The Potential Impact of Neuromorphic Computing on Radio Telescope Observatories

TL;DR

This work analyzes how neuromorphic computing and Spiking Neural Networks can transform radio astronomy processing by enabling real-time, power-efficient handling of RFI, transients, and imaging across current and near-term telescopes. It proposes a practical progression from FPGA-based near-term deployments to neuromorphic ASICs, supported by system-level energy and latency modeling for instruments such as MWA, ASKAP, LOFAR, SKA-Low, SKA-Mid, and ngVLA. Key findings indicate potential energy savings of up to three orders of magnitude for several processing steps, with RFI detection identified as the most promising near-term application. The study argues that neuromorphic hardware could dramatically reduce operational costs and enable future telescope scales, framing radio astronomy as a major domain for in-sensor computing and large-scale spectrographic processing.

Abstract

Radio astronomy relies on bespoke, experimental and innovative computing solutions. This will continue as next-generation telescopes such as the Square Kilometre Array (SKA) and next-generation Very Large Array (ngVLA) take shape. Under increasingly demanding power consumption, and increasingly challenging radio environments, science goals may become intractable with conventional von Neumann computing due to related power requirements. Neuromorphic computing offers a compelling alternative, and combined with a desire for data-driven methods, Spiking Neural Networks (SNNs) are a promising real-time power-efficient alternative. Radio Frequency Interference (RFI) detection is an attractive use-case for SNNs where recent exploration holds promise. This work presents a comprehensive analysis of the potential impact of deploying varying neuromorphic approaches across key stages in radio astronomy processing pipelines for several existing and near-term instruments. Our analysis paves a realistic path from near-term FPGA deployment of SNNs in existing instruments, allowing the addition of advanced data-driven RFI detection for no capital cost, to neuromorphic ASICs for future instruments, finding that commercially available solutions could reduce the power budget for key processing elements by up to three orders of magnitude, transforming the operational budget of the observatory. High-data-rate spectrographic processing could be a well-suited target for the neuromorphic computing industry, as we cast radio telescopes as the world's largest in-sensor compute challenge.

Figures (7)

• Figure 1: Predicted relative (circles, left axis) and absolute (bars, right axis, log-scale) power consumption for several radio astronomy processing tasks and several radio telescopes. RFI detection at the receiving stage with purely neuromorphic computing tasks would consume up to three orders of magnitude less energy for MWA, ASKAP and SKA-Mid instruments. Correlator-scale RFI detection would provide almost the same relative improvement. Post-correlector RFI-detection provides around one order magnitude improvement, as would transient detection and imaging.
• Figure 2: High-level data flow diagram for the MWA telescope. Cyan blocks represent data sources or sinks, yellow blocks represent FPGA processing, blue blocks represent GPU processing, and orange blocks represent CPU processing.
• Figure 3: High-level data flow diagram for the ASKAP telescope. Diagram conventions from Figure \ref{['fig:dataflow:MWA']} apply.
• Figure 4: High-level data flow diagram for the LOFAR telescope. Diagram conventions from Figure \ref{['fig:dataflow:MWA']} apply.
• Figure 5: High-level data flow diagram for the SKA-Low telescope. Diagram conventions from Figure \ref{['fig:dataflow:MWA']} apply.