A team of researchers from Northwestern University, Boston College, and the Massachusetts Institute of Technology (MIT) has exploited the complex workings of the human brain to create an innovative synaptic transistor. This advanced device not only processes information but also stores it, mirroring the versatility of the human brain. The team's recent experiments have shown that the transistor can not only perform simple machine learning tasks, but also classify data and perform associative learning.
Although previous research has used similar strategies to develop brain-inspired computing devices, these transistors were unable to operate in cryogenic environments. In contrast, the new device operates stably at room temperature. It also runs very fast, consumes very little energy, and retains stored information even when the power is off, making it ideal for practical applications.
The research was recently published in the journal Nature.
Mimic the efficiency of the brain
"In a digital computer, data moves back and forth between the microprocessor and memory, which consumes a lot of energy and creates bottlenecks when trying to perform multiple tasks simultaneously. In the brain, on the other hand, memory and information processing are co-located and fully integrated, so energy efficiency is orders of magnitude higher. Our synaptic transistor also enables simultaneous memory and information processing, thus more faithfully simulating the brain." Mark C. Hersam of Northwestern University said he was a co-leader of the study.
Hersam is the Walter P. Murphy Professor in the Department of Materials Science and Engineering at Northwestern University's McCormick School of Engineering. He is also chair of the Department of Materials Science and Engineering, director of the Center for Materials Research Science and Engineering, and a member of the International Institute of Nanotechnology. Hersam co-led the research with Qiong Ma of Boston College and Pablo Jarillo-Herrero of MIT.
The driving force behind development
Recent advances in artificial intelligence (AI) have led researchers to develop computers that operate more like the human brain. Traditional digital computing systems have separate processing and storage units, resulting in data-intensive tasks consuming large amounts of energy. As smart devices continue to collect vast amounts of data, researchers are working to find new ways to process this data without consuming more and more power. Currently, memristors, or "memristors," are the state-of-the-art technology that can perform both processing and storage functions. However, memristors still have the problem of high switching energy consumption.
"For decades, the paradigm in electronics has been to make everything with transistors and use the same silicon architecture," Hessam said. "We have come a long way by integrating more and more transistors in integrated circuits. The success of this strategy cannot be denied, but it comes at the cost of high energy consumption, especially in the current era of big data, where digital computing is moving towards overwhelming the power grid. We must rethink computing hardware, especially for artificial intelligence and machine learning tasks."
Innovative design using moiré pattern
To rethink this paradigm, Hessam and his team explored new advances in moiré physics. Moiré is a geometric design that appears when two patterns are layered on top of each other. When two-dimensional materials are stacked together, new properties are created that are not available in a single layer of material. When these layers are twisted to form moiré patterns, unprecedented tunability of electronic properties is possible.
In the new device, the researchers combined two different types of atomically thin materials: double-layer graphene and hexagonal boron nitride. When these two materials are stacked together and intentionally twisted, moire patterns form. By rotating one layer relative to another, researchers can achieve different electronic properties in each layer of graphene, even though they are only atomic-scale apart. By choosing the right twist, the researchers exploited moiré physics to achieve neuromorphic functionality at room temperature.
"With the twist as a new design parameter, there are a lot of permutations," Hessam said. "Graphene and hexagonal boron nitride are very similar in structure, but different enough to produce an unusually strong moiré effect."
Advanced features and testing
To test the transistor, Hessam and his team trained it to recognize similar, but not identical, patterns. Just earlier this month, Hessam unveiled a new nanoelectronic device capable of analyzing and classifying data in an energy-efficient way, but his new synaptic transistor takes machine learning and artificial intelligence one step further.
"If the purpose of AI is to imitate human thinking, one of the lowest-level tasks is to classify data, which is simply to classify it into categories," Hessam said. "Our goal is to push AI technology toward higher-level thinking. Real-world conditions are often more complex than current AI algorithms can handle, so we tested our new device under more complex conditions to verify its advanced capabilities."
First, the researchers showed the device a pattern: 000 (three zeros in a row). They then asked the AI to identify similar patterns, such as 111 or 101. "If we train it to detect 000 and then give it 111 and 101, it will learn that 111 is more similar to 000 than 101 is. 000 and 111 are not exactly the same, but they are both consecutive three-digit numbers. Recognizing this similarity is a higher-level form of cognition known as associative learning," Hessam explained.
In experiments, the new synaptic transistor successfully recognized similar patterns, demonstrating its associative memory capabilities. Even when the researchers threw hard problems at it - such as giving it incomplete patterns - it still managed to demonstrate associative learning.
Current AI can be easily confused, which can cause major problems in some cases. Imagine if you were using a self-driving car and weather conditions worsened. Vehicles may not be able to interpret more complex sensor data as well as human drivers. But even if we give the transistor imperfect input, it can still recognize the correct response.
Compiled source: ScitechDaily