- Future memory could be more durable thanks to developments with wurtzite ferroelectrics materials
- A new structural design has enabled a demonstration of 10 billion writing cycles, around 100 times the previous limit
- The breakthrough was made by scientists from the Xidian University working with the City University of Hong Kong and Fudan University.
Next generation memory for future AI systems will need far faster and more durable performance, and scientific researchers in China seem to have made a breakthrough, demonstrating a new design which can complete some 10 billion writing cycles, 100 times more than the previous limit.
This development – by teams at the Xidian University working with the City University of Hong Kong and Fudan University – is one of the most significant breakthroughs in memory development, and relies on the discovery of a quirk in the behavior of Nitrogen atoms.
Dense, low-power memory chips now seem to be a realistic possibility for a future demanding ever-scaling AI processing.
Wurtzite ferroelectrics
Creating faster memory and storage with greater capacity while keeping the energy requirements low is the holy grail of computing.
Over the past few years, research has increasingly focused on wurtzite ferroelectrics (polar crystal materials) such as aluminium scandium nitride (AlScN), which has the advantage of being compatible with semiconductor manufacturing processes currently in use.
It also offers that low energy target, along with the superior switching speed required by fast storage. Until now, the main challenge had been reliability. Testing of repeated switching in previously developed AlScN chips had resulted in failure after around 100 million writing cycles, but the researchers have found a solution.
Nitrogen cornfields
So, why has this breakthrough been made now? The previous failed attempts led researchers into deeper investigation as to why the AlScN chips would top out at 100 million cycles, which is where they found the problem: missing nitrogen atoms.
Wang Ruiqing, is a doctoral researcher at Xidian University and one of the authors of the research paper on the study, recently published in journal Science.
“Think of ferroelectric material as a neatly planted cornfield,” Wang said, “with nitrogen vacancies representing spots where seedlings are missing.” The defects were not the problem itself, however, but rather their behavior during switching. Rather than remain stationary, the nitrogen vacancies would move, and even group together. The consequence of this was the creation of “pathways” that allowed electricity to leak.
Upon discovering this, the team designed a structure to minimise the nitrogen vacancies, slowing the deterioration of the material.
“In the past, researchers knew that the devices failed and could observe some of the symptoms, but no one had been able to explain at the atomic scale what was moving, how they were moving and how that movement ultimately caused failure.” Wang added.
If successful, AlScN-based RAM and storage could form the backbone of future server farms and provide the framework for greater LLM and artificial intelligence adoption.
