Highly Scalable (30 nm) and Ultra-low-energy (~5fJ/pulse) Vertical Sensing ECRAM with Ideal Synaptic Characteristics Using Ion-permeable Graphene Electrodes
Jongwon Lee, Revannath Dnyandeo Nikam, Dongmin Kim, Hyunsang Hwang
Abstract
We demonstrate vertical sensing Electrochemical Random-Access Memory (VS-ECRAM) with ideal synaptic characteristics for energy-efficient neuromorphic computing. In VS-ECRAM, excellent weight-update linearity was secured through precisely controlled ion injection using monolayer graphene, which acts as a drain electrode and a barrier layer. In addition, strong endurance and retention characteristics were confirmed thanks to the graphene barrier, which prevents ion intermixing at the channel/electrolyte interface. Furthermore, read and write energy $(\sim 5$ fJ/pulse) was reduced while maintaining the ideal synaptic behaviors in the device scaled down to 30x30 nm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . Also, the read latency was enhanced (*20) due to the vertical channel structure. Finally, a high MNIST pattern recognition accuracy was evaluated in a 1k-bit VS-ECRAM array (4F $^{2})$ with near-ideal synaptic characteristics.