Projects per year
Abstract
We present the 1st embedded 14nm FeFinFET NVMs, 2T1CFE. Compared to 1T1CFE array, 2T1CFE one shows 25x↑of the SN ratio, 104x↓of the PGM power, and 103x↓of READ power. The gradual tuning of conductance for 2T1CFE has been tuned linearly and symmetrically with 3x104x of the window, where 8 separated conductance states can be stored with crystal gaps in between. Continual endurance cycles for 8 states are >107 times. The retention of 8 states has been baked in 85°C for >1 month. The 8 states have also passed the PGM disturbance test. Finally, this 2T1CFE array is as electrical synapses in a very deep neural network (DNN) with 23.8 million parameters for practice, and most importantly, the 2T1CFE is also as an activation, Rectified Linear Unit (ReLU), for the first time. Compared to the 5T-CMOS ReLU, the 2T1CFE ReLU shows 70%↑of inference accuracy. Overall, 2T1CFE shows strong potential as key components in high-performance low-power inference-accelerators.
Original language | English |
---|---|
Title of host publication | 2021 Symposium on VLSI Technology, VLSI Technology 2021 |
Publisher | Institute of Electrical and Electronics Engineers Inc. |
ISBN (Electronic) | 9784863487802 |
State | Published - 2021 |
Event | 41st Symposium on VLSI Technology, VLSI Technology 2021 - Virtual, Online, Japan Duration: 13 Jun 2021 → 19 Jun 2021 |
Publication series
Name | Digest of Technical Papers - Symposium on VLSI Technology |
---|---|
Volume | 2021-June |
ISSN (Print) | 0743-1562 |
Conference
Conference | 41st Symposium on VLSI Technology, VLSI Technology 2021 |
---|---|
Country/Territory | Japan |
City | Virtual, Online |
Period | 13/06/21 → 19/06/21 |
Fingerprint
Dive into the research topics of 'The First Embedded 14nm FeFinFET NVM: 2T1CFE Array as Electrical Synapses and Activations for High-performance and Low-power Inference Accelerators'. Together they form a unique fingerprint.Projects
- 1 Finished
-
仿生運算於AI晶片之開發與應用:建構通用型類神經網路技術平台與異質晶片整合於圖片辨識之應用(2/3)
Hsieh, E.-R. (PI)
1/02/21 → 31/01/22
Project: Research