While AI is often seen by the public as being affiliated with software, researchers at Han Wang’s Nanoscale Materials and Devices Lab of USC’s Department of Electrical and Computer Engineering Ming Hsieh and in the Mork family’s chemistry department are focused on improving the performance of AI and machine learning. by the material. The lab, whose work is focused on neuromorphic computing or brain-inspired computing, is conducting new research that introduces hardware improvements by exploiting a quality known as “randomness” or “stochasticity”. Their research, now published in Nature Communications, contradicts the perception of randomness as a quality that will have a negative impact on the results of calculations and demonstrates the use of finely controlled stochastic characteristics in semiconductor devices to improve the optimization of data. performance.
In the brain, chance plays an important role in human thought or calculation. It was born from billions of neurons that multiply in response to input stimuli and generate many signals that may or may not be relevant. The decision-making process is perhaps the best-studied example of how our brains use chance. It allows the brain to take a detour from past experiences and explore a new solution when making a decision, especially in a difficult and unpredictable situation.
“Neurons exhibit stochastic behavior, which can help certain computational functions,” said Jiahui Ma, a doctoral student at USC, and lead author Xiaodong Yan (both equally as first authors). The team wanted to emulate neurons as much as possible and designed a circuit to solve combinatorial optimization problems, which are one of the most important tasks for computers.
The idea is that for computers to do this effectively, they have to behave more like the human brain (on super steroids) in terms of processing stimuli and information, as well as making decisions.
In much simpler terms, we need computers to converge on the best possible solution. According to the researchers, “the randomness introduced into the new device demonstrated in this work may prevent it from getting stuck on a not-so-viable solution, and instead keep looking until it finds a result close to. the optimal. This is especially important for optimization problems, says the corresponding author, Professor Wang, “If we can dynamically adjust the random characteristics, the machine for performing the optimization can operate more efficiently as we do. wish. “
Researchers achieve this dynamic “tuning” by creating a specialized device, a hetero-memristor. Unlike transistors which are logical switches inside an ordinary computer chip, the hetero-memristor combines memory and computing. Memristors have been developed previously, normally with a two terminal structure. The innovation of the Viterbi team is to add a third electrical terminal and modulate its voltage to activate the neuron-like device and to dynamically adjust the stochastic characteristics of its output, much like heating a pot of water and adjusting dynamically the temperature to control the activity of the water molecules, thus allowing the so-called simulated “cooling”. This provides a level of control that older memristors do not have.
The researchers say: “This method emulates the stochastic properties of the activity of neurons.” In fact, the activity of neurons is perceived as random, but can follow a certain pattern of probability. The hetero-memristors they developed introduce such probability-ruled randomness into a neuromorphic computer circuit by the reconfigurable tuning of the device’s intrinsic stochastic property.
So this is a more sophisticated building block for creating computers that can solve sophisticated optimization problems, which can potentially be more efficient. In addition, they can consume less energy.
The full research team includes Xiaodong Yan, Jiahui, Ma Tong Wu, Aoyang Zhang,
Jiangbin Wu, Matthew Chin, Zhihan Zhang, Madan Dubey, Wei Wu, Mike Shuo-Wei Chen, Jing Guo and Han Wang.
The research was carried out in collaboration with the Army Research Laboratory, the University of Florida and Georgia Tech.
Posted on November 21, 2021
Last updated on November 21, 2021
Comments are closed.