Efficient Synaptic Delay Implementation in Digital Event-Driven AI Accelerators
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866929684835467264 |
|---|---|
| author | Meijer, Roy Detterer, Paul Yousefzadeh, Amirreza Patino-Saucedo, Alberto Tang, Guanghzi Vadivel, Kanishkan Xu, Yinfu Gomony, Manil-Dev Corradi, Federico Linares-Barranco, Bernabe Sifalakis, Manolis |
| author_facet | Meijer, Roy Detterer, Paul Yousefzadeh, Amirreza Patino-Saucedo, Alberto Tang, Guanghzi Vadivel, Kanishkan Xu, Yinfu Gomony, Manil-Dev Corradi, Federico Linares-Barranco, Bernabe Sifalakis, Manolis |
| contents | Synaptic delay parameterization of neural network models have remained largely unexplored but recent literature has been showing promising results, suggesting the delay parameterized models are simpler, smaller, sparser, and thus more energy efficient than similar performing (e.g. task accuracy) non-delay parameterized ones. We introduce Shared Circular Delay Queue (SCDQ), a novel hardware structure for supporting synaptic delays on digital neuromorphic accelerators. Our analysis and hardware results show that it scales better in terms of memory, than current commonly used approaches, and is more amortizable to algorithm-hardware co-optimizations, where in fact, memory scaling is modulated by model sparsity and not merely network size. Next to memory we also report performance on latency area and energy per inference. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2501_13610 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Efficient Synaptic Delay Implementation in Digital Event-Driven AI Accelerators Meijer, Roy Detterer, Paul Yousefzadeh, Amirreza Patino-Saucedo, Alberto Tang, Guanghzi Vadivel, Kanishkan Xu, Yinfu Gomony, Manil-Dev Corradi, Federico Linares-Barranco, Bernabe Sifalakis, Manolis Neural and Evolutionary Computing Artificial Intelligence Synaptic delay parameterization of neural network models have remained largely unexplored but recent literature has been showing promising results, suggesting the delay parameterized models are simpler, smaller, sparser, and thus more energy efficient than similar performing (e.g. task accuracy) non-delay parameterized ones. We introduce Shared Circular Delay Queue (SCDQ), a novel hardware structure for supporting synaptic delays on digital neuromorphic accelerators. Our analysis and hardware results show that it scales better in terms of memory, than current commonly used approaches, and is more amortizable to algorithm-hardware co-optimizations, where in fact, memory scaling is modulated by model sparsity and not merely network size. Next to memory we also report performance on latency area and energy per inference. |
| title | Efficient Synaptic Delay Implementation in Digital Event-Driven AI Accelerators |
| topic | Neural and Evolutionary Computing Artificial Intelligence |
| url | https://arxiv.org/abs/2501.13610 |