Technology is getting closer to the super-speed world of computing with artificial intelligence. However, the world may not have the necessary hardware to handle the workload of new AI breakthroughs.
According to Erica Carlson, 150th Anniversary Professor of Physics and Astronomy at Purdue University, the current silicon computer architectures were not designed for the brain-inspired codes of the AI revolution.
A joint effort between Physicists from Purdue University, University of California San Diego (USCD), and École Supérieure de Physique et de Chimie Industrielles (ESPCI) in Paris, France, has discovered a way to rework the hardware by mimicking the synapses of the human brain. Their findings were published in the October 2023 edition of Advanced Electronic Materials.
Carlson, the lead theoretical scientist of this research, believes that new paradigms in hardware will be necessary to handle the complexity of future computational advances. She states that “neuromorphic architectures hold promise for lower energy consumption processors, enhanced computation, fundamentally different computational modes, native learning, and enhanced pattern recognition.”
Essentially, neuromorphic architecture refers to computer chips that mimic brain behavior. Neurons transmit information in the brain and have synapses, which are small gaps that allow signals to pass from one neuron to the next. In biological brains, synapses encode memory. The team of scientists concludes that vanadium oxides show promise for neuromorphic computing as they can be used to make artificial neurons and synapses.
Carlson explains that the dissonance between hardware and software is the reason for the high energy cost of training large language models like ChatGPT. Neuromorphic architectures, on the other hand, mimic the basic components of a brain, such as neurons and synapses, leading to lower energy consumption. However, silicon is not suitable for neuron-like behavior, so materials with different behavior from silicon are needed for efficient neuromorphic hardware solutions. Only a few materials, mostly quantum materials, have demonstrated the ability to mimic both synapses and neurons.
The team relied on a recently discovered type of non-volatile memory in vanadium oxides, which is driven by repeated partial temperature cycling through the insulator-to-metal transition.
Alexandre Zimmers, lead experimental scientist from Sorbonne University and École Supérieure de Physique et de Chimie Industrielles, explains that only a few quantum materials are suitable for future neuromorphic devices. In vanadium dioxide, they can observe optically what changes in the material as it operates as an artificial synapse. The memory accumulates throughout the entirety of the sample, presenting new opportunities for control.
According to Carlson, the memory appears as shifts in the local temperature at which the material transitions from insulator to metal or vice versa. The team proposes that these changes accumulate due to the preferential diffusion of point defects into the metallic domains during the transition.
Having established that vanadium oxides are potential candidates for future neuromorphic devices, the team plans to continue their research by exploring ways to enhance the synaptic behavior of the material through ion bombardment and observation of its effects.