Samsung Develops a New Type of Computing for Artificial Intelligence

Samsung demonstrated the world’s first in-memory computing based on MRAM. The company says this will enable the integration of memory and system semiconductors to create AI chips of the next generation.

To achieve this, researchers replaced the standard in-memory computation architecture based on the “sum of currents” with a new one based on the “sum of resistances.”

Performance testing of AI deployed on the compute-in-memory chip showed 98% accuracy in handwritten digit classification and 93% in face detection.

This work expands the boundaries of next-generation AI chip technologies with low power consumption,” the company said.

Researchers also suggested that the new MRAM chip could be used not only for in-memory computing but also as a platform for loading biological neural networks. This aligns with the neuromorphic electronics concept proposed by Samsung researchers in статье, published in Nature Electronics in September 2021.

In standard computer architecture data are stored in memory chips and processing is performed on the central processor. In-memory computing is a new paradigm. It aims to carry out both storage and computations within the memory network, without the need to move information.

This could reduce the electricity consumption of computing power when processing large data sets, scientists say.

Earlier in October 2021, Samsung unveiled brain-copying technology in a neural chip.

In August it was reported that the South Korean tech giant employed artificial intelligence in the design of processors for mobile devices.

Subscribe to ForkLog AI news on Telegram: ForkLog AI — all the news from the world of AI!