LANL researchers design novel artificial synapses for neuromorphic computing

June 2, 2023 – The human brain has been called the most complicated object in the universe. Seeking to replicate that as yet unmatched computing power, scientists at Los Alamos National Laboratory have built a new interface-like memristive device, which their results suggest could be used to build artificial synapses for next-generation neuromorphic computing. Memristive devices, or memristors, represent a long-sought circuit technology that, unlike current resistor technology, has both programming and memory capabilities, memristors could remember what electrical state they were in when turned off, an ability similar to the human brain that opens up new possibilities for computing and devices.

Tested on a dataset of handwritten images from the Modified National Standards and Technology database, the interface-type memristors realized a high image recognition accuracy of 94.72%. Image: LANL.

Data processing is an essential part of science today, with machine learning, artificial intelligence and artificial neural networks being used to address pressing issues in everything from climate science to national security applications, said Aiping Chen , laboratory scientist at the Center for Integrated Nanotechnologies. But conventional computing architecture consumes a great deal of energy and is less and less able to scale to meet increasingly large data challenges. Neuromorphic computing, which mimics the unmatched architecture and data storage and processing capabilities of the human brain, offers a path to continue extending computing performance.

Conventional computation is constrained by the so-called von Neumann bottleneck, where computation and memory are separated. Processing advanced tasks such as machine learning and image recognition on digital computers consumes a significant amount of energy and time due to data transfer back and forth between a central processing unit and memory. Data center energy consumption has increased rapidly in recent years, with projections that approximately 8% of the world’s electricity will be used by data centers by 2030.

Also, in conventional computer architecture, billions of transistors on silicon-based microchips act as switches for a computer’s binary code. The physical limits on the miniaturization of those transistors helped spell the end of Moore’s Law, a maxim that computing power should double roughly every two years.

In-Memory Computing: Just like a brain

By co-locating information storage and processing in the synapses, which connect the 100 billion neurons that send and receive chemical information, the human brain’s in-memory processing saves time and energy. Neuromorphic computing relies on emerging devices such as memristors, switches between two terminals that control and remember flowing charge, to replicate the structure and function of synapses.

In the rapidly evolving field of neuromorphic computing, memristor designs have included filament systems, where a charge is delivered across devices. But, prone to overheating, filament systems lack stability and reliability.

Chen and his colleagues are working on a different approach called an interface-type memristor and have produced a reliable, high-performance device with a simple structure based on an Au/Nb-doped SrTiO3interface essentially gold and other semiconductor materials. Interface-type memristors can, in principle, be reduced to nanometer sizes that even filament-based memristor technology cannot achieve. (By contrast, a human hair is about 100,000 nanometers thick.) And especially in contrast to transistor-based neuromorphic chips, the interface-type memristive device needs much less energy to power its processing.

Different from digital computing with a von Neumann architecture, neuromorphic computing, inspired by biological systems, works just like a brain, Chen said. The advantages of this design include low power consumption, high parallelism, and excellent fault tolerance. After all, the human brain runs on just 20 watts, yet it learns extremely effectively. These benefits make it great for advanced computing tasks like learning, recognition, and decision making.

Excel in advanced computing tasks

The team used artificial neural network simulation to study the computational performance of the interface-type memristor, testing it against a handwritten image dataset from the Modified National Standards and Technology database maintained by the National Institute of Standards and Technology. Demonstrating excellent uniformity, programmability, and reliability, the device achieved a recognition accuracy of 94.72%.

These performances lead the team to believe that these new interface-like memristive devices could be a critical hardware component for next-generation neuromorphic computing.

The capabilities they were seeing suggest that neuromorphic chips, like human brains, will be good at advanced tasks that include real-time learning and decision-making, Chen said. We could see the neuromorphic computer enabling many applications that require intelligence, from self-driving cars to drones to security cameras. Basically, a lot of things that people are able to do, these types of devices will be able to do.

The team plans to continue developing the technology with an emphasis on the need for co-engineering hardware design based on algorithmic approaches offered by computer scientists.

Paper:An interface-type memristive device for artificial synapses and neuromorphic computing, advanced intelligent systems. DOI:10.1002/aisy.202300035

Financing:The work was supported by the laboratory-led research and development program at Los Alamos National Laboratory, NNSA, and funding from the US Department of Energy’s Office of Sciences Center for Integrated Nanotechnology.

Source: LANL

#LANL #researchers #design #artificial #synapses #neuromorphic #computing

Leave a Comment