Abstract
We demonstrate a monolithic 3D integration of Si-based CMOS logic, resistive random-access memory (RRAM) based computing-in-memory (CIM) and ternary content-addressable memory (TCAM) layers, namely M3D-LIME, to implement one-shot learning. The first layer of Si MOSFETs was designed and fabricated using a standard CMOS process and served as control logic. The second layer of 1 T1R array was fabricated with HfAlOx-based analog RRAM using a low-temperature (≤ 300°C) back-end-of-line (BEOL) process to implement CIM for feature extractions. The third layer of 2T2R-based TCAM was fabricated with carbon nanotube field-effect transistors (CNTFETs) and Ta <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> O <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">5</inf> -based RRAM to perform template storing and matching. Extensive structural analysis and electrical measurements were carried out to validate the integrity and proper function of the fabricated M3D-LIME chip. As a demonstration, GPU-equivalent classification accuracy up to 97.8% was achieved in the one-shot/few-shot learning task on the Omniglot dataset with 162x lower energy consumption. Our work demonstrates the feasibility and great potential of M3D chips consisted of logic, memory and CIM for emerging applications such as artificial intelligence (AI) and high-performance computing (HPC).
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.