A concept picture of chip Photo: VCG

Analog computing has attracted growing attention in the AI hardware field in recent years for its remarkable advantages in energy efficiency and processing speed. A research team from Nanjing University developed a high-precision analog in-memory computing scheme and, based on it, an analog in-memory computing chip, which achieved the highest computational precision ever reported in the field of analog in-memory computing. The findings were recently published in Science Advances, according to the Science and Technology Daily.

Although analog computing hardware offers high energy efficiency and strong parallelism, it still faces challenges such as low computational accuracy and limited stability, said Professor Miao Feng, corresponding author of the paper, to the Science and Technology Daily.

In this study, the research team proposed a high-precision analog in-memory computing framework and built a corresponding circuit architecture, which they verified through chip fabrication and testing. 

The team adopted a weight remapping technique to further enhance the chip's computational precision, Miao explained in the report. In parallel vector-matrix multiplication operations, the chip achieved a root-mean-square error of only 0.101 percent, setting a new record for accuracy in analog vector - matrix multiplication.

Another co-corresponding author Professor Liang Shijun said the chip remained stable under extreme temperatures ranging from -78.5°C to 180°C.

Tests conducted in strong magnetic fields showed that the output current of the chip's core unit did not deviate by more than 0.21 percent compared with its operation under normal magnetic conditions.

These findings confirm the reliability of the high-precision analog computing scheme in extreme environments, Miao said. He added that this breakthrough marks a significant step toward the practical application of analog in-memory computing technology, with promising potential to advance low-power, high-precision AI hardware innovations.

Global Times