COFFEE TERMINATED

Project "COFFEE" data sheet

The following table provides information about the project.

 Partnership

Take a look of project's partnership.

Map

 Project objective

With modern data demands and computational burdens rapidly expanding, technology must quickly move beyond the traditional von Neumann architecture that has driven computational advances since the 20th century. Taking its inspiration from the remarkable plasticity and power efficiency of the human brain, neuromorphic computing offers a promising approach to overcome the fundamental limitations imposed by the von Neumann architecture and the imminent demise of Moore’s Law. One notable formulation of neuromorphic hardware relies on analog memory elements called memristors (resistive switching devices). While resistive switching is a well-known phenomenon, its implementation in neuromorphic computing currently suffers from several serious issues, including significant device-to-device variations, binary (as opposed to analog) switching and cycle-to-cycle variability. In COFFEE (Controlling and Observing Filaments For Enhanced memristive Elements), we seek to overcome these shortcomings by studying the fundamental materials physics of conductive filaments as well as through iterative and targeted device optimization efforts. We will utilize novel experimental techniques, including in operando transmission electron microscopy (TEM) and scanning thermal microscopy (SThM), to visualize the formation and behavior of conductive filaments in practical devices. Insights gained from filament visualization experiments will be used to modify device design through geometric, chemical, and electrode engineering in the hopes of improving device performance. Improved memristors will be used for the fabrication of cross-bar arrays to perform benchmark computational tasks in neural network hardware and for neural network simulations. Through the study of conductive filaments and targeted engineering efforts, the performance of filamentary memristors can likely be dramatically improved and their implementation in viable neuromorphic technologies can move closer to reality.