Download

AutoPNN Open Source Code

Download the complete AutoPNN framework source code

⬇ Download AutoPNN Source Code

Software

Check out our github repositories!

  • EDA-Driven ML Circuits for Flexible Electronics (ISLPED’25)

    Code and scripts for low-power flexible stress-monitoring classifiers targeting conformal wearables; framework referenced by the ISLPED’25 paper.

    EDA-Driven-ML-Circuits-for-Flexible-Electronics

  • Enabling Printed Multilayer Perceptrons via Area-Aware Neural Minimization (TC’25)

    Area-aware methodology for printed MLPs, including HW-aware DSE notebooks, pretrained models, and generated Verilog for UCI datasets.

    Printed_MLPs

  • Sequential Support Vector Machine Circuits for Printed Electronics (ISCAS’25 / DATE’25-LBR)

    SystemVerilog implementation & generator for compact, accurate printed SVM classifiers using a single-MAC sequential engine and bespoke storage/control.

    Support-Vector-Machine-Circuits-Targeting-Printed-Electronics

  • Flex-SVM: Support Vector Machines for Flexible/Printed Electronics

    Repository with SVM designs and flows tailored to flexible/printed technologies, focusing on compact RTL and evaluation scripts.

    Flex-SVM

  • Approximate Popcount & Popcount-Compare for Ternary Neurons (ICCAD’24)

    Library of approximate PC/PCC circuits (HW + SW) used to build printed Ternary Neural Networks; includes multi-objective optimization assets.

    approximate-popcount

  • Approximate Popcount & Popcount-Compare for Ternary Neurons (ICCAD’24)

    Library of approximate PC/PCC circuits (HW + SW) used to build printed Ternary Neural Networks; includes multi-objective optimization assets.

    approximate-popcount

  • Fault Sensitivity Analysis of Printed Bespoke Multilayer Perceptron Classifiers

    Dive into our exploration of fault sensitivity in printed multilayer perceptron (MLP) classifiers tailored for Printed Electronics (PE) applications. Our study delves into the analysis of various digital and analog realizations of printed MLPs, assessing their reliability across diverse classification tasks. We evaluate different digital architectures, including generic, bespoke, and approximate designs, to provide comprehensive insights into fault sensitivity on benchmark datasets. Explore how our findings contribute to enhancing the reliability and robustness of MLPs in emerging PE technologies.

    Fault_Sensitivity_Analysis

  • On-sensor Printed Machine Learning Classification via Bespoke ADC and Decision Tree Co-Design

    For insights into on-sensor printed machine learning classification facilitated by bespoke ADC and Decision Tree co-design, visit our github.

    Ax-Printed-ML-Classifiers

  • Embedding Hardware Approximations in Discrete Genetic-based Training for Printed MLPs

    Explore how integrating hardware approximation techniques into Multilayer Perceptron (MLP) training optimizes costs and enhances performance. Discover our genetic-based, hardware-aware approach tailored for printed MLPs

    Approximation-Techniques-Targeting-Printed-MLPs

  • Bespoke Approximation of Multiplication Accumulation and ActivationTargeting Printed Multilayer Perceptrons

    Discover how we overcome constraints using Approximate Computing and Bespoke design principles. Explore our innovative automated framework for crafting ultra-low power Multilayer Perceptron (MLP) classifiers, introducing a holistic approximation approach to all neuron components; multiplier, accumualor and activation.

    Approximation-Techniques-Targeting-Printed-MLPs