increase of 2% month-on-month, ending the previous two consecutive quarters of recession.

The silicon wafer shipment area is better than the shipment performance of major wafer foundries. The legal person estimates that the foundries still require the foundries to pull goods according to the long-term contract, resulting in a relatively good performance of the silicon wafer factory in the second quarter. The outlook for the third quarter of major fabs is generally weak, and the quarterly revenue growth should be flat or slightly increased, with no obvious peak season effect. The legal person expects that the adjustment time of the semiconductor cycle will be prolonged. Under the condition that the inventory level of customers is still high, the purchase of silicon wafers will be slowed down, and the short-term operation of silicon wafer factories will face challenges.

5. It is rumored that the UK has spent huge sums of money to purchase thousands of high-performance AI chips

According to foreign media sources quoted by TechWeb, in order to catch up in the global computing power race, British Prime Minister Sunak plans to spend 130 million US dollars to purchase thousands of high-performance AI chips from Nvidia, AMD and Intel. It is reported that UK research and innovation, the UK science funding agency, has entered the advanced stage of ordering 5000 Nvidia GPUs.

According to recent media reports, Saudi Arabia has also purchased more than 3,000 Nvidia H100 processors, which are high-end components used by the company to train artificial intelligence, and cost $40,000 each.