Toward Wearable Artificial Intelligence: Flexible Nanocrystal Memristors That Mimic Multilevel Synaptic Functions
Xiaojie Xu | 20-LW-018
A memristor, which is a resistor with a memory effect, is analogous to a biological synapse and is considered as a leading candidate for neuromorphic (brain-like) computing and hardware-based artificial intelligence. Efforts to develop an artificial synapse with memristors have thus been boosted but now have hit a bottleneck due to the so called "stochastic conductive filament effect," where uncontrollable filament growth leads to large device performance variability and poor switching endurance.
We proposed a novel approach to realizing a better control over the conductive filaments through material designing and grain boundary engineering. We investigated two material systems as the active switching layer: nanocrystals and polymers. For the nanocrystal memristors, we demonstrated the first fiber-shaped nanocrystal memristors with analog behaviors based on lead selenide (PbSe) nanocrystals. For the polymer memristors, we demonstrated a flexible polymer memristor based on FK800 (a copolymer of chlorotrifluoroethylene (CTFE) and vinylidene fluoride (VF2)) with an outstanding switching cycle of greater than 1,000,000, which is two to three orders of magnitude higher than all the other polymer/organic memristors in literature, and comparable to that of the state-of-art inorganic memristors. We also developed the first bio-inspired injury response system based on the effective integration of a triboelectric generator (an artificial mechanoreceptor), a nociceptive memristor (an artificial nociceptor), and a light emitting diode (an artificial bruise), which mimicked a sense of pain, signs of injury, and the healing process of the human body in response to noxious stimuli.
This research, which is to develop the fundamental hardware for neuromorphic computing, helps advance Lawrence Livermore National Laboratory's missions in high-performance supercomputers and advanced materials and manufacturing and aligns with DOE's call on neuromorphic computing as a basic science need. This project is the first memristor research in at the Laboratory, which not only enables new research directions and capabilities for Livermore in developing high-performance fundamental devices for neuromorphic computing, but also opens new areas for exploration of next-generation wearable electronic devices.
Publications, Presentations, and Patents
Xu, X. 2021. "An artificial nociceptor mimicked by a flexible polymer memristor," Weapons and Complex Integration seminar, Lawrence Livermore National Laboratory, Livermore, CA, June, 2021. LLNL-PRES-823325.