High-Density Hybridized CMOS-RRAM Integration and Low-Noise Characterization of Large-Scale Neuromorphic Compute-in-Memory Arrays
Abstract
Unmanned fleet/swarm Navy operations require persistent autonomy and resilience in adaptive natural intelligence on the edge, enabling their effective use in environments where direct human participation poses great risk to warfighter health and mission effectiveness. For greatest mobility and agility, these need to operate at extremely low size, weight and power (SWAP). This calls for a concerted effort in the fabrication and characterization of custom reconfigurable large-scale neuromorphic hardware at extreme efficiency and density. This acquisition and procurement of infrastructure equipment facilities and microfabrication in the Defense University Research Instrumentation Program is in support of research being conducted as part of the Office of Naval Research Science of Artificial Intelligence Program towards large-scale spike-based neuromorphic systems with persistent online deep learning. To attain record levels in high-density, low-energy operation of massively parallel neuromorphic hardware developed in this research, we target custom fabrication and experimental characterization of vertically integrated complementary metal-oxide semiconductor (CMOS) neuron arrays and emerging technology memristive crossbar synapse arrays. Specifically this DURIP will provide the following instrumentationresources to this research: 1) a low-noise, precisely temperature controlled probe station for acquiring and analyzing current-voltage characteristics across very large crossbar arrays of memristive elements such as the resistive random-access memory (RRAM) adaptable synapses under development; and 2) custom high-density hybridized integration of RRAM synapse arrays at 380nm pitch on top of CMOS wafers in 22nm fully-depleted silicon-on-insulator (FDSOI) technology. The combination of these two will provide a one-of-its-kind experimental platform, at unprecedented high density and high accuracy, in evaluating compute-in-memory architectures for neuromorphic adaptive AI onthe edge in current and future ONR programs.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2023
- Source ID
- N000142312162
Entities
People
- Gert Cauwenberghs
Organizations
- Office of Naval Research
- United States Navy
- University of California, San Diego