Inspired by the human brain, researchers have developed a new type of synaptic transistor with higher-order thinking abilities.
The device, designed by researchers at Northwestern University, Boston College and the Massachusetts Institute of Technology (MIT), processes and stores information at the same time as the human brain. In the latest experiments, the researchers demonstrated that the transistor can not only perform simple machine learning tasks to classify data, but also perform associative learning.
Although previous research has used similar strategies to develop computing devices that mimic the human brain, those transistors cannot operate in environments other than cryogenic temperatures. In contrast, this new device can also work stably at room temperature. It operates fast, consumes very little energy, and retains stored information even when the power is off, making it ideal for real-world applications.
The study will be published in the journal Nature on Wednesday, December 20.
"The structure of the brain is fundamentally different from that of a digital computer," said Northwestern's Mark Brown. C. Hessam said he was a co-leader of the study. "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 at the same time. In the brain, on the other hand, memory and information processing are co-located and fully integrated, which results in energy efficiency orders of magnitude higher than that of digital computers. Our synaptic transistors also achieve parallel memory and information processing functions, more faithfully mimicking the brain. ”
Hessam is a professor of materials science and engineering at Northwestern University's McCormick School of Engineering. Professor P. Murphy. He is also the Chair of the Department of Materials Science and Engineering, the Director of the Center for Materials Research Science and Engineering, and a member of the International Institute of Nanotechnology. Hessam co-led the study with Qiong Ma of Boston College and Pablo Jarillo-Herrero of MIT.
Recent advances in artificial intelligence (AI) have inspired researchers to develop computers that function more like the human brain. Traditional digital computing systems have separate processing and storage units, resulting in a large amount of energy consumption for processing data-intensive tasks. As smart devices continue to collect vast amounts of data, researchers are eager to find new ways to process this data without consuming more and more energy. Currently, memory resistors, or "memristors", are the most established technology capable of performing combined processing and memory functions. However, memristors still have the problem of high energy consumption.
"For decades, the design paradigm for electronics has been to build everything with transistors and use the same silicon architecture," Hessam said. "Significant progress has been made by simply fitting more and more transistors into integrated circuits. You can't deny the success of this strategy, but it comes at the cost of high energy consumption, especially in the current era of big data, where digital computing is expected to overwhelm the grid. We must rethink computing hardware, especially for AI and machine learning tasks. ”
To rethink this paradigm, Hessam and his team explored new advances in the physics of moire patterns, a geometric design that arises when two patterns are superimposed. When 2D materials are stacked, some new properties appear that do not exist in a single layer. And when these layers are twisted to form a moire pattern, an unprecedented tunability of electronic properties becomes possible.
For the new device, the researchers combined two different types of atomic-thin materials: bilayer graphene and hexagonal boronitride. When these materials are superimposed and purposefully twisted, they form a moire pattern. By rotating one layer relative to each other, researchers can even achieve different electronic properties in each layer of graphene, even though they are only atomic-level distances from each other. By choosing the right twist angle, the researchers used Moire physics to achieve brain-like function at room temperature.
"The possibilities for permutations and combinations of twists as a new design parameter are enormous," Hessam said. "Graphene and hexagonal boron nitride are very similar in structure, but there are enough differences that you can get a very strong moire effect. ”
To test the transistor, Hessam and his team trained it to recognize similar, but not identical, patterns. Just earlier this month, Hessam introduced a new nanoelectronic device capable of analyzing and classifying data in an energy-efficient way, but his new synaptic transistors take machine learning and AI technology a step further.
"If AI is designed to mimic the human mind, one of the most fundamental tasks is to classify the data, that is, simply categorize it," Hessam said. "Our goal is to push AI technology to the next level of thinking. Real-world conditions tend to be more complex than current AI algorithms can handle, so we tested our new devices under more complex conditions to validate their advanced capabilities. ”
The researchers first showed the device a pattern: 000 (three zeros in a row). They then asked the AI to recognize similar patterns, such as 111 or 101. "If we train it to detect 000 and then give it 111 and 101, it knows that 111 is more similar to 000 than 101," Hessam explained. "000 and 111 are not exactly the same, but they are both three numbers in a row. Recognizing this similarity is a higher-level form of cognition known as associative learning. ”
In the experiment, the new synaptic transistor successfully recognized similar patterns,
Demonstrates its associative memory capabilities. Even though the researchers gave it an incomplete pattern, it still managed to demonstrate the ability of associative learning.
"Current AI is easily confused, which can lead to serious problems in some cases," Hessam said. "Imagine if you're using a self-driving vehicle and the weather conditions get worse. Vehicles may not be able to interpret more complex sensor data as accurately as human drivers. But even if we give our transistor imperfect input, it is still able to recognize the correct response. ”
The study, "Moiré Synaptic Transistors with Room Temperature Brain-like Function", was largely supported by the National Science Foundation.