Twinning towards excellence for Privacy Enhancing Technologies leveraging Homomorphic Encryption

Projektinformationen

enCRYPTON

ID Finanzhilfevereinbarung: 101079319

DOI

10.3030/101079319

Projekt abgeschlossen

EK-Unterschriftsdatum 19 Juli 2022

Startdatum 1 Oktober 2022

Enddatum 31 Dezember 2025

Finanziert unter

• Widening participation and spreading excellence

Gesamtkosten

€ 1 515 125,00

EU-Beitrag

€ 1 515 125,00

Koordiniert durch

SABANCI UNIVERSITESI
Türkiye

Periodic Reporting for period 1 - enCRYPTON (Twinning towards excellence for Privacy Enhancing Technologies leveraging Homomorphic Encryption)

Berichtszeitraum: 2022-10-01 bis 2025-12-31

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

Homomorphic encryption (HE) is a major and powerful tool within the context of privacy-enhancing technologies (PET). It can help to achieve end-to-end data security, facilitate outsourced computation over encrypted data, enhance data sharing and collaboration (e.g. federated learning), and ensure compliance with data protection regulations such as GDPR, HIPAA, KVKK, etc., in addition to privacy-preserving. However, there are certain challenges in its implementation and deployment in production environments for industrial use cases pertaining to aspects such as FL to facilitate real-world applications. The enCRYPTON project focuses on overcoming these implementation and deployment challenges of HE.

The main goal of the enCRYPTON project is to enhance the innovation capacity of Sabanci University and Turkey in the field of PET. PETs have gained significant importance over the last decade, as the extraction and processing of data have emerged as a significant driver of value. However, this trend has raised privacy concerns regarding the use of sensitive data. The project aims to elevate Sabanci University's capabilities in scientific, technical, and management domains to the level of the EU while seeking to enhance the scientific research quality and excellence of the consortium within PET. The project aims to establish SU as an international PET excellence hub focusing on HE, and its applications in federated learning (FL) and secure multi-party computation (SMC).

The project’s expected outcomes include hardware and software acceleration of HE primitives, as well as, a prototype FL/SMC protocol implementation. Furthermore, the career development of Early-Stage Researchers (ESRs) is planned to be boosted by providing guidance and support which is to be facilitated by providing mentorship and industrial collaboration opportunities, specialized graduate courses, and least of all summer schools. Moreover, SU's research management and administrative skills will be enhanced through the organization of knowledge exchange workshops and tailored courses. In addition, the project aims to enhance the scientific visibility and knowledge dissemination capability of all partners involved in the consortium, establish a leading position in the field of data privacy preservation (DPP) in Turkey and Europe, and attract high-profile researchers.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

The main achievements and activities performed can be divided into four tasks: i) Creating customized hardware (FPGA) and software (GPU/CPU) solutions for the acceleration of HE, ii) Developing light-weight implementations of HE primitives of interest, iii) Developing secure implementation of HE technology, and iv) Implementing a FL application using HE-based SMC. For hardware acceleration, NTT transforms in both forward and inverse directions (NTT and INTT, respectively) are utilized. Specifically, we designed an FPGA-based architecture, which has shown promising results in batch NTT scenarios. In parallel, we explored the use of GPUs. Instead of targeting specific primitives such as NTT, our GPU implementation focused on complete FHE schemes. This approach led to outstanding results. Our solution, referred to as HEonGPU, attracted significant attention from the community, as evidenced by open-source repository statistics. These acceleration achievements enabled practical deployment in federated learning and secure multi-party computation scenarios. A working prototype of a FL protocol, as well as an AI model fine-tuning protocol, was implemented, demonstrating the applicability of the proposed methods in real-world industrial use cases. For the lightweight implementation of HE primitives, we developed a hardware-software co-design framework, referred to as RISQrypt. The proposed framework demonstrated significant advantages compared to existing solutions targeting lattice-based cryptography, particularly in terms of efficiency and flexibility. Regarding secure implementations of HE technology, we investigated potential attack vectors targeting NTT-based polynomial multiplication operations. The insights gained from this analysis were leveraged to strengthen and secure the developed lightweight implementation, contributing to improved resilience against implementation-level attacks.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The developed acceleration approaches demonstrated major performance gains for FHE workloads. In particular, the GPU-based acceleration framework, HEonGPU, evolved into a state-of-the-art FHE library. This solution significantly advances the performance frontier of FHE implementations and acts as a key enabler for practical privacy-preserving applications, including federated learning, secure analytics, and encrypted AI model training. HEonGPU is actively maintained and has fostered a growing open-source community, with external developers contributing to its ongoing development and optimisation. In addition, the lightweight framework RISQrypt has proven to be one of the most comprehensive solutions supporting FHE and PQC, delivering state-of-the-art performance, provable security, and high flexibility compared to existing approaches.

Beyond the technical results, the established scientific network continues beyond the project lifetime, ensuring sustained knowledge exchange and joint research in privacy-enhancing technologies. Future exploitation include continued optimisation and large-scale benchmarking, demonstration in industrial environments, and active participation in standardisation efforts. Follow-up research funding, innovation investment, and structured IPR and commercialisation support—particularly through the enCRYPTON Labs initiative—will further accelerate technology transfer and market readiness.