A Photonic Transformer Processor: Thermal-Optical Screening with Direct CMOS Detection

We present a photonic Transformer attention processor based on athermal-optical screening architecture with direct CMOS photodetection.Unlike silicon-photonic MZI mesh approaches, our design uses free-spaceoptical paths where light passes through thermo-optic phase shiftersencoding query vectors, is attenuated by electro-optic modulatorsencoding key vectors, and is directly detected by a CMOS Ge-on-Siphotodetector array. Bipolar differential detection provides intrinsiccommon-mode noise rejection. We verify 25 validation items spanningdevice physics, algorithm robustness, thermal-field simulation,electro-optic bandwidth, temperature drift, detector physics, opticallink budget, and full-system Monte Carlo. At D=128, bipolar dot-productSpearman rho=0.9971, and complete Transformer attention achievesrho>=0.995 across all configurations.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-06-14
DOI
https://doi.org/10.5281/zenodo.20690521
Primary Topic
Neural Networks and Reservoir Computing
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
preprint

A Photonic Transformer Processor: Thermal-Optical Screening with Direct CMOS Detection

Wayne
Zenodo (CERN European Organization for Nuclear Research)
Neural Networks and Reservoir Computing
preprint

A Photonic Transformer Processor: Thermal-Optical Screening with Direct CMOS Detection

Wayne
preprint en

Abstract

We present a photonic Transformer attention processor based on athermal-optical screening architecture with direct CMOS photodetection.Unlike silicon-photonic MZI mesh approaches, our design uses free-spaceoptical paths where light passes through thermo-optic phase shiftersencoding query vectors, is attenuated by electro-optic modulatorsencoding key vectors, and is directly detected by a CMOS Ge-on-Siphotodetector array. Bipolar differential detection provides intrinsiccommon-mode noise rejection. We verify 25 validation items spanningdevice physics, algorithm robustness, thermal-field simulation,electro-optic bandwidth, temperature drift, detector physics, opticallink budget, and full-system Monte Carlo. At D=128, bipolar dot-productSpearman rho=0.9971, and complete Transformer attention achievesrho>=0.995 across all configurations.

Zenodo (CERN European Organization for Nuclear Research)
Neural Networks and Reservoir Computing
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.