Samsung begins production of 12-layer HBM3E chips for Nvidia

Samsung has reportedly started the mass production of 12-layer HBM3E chips for Nvidia. The expansion of production reportedly started early this year, and the company's manufacturing line that was being operated at a low rate has been converted into a high volume production line.

High-bandwidth chips

According to a report from ZDNet Korea, Samsung is confident that it will get Nvidia'a approval for its new 12-layer HBM3E chips. The South Korean firm has thus started the mass production of high-end memory chips that will be used in Nvidia's AI accelerators used by several tech firms for AI processing.

The mass production of Samsung's improved 12-layer HBM3E chips reportedly started in February. These chips are based on the company's 1a DRAM (fifth-generation 10nm class), which might offer better performance than Samsung's previous HBM3E chips that failed Nvidia's tests in the second half of last year, causing major losses.

Apparently, Samsung can now make 120,000 to 130,000 units of HBM3E chips per month. It usually takes around 6 months for the entire production process. If Samsung receives Nvidia's approval in June or July of this year, it would be hard for the company to start the mass production then, as it could be too late.

By the end of this year, Nvidia could switch to HBM4 chips for its new AI accelerator called Rubin. So, the company has started mass production of 12-layer HBM3E chips already, hoping to have enough stock ready by the time it receives approval from Nvidia.

Samsung is also planning to double the production of HBM chips form 4 billion Gb (gigabits) last year to 8 billion Gb (gigabits) this year. The production was just 600-800 million Gb in the first quarter of this year. So, the company now needs to increase the pace of production of it wants to meet the target.

Image Credits: Samsung