Cao, T., Zhang, Z., Goh, W. L., Liu, C., Zhu, Y., & Gao, Y. (2023, June 11). A Ternary Weight Mapping and Charge-mode Readout Scheme for Energy Efficient FeRAM Crossbar Compute-in-Memory System. 2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS). https://doi.org/10.1109/aicas57966.2023.10168639
Abstract:
This work presents an edge-AI system built on capacitive ferroelectric random-access memory (FeRAM) crossbar array, which is compatible with CMOS backend-of-line (BEOL) fabrication process. A novel capacitive crossbar circuit and a ternary mapping technique are proposed. Compared to the conventional binary representation, the proposed ternary mapping improves the storage efficiency exponentially in weight resolution. The feasibility of neuromorphic computing system implemented on FeRAM crossbar array is explored with speech command classification task. A ResNet-32 model with 0.45M parameters is implemented on 64 × 64 FeRAM crossbar array with the measured FeRAM model. It achieved 97.12% inference accuracy with 2 ternary digits and 5% device variation on Google Speech Command dataset 35-command classification task.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the A*STAR - Nanosystems at the Edge (WP03)
Grant Reference no. : A18A4b0055
This research / project is supported by the A*STAR - Ferroelectric Aluminum Scandium Nitride (Al1-xScxN) Thin Films and Devices for mm-Wave and Edge Computing programme
Grant Reference no. : A20G9b0135