TLDR

  • Samsung (SSNLF) and Nvidia (NVDA) are collaborating to develop the next-generation ferroelectric NAND flash memory chips.
  • Together with researchers from the Georgia Institute of Technology, the two companies have co-developed an AI model named Physics-Informed Neural Operator, or PINO.
  • The PINO model can analyze the performance of ferroelectric NAND devices more than 10,000 times faster than traditional methods.
  • Standard chip simulation tools (TCAD) take about 60 hours for each run, while the new AI system reduces that to less than 10 seconds.
  • Last year, Samsung announced that ferroelectric NAND can cut power consumption by 96% compared to conventional NAND.

Samsung and Nvidia are working jointly on a new kind of memory chip that might alter the way AI hardware is constructed and powered.

The two companies, along with researchers from the Georgia Institute of Technology, have co-developed an AI model called Physics-Informed Neural Operator, or PINO, with the aim of accelerating research into ferroelectric NAND flash memory.

Ferroelectric NAND is made using ferroelectric materials rather than conventional silicon. These materials can store data without requiring a continuous power supply, making them appealing for energy-efficient chips.

Samsung SDI Co., Ltd., 0L2T.L
0L2T.L Stock Card

Samsung has been leading the research on this technology for some time. Last year, the company published findings in the journal Nature indicating that ferroelectric NAND can reduce power consumption by 96% compared to standard NAND.

That’s a figure worth pausing over. For AI workloads that consume a large amount of energy, a 96% reduction in memory power usage would be a significant development.

The issue has always been the research speed. Simulating and enhancing chip performance characteristics, such as threshold voltage and data retention, requires a tool called Technology Computer-Aided Design, or TCAD.

TCAD typically takes around 60 hours for each operation. This bottleneck slows down the speed at which engineers can test and refine chip designs.

The PINO model developed by the joint team changes this. It can perform the same analysis in less than 10 seconds.

A Partnership That Makes Sense

Nvidia is Samsung’s largest memory customer. The two companies already have a deep commercial relationship, which makes this R&D collaboration a natural extension.

SK Hynix, Samsung’s South Korean rival, currently supplies most of Nvidia’s high-bandwidth memory (HBM) chips. Samsung has been working to narrow that gap, and a deeper partnership on next-generation memory technology aligns with that strategy.

Micron Technology (MU) also competes in this area, manufacturing HBM chips for the AI chip market.

The research findings from the PINO project have been made public. The work is now moving towards commercialization, with Samsung expected to partner with Nvidia in the next phase of ferroelectric NAND development.

What the Numbers Say

Samsung’s financials remain strong. The company reported trailing twelve-month revenue of $234.73 billion with an operating margin of 13.07%.

Its debt-to-equity ratio is only 0.06, and its current ratio is 2.33, indicating a well-capitalized balance sheet with room for investment in long-term R&D.

The P/E ratio is 28.43, which is on the higher side historically. RSI readings suggest that the stock is currently in overbought territory, something traders might want to keep an eye on.

Institutional ownership is low at 2%, and no insider trading activity has been reported in the past year.

Neither Nvidia nor Samsung responded to media requests for comment at the time of reporting.