A physics-based compact model of phase change for the design of cross-point storage-class memories

• Published in: Solid-state electronics 2021-11, Vol.185, p.107955, Article 107955
• Main Authors: Kim, Donguk, Tae Jang, Jun, Myong Kim, Dong, Choi, Sung-Jin, Ban, Sanghyun, Shin, Minchul, Lee, Hanwool, Dong Lee, Hyung, Mo, Hyun-Sun, Hwan Kim, Dae
• Format: Article
• Language: English
• Subjects: Ovonic threshold switching; PcRAM; physics-based compact model; Poole–Frenkel emission GST crystal growth
• ISSN: 0038-1101; 1879-2405
• DOI: 10.1016/j.sse.2020.107955

•We proposed a physics-based compact model for phase-change random access memory (PcRAM).•The ratio of vertical-to-lateral crystal growth rate (α) was incorporated into HSPICE via Verilog-A.•The proposed model was verified by using the experimental results taken from the 256 × 256 cross-point (X-point) PcRAM cell array with the Ge2Sb2Te5 2z-nm technology node.•Furthermore, we found that the SET pulse-dependent abrupt/gradual change of PcRAM resistance is sensitive to α. A physics-based compact model for phase-change random access memory (PcRAM) was proposed considering the ratio of vertical-to-lateral crystal growth rate (α), and it was incorporated into HSPICE via Verilog-A. The proposed model was verified using the experimental results taken from the 256 × 256 cross-point (X-point) PcRAM cell array with the Ge2Sb2Te5 20–22 nm ITRS technology node. The proposed compact model successfully reproduced the measured PcRAM cell resistance (RC) depending on the SET pulse width and amplitude after a background RESET, which is a challenging issue that holds back the X-point PcRAM as a promising candidate for a modern storage-class memory in perspective of the write latency and power consumption, without heavy computational burden while capturing the essence of physical meaning via the multidomain simulation which includes the threshold switching, electrical, thermal, and phase-change modules. The extracted α value was 1.55. Furthermore, it was found that the SET pulse-dependent abrupt/gradual change of RC is sensitive to α. This suggests that α should be carefully optimized for PCM-based neuromorphic applications