| Nov 07, 2025 |
Brain-inspired AI moves closer as researchers build high-yield memristor chipsResearchers achieved wafer-scale, 95% yield memristor circuits, unlocking new potential for dense, efficient AI chips modeled on the human brain.(Nanowerk News) A research team at DGIST in South Korea has taken a major step toward building chips that work more like the human brain. Led by Professor Sanghyeon Choi from the Department of Electrical Engineering and Computer Science, the group has developed a highly integrated memristor device at full wafer scale, a milestone for next-generation AI hardware (Nature Communications, "Wafer-scale fabrication of memristive passive crossbar circuits for brain-scale neuromorphic computing"). |
 |
| The figure shows how researchers built and tested a wafer-scale array of tiny memory-and-processing devices called memristors, from fabrication steps to a completed chip ready for measurement. (Image: Reproduced from DOI:10.1038/s41467-025-63831-2, CC BY) (click on image to enlarge) |
| The brain’s power comes from roughly 100 billion neurons linked by about 100 trillion synapses, giving it massive processing ability in a compact space. Engineers have long tried to copy this structure in AI chips, but today’s designs are still bulky and power-hungry. |
| That’s where the memristor comes in. This tiny component can both store data and perform calculations, thanks to its ability to “remember” how much electric current has passed through it. Its streamlined design allows for far greater density than conventional chips. When arranged in a crossbar configuration, memristors can pack dozens of times more information into the same area than standard SRAM chips. |
| Until now, though, memristors have been held back by production challenges like low yields, high energy loss, and tricky fabrication steps. The new study addresses those limits by using a method called “co-design,” which synchronizes materials, components, circuits, and algorithms from the start. The result: a memristor crossbar circuit with an impressive 95 percent yield on a 4-inch wafer, all without complex manufacturing. |
| The team also stacked the memristors in a 3D structure, a design that supports expansion into larger AI systems. When they tested the technology with a spiking neural network, it delivered stable, efficient performance in real AI tasks. |
| “This study proposed a method for improving memristor integration technology, which had been limited in the past,” said Professor Choi. “We are expecting it to lead to the development of a next-generation semiconductor platform in the future.” |
| Source: Daegu Gyeongbuk Institute of Science and Technology (Note: Content may be edited for style and length) |
