TY - GEN
T1 - Analog spike processing with high scalability and low energy consumption using thermal degree of freedom in phase transition materials
AU - Yajima, T.
AU - Nishimura, T.
AU - Toriumi, A.
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2018/10/25
Y1 - 2018/10/25
N2 - Spike integration and threshold processing are the basic signal processing in brain-inspired computing, such as deep learning, reservoir computing etc. In such processes, analog technology is essential for suppressing energy consumption. However, analog technology often faces problems in miniaturization due to deteriorated noise tolerance by scaling and intrinsically large analog elements such as capacitors. Here, we propose to exploit a thermal degree of freedom in phase transition materials for scalable and noise-tolerant analog spike processing. We focus on a two-terminal metal-insulator-transition VO2 device, where quasi-adiabatic Joule heating enables efficient spike integration, and metal-insulator transition implements threshold processing. This VO2 device is highly scalable, consuming only ∼1fJ/spike (smallest so far) according to the simulation. By using this device, fully autonomous spike integration and threshold processing are also demonstrated. Exploiting the quasi-adiabatic thermal degree of freedom will facilitate scalable and energy-efficient analog implementation for a wide range of brain-inspired computing.
AB - Spike integration and threshold processing are the basic signal processing in brain-inspired computing, such as deep learning, reservoir computing etc. In such processes, analog technology is essential for suppressing energy consumption. However, analog technology often faces problems in miniaturization due to deteriorated noise tolerance by scaling and intrinsically large analog elements such as capacitors. Here, we propose to exploit a thermal degree of freedom in phase transition materials for scalable and noise-tolerant analog spike processing. We focus on a two-terminal metal-insulator-transition VO2 device, where quasi-adiabatic Joule heating enables efficient spike integration, and metal-insulator transition implements threshold processing. This VO2 device is highly scalable, consuming only ∼1fJ/spike (smallest so far) according to the simulation. By using this device, fully autonomous spike integration and threshold processing are also demonstrated. Exploiting the quasi-adiabatic thermal degree of freedom will facilitate scalable and energy-efficient analog implementation for a wide range of brain-inspired computing.
UR - http://www.scopus.com/inward/record.url?scp=85056865865&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85056865865&partnerID=8YFLogxK
U2 - 10.1109/VLSIT.2018.8510649
DO - 10.1109/VLSIT.2018.8510649
M3 - Conference contribution
AN - SCOPUS:85056865865
T3 - Digest of Technical Papers - Symposium on VLSI Technology
SP - 27
EP - 28
BT - 2018 IEEE Symposium on VLSI Technology, VLSI Technology 2018
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 38th IEEE Symposium on VLSI Technology, VLSI Technology 2018
Y2 - 18 June 2018 through 22 June 2018
ER -