Paper Overview
- Field: Machine Learning
- Authors: Solomon Micheal Serunjogi, Rachmad Vidya Wicaksana Putra, Ayat Taha, Muhammad Shafique, Mahmoud Rasras
- Published: 2026-07-28
- arXiv: 2607.26016
- MDTransformer performs complex matrix operations using spatial-mode interference, leveraging inverse-designed multi-mode couplers, crossings, and Mach-Zehnder IQ modulators integrated into a compact mode-division photonic tensor core (MPTC) that executes matrix multiplication in the optical domain.
- Each guided-wave mode (TE0-TE3) serves as an independent computation channel, delivering 4x parallelism per waveguide without spectral filtering or free-spectral-range (FSR) limitations.
- Coherent detection and IQ modulation jointly encode amplitude and phase, enabling complex-valued arithmetic for the full range of operations in Transformers.
- The design offers sub-4-bit effective precision analog multiplication with inter-mode crosstalk below -30 dB.
- The inverse-design methodology provides scalability and full compatibility with 1550 nm single-laser continuous-wave operation.
- 40.4% area reduction
- 63.6% power savings
- 40.6% energy savings
- Comparable latency
Abstract
Recently, photonic transformer accelerators (PTAs) have successfully achieved significant speedup and energy efficiency improvements over electronic accelerators for expediting Transformer inference. However, state-of-the-art solutions rely on expensive multi-wavelength light generation and large dot-product units due to active phase-shifter components, making them inefficient and impractical. To address this, the authors propose MDTransformer, a novel hardware-software co-design of a PTA based on mode-division optical dataflow and operations.
Key Design
Results
Across diverse workloads (DeiT-Tiny/Small/Base and BERT-Base/Large), MDTransformer achieves compared to state-of-the-art PTAs:
---
*Source: arXiv:2607.26016, auto-collected on 2026-07-30.*