BraiNEdarwin/brains-py

A python package to support the study of Dopant Network Processing Units as hardware accelerators for non-linear operations. Its aim is to support key functions for hardware setups and algorithms related to searching functionality on DNPUs and DNPU architectures both in simulations and in hardware.

brains-py

A python package based on PyTorch and NumPy to support the study of Dopant Network Processing Units (DNPUs) [1][2] as low-power computational units in the context of neural networks. The aim of the package is to support key functions for hardware setups and algorithms related to searching functionality on DNPUs and DNPU architectures both in simulations and in hardware. The package is part of the brains-py project, a set of python libraries to support the development of nano-scale in-materio hardware for designing hardware accelerators for neural-network like operations. This package is based on several peer-reviewed scientific publications.

1. Instructions

You can find detailed instructions for the following topics on the wiki:

• Introduction: Provides a general description behind the background of this project project. These instructions are strongly recommended for new students joining the research group.
• Package description: Gives more information on how the package is structured and the particular usage of files.
• Installation instructions: How to correctly install this package
• User instructions and usage examples: Instructions for users and examples on how brains-py can be used for different purposes
• Developer instructions: Instructions for people wanting to develop brains-py

2. License and libraries

This code is released under the GNU GENERAL PUBLIC LICENSE Version 3. Click here to see the full license. The package relies on the following libraries:

• General support libraries:
  • PyTorch, NumPy, tensorboard, tqdm and matplotlib
• Drivers and hardware setups:
  • nidaqmx, pypiwin32
• Configurations:
  • pyyaml
• Linting, and testing:
  • unittest, mypy, flake8, coverage

3. Related scientific publications

[1] Chen, T., van Gelder, J., van de Ven, B., Amitonov, S. V., de Wilde, B., Euler, H. C. R., ... & van der Wiel, W. G. (2020). Classification with a disordered dopant-atom network in silicon. Nature, 577(7790), 341-345. https://doi.org/10.1038/s41586-019-1901-0

[2] HCR Euler, U Alegre-Ibarra, B van de Ven, H Broersma, PA Bobbert and WG van der Wiel (2020). Dopant Network Processing Units: Towards Efficient Neural-network Emulators with High-capacity Nanoelectronic Nodes. https://arxiv.org/abs/2007.12371](https://arxiv.org/abs/2007.12371)

[3] HCR Euler, MN Boon, JT Wildeboer, B van de Ven, T Chen, H Broersma, PA Bobbert, WG van der Wiel (2020). A Deep-Learning Approach to Realising Functionality in Nanoelectronic Devices. https://doi.org/10.1038/s41565-020-00779-y

4. Acknowledgements

This package has been created and it is maintained by the Brains team of the NanoElectronics research group at the University of Twente. It has been designed by:

• Dr. Unai Alegre-Ibarra, @ualegre (u.alegre@utwente.nl): Project lead, including requirements, design, implementation, maintenance, linting tools, testing and documentation (Jupyter notebooks, Wiki and supervision of file by file documentation).
• Dr. Hans Christian Ruiz-Euler, @hcruiz (h.ruiz@utwente.nl): Initial design and implementation of major features both in this repository and in the legacy SkyNEt repository and in this one.

With the contribution of:

• Humaid A. Mollah, @hnm27 (h.a.mollah@student.utwente.nl) : Testing, identification of bugs, linting tools, continuous integration and documentation.
• Bozhidar P. Petrov, @bozhidar-petrov (b.p.petrov@student.utwente.nl) : Testing, identification of bugs, linting tools and documentation.
• Marcus Boon: @Mark-Boon: The on-chip gradient descent. The initial structure for the CDAQ to NiDAQ drivers in the legacy SkyNEt repository.
• Dr. ir. Michel P. de Jong @xX-Michel-Xx (m.p.dejong@utwente.nl): Testing and identification of bugs, especially on the installation procedure.
• Florentina Min Joo Uitzetter: The genetic algorithm as shown in the legacy SkyNEt repository.
• Michelangelo Barocci @micbar-21. The usage of multiple DNPUs simultaneously on hardware and the creation of an improved PCB for measuring DNPUs.
• Antonio J. Sousa de Almeida @ajsousal (a.j.sousadealmeida@utwente.nl): Checking and upgrading drivers and National Instruments equipment from the labs.
• Srikumar Sastry, Vishu26 (s.s.sastry@student.utwente.nl) : Testing and identification of bugs.
• Bram van de Ven, @bbroo1 (b.vandeven@utwente.nl) : General improvements and testing of the different hardware drivers and devices and documentation.
• Mohamadreza Zolfagharinejad @mamrez (m.zolfagharinejad@utwente.nl): Writing of some of the examples in Jupyter notebooks (IV curves and surrogate model generation).
• Jochem Wildeboer @jtwild Nearest neighbour loss functions.

Other minor contributions might have been added, in form of previous scripts that have been improved and restructured from SkyNEt, and the authorship remains of those people who collaborated in it.

This project has received financial support from:

• University of Twente
• Dutch Research Council
  • HTSM grant no. 16237
  • Natuurkunde Projectruimte grant no. 680-91-114
• Horizon Europe research and innovation programme
  • Grant no. 101046878
• Toyota Motor Europe N.V.
• Deutsche Forschungsgemeinschaft
  • Project 433682494 – SFB 1459