REBECCA: Reconfigurable heterogeneous highly parallel processing platform for safe and secure AI
(2026) In Microprocessors and Microsystems 122.- Abstract
- An ever increasing number of applications is starting to adopt AI/ML components to unlock advanced capabilities and functionalities that are extremely hard or even not feasible at all to implement with more traditional approaches. The cost to be paid for these advancements lies in their computational requirements, which is often so significant that dictates their offloading to dedicated powerful compute systems placed in datacenter environments. This distribution of computational tasks, although convenient for a number of applications, is prohibitive for a lot of use cases (e.g. automotive and healthcare) where latency, reliability, security, privacy and other concerns pose specific restrictions and mandate that all computations are... (More)
- An ever increasing number of applications is starting to adopt AI/ML components to unlock advanced capabilities and functionalities that are extremely hard or even not feasible at all to implement with more traditional approaches. The cost to be paid for these advancements lies in their computational requirements, which is often so significant that dictates their offloading to dedicated powerful compute systems placed in datacenter environments. This distribution of computational tasks, although convenient for a number of applications, is prohibitive for a lot of use cases (e.g. automotive and healthcare) where latency, reliability, security, privacy and other concerns pose specific restrictions and mandate that all computations are carried out either at the source of data or very close to it. This is the main driver for all efforts that try to enable AI-related computations at client devices and the edges of the network. The EU-funded REBECCA project has emerged as part of those efforts. Its main goal is to develop a novel system-on-chip for IoT and edge systems that will be able to handle advanced AI tasks at suitable energy/power budgets. REBECCA will integrate multiple RISC-V general purpose CPU cores along with advanced AI/ML and security accelerator engines in a single package comprised of two chiplets and the capability to tightly interconnect to reconfigurable devices. The adoption of reconfigurable hardware as a key element of the REBECCA platform will provide the required flexibility to address different markets without sacrificing the energy efficiency levels it targets through the use of custom application-specific accelerators. Besides hardware components, REBECCA will also develop the required software support in terms of operating systems, hypervisors and libraries to enable the full potential of its hardware platform. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/6ac4d2ee-a8ea-45c7-8238-40c0da969a49
- author
- organization
- publishing date
- 2026-05-15
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- SoC, Edge AI, Reconfigurable, Hardware, RISC-V
- in
- Microprocessors and Microsystems
- volume
- 122
- pages
- 11 pages
- publisher
- Elsevier
- external identifiers
-
- scopus:105039119514
- ISSN
- 0141-9331
- DOI
- 10.1016/j.micpro.2026.105279
- language
- English
- LU publication?
- yes
- id
- 6ac4d2ee-a8ea-45c7-8238-40c0da969a49
- date added to LUP
- 2026-06-26 15:12:05
- date last changed
- 2026-08-12 14:03:37
@article{6ac4d2ee-a8ea-45c7-8238-40c0da969a49,
abstract = {{An ever increasing number of applications is starting to adopt AI/ML components to unlock advanced capabilities and functionalities that are extremely hard or even not feasible at all to implement with more traditional approaches. The cost to be paid for these advancements lies in their computational requirements, which is often so significant that dictates their offloading to dedicated powerful compute systems placed in datacenter environments. This distribution of computational tasks, although convenient for a number of applications, is prohibitive for a lot of use cases (e.g. automotive and healthcare) where latency, reliability, security, privacy and other concerns pose specific restrictions and mandate that all computations are carried out either at the source of data or very close to it. This is the main driver for all efforts that try to enable AI-related computations at client devices and the edges of the network. The EU-funded REBECCA project has emerged as part of those efforts. Its main goal is to develop a novel system-on-chip for IoT and edge systems that will be able to handle advanced AI tasks at suitable energy/power budgets. REBECCA will integrate multiple RISC-V general purpose CPU cores along with advanced AI/ML and security accelerator engines in a single package comprised of two chiplets and the capability to tightly interconnect to reconfigurable devices. The adoption of reconfigurable hardware as a key element of the REBECCA platform will provide the required flexibility to address different markets without sacrificing the energy efficiency levels it targets through the use of custom application-specific accelerators. Besides hardware components, REBECCA will also develop the required software support in terms of operating systems, hypervisors and libraries to enable the full potential of its hardware platform.}},
author = {{Brokalakis, Andreas and Malakonakis, Pavlos and Harteros, Konstantinos and Andronikou, Dimitris and Galanomatis, Ioannis and Savvakos, Charalampos and Tang, Guantzi and Vadivel, Kanishkan and van Schaik, Gert-Jan and Ioannidis, Sotiris and Georgopoulos, Konstantinos and Chrysos, Grigoris and Allfjord, Alex and Nouripayam, Masoud and Prieto Llorens, Arturo and Rodrigues, Joachim and Mavroidis, Iakovos and Papaefstathiou, Ioannis}},
issn = {{0141-9331}},
keywords = {{SoC; Edge AI; Reconfigurable; Hardware; RISC-V}},
language = {{eng}},
month = {{05}},
publisher = {{Elsevier}},
series = {{Microprocessors and Microsystems}},
title = {{REBECCA: Reconfigurable heterogeneous highly parallel processing platform for safe and secure AI}},
url = {{http://dx.doi.org/10.1016/j.micpro.2026.105279}},
doi = {{10.1016/j.micpro.2026.105279}},
volume = {{122}},
year = {{2026}},
}