TensorPool: A 3D-Stacked 8.4TFLOPS/4.3W Many-Core Domain-Specific Processor for AI-Native Radio Access Networks

Marco Bertuletti, Yichao Zhang, Diyou Shen, Alessandro Vanelli-Coralli, Frank K. Gürkaynak, Luca Benini · 2026

Read TensorPool: A 3D-Stacked 8.4TFLOPS/4.3W Many-Core Domain-Specific Processor for AI-Native Radio Access Networks by Marco Bertuletti, Yichao Zhang, Diyou Shen, Alessandro Vanelli-Coralli, Frank K. Gürkaynak, Luca Benini (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-02291.txt.

Description

Bertuletti and colleagues present TensorPool, a 3D stacked many core processor for AI native radio networks. A cluster of 256 programmable cores plus 16 tensor engines tackles AI physical layer models under tight real time and power limits for 6G base stations. Built in TSMC N7, it delivers over 8 teraflops at about 4 watts and hits high tensor unit use on AI radio tasks, beating a core only cluster on efficiency. Stacking compute blocks in 3D shrinks footprint by over two times with no frequency loss versus a flat design.

Catalog record

CREATORMarco Bertuletti, Yichao Zhang, Diyou Shen, Alessandro Vanelli-Coralli, Frank K. Gürkaynak, Luca Benini
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51976
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2604.02291

Browse the full library · About the Tower of Babel