WIRE: Write Energy Reduction via Encoding in Phase Change Main Memories (PCM)
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866914142212849664 |
|---|---|
| author | Desai, Mahek Rumale, Apoorva Asadinia, Marjan Bogle, Sherrene |
| author_facet | Desai, Mahek Rumale, Apoorva Asadinia, Marjan Bogle, Sherrene |
| contents | Phase Change Memory (PCM) has rapidly progressed and surpassed Dynamic Random-Access Memory (DRAM) in terms of scalability and standby energy efficiency. Altering a PCM cell's state during writes demands substantial energy, posing a significant challenge to PCM's role as the primary main memory. Prior research has explored methods to reduce write energy consumption, including the elimination of redundant writes, minimizing cell writes, and employing compact row buffers for filtering PCM main memory accesses. However, these techniques had certain drawbacks like bit-wise comparison of the stored values, preemptive updates increasing write cycles, and poor endurance. In this paper, we propose WIRE, a new coding mechanism through which most write operations force a maximum of one-bit flip. In this coding-based data storage method, we look at the frequent value stack and assign a code word to the most frequent values such that they have a hamming distance of one. In most of the write accesses, writing a value needs one or fewer bit flips which can save considerable write energy. This technique can be augmented with a wear-leveling mechanism at the block level, and rotating the difference bit in the assigned codes, increasing the lifetime of the PCM array at a low cost. Using a full-system evaluation of our method and comparing it to the existing mechanisms, our experimental results for multi-threaded and multi-programmed workloads revealed considerable improvement in lifetime and write energy as well as bit flip reduction. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_04928 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | WIRE: Write Energy Reduction via Encoding in Phase Change Main Memories (PCM) Desai, Mahek Rumale, Apoorva Asadinia, Marjan Bogle, Sherrene Emerging Technologies Phase Change Memory (PCM) has rapidly progressed and surpassed Dynamic Random-Access Memory (DRAM) in terms of scalability and standby energy efficiency. Altering a PCM cell's state during writes demands substantial energy, posing a significant challenge to PCM's role as the primary main memory. Prior research has explored methods to reduce write energy consumption, including the elimination of redundant writes, minimizing cell writes, and employing compact row buffers for filtering PCM main memory accesses. However, these techniques had certain drawbacks like bit-wise comparison of the stored values, preemptive updates increasing write cycles, and poor endurance. In this paper, we propose WIRE, a new coding mechanism through which most write operations force a maximum of one-bit flip. In this coding-based data storage method, we look at the frequent value stack and assign a code word to the most frequent values such that they have a hamming distance of one. In most of the write accesses, writing a value needs one or fewer bit flips which can save considerable write energy. This technique can be augmented with a wear-leveling mechanism at the block level, and rotating the difference bit in the assigned codes, increasing the lifetime of the PCM array at a low cost. Using a full-system evaluation of our method and comparing it to the existing mechanisms, our experimental results for multi-threaded and multi-programmed workloads revealed considerable improvement in lifetime and write energy as well as bit flip reduction. |
| title | WIRE: Write Energy Reduction via Encoding in Phase Change Main Memories (PCM) |
| topic | Emerging Technologies |
| url | https://arxiv.org/abs/2511.04928 |