Spiking Neural Networks

Spiking neurons with a sense of time

With spiking neural networks (SNNs), Artificial Intelligence becomes even more similar to the human brain. Unlike traditional artificial neural networks, where signals are exchanged continuously, SNNs only transmit relevant data in the form of short electrical pulses. Like their biological counterparts, the artificial neurons have their own sense of time: they only become active when a critical threshold of signals is exceeded.

While classical AI models increasingly demand more computing power, spiking neural networks resolve the tension between energy efficiency and real-time capability. Their structure allows for the processing of massive amounts of data in a power-saving and fast manner, without losing performance. SNNs particularly showcase their strengths when Artificial Intelligence needs to be directly integrated into end devices.

We aim to bring SNNs to breakthrough and implement them in practice. Therefore, we not only explore their topology but also develop customized hardware designs for ultra-low-power and ultra-low-latency applications.

Spiking neural networks – a technology with clear advantages

Energy efficiency

 

The communication between artificial neurons is highly energy-intensive. In particular, the continuous and clocked transmission of data packages leads to significant efficiency losses in large neural networks. Spiking neural networks address this issue through asynchronous pulses, which limit information exchange to the essentials—thereby significantly reducing power consumption.

Latency

 

Microprocessors are based on the Von Neumann architecture, where data processing and data storage are strictly separated. However, this creates additional communication overhead that causes latencies. In contrast, the neurons in SNNs can not only process information but also store it over time at the location where it is needed. This alleviates the bottleneck, reduces latencies, and enables a high processing rate, which is essential for real-time applications.

Robustness

 

Neural networks are often exposed to interferences or adversarial attacks. Even small deviations in the input signal can generate fundamentally different output signals. In contrast, SNNs create a protective shield: Since the transmission of pulses is based on statistical patterns, the exact time at which a signal arrives is irrelevant. The potential influence of individual errors decreases and the reliability of the neural network is significantly increased.

An overview of our SNN portfolio

We support companies that want to implement SNNs in their own products. We assist you with all questions related to the complex topic of spiking neural networks and develop software and hardware solutions that are optimally tailored to your needs.

  • Consulting and customized training
  • Feasibility and technology studies
  • R&D projects
  • Provision of simulation, visualization, and training tools
  • IP licensing of hardware IP cores

Spiking neural networks provide the crucial added value

While deep neural networks currently lead in image recognition and analysis, SNNs excel in the field of time series analysis. These arise wherever sensor data is evaluated, making SNNs a tool used in a variety of application scenarios. 

Communication systems

 

In mobile and wireless communication systems, speed, energy efficiency, and robustness are crucial requirements that SNNs fulfill ideally. They can efficiently filter and process signals, as well as improve their quality through noise reduction. Additionally, they support adaptive modulation and coding schemes, which can further optimize transmission efficiency.

Closed-loop control systems

 

Motor movements need to be controlled precisely and reliably. For this, sensor and control data in motor control systems must be processed in real time. SNNs can implement adaptive control algorithms that adjust to changing conditions at any time. This benefits industries such as the automotive sector, allowing them to optimize the efficiency and performance of their developed engines.

Robotics

 

Robots designed to make autonomous decisions must be able to perceive sudden changes in their environment immediately and respond accordingly. With SNNs, data from sensors and actuators can be analyzed in real time. An example of this is robots equipped with event-based cameras, which operate on principles similar to those of the human retina.

Radar systems

 

In radar systems, everything revolves around the detection and analysis of objects. To enable this to be done powerfully and precisely, SNNs aid in the preprocessing and filtering of received signals, target detection and tracking, as well as the classification of objects based on radar signatures. With its energy efficiency and speed, pulse-based communication is particularly suitable for mobile and real-time radar applications.

 

Healthcare

 

In disease diagnosis, every second counts. With the help of SNNs, medical data can be quickly collected and analyzed. Their applications in healthcare include, for example, electroencephalograms (EEG) and electrocardiograms (ECG). Additionally, neural networks can be implemented in wearable health devices to continuously monitor biometric data and identify anomalies early.

