Artificial neurons mimic complex brain capabilities for next-generation artificial intelligence computation

Researchers have created atomically thin artificial neurons that can process both light and electrical signals for computation. The material allows for the simultaneous existence of separate feedforward and feedback paths within a neural network, increasing the ability to solve complex problems.

For decades, scientists have studied how to recreate the versatile computational capabilities of biological neurons to develop faster and more energy-efficient machine learning systems. One promising approach involves the use of memristors: electronic components that can store a value by changing their conductance and then use that value for processing in memory.

However, a key challenge in replicating the complex processes of neurons and biological brains using memristors has been the difficulty of integrating feedforward and feedback neuronal signals. These mechanisms underlie our cognitive ability to learn complex tasks, using rewards and errors.

A team of researchers from the University of Oxford, IBM Research Europe and the University of Texas have announced a major feat: the development of atomically thin artificial neurons created by stacking two-dimensional (2D) materials. The results have been published in Nature Nanotechnology.

In the study, the researchers expanded the functionality of electronic memristors by making them sensitive to optical as well as electrical signals. This allowed for separate feedforward and feedback paths to exist simultaneously within the network. The progress has allowed the team to create winning neural networks: computational learning programs with the potential to solve complex problems in machine learning, such as unsupervised learning in clustering and combinatorial optimization problems.

2D materials are made up of just a few layers of atoms, and this fine scale gives them various exotic properties, which can be fine-tuned depending on how the materials are layered. In this study, the researchers used a stack of three 2D materials, graphene, molybdenum disulfide and tungsten disulfide, to create a device that exhibits a change in its conductance depending on the strength and duration of the light/electricity being projected. on it.

Unlike digital storage devices, these devices are analog and function similar to the synapses and neurons in our biological brain. The analog feature allows for calculations, where a sequence of electrical or optical signals sent to the device produces gradual changes in the amount of stored electronic charge. This process forms the basis for threshold modes for neural computations, analogous to the way our brain processes a combination of excitatory and inhibitory signals.

The lead author, Dr. Ghazi Sarwat Syed, research staff member at IBM Research Europe Switzerland, said: “This is a very exciting development. Our study has introduced a new concept that goes beyond the fixed feedforward operation typically used in current artificial neural networks. Plus potential applications in AI hardware, these current test results demonstrate an important scientific advance in the broader fields of neuromorphic engineering and algorithms, enabling us to better emulate and understand the brain.”

Dr Syed and Dr Yingqiu Zhou (who were DPhil students and lab colleagues at Oxford) conducted the experimental work. According to Dr Zhou, now a postdoctoral researcher at Denmark Technical University, their implementation captures the essential components of a biological neuron through the optoelectronic physics of low-dimensional systems.

They note that we have created atomically abrupt semiconductor junctions through the design of our heterostructure stack. The stack, in particular, provides a heterojunction that acts as a neuronal membrane, while the graphene electrodes that contact the heterojunction act as a neuronal soma. In this way the neuronal state is represented in the soma, but modified by membrane changes, just like in real neurons.

As the advancement of artificial intelligence applications has grown exponentially, the required computing power has exceeded the development of new hardware based on traditional processors. There is an urgent need to research new techniques, including the work of co-lead author Professor Harish Bhaskaran at the Advanced Nanoscale Engineering Laboratory, University of Oxford, and at the IBM Research laboratory in Zurich.

Professor Bhaskaran said: ‘The whole field is hugely exciting, as material innovations, device innovations and new insights into how they can be applied creatively must all come together. This work represents a new toolkit, which explores the power of 2D materials, not in transistors, but for new computing paradigms.”

Co-author Professor Jamie Warner, from the University of Texas at Austin, said: ‘The use of such 2D structures in computing has been talked about for years, but we are only now finally seeing the result after spending over seven years in the development of wafer-scale 2D monolayers in complex ultrathin optoelectronic devices, this will enable the initiation of new information processing approaches using 2D materials based on industrially scalable fabrication methods.”

“Our findings are more exploratory in nature than actual system-level demonstrations,” says Dr. Syed. “Although we aim to expand on this concept in the future, we are convinced that our current proof-of-principle results demonstrate important scientific interest in the broader fields of neuromorphic engineering, enabling us to better emulate and understand the brain.”

Professor Bhaskaran points out that exciting research developments are important for future innovation, but this is not the technology one should expect in their mobile phones in the next couple of years.