DYNAP technology

The Dynamic Neurormorphic Asynchronous Processor (DYNAP) line facilitates implementation of reconfigurable, general-purpose, real-time neural networks based on spiking neurons.

Our patented event-routing technology enables us to develop efficient ultra-low-power, ulta-low-latency neuromorphic solutions for a variety of artificial intelligence edge computing applications including autonomous robots, always-on co-processors for mobile and embedded devices, wearable healthcare systems, security, IoT applications, and computing at the network edge.

AI Processor Solutions

DYNAP-SE2

Our latest SE family neuromorphic chip DYNAP-SE2 combines the impeccable energy efficiency of its previous generations with new features for low latency, real-time end-to-end applications using general purpose feed-forward, recurrent and reservoir networks. It’s integrated bio-signal amplifiers make the DYNAP-SE2 a perfect fit for mobile health and robotic applications. Each Chip features 1k redesigned adaptive exponential integrate-and-fire analog ultra low-power spiking neurons and 65k enhanced synapses with configurable delay, weight and short term plasticity. The innovative asynchronous low latency communication infrastructure (patented) enables each neuron to communicate with up to 230k surrounding neurons and infinite scalability via relay neurons for truly large scale networks.

DYNAP-SEL

Our next-generation neuromorphic chip, DYNAP-SEL, features 1k analog low-power spiking neurons and upto 80k configurable synaptic connections, including 8k synapses with integrated spike-based learning rules. The fan-in and fan-out capabilities of DYNAP-SEL are comparable to the connectivity architectures observed in mammalian cerebral cortex, and permit networks with biologically realistic connectivity to be emulated using DYNAP-SEL.Our technology enables implementation of spiking neural networks with on-chip learning and large fan-in/out network connectivity at an extremely low power-budget in real-time on DYNAP-SEL.

DYNAP-CNN

DYNAP-CNN is a scalable, fully-configurable digital event-driven neuromorphic processor with 1M ReLU spiking neurons per chip for implementing Spiking Convolutional Neural Networks (SCNN). This technology is ideal for always-on, ultra-low power and ultra-low latency event-driven sensory processing applications. With a dedicated interface for dynamic-vision-sensors, it allows direct input of event streams from most of the advanced dynamic-vision-senors in the world, enabling seamless integration and rapid prototyping of models. DYNAP-CNN is fully configurable and supports various types of CNN layers (like ReLU, Cropping, Padding and Pooling) and network models (like LeNet, ResNet and Inception). It provides complete control of your models with extensive programmablility of all of its parameters. In addition, DYNAP-CNN is scalable, enabling implementation of deep neural networks with unlimited number of layers over multiple interconnected DYNAP-CNNs.

Speck

Speck is a comprehensive, multicore spiking neural network processing chip. Speck is able to support large-scale spiking convolutional neural network (SCNN) with an fully asynchronous chip architecture. Speck is fully configurable with the spiking neuron capacity of 32 million. Furthermore, it integrates the state-of-art dynamic vision sensor (DVS) that enables fully event-driven based, real-time, highly integrated solution for varies dynamic visual scene. For classical applications, speck can provide intelligence upon the scene at only mWs with a response latency in few ms.

Xylo

Xylo is a ultra-low power, always on, low dimentional signal processing dedicated processor. Xylo combines the analog front end(AFE) that can efficiently provide pre-processing functionality to input analog signals. Xylo has is highly re-configurable and scalable, which supports feed-forward, recurrent and reservior and other complex neural network structure. Xylo can be easily combined with various common sensors such as: MEMS microphone , thermal sensor, pressure sensor, vibration sensor, IMU, Gyro, PPG sensor etc.

Software solutions

Samna

Samna is the developer interface to the SynSense toolchain and run-time environment for interacting with our devices. Written in C++, it provides a Python API and data visualization tools for working with spiking neural networks and for processing streams of event based data. Users can easily manipulate and upload networks generated by Rockpool or Sinabs to Synsense neuromorphic chips with the help of Samna.

Rockpool

Rockpool is an open source Python package for developing signal processing applications with spiking neural networks. Rockpool allows you to build networks, simulate, train and test them, deploy them either in simulation or on event-driven neuromorphic compute hardware. Rockpool provides layers with a number of simulation backends, including Brian2, NEST, Torch, JAX, Numba and raw numpy. Rockpool is designed to make machine learning based on SNNs easier. It is not designed for detailed simulation of biological networks.
 

Sinabs

sinabs – an open source pytorch based library – is developed to design and implement Spiking Convolutional Neural Networks (SCNNs).
The library implements several layers that are spiking equivalents of CNN layers.
In addition it provides support to import CNN models implemented in keras conveniently to test their spiking equivalent implementation.
 

Selected Publications

Indiveri, G., Corradi, F., Qiao, N. “Neuromorphic architectures for spiking deep neural networks. In Electron Devices Meeting (2015).
Brader, J., Senn, W., and Fusi, S., “Learning real world stimuli in a neural network with spike-driven synaptic dynamics”. Neural Comput. 19, 2881–2912 (2007).
Moradi, S, Qiao, N, Stefanini, F, Indiveri, G. “A scalable multicore architecture with heterogeneous memory structures for Dynamic Neuromorphic Asynchronous Processors (DYNAPs).” IEEE Transactions on Biomedical Circuits and Systems (2017).
Qiao, N, et al. “A reconfigurable on-line learning spiking neuromorphic processor comprising 256 neurons and 128K synapses.” Frontiers in neuroscience 9 (2015).
Giulioni, M., Corradi, F., Dante, V., Del Giudice, P., “Real time unsupervised learning of visual stimuli in neuromorphic VLSI systems.”, Scientific Report (2015).

Zurich

Thurgauerstrasse 40, 8050 Zurich, Switzerland

Chengdu

No. 1999-8-5, Yizhou Avenue, Gaoxin District, Chengdu, Sichuan, PR China

Nanjing

No. 22-98, Dangui Road, Pukou District, Nanjing, PR China

Shanghai

No. 302, Building 21, ZJ AI Island, Pudong Xin District, Shanghai, PR China

Suzhou

No.398 Ruoshui Road, Suzhou Industrial Park, Suzhou City, Jiangsu Province, PR China

We don’t develop technologies or accept funding for military purposes.