Reference projects

  • DSAI – Center for Digital Signal Processing using Artificial Intelligence

    Project duration: 1.10.2020 – 30.9.2025
    Funding: Bavarian Ministry of Economic Affairs, Regional Development and Energy

    The main goal of the project is to explore technologies in the field of digital signal processing using Artificial Intelligence. The DSAI relies on the combination of new AI techniques with existing knowledge of signal processing and successful application transfer.

    The DSAI addresses three economically significant areas where Fraunhofer IIS has its core competencies: machine vision (computer vision), speech signal processing, and signal processing for data transmission. A key focus is on using spiking neural networks for efficient signal processing and transmission. As part of the project, tools for the simulation and training of SNNs, as well as networks for the application of pulse-based communication, have been developed.

  • MANOLO – Trustworthy Efficient AI for Cloud-Edge Computing

    Project duration: 1.1.2024 – 31.12.2026 
    Consortium: 2 partners from Germany, further 16 European partners 
    Funding: Horizon Europe framework programme of the European Union
    Project website: https://manolo-project.eu/

    The vision of MANOLO is to deliver a complete and trustworthy stack of algorithms and tools to help AI systems reach better efficiency and seamless optimization in their operations. The focus is on energy-efficient training of AI models with quality-checked data and the execution of resource-efficient AI models on a wide range of devices for use on the edge and in the cloud.

    In the project, Fraunhofer IIS is focusing on bringing AI applications to the edge. To this end, algorithms and tools are being developed that search for and optimize suitable neural networks automatically (Neural Architecture Search, NAS). In this context, Fraunhofer IIS also explores methods to generate spiking neural networks (SNNs) from DNN models, simplifying the use of novel neuromorphic algorithms. 

  • NEUROKIT2E – Open-Source Deep Learning Platform for Embedded Hardware in Europe

    Project duration: 1.6.2023 – 31.05.2026​
    Consortium: Four partners from Germany, 22 others from Europe
    Funding: KDT Joint Undertaking Initiative of the EU and the Federal Ministry of Education and Research (BMBF)

    NEUROKIT2E aims to develop an independent open-source framework for edge/embedded AI, supporting an international community of users and a wide range of applications. This European framework, specifically for embedded AI, is intended to be compatible with existing frameworks and facilitate the development and implementation of AI applications on embedded hardware.

    At Fraunhofer IIS, tools are specifically developed that allow for the efficient mapping of spiking neural networks onto embedded hardware accelerators.

  • SEC-Learn – Sensor edge cloud for federated learning

    Project duration: 1.7.2020 – 31.12.2024
    Consortium: 11 Fraunhofer Institutes from the Groups for Microelectronics and ICT
    Funding: until 2021: InnoPush Program of the German Federal Ministry of Education and Research (BMBF); from 2022: Fraunhofer Executive Board Project

    In the SEC-Learn project, a system of distributed energy-efficient edge devices is being created that learn together to solve a complex signal processing problem using machine learning. The focus of the project is on the development of fast, energy- and space-efficient hardware accelerators for spiking neural networks (SNNs) on the one hand, and on their interconnection to form a federated system on the other hand, in which each device can act and learn autonomously, but shares its learning successes with all other devices through federated learning.

    This concept enables numerous applications, from autonomous driving to condition monitoring, where decentralized data processing through AI needs to be connected to a centralized system for training – without violating privacy or causing excessive power consumption and data traffic.

    The hardware accelerators used in the project are being developed under the coordination of Fraunhofer IIS in close cooperation with Fraunhofer EMFT and the EAS division of Fraunhofer IIS. To this end, Fraunhofer IIS is developing neuromorphic mixed-signal circuits for specialized neuron and synapse models at its Erlangen site, the associated software tools for hardware-aware training and simulation, and a scalable chip architecture that should make it possible to serve a wide variety of application problems in the future.

    More information about the SEC-Learn project

You may also be interested in

 

Artificial intelligence

Go back to business area “Artificial intelligence“