New computing device can be trained to learn like the human brain

April 30, 2021

A new transistor gets computers closer to thinking like humans. (Northwestern University)

Researchers from Northwestern University and the University of Hong Kong have created an electrochemical transistor that mimics brain synapse functions to form a circuit made of soft plastic, ultimately producing a novel device capable of learning by association in much the same way as our brains.

In a paper published Friday in Nature Communications, the group of engineers detailed how they enabled their circuit to simultaneously process and store information like the human brain, conditioning it to associate light with pressure. The transistor, an electronic device that can amplify or switch electronic signals and power, functions similarly to synapses, which are structures in the brain that allow neurons to pass signals to other neurons using neurotransmitters.

Their "synaptic transistor" could be trained to learn associatively, meaning that it could build on memories to learn over time. The most well-known example of associative learning is Ivan Pavlov's dog experiments, in which dogs were conditioned to associate a ringing bell with food. Eventually the dogs salivated whenever they heard the ringing bell in anticipation of being given food. The scientists who develop computer technologies have always been inspired by the human brain, but recently there has been more research on replicating the way brains combine their computing and storage processes within devices, according to the paper. 

Jonathan Rivnay, a senior author of the paper and assistant professor of biomedical engineering at Northwestern University, and Xudong Ji, the paper's first author and a postdoctoral researcher in Rivnay's research group, told The Academic Times that, "The key for a device to perform brain-like computing is to process and store data together, which requires the device [to have] non-volatile memory behavior."

"The way our current computer systems work is that memory and processing units are physically separated. The computer performs computations and then sends that information to a memory unit," they continued. "This means that you need to recall the information from the memory unit every time you want to retrieve that information, which causes huge energy consumption."

The researchers' proposed synaptic transistor device brings these two functions together. Currently, the most common technology used to combine processing and memory functions is the memory resistor, or "memristor." But memristors are energy-costly and not very biocompatible, meaning the materials from which they are typically constructed are not highly compatible with living tissue. By contrast, this team's organic electrochemical synaptic transistor operates with low voltages and is far more compatible with biological systems.

"Even high-performing organic electrochemical synaptic transistors require the write operation to be decoupled from the read operation," Rivnay said, referring to the computer operations in which data is transferred between the memory and processing units. "So if you want to retain memory, you have to disconnect it from the write process, which can further complicate integration into circuits or systems."

"Our device is an organic electrochemical transistor with an organic active [material] that can trap ions during operation," the authors said. This design allows the memory effect to be maintained while the write and read operations are still coupled. The ions behave similarly to neurotransmitters, and because the transistor retains stored data from the trapped ions, it can remember previous activities. This feature mimics both the short-term and long-term plasticity behavior of a synapse in the brain. Synaptic plasticity refers to how synapses strengthen or weaken over time based on their level of activity. 

The memory effect allowed the researchers to test the circuit's brain-like abilities through an associative learning task. At the beginning of the experiment, the circuit only responded to the presence of pressure from a finger press, the unconditioned stimulus. The circuit was then trained with pressure and light from an LED bulb, the conditioned stimulus, together. The circuit was eventually able to associate the two, and the light alone could trigger a signal in the circuit in what is called the unconditioned response.

As the authors note in the paper, the human brain can easily outperform computers in tasks such as pattern recognition, motor control and multisensory integration. But the demonstration of associative learning from their device acts as the "first important step" that their circuit can bridge the gap between brain and computer in artificial intelligence products. What's more, as compared to traditional computers, their device is more energy efficient, has a higher fault tolerance and is able to perform more tasks at the same time.

Thus far, the circuit is a proof-of-concept device, but it can be further extended to include more sensory inputs and integrated with other electronics for more efficient computation, according to the authors. The circuit is also made of soft organic polymers, which means it may eventually be used in soft, wearable electronics, smart robotics and implantable devices.

Future work on the device might involve scaling down the size and collaborating with other researchers and computer scientists to develop new architectures and algorithms for its potential applications. The authors are currently working on the device's implementation and further understanding its materials as a follow-up to this paper. 

The study, "Mimicking associative learning using an ion-trapping non-volatile synaptic organic electrochemical transistor" published April 30 in the Nature Communications journal, was authored by Xudong Ji, the University of Hong Kong and Northwestern University; Bryan D. Paulsen, Ruiheng Wu and Jonathan Rivnay, Northwestern University; Gary K. K. Chik and Paddy K. L. Chan, Advanced Biomedical Instrumentation Centre at the Hong Kong Science Park and the University of Hong Kong; and Yuyang Yin, the University of Hong Kong.

We use cookies to improve your experience on our site and to show you relevant advertising.