As memory shortages at artificial intelligence (AI) data centres worsen, European semiconductor research group imec has unveiled next-generation ferroelectric memory technology that could replace DRAM and NAND flash. It is still at the research stage, but is drawing attention as a new candidate to solve memory supply shortages in the AI era.

On June 24 local time, IT outlet TechRadar reported that imec presented its next-generation ferroelectric memory research at the 2026 IEEE-JSAP VLSI Technology and Circuits Symposium. The research is being assessed as technological progress that could accelerate commercialisation of a new memory architecture within the next 10 years.

There are two main points. The first is a ferroelectric capacitor that operates at low voltage. imec said the structure showed the potential to replace DRAM by maintaining high durability even under repeated writes while also securing data retention performance.

The second is a vertically stacked transistor structure. The technology applies a backgate modification technique to improve data-erasure problems in existing structures while achieving high density like NAND flash.

The research is drawing attention amid a worsening memory supply crunch centred on AI data centres. TechRadar reported that data centres are consuming about 70 percent of global memory output this year, sustaining the most severe supply pressure in the past 15 years. The impact is also putting consumer PC memory, as well as high-density NAND flash used in SSDs, under supply shortages and upward price pressure.

Memory makers also expect the supply crunch to be prolonged. Micron has projected that the consumer memory market may have to wait until 2028 to normalise. As a result, the data centre industry is accelerating efforts to secure faster and cheaper next-generation memory technologies, in addition to expanding existing memory.

Ferroelectric memory itself is not a new concept. It was first proposed in 1952, but remained at laboratory level for a long time and did not lead to commercialisation. But as memory shortages intensify with the expansion of AI infrastructure, it is again drawing interest as a next-generation memory candidate.

Maarten Rossmullen (마르턴 로스뮬런), programme director at imec, said, "We are solving the most urgent challenges facing memory technology by combining a range of research capabilities, from materials science to advanced three-dimensional (3D) integration technology." He added, "We are researching various next-generation memory technologies to support the rapid growth of AI and data-intensive applications."

imec is also drawing attention for its broad cooperation with major global semiconductor companies. Partners include Nvidia, ASML and TSMC, as well as Samsung Electronics, Intel, Micron, Qualcomm, AMD and Apple. This means that if the research results are commercialised, they could spread across global semiconductor companies and AI data centres.

Still, commercialisation will take time. imec also acknowledged that the research is at the proof-of-concept (PoC) stage and that technical challenges remain to be solved.

The industry does not see it as a technology that can resolve the memory supply crunch in the short term, but is assessing it as meaningful progress because it shows that AI-era memory demand has begun to change technology roadmaps centred on DRAM and NAND.

A key point to watch is whether ferroelectric memory can secure mass-production-level yield and cost competitiveness and establish itself in the next-generation memory market.