# CUTIE: Beyond PetaOp/s/W Ternary DNN Inference Acceleration with Better-than-Binary Energy Efficiency

arXiv: https://arxiv.org/abs/2011.01713 · HTML: https://mlsanity.com/paper/2011.01713/

## Key facts

- Cited by 8 papers in the map (data through 7 Oct 2026).
- First posted on arXiv 3 Nov 2020.

## Built on by (most cited first, 8 of 8)

- [Sub-mW Keyword Spotting on an MCU: Analog Binary Feature Extraction and Binary Neural Networks](https://mlsanity.com/paper/2201.03386/) (2022)
- [Efficient Deep Learning Infrastructures for Embedded Computing Systems: A Comprehensive Survey and Future Envision](https://mlsanity.com/paper/2411.01431/) (2024)
- [xTern: Energy-Efficient Ternary Neural Network Inference on RISC-V-Based Edge Systems](https://mlsanity.com/paper/2405.19065/) (2024)
- [Vau da muntanialas: Energy-efficient multi-die scalable acceleration of RNN inference](https://mlsanity.com/paper/2202.07462/) (2022)
- [How to train accurate BNNs for embedded systems?](https://mlsanity.com/paper/2206.12322/) (2022)
- [Stella Nera: A Differentiable Maddness-Based Hardware Accelerator for Efficient Approximate Matrix Multiplication](https://mlsanity.com/paper/2311.10207/) (2023)
- [A Construction Kit for Efficient Low Power Neural Network Accelerator Designs](https://mlsanity.com/paper/2106.12810/) (2021)
- [Full-Stack Optimization for CAM-Only DNN Inference](https://mlsanity.com/paper/2401.12630/) (2024)

## Abstract (arXiv)

We present a 3.1 POp/s/W fully digital hardware accelerator for ternary neural networks. CUTIE, the Completely Unrolled Ternary Inference Engine, focuses on minimizing non-computational energy and switching activity so that dynamic power spent on storing (locally or globally) intermediate results is minimized. This is achieved by 1) a data path architecture completely unrolled in the feature map and filter dimensions to reduce switching activity by favoring silencing over iterative computation and maximizing data re-use, 2) targeting ternary neural networks which, in contrast to binary NNs, allow for sparse weights which reduce switching activity, and 3) introducing an optimized training method for higher sparsity of the filter weights, resulting in a further reduction of the switching activity. Compared with state-of-the-art accelerators, CUTIE achieves greater or equal accuracy while decreasing the overall core inference energy cost by a factor of 4.8x-21x.