- Memristors to bring brain-like computing to artificial intelligence systems
- Atomically tunable devices offer energy-efficient AI processing
- Neuromorphic circuits open new possibilities for artificial intelligence
A new frontier in semiconductor technology could be closer than ever after the development of atomically tunable “memristors,” which are cutting-edge memory resistors that emulate the neural network of the human brain.
With funding from the National Science Foundation’s FuSe2 program, this initiative aims to create devices that enable neuromorphic computing, a next-generation approach designed for high-speed, energy-efficient processing that mimics the brain’s ability to learn and adapt. .
At the core of this innovation is the creation of ultra-thin memory devices with atomic-scale control, which will potentially revolutionize AI by allowing memristors to act as artificial synapses and neurons. These devices have the potential to significantly improve computing power and efficiency, opening up new possibilities for artificial intelligence applications, while training a new generation of semiconductor technology experts.
Challenges of neuromorphic computing
The project focuses on solving one of the most fundamental challenges of modern computing: achieving the precision and scalability needed to bring brain-inspired AI systems to life.
To develop high-speed, energy-efficient networks that function like the human brain, memristors are the key components. They can store and process information simultaneously, making them particularly suitable for neuromorphic circuits where they can facilitate the type of parallel data processing seen in biological brains, potentially overcoming the limitations of traditional computing architectures.
The joint research effort between the University of Kansas (KU) and the University of Houston led by Judy Wu, Distinguished Professor of Physics and Astronomy at KU, is supported by a $1.8 million grant from FuSe2.
Wu and his team have pioneered a method to achieve sub-2 nanometer thickness in memory devices, with film layers approaching a staggering 0.1 nanometer, about 10 times thinner than the average nanometer scale. .
These advances are crucial for the semiconductor electronics of the future, as they enable the creation of devices that are extremely thin and capable of performing precise functionality, with large area uniformity. The research team will also use a co-design approach that integrates design, manufacturing and materials testing.
In addition to its scientific goals, the project also has a strong focus on workforce development. Recognizing the growing need for trained professionals in the semiconductor industry, the team has designed an educational outreach component led by experts from both universities.
“The overall goal of our work is to develop atomically ‘tunable’ memristors that can act as neurons and synapses in a neuromorphic circuit. By developing this circuit, we aim to enable neuromorphic computing. “This is the main focus of our research,” Wu said.
“We want to mimic how our brain thinks, calculates, makes decisions, and recognizes patterns—essentially everything the brain does with high speed and high energy efficiency.”
