Programmable Accelerators for Lattice-based Public Key Protocols
Post Quantum Lattice-Based Cryptography (LBC) schemes are increasingly gaining attention in traditional and emerging security problems, such as encryption, digital signature, key exchange, homomorphic encryption etc, to address security needs of both short and long-lived devices — due to their foundational properties and ease of implementation. However, LBC schemes induce higher computational demand compared to classic schemes (e.g., DSA, ECDSA) for equivalent security guarantees, making domain-specific acceleration a viable option for improving security and favor early adoption of LBC schemes by the semiconductor industry.
Scale Down Neural Network Models Considering HW Constraints
Traditionally machine learning (ML) computations have been performed on resourceful servers due to the high computational demands of these ML techniques. However, when these ML techniques are deployed for emerging applications that are heavily resource-constrained (e.g., smartphones, mobile platforms, IoT devices), time-critical (e.g., self-driving cars), or in environments where cloud connectivity is not reliably available, there is a need to perform ML computation/acceleration on the device itself. Furthermore, consumers are increasingly concerned about the privacy of their data when stored on public clouds. All of these concerns pose daunting challenges for ML formulations and on-device acceleration. On one hand, due to the limited resources, on-device AI should scale-down the network models (e.g. changing the number of layers, neurons per layer, etc.). On the other hand, on-device AI acceleration must simultaneously satisfy multiple constraints including power consumption, latency, privacy and accuracy of the prediction.
Design Automation Conference (DAC), 2020, (Best paper award finalist).
Data Reuse for Accelerated Approximate Warps Journal Article
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD), 2020.
International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2020, (Best paper award finalist).
Transactions on Embedded Computing Systems, 2020.
IEEE/ACM Design Automation Conference (DAC), 2019.
Small Memory Footprint Neural Network Accelerators Inproceedings
International Symposium on Quality Electronic Design (ISQED), 2019.
ACM Computing Survey, 51 (6), 2019, ISSN: 0360-0300.
Cryptology ePrint Archive, Report 2018/608, 2018, (urlhttps://eprint.iacr.org/2018/608).
Proceedings of the Twelfth IEEE/ACM/IFIP International Conference on Hardware/Software Codesign and System Synthesis Companion, pp. 6:1–6:3, ACM, Seoul, Republic of Korea, 2017, ISBN: 978-1-4503-5185-0.
University of California Irvine, (CECS TR 17-04), 2017.