This paper presents a millimeter-scale, robust bilayer-integrated diffractive neural network (DNN) chip for optical inference in extreme environments. The chip, fabricated by engraving two binary phase-modulated diffractive layers onto a quartz wafer, achieved 82% accuracy in handwritten digit recognition. Its high-temperature stability and estimated room-temperature lifetime of 1.84 × 10²³ trillion years make it suitable for extreme environment applications.

Yibo Dong, Dajun Lin, Long Chen, Baoli Li, Xi Chen, Qiming Zhang, Haitao Luan, Xinyuan Fang, Min Gu