Abstract
Abstract
Today, continued miniaturization in electronic integrated circuits (ICs) appears to have reached its fundamental limit at ∼2 nm feature-sizes, from originally ∼1 cm. At the same time, energy consumption due to communication becomes the dominant limitation in high performance electronic ICs for computing, and modern computing concepts such neural networks further amplify the challenge. Communication based on co-integrated photonic circuits is a promising strategy to address the second. As feature size has leveled out, adding a third dimension to the predominantly two-dimensional ICs appears a promising future strategy for further IC architecture improvement. Crucial for efficient electronic–photonic co-integration is complementary metal-oxide-semiconductor (CMOS) compatibility of the associated photonic integration fabrication process. Here, we review our latest results obtained in the FEMTO-ST RENATECH facilities on using additive photo-induced polymerization of a standard photo-resin for truly three-dimensional (3D) photonic integration according to these principles. Based on one- and two-photon polymerization (TPP) and combined with direct-laser writing, we 3D-printed air- and polymer-cladded photonic waveguides. An important application of such circuits are the interconnects of optical neural networks, where 3D integration enables scalability in terms of network size versus its geometric dimensions. In particular via flash-TPP, a fabrication process combining blanket one- and high-resolution TPP, we demonstrated polymer-cladded step-index waveguides with up to 6 mm length, low insertion (∼0.26 dB) and propagation (∼1.3 dB mm–1) losses, realized broadband and low loss (∼0.06 dB splitting losses) adiabatic 1 to M couplers as well as tightly confining air-cladded waveguides for denser integration. By stably printing such integrated photonic circuits on standard semiconductor samples, we show the concept’s CMOS compatibility. With this, we lay out a promising, future avenue for scalable integration of hybrid photonic and electronic components.
Funder
Agence Nationale de la Recherche
Volkswagen Foundation
European Commission
RENATECH
Bourgogne Franche-Comte
Subject
Electrical and Electronic Engineering,Mechanical Engineering,Mechanics of Materials,General Materials Science,General Chemistry,Bioengineering
Reference73 articles.
1. Optical neural networks: The 3D connection
2. Dendrocentric learning for synthetic intelligence
3. Planar and channel optical waveguides utilizing silicon technology;Boyd,1985
4. Hardware-enabled artificial intelligence;Dally,2018
Cited by
6 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献