Transistor Mimics Human Brain Intelligence

In yet another example of how artificial intelligence (AI) technology is evolving at an unprecedented pace, researchers at three Massachusetts universities have developed a new synaptic transistor that mimics the human brain to engage in higher-level thinking.

The device—designed by a team of collaborators from Northwestern University, Boston College, and the Massachusetts Institute of Technology (MIT)—goes beyond simple machine-learning to categorize data and perform associative learning, as well as to simultaneously process and store information.

While scientists already have explored the development of other brain-like computing devices that behave similarly, this one has several characteristics that set it apart from those, the researchers said. For one, other devices could rarely function outside of cryogenic temperatures, making them severely limited for any type of real-world application, they said. 

The new device, on the other hand, is stable at room temperature and has additional, more evolved characteristics, they reported. Among them are that it can operate at fast speeds, consume very little energy, and retain stored information even without a connection to power.

Changing the Game for AI

Recent advances in AI like ChatGPT and others have motivated researchers to develop computers that operate more like the human brain. But the challenge has always been and remains that “the brain has a fundamentally different architecture than a digital computer,” said Mark Hersam, co-leader of the research and a professor of materials science and engineering, a professor of medicine, and a professor of chemistry at Northwestern University.

Related:MathWorks Exec Expects AI to Mature in 2024

"In a digital computer, data move back and forth between a microprocessor and memory, which consumes a lot of energy and creates a bottleneck when attempting to perform multiple tasks at the same time," he said. "On the other hand, in the brain, memory and information processing are co-located and fully integrated, resulting in orders of magnitude higher energy efficiency."

At the same time, the researchers had to rethink the paradigm of decades of electrical science, which “has been to build everything out of transistors and use the same silicon architecture,” he said. Currently, the memory resistor, or “memristor,” is the most well-developed technology that can perform combined processing and memory function. However, memristors still suffer from costly switching.

“Significant progress has been made by simply packing more and more transistors into integrated circuits," Hersam explained. "You cannot deny the success of that strategy, but it comes at the cost of high power consumption, especially in the current era of big data where digital computing is on track to overwhelm the grid. We have to rethink computing hardware, especially for AI and machine-learning tasks.”

Related:Balancing the Promise, Progress, & Problems of AI

Finding the Solution

To rethink how to build a smart transistor, Hersam, his co-leaders—Professors Qiong Ma of Boston College and Pablo Jarillo-Herrero of MIT—and their team explored new advances in the physics of moiré patterns, a type of geometrical design that arises when two patterns are layered atop one another. 

This layering causes the emergence of new properties that do not exist in one layer alone. Further, when those layers are twisted to form a moiré pattern, the modification paves the way for unprecedented tunability of electronic properties, the researchers said.

The team combined two different types of atomically thin materials to create the device—bilayer graphene and hexagonal boron nitride. They stacked and twisted them in such a way to form a moiré pattern, achieving different electronic properties in each graphene layer even though they are separated by mere atomic-scale dimensions, the researchers said. By using this type of twisting physics, the researchers created a device with neuromorphic functionality at room temperature, Hersam said. 

“With twist as a new design parameter, the number of permutations is vast,” he said. “Graphene and hexagonal boron nitride are very similar structurally but just different enough that you get exceptionally strong moiré effects.”

Testing AI for Design Success

The team published a paper on its work in the journal Nature. The researchers trained the device to recognize similar, but not identical patterns, and set about testing its performance.

In experiments, the new synaptic transistor successfully recognized similar patterns, demonstrating its capability for associative memory. Even when the researchers tried to trick the device — like giving it incomplete patterns — it still successfully demonstrated associative learning, they said.

“Current AI can be easy to confuse, which can cause major problems in certain contexts,” Hersam said. He used a self-driving vehicle in deteriorating weather conditions as an example. 

“The vehicle might not be able to interpret the more complicated sensor data as well as a human driver could," Hersam said. "But even when we gave our transistor imperfect input, it could still identify the correct response.”

The ultimate goal of the team's work is to create AI to handle such real-world conditions, which are often more complicated than current machine-learning algorithms can handle, Hersam said.

“If AI is meant to mimic human thought, one of the lowest-level tasks would be to classify data, which is simply sorting into bins,” he said. “Our goal is to advance AI technology in the direction of higher-level thinking."