Researchers at Postech have found ways to speed up artificial intelligence (AI) technology. Their work focuses on Electrochemical Random-Access Memory, or ECRAM, a new type of memory for AI. ECRAM is a device that stores and processes data using ion movement, allowing flexible, analog-type data storage.

This research, published in Nature Communications, could make AI faster and more energy-efficient. AI needs to process huge amounts of data quickly. Current computers separate memory, where data is stored, from processors, where data is calculated. This separation causes delays and uses extra energy because data must move between them. In-Memory Computing solves this by doing calculations inside the memory itself, speeding up processes and saving power. ECRAM supports this by storing data in a way that allows direct calculations.

Unlocking ECRAM’s potential

Understanding ECRAM’s complex structure has been difficult, slowing its use in real products. The researchers created an ECRAM device with multiple terminals using tungsten oxide. They used a tool called the Parallel Dipole Line Hall System, a setup with two rotating magnets that creates a strong magnetic field to study electron behavior. This let them observe how electrons move inside ECRAM at very cold temperatures, down to -223°C, and at room temperature. They found that tiny gaps, called oxygen vacancies, create paths for electrons to travel easily. These paths make ECRAM work efficiently without needing more electrons. The device stayed stable even in extreme cold, proving its reliability. This finding shows how ECRAM can handle AI tasks effectively.

“This research is significant as it experimentally clarified the switching mechanism of ECRAM across various temperatures,” said researcher Seyoung Kim in a Postech press release. “Commercializing this technology could lead to faster AI performance and extended battery life in devices such as smartphones, tablets, and laptops.”

This advance brings ECRAM closer to real-world use, potentially transforming AI hardware by making it quicker and less power-hungry.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.