Chiplet network links brainscaleS neuromorphic units for larger neural emulation
A Unified Interconnection Network for Chiplet-Based Scaling of the BrainScaleS Neuromorphic System
Hardware Architecture
Summary
Simulating large brain-like neural networks on special chips is hard because the chips have limited fixed hardware parts. The authors designed a new network chip that connects multiple smaller brainlike chips (called chiplets) so they work together as a bigger system. This network handles two kinds of data: error-tolerant signals like neural spikes and error-sensitive data like configuration info. The design uses different methods to send each type efficiently and reliably, letting many chiplets communicate with good timing and low errors even with some faults.
What this means in practice
- •For neuromorphic hardware engineers: Connect multiple BrainScaleS-2 chiplets to build scalable spiking neural network systems with mixed traffic handling.
- •For chiplet system designers: Use a unified interconnect to manage error-tolerant and error-sensitive data over a single link in chiplet-based analog-digital hybrid systems.
Authors
Robin Heinemann, Johannes Schemmel
Abstract
The BrainScaleS-2 (BSS-2) neuromorphic architecture combines analog emulation of spiking neural network (SNN) primitives with tightly coupled ADCs and digital processing units. These analog SNN primitives are fixed hardware resources that cannot be multiplexed, limiting the emulated network size to the number of physical hardware copies. To overcome the challenges of scaling analog designs, chiplet-based designs offer a promising approach with cost and flexibility advantages over monolithic scaling. Implementing a chiplet-based BSS-2 architecture requires an interconnection network that handles two distinct classes of traffic: error-tolerant traffic such as spikes and error-intolerant traffic like configuration data or data exchanged by the processing units. This work presents the design of a routing chiplet for the BSS-2 architecture that enables interconnection of multiple BSS-2 units in a 2D mesh topology. Both traffic classes are multiplexed over a single wide parallel die-to-die link. Exploiting the fault tolerance of SNNs, spikes are transmitted unsecured and synchronously, with the arrival time on the receiving side directly determining the pre-synaptic time of the spike. Conversely, error-intolerant data transmission is secured by a point-to-point Automatic Repeat Request protocol and uses credit-based flow control. These design choices are validated in simulation, where the proposed design can sustain $95\,\%$ link bandwidth utilization for the use case of surrogate gradient training across a wide range of spike-to-secured traffic ratios under a $1 \times 10^{-10}$ bit error rate with little impact on spike timing jitter.