Abstract
State-of-the-art large-scale neuromorphic systems require sophisticated spike event communication between units of the neural network. We present a high-speed communication infrastructure for a waferscale neuromorphic system, based on application-specific neuromorphic communication ICs in an field programmable gate arrays (FPGA)-maintained environment. The ICs implement configurable axonal delays, as required for certain types of dynamic processing or for emulating spike-based learning among distant cortical areas. Measurements are presented which show the efficacy of these delays in influencing behavior of neuromorphic benchmarks. The specialized, dedicated address-event-representation communication in most current systems requires separate, low-bandwidth configuration channels. In contrast, the configuration of the waferscale neuromorphic system is also handled by the digital packet-based pulse channel, which transmits configuration data at the full bandwidth otherwise used for pulse transmission. The overall so-called pulse communication subgroup (ICs and FPGA) delivers a factor 25–50 more event transmission rate than other current neuromorphic communication infrastructures.
Highlights
The last years have seen a steady increase in the size of neuromorphic systems in order to handle progressively more advanced computational tasks (Giulioni et al, 2008; Serrano-Gotarredona et al, 2009; Schemmel et al, 2010)
A combination of these characteristics is especially required when moving to a large-scale hardware system, such as the waferscale neuromorphic hardware depicted in Figure 1, which has been developed in the FACETS project (Ehrlich et al, 2007) and is currently being completed in the follow-on project BrainScaleS
We have developed a communication infrastructure (Hartmann et al, 2010) for this waferscale neuromorphic system centered around a application-specific digital neuromorphic communication IC, called digital network chip (DNC; Scholze et al, 2011)
Summary
The last years have seen a steady increase in the size of neuromorphic systems in order to handle progressively more advanced computational tasks (Giulioni et al, 2008; Serrano-Gotarredona et al, 2009; Schemmel et al, 2010). Several communication/interface boards based on commercial field programmable gate arrays (FPGA) have been developed in recent years, commonly employing the address-event-representation (AER) protocol for pulse transmission (Berge and Häfliger, 2007; Fasnacht et al, 2008; SerranoGotarredona et al, 2009) Those designs were predominantly optimized for asynchronous operation and low latency, whereas demands on integration density and bandwidth were relatively relaxed. A combination of these characteristics is especially required when moving to a large-scale hardware system, such as the waferscale neuromorphic hardware depicted, which has been developed in the FACETS project (Ehrlich et al, 2007) and is currently being completed in the follow-on project BrainScaleS This system employs waferscale integration technology to gain a high connection density thereby implementing 40 million synapses and up to 180 k neurons. It is designed for operating at a speed-up factor of 104 compared to biological real-time, which increases simulation speed and integration density of the analog neuron and synapse circuits at the same time (Schemmel et al, 2010)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have