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.
@article{arxiv.2604.09759,
title = {Sustainable Transformer Neural Network Acceleration with Stochastic Photonic Computing},
author = {S. Afifi and O. Alo and I. Thakkar and S. Pasricha},
journal= {arXiv preprint arXiv:2604.09759},
year = {2026}
}