Academic Publications

FRACTAL Project: Research Articles

The list of academic publications done during the project for the support and sharing of the FRACTAL technology development.

“A 10-core SoC with 20 Fine-Grain Power Domains for Energy-Proportional Data-Parallel Processing over a Wide Voltage and Temperature Range”
Submitted to: Integrated Systems Laboratory (IIS)
Responsible partner: ETH Zurich (ETHZ)
FRACTAL project relevance: Power gating is a well-known technique to reduce leakage power in large System-on-Chips, in this paper, FRACTAL partner ETH Zurich has explored a fine-grained power gating technique with 20 individually controlled power islands that can be turned on and off on demand saving up to 41% power.

“EDISON: An Edge-native Method and Architecture for Distributed Interpolation 
Submitted to: Sensors 21, no. 7: 2279. Sensors and Smart Devices at the Edge: IoT Meets Edge Computing.
Responsible partner: University of Oulu (UOULU)
FRACTAL project relevance: In their article, researchers from the University of Oulu present EDISON – a novel edge-native, distributed interpolation architecture for the smart city networking environment. EDISON’s device layer comprises fixed sensors as well as mobile sensors mounted on vehicles. IoT gateways provide connectivity, store mobile sensor observations, and provide local computational capabilities. The edge layer has edge servers and enhances the fixed sensors with connectivity and further computational capacity. Cloud provides coordination and centralized processing. Is part of the academic study on FRACTAL AI.

On-Demand Redundancy Grouping: Selectable Soft-Error Tolerance for a Multicore Cluster
Submitted to: ISVLSI2022 (IEEE Computer Society Annual Symposium on VLSI)
Responsible partner: ETH Zurich (ETHZ)
FRACTAL project relevance: With the shrinking of technology nodes and the use of parallel processor clusters in hostile and critical environments, such as space, run-time faults caused by radiation are a serious cross-cutting concern, also impacting architectural design. This paper introduces an architectural approach to run-time config- urable soft-error tolerance at the core level, augmenting a six-core open-source RISC-V cluster with a novel On-Demand Redun- dancy Grouping (ODRG) scheme.

“MiniFloat-NN and ExSdotp: An ISA Extension and a Modular Open Hardware Unit for Low-Precision Training on RISC-V Cores”
Submitted to: ARITH conference, Sep12-14 2022
Responsible partner: ETH Zurich (ETHZ)
FRACTAL project relevance: Low-precision formats have recently driven major breakthroughs in neural network (NN) training and inference by reducing the memory footprint of the NN models and improving the energy efficiency of the underlying hardware architectures. Narrow integer data types have been vastly investigated for NN inference and have successfully been pushed to the extreme of ternary and binary representations. In contrast, most training- oriented platforms use at least 16-bit floating-point (FP) formats. Lower-precision data types such as 8-bit FP formats and mixed- precision techniques have only recently been explored in hardware implementations. We present MiniFloat-NN, a RISC-V instruction set architecture extension for low-precision NN training, providing support for two 8-bit and two 16-bit FP formats and expanding op- erations.

“Autonomy and Intelligence in the Computing Continuum: Challenges, Enablers, and Future Directions for Orchestration”
Submitted to: Proceedings of the IEEE, May 2022
Responsible partner: University of Oulu (UOULU)
FRACTAL project relevance: The main focus of the article is on the AI-based orchestration in the device-edge-cloud computing continuum.

“Reinforcement Learning based Thermal Management for Cabin Heating Mode Selection”
Presented at: 15th Graz Symposium Virtual Vehicle, 31/08/2022 – 01/09/2022, Graz, Austria
Responsible partner: AVL
FRACTAL project relevance: Existing automotive control strategies are fully reliant on model-based strategies. These techniques imply a high calibration effort and the ability to perform self-learning through observations is very limited. In this work we focused on the thermal management application for cabin heating mode selection. This use case will, therefore, contribute to integrate the environmental influences and changes as a fundamental part of the system, among other benefits, like potentially increased product quality and increased efficiency for the development of customized controllers.

“AI-Based Scheduling for Adaptive Time-Triggered Networks”
Presented at: MECO2022 & CPSIoT 2022
Responsible partner: University of Siegen
FRACTAL project relevance: Time-triggered systems are ideal for safety-critical systems due to the inherent determinism and better fault tolerance. However, the current trend of adaptation in time-triggered systems is typically limited to switching between a small number of precomputed schedules. Artificial neural networks (ANNs) have the potential to overcome this limitation. In this paper, an ANN is implemented to learn schedules to provide adaptation for time-triggered systems while ensuring that collision and precedence constraints are met. In our evaluation, the AI-based scheduler is compared with conventional scheduling algorithms such as list scheduling and genetic algorithm in terms of makespan and computation time. The results show the AI-based scheduler’s potential when increasing the scheduling problem’s complexity.

“Triggering Conditions Analysis and Methodology for validation of ADAS/AD functions: A use case study”
Presented at: IEEE Inteligent Vehicles Symbosium in Aachen, Germany, at June 5th-9th 2022
Responsible partner: Virtual Vehicle (VIF)
FRACTAL project relevance: Covers safety analysis topics related to UC7 (SPIDER autonomous robot).

“Improving the Robustness of Redundant Execution with Register File Randomization”
Presented at: ICCAD conference, November 2021
Responsible partner: Valencia Polytechnic University (UPV)
FRACTAL project relevance: This article proposes RFR, a new technique to increase the robustness of redundant execution in the context of safety-critical applications. RFR randomizes the mapping of processors architectural registers to avoid common cause faults in the FRACTAL-based RISCV computing nodes. It supports project activities on HW support for safety.

“Weathering the Reallocation Storm: Large-Scale Analysis of Edge Server Workload”
Presented at: Joint EuCNC & 6G Summit 2021
Responsible partner: Univesity of Oulu (UOULU)
FRACTAL project relevance: The topic of the paper is edge orchestration. In more detail, the article analyses the workloads accumulating on a large deployment of edge servers in a number of scenarios and with different workload reallocation / migration strategies. We identify a phenomenon (named the “reallocation storm”) where a high number of reallocations on the edge network are superfluous, that is, happening unnecessarily. We study the causes behind reallocation storms, and look for ways to avoid them.

“The ECSEL FRACTAL Project: A Cognitive Fractal and Secure edge based on a unique Open-Safe-Reliable-Low Power Hardware Platform
Responsible partner: IKERLAN (IKER)
FRACTAL project relevance: This paper has presented the ECSEL FRACTAL project. The aim of the project is to create a reliable computing platform node, realizing a so-called Cognitive Edge under industry standards. The project builds on knowledge of partners gained in current or former EU projects and will demonstrate the newly conceived approaches to co-engineering across use cases spanning Transport and Industrial Control. As the paper is written at the beginning of the project, it focuses rather on the project introduction.