Quantization Effects of Artificial Neural Networks for Embedded Edge-Computing Applications

JuSER (Forschungszentrum Jülich)

Abstract

This paper examines the use of Quantized Neural Networks (QNNs) for two resource-constrained scientific applications: automated calibration of semi- conductor quantum bits (qubits) and scientific particle detectors. We evaluate the trade-offs between Post-Training Quantization (PTQ), Quantization-Aware Train- ing (QAT), and ultra-low-bit Binary Neural Networks (BNNs) with respect to la- tency and resource usage. Our results demonstrate that PTQ and QAT are easily implementable solutions achieving a four-fold reduction in memory usage for U- shaped CNN (U-Net) architectures, whereas BNNs are better suitable for use cases with challenging latency requirements. For the training of non-differentiable custom BNNs , we propose a novel, hardware-constrained learning approach using Genetic Algorithms (GAs). We showcase a Look-Up Table (LUT)-based BNN architecture suitable for direct conversion to Very High-Speed Integrated Circuit Hardware De- scription Language (VHDL) via the HCL4BNN framework. This method achieves nanosecond-scale inference latencies at 10 ns to 15 ns without requiring specialized Digital Signal Processor (DSP) or Block RAM (BRAM) resources.

Authors 9

1. Alperen Aksoy

   University of Siegen · Folkwang University of the Arts

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

2. Ilja Bekman

   University of Siegen · Folkwang University of the Arts

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

3. Vesselin Dimitrov

   Forschungszentrum Jülich · University of Siegen · Folkwang University of the Arts

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

   Faculty of Engineering , JARA-FIT Institute for Quantum Information , Forschungszentrum Jülich GmbH and

4. Qader Dorosti

   Forschungszentrum Jülich

   Affiliation as printed

   Faculty of Engineering , JARA-FIT Institute for Quantum Information , Forschungszentrum Jülich GmbH and

5. Chimezie Eguzo

   University of Siegen · Folkwang University of the Arts

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

6. Sarah Fleitmann

   University of Siegen · Folkwang University of the Arts

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

7. Fabian Hader

   Forschungszentrum Jülich · University of Duisburg-Essen

   Affiliation as printed

   Peter Grünberg Institute (PGI) , Integrated Computing Architectures (ICA | PGI-4) , Forschungszentrum Jülich GmbH , Germany

   University Duisburg-Essen , Germany

8. Andre Zambanini

   University of Siegen · Folkwang University of the Arts

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

9. Stefan van Waasen Aachen

   University of Siegen · Folkwang University of the Arts · RWTH Aachen University

   Affiliation as printed

   Center for Particle Physics Siegen , Department für Physik , Universität Siegen , Germany

   RWTH Aachen University , Germany