The BrainScaleS-2 Accelerated Neuromorphic System With Hybrid Plasticity

Abstract

Since the beginning of information processing by electronic components, the nervous system has served as a metaphor for the organization of computational primitives. Brain-inspired computing today encompasses a class of approaches ranging from using novel nano-devices for computation to research into large-scale neuromorphic architectures, such as TrueNorth, SpiNNaker, BrainScaleS, Tianjic, and Loihi. While implementation details differ, spiking neural networks—sometimes referred to as the third generation of neural networks—are the common abstraction used to model computation with such systems. Here we describe the second generation of the BrainScaleS neuromorphic architecture, emphasizing applications enabled by this architecture. It combines a custom analog accelerator core supporting the accelerated physical emulation of bio-inspired spiking neural network primitives with a tightly coupled digital processor and a digital event-routing network.

A Scalable Approach to Modeling on Accelerated Neuromorphic Hardware

“…This work describes the software environment for the latest BrainScaleS (BSS) neuromorphic architecture (Pehle et al, 2022 ): the BrainScaleS-2 (BSS-2) operating system. In Müller et al ( 2020b ), we introduced the operating system for the BrainScaleS-1 (BSS-1) wafer-scale neuromorphic hardware platform.…”

“…Over the past decade neuromorphic systems evolved from intricate lab setups toward back ends for the more comfortable execution of spiking neural networks (Indiveri et al, 2011 ; Furber et al, 2012 ; Benjamin et al, 2014 ; Davies et al, 2018 ; Pehle et al, 2022 ). One major step along this development path is to provide users with seamless access to the systems.…”

“…The environment simulation as well as the interaction with the insect is performed by a program running on the on-chip processor. Figure taken from Pehle et al ( 2022 ).…”

“…It provides a neuromorphic substrate for neuroscientific modeling as well as neuro-inspired machine learning. Earlier work shows its scalability in wafer-scale applications, emulating up to 200 k neurons and 40 M synapses (Schmitt et al, 2017 ; Kungl et al, 2019 ; Müller et al, 2020b ; Göltz et al, 2021 ), as well as its energy-efficient application as standalone system with 512 neurons and 128 k synapses in use cases related to edge computing (Stradmann et al, 2021 ; Pehle et al, 2022 ). Compared to the biological time domain, the model dynamics evolve on a 1.000-fold accelerated time scale making the system interesting for iterative and long-running experiments.…”

“…In this section, we introduce the BSS-2 system and highlight the basic hardware design which is guiding the development of the accompanying software stack. For a more in depth description of the hardware aspects of the BSS-2 system refer to Aamir et al ( 2018 ), Schemmel et al ( 2020 ), and Pehle et al ( 2022 ).…”