High Bandwidth Memory with Processing-in-Memory (HBM-PIM) offers an opportunity to reduce data movement by executing computation directly inside memory, but current commercial platforms expose limited instruction sets and require specialized software stacks. In this work, we investigate whether HBM-PIM can serve as a backend for ISA-level matrix acceleration, using the RISC-V Attached Matrix Extension (AME) as a semantic reference. We propose a PEP-based execution model that maps AME element-wise and matrix instructions to HBM-PIM micro-kernels and data instructions in memory operations. Differently from SoA HBM-PIM, we introduce a reduction-free outer-product dataflow that enables accumulation entirely within memory despite the lack of native reduction support. Our approach supports end-to-end execution of element-wise operations, GEMV, and GEMM in PIM mode, minimizing host involvement and off-chip transfers. An experimental evaluation on Samsung Aquabolt-XL shows that AME matrix tile multiplication achieves up to 14.9 GFLOP/s (59.4 FLOP/cycle) on a single HBM pseudo-channel.

Venieri, E., Manoni, S., Florian, A., Park, J., Sohn, K., Bartolini, A. (2026). AME-PIM: Can Memory be Your Next Tensor Accelerator? [10.1145/3801487.3806067].

AME-PIM: Can Memory be Your Next Tensor Accelerator?

Venieri, Emanuele
;
Manoni, Simone;Bartolini, Andrea
Ultimo
2026

Abstract

High Bandwidth Memory with Processing-in-Memory (HBM-PIM) offers an opportunity to reduce data movement by executing computation directly inside memory, but current commercial platforms expose limited instruction sets and require specialized software stacks. In this work, we investigate whether HBM-PIM can serve as a backend for ISA-level matrix acceleration, using the RISC-V Attached Matrix Extension (AME) as a semantic reference. We propose a PEP-based execution model that maps AME element-wise and matrix instructions to HBM-PIM micro-kernels and data instructions in memory operations. Differently from SoA HBM-PIM, we introduce a reduction-free outer-product dataflow that enables accumulation entirely within memory despite the lack of native reduction support. Our approach supports end-to-end execution of element-wise operations, GEMV, and GEMM in PIM mode, minimizing host involvement and off-chip transfers. An experimental evaluation on Samsung Aquabolt-XL shows that AME matrix tile multiplication achieves up to 14.9 GFLOP/s (59.4 FLOP/cycle) on a single HBM pseudo-channel.
2026
Proceedings of the 23rd ACM International Conference on Computing Frontiers 2026(CF 2026)
232
240
Venieri, E., Manoni, S., Florian, A., Park, J., Sohn, K., Bartolini, A. (2026). AME-PIM: Can Memory be Your Next Tensor Accelerator? [10.1145/3801487.3806067].
Venieri, Emanuele; Manoni, Simone; Florian, Alberto; Park, Jaehyun; Sohn, Kyomin; Bartolini, Andrea
File in questo prodotto:
File Dimensione Formato  
3801487.3806067.pdf

accesso aperto

Tipo: Versione (PDF) editoriale / Version Of Record
Licenza: Licenza per Accesso Aperto. Creative Commons Attribuzione (CCBY)
Dimensione 5.39 MB
Formato Adobe PDF
5.39 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11585/1069857
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
  • OpenAlex ND
social impact