Sustainable Transformer Neural Network Acceleration with Stochastic Photonic Computing
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915931643445248 |
|---|---|
| author | Afifi, S. Alo, O. Thakkar, I. Pasricha, S. |
| author_facet | Afifi, S. Alo, O. Thakkar, I. Pasricha, S. |
| contents | Transformers achieve state-of-the-art performance in natural language processing, vision, and scientific computing, but demand high computation and memory. To address these challenges, we present ASTRA, the first silicon-photonic accelerator leveraging stochastic computing for transformers. ASTRA employs novel optical stochastic multipliers and unary/analog homodyne accumulation in a crosstalk-minimal organization to efficiently process dynamic tensor computations. Evaluations show at least 7.6x speedup and 1.3x lower energy overheads compared to state-of-the-art accelerators, highlighting ASTRA's potential for efficient, scalable, and sustainable transformer inference. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_09759 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Sustainable Transformer Neural Network Acceleration with Stochastic Photonic Computing Afifi, S. Alo, O. Thakkar, I. Pasricha, S. Hardware Architecture Machine Learning Transformers achieve state-of-the-art performance in natural language processing, vision, and scientific computing, but demand high computation and memory. To address these challenges, we present ASTRA, the first silicon-photonic accelerator leveraging stochastic computing for transformers. ASTRA employs novel optical stochastic multipliers and unary/analog homodyne accumulation in a crosstalk-minimal organization to efficiently process dynamic tensor computations. Evaluations show at least 7.6x speedup and 1.3x lower energy overheads compared to state-of-the-art accelerators, highlighting ASTRA's potential for efficient, scalable, and sustainable transformer inference. |
| title | Sustainable Transformer Neural Network Acceleration with Stochastic Photonic Computing |
| topic | Hardware Architecture Machine Learning |
| url | https://arxiv.org/abs/2604.09759 |