Research Projects

Artificial intelligence (AI) is becoming a key driver in all economic sectors, including technology, manufacturing, healthcare, finance/banking, robotics, transportation, etc. There is an unfortunate downside to this revolution in terms of unsustainable global energy demands that could eventually dampen the prospects, as AI uses today’s computers with von Neumann architecture and separate memory/logic processors, suffering from massive energy inefficiency. While some commercial neuromorphic processors offer parallel logic/memory, they still use silicon CMOS technology. True brain-like computing enabled by neuromorphic devices is considered to be the only viable alternative. The brain consumes about 10 fJ per synaptic event (about 20 W), and the current AI chips exceed this by many orders of magnitude. Thus, there is an intensive global research on constructing artificial neurons and synapses using emerging nanomaterials. The success at the device level via identifying best candidate materials, viable manufacturing processes and demonstration of utility in neuromorphic computing and artificial visual perception systems can lead to system integration to a true neuromorphic computer or processor. This is the broad area of this proposal with a specific focus on optoelectronic synapse (OES) development. There have been many efforts to develop artificial synapses using various materials. Most are electrically driven, but limited in their ability to meet the 10 fJ goal. Optically driven synapses offer advantages of low power consumption, wide bandwidth, low crosstalk, and absence of RC delay. Thus, our effort here focuses on OES using 2D Moiré structures. There are two guiding and uncompromising principles driving our effort here. 1. We aim at rivaling the brain in energy consumption at 10 fJ per event or lower. Most previous studies are only at the pJ and nJ level. 2. The persistent photoconductivity effect - the backbone of OES - in most materials depends on defects, e.g., vacancies. This is contrary to the practice in the last 50 years of the electronics revolution, creating defect-free substrates and active layers. But OES critically relies on defects and thus, the ability to control them, i.e. defect engineering in 2D moiré materials, is critical, which is a new paradigm and unchartered avenue. The proposed effort will use 2D Bi2Se3, Bi2Te3, and In2Se3 to achieve the above energy goal with a broadband OES operation, including near-infrared (NIR), which has been elusive. In particular, the Moiré structures will be examined due to their promise for low energy and broadband operation. The monolayers typical in OES fabrication push the bandgap above 1 eV even when starting with a narrow bandgap material, thus pushing the peak response to the visible region. Simply making thicker multilayers of the same material presents fewer vacancies, leading only to photodetector operation. Employing the Moiré structure, which is truly novel, would create multilayers capable of persistent photoconductivity and reduce the bandgap, leading to broadband OES operation with ultralow energy consumption. The proposed work will include controlled growth of active layers by chemical vapor deposition, preparation of Moiré structures, identification/quantification of defects, careful defect engineering, two-terminal OES fabrication and testing, density functional theory simulations to understand mechanisms and five application demonstrations: handwritten digit/alphabet recognition, facial recognition, urban street scene segmentation, logic gates and Morse-code-based optical communication. Achieving the goals would help with low-energy (less than 10 fJ) optoelectronic synapses for neuromorphic computing and artificial visual perception systems, which could prevent a potential energy crisis with the present computing/AI systems. A broadband OES is critical for image sensing, preprocessing, transmitting signals and neuromorphic computing.