Today's computing systems that are capable of processing at phenomenal speed and efficiency are based on an architecture that was first proposed by John Von Neumann in 1945, before computers existed. Researchers who have searched for better ways to perform computing have been inspired by the human brain "architecture" that represents the ultimate in processing efficiency for applications such as image processing and pattern recognition. Attempts to create neurocomputers and associative memory circuits that mimic the brain via a network of coupled artificial neurons and synapses have been proposed. However, such implementations based on conventional silicon technology have largely been viewed as impractical and inferior to traditional computing hardware in terms of complexity and power consumption. This project will demonstrate a novel neurocomputing system based on the co-design of the system architecture and the nanoscale aluminum nitride (AlN) resonator devices that act as thermally-tunable elements for implementation of artificial neurons and synapses. The goal of this project is to demonstrate the functionality of such a system for a pattern recognition problem.
This project combines two exciting technologies, namely, modeling the function of the brain and piezoelectric materials, both of which are of interest to a broad community. Piezoelectric materials, such as crystals, are an exciting demonstration of converting mechanical energy into electrical energy. The same materials used in this project will be incorporated into the CMU undergraduate course 18-220, which is an introduction to circuits, to demonstrate energy harvesting and the important concept of resonance. A lab assignment based on converting vibrational energy to electrical energy will encourage the students to explore techniques for harvesting energy to support future electronic systems. Simpler samples of piezoelectric materials that react to being squeezed and vibrated will be used to light LEDs for demonstration to the C-Mite K through 10th classes at Carnegie Mellon (www.cmu.edu/cmites/ ). The brain-inspired electronic systems will be incorporated into graduate course projects for digital integrated circuit design at Carnegie Mellon. The results of this research will also be collaboratively exchanged with researchers from industry, specifically Intel and Qualcomm, to explore the benefits of neurocomputing-based associative memories for use in portable electronic systems.
