Dynamic Properties of 3D RRAM Memory Elements for In-Memory and Neuromorphic Computation

Integration of memory and logic is key to reduce power consumption in machine learning and neural network applications.

Data transfer between storage and computation units costs energy and adds to latency, known as the memory wall present in the von Neumann architecture.

New device technologies such as Resistive RAM (RRAM) provide opportunities for non-volatile operation combining speed and energy advantages.

However, fundamental scientific and technical questions remain that needs to be addressed in order to benefit with future energy savings.In this project, we will investigate the dynamic properties and computation capabilities of a new class of ideal 3D RRAM elements integrated on III-V nanowire MOSFETs used as selectors.

Building on demonstrated 1st generation prototypes, we will explore the potential in this new structure by integrating multilevel RRAM elements and investigate the ultimate switching limits using the MOSFET as a high performance local clock.

We will investigate their multistate operation and explore innovative ways to configure it for in-memory computation and as a memristive cell in neuromorphic applications.The RRAM element has potential as a non-volatile memory technology with very low power consumption (~1 pJ) and very fast switching speeds (~1 ns).

The unique technology in this project offers a new path for integration and leads to a set of very intriguing research questions related to in-memory and neuromorphic computing that we will address.