Hardware
Sustainable Transformer Neural Network Acceleration with Stochastic Photonic Computing
arXiv:2604.09759v1 Announce Type: cross Abstract: Transformers achieve state-of-the-art performance in natural language processing, vision, and scientific computing, but demand high computation and me
arXiv:2604.09759v1 Announce Type: cross Abstract: 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.
Related
- Integrated electro-optic attention nonlinearities for transformers
- Tessera: Unlocking Heterogeneous GPUs through Kernel-Granularity Disaggregation
- Deep Optimizer States: Towards Scalable Training of Transformer Models Using Interleaved Offloading
- CodeQuant: Unified Clustering and Quantization for Enhanced Outlier Smoothing in Low-Precision Mixture-of-Experts
- Characterizing Performance-Energy Trade-offs of Large Language Models in Multi-Request Workflows
Source: arXiv cs.LG | 2026-04-14