Hardware Implementation of Ring Oscillator Networks Coupled by BEOL Integrated ReRAM for Associative Memory Tasks

TL;DR

The paper tackles the ONN scalability problem, where coupling connections scale as $O(N^2)$ with network size. It introduces a BEOL-integrated ReRAM crossbar that provides dense, programmable coupling for CMOS ring-oscillator neurons, enabling in-memory, phase-encoded computation. The experimental demonstration with a conductive-metal-oxide ReRAM (CMO/HfOx) in a $5×5$ 1T1R array validates analog programming, robust retention, and a $2×2$ associative memory network that retrieves patterns via phase locking. The results suggest a viable path toward large-scale, on-chip ONNs with online learning capabilities.

Abstract

We demonstrate the first hardware implementation of an oscillatory neural network (ONN) utilizing resistive memory (ReRAM) for coupling elements. A ReRAM crossbar array chip, integrated into the Back End of Line (BEOL) of CMOS technology, is leveraged to establish dense coupling elements between oscillator neurons, allowing phase-encoded analog information to be processed in-memory. We also realize an ONN architecture design with the coupling ReRAM array. To validate the architecture experimentally, we present a conductive metal oxide (CMO)/HfOx ReRAM array chip integrated with a 2-by-2 ring oscillator-based network. The system successfully retrieves patterns through correct binary phase locking. This proof of concept underscores the potential of ReRAM technology for large-scale, integrated ONNs.

Figures (7)

• Figure 1: (a) Quadratic increase in coupling connections with network size growth. (b) Two coupling methods: conventional passive components vs. ReRAM devices.
• Figure 2: (a) The fabrication of CMO/HfO$_{\rm x}$ ReRAM chip, including (b) 1T1R arrays. (c) The distribution of initial forming voltages and (d) the following resistive switching of ReRAM devices on the wire-bonded $5\times5$ array. (e) The analog programming capability with 60 ns fast pulses is also demonstrated.
• Figure 3: (a) 2-by2 ONN and (b) its architecture implementation with ReRAM crossbar array. The experimental retention data of the implemented coupling resistance are shown in (c).
• Figure 4: Examples of resulting phase locking depending on (a) positive and (b) negative coupling between two CMOS ring oscillators (9-stage).
• Figure 5: Operation flow of the experiments and assembled ONN hardware system.