Breakthrough in Analog Hardware Enhances AI Computational Efficiency

Researchers develop ECRAM-based technology to overcome limitations of digital hardware, achieving high accuracy and low power consumption.

The research team from POSTECH and Korea University successfully fabricated ECRAM devices in a 64×64 array.
ECRAM technology uses a three-terminal structure, allowing efficient data processing with low power.
The Tiki-Taka algorithm was applied to maximize AI neural network training accuracy.
The new hardware demonstrates superior electrical and switching characteristics, with high yield and uniformity.
This advancement marks the largest scale implementation of ECRAM devices for AI computation to date.