STDP Based Online Learning for a Current-Controlled Memristive Synapse

• Published in: 2022 IEEE 65th International Midwest Symposium on Circuits and Systems (MWSCAS) pp. 1 - 4
• Main Authors: Weiss, Ryan, Das, Hritom, Chakraborty, Nishith N., Rose, Garrett S.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.08.2022
• Subjects: current-controlled circuit; Integrated circuit reliability; learning; LRS; memristor; Memristors; Neural networks; Reliability engineering; Resistance; Simulation; spiking neural network; STDP; synapse; Timing
• ISSN: 1558-3899
• DOI: 10.1109/MWSCAS54063.2022.9859294

Spike-timing-dependent plasticity (STDP) is a popular approach for online learning that determines synaptic weight updates based on the relative timing of temporal events of pre-synaptic and post-synaptic spikes. Online learning is very effective for fast and low power processing of locally sensed signals. Moreover, the memristor (or "memory resistor") has garnered attention as a key component in emerging synaptic circuits, due in large part to the inherent plasticity of the device. Circuits that leverage metal-oxide memristors, including HfO 2 , are promising but must be carefully designed to account for the non-idealities of the memristor itself. STDP circuits for memristor-based synapses can be designed based on either voltage- or current-controlled mechanisms. In this paper, a novel current-controlled memristive synapse is presented with an STDP online learning circuit designed for improved reliability. This circuit specifically limits the range of operation to resistance states near the low resistance state (LRS). It utilizes this property to achieve online learning with STDP characteristics. Simulation result are provided that show STDP operation of this circuit, including demonstrations of controlled operation using the more reliable switching characteristics of near-LRS resistance states.