Mixed-Precision Training and Compilation for RRAM-based Computing-in-Memory Accelerators
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914407883210752 |
|---|---|
| author | Pelke, Rebecca Klein, Joel Cubero-Cascante, Jose Bosbach, Nils Joseph, Jan Moritz Leupers, Rainer |
| author_facet | Pelke, Rebecca Klein, Joel Cubero-Cascante, Jose Bosbach, Nils Joseph, Jan Moritz Leupers, Rainer |
| contents | Computing-in-Memory (CIM) accelerators are a promising solution for accelerating Machine Learning (ML) workloads, as they perform Matrix-Vector Multiplications (MVMs) on crossbar arrays directly in memory. Although the bit widths of the crossbar inputs and cells are very limited, most CIM compilers do not support quantization below 8 bit. As a result, a single MVM requires many compute cycles, and weights cannot be efficiently stored in a single crossbar cell. To address this problem, we propose a mixed-precision training and compilation framework for CIM architectures. The biggest challenge is the massive search space, that makes it difficult to find good quantization parameters. This is why we introduce a reinforcement learning-based strategy to find suitable quantization configurations that balance latency and accuracy. In the best case, our approach achieves up to a 2.48x speedup over existing state-of-the-art solutions, with an accuracy loss of only 0.086 %. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_21737 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Mixed-Precision Training and Compilation for RRAM-based Computing-in-Memory Accelerators Pelke, Rebecca Klein, Joel Cubero-Cascante, Jose Bosbach, Nils Joseph, Jan Moritz Leupers, Rainer Machine Learning Emerging Technologies Computing-in-Memory (CIM) accelerators are a promising solution for accelerating Machine Learning (ML) workloads, as they perform Matrix-Vector Multiplications (MVMs) on crossbar arrays directly in memory. Although the bit widths of the crossbar inputs and cells are very limited, most CIM compilers do not support quantization below 8 bit. As a result, a single MVM requires many compute cycles, and weights cannot be efficiently stored in a single crossbar cell. To address this problem, we propose a mixed-precision training and compilation framework for CIM architectures. The biggest challenge is the massive search space, that makes it difficult to find good quantization parameters. This is why we introduce a reinforcement learning-based strategy to find suitable quantization configurations that balance latency and accuracy. In the best case, our approach achieves up to a 2.48x speedup over existing state-of-the-art solutions, with an accuracy loss of only 0.086 %. |
| title | Mixed-Precision Training and Compilation for RRAM-based Computing-in-Memory Accelerators |
| topic | Machine Learning Emerging Technologies |
| url | https://arxiv.org/abs/2601.21737 |