2026 Mahowald-Mead Prizes

2 May 2026

The 2026 Mahowald-Mead Prize is awarded for "The SpiNNaker Ecosystem" project, to teams in Dresden and the University of Manchester, "for developing a large-scale neuromorphic computing platform that supports neuromorphic researchers and commercial applications".

The 2026 Mahowald Early Career Award is awarded to  Dr. Mohamadreza Zolfagharinejad from the University of Twente, for the project "Analogue speech recognition based on physical computing", "for leveraging the intrinsic physics of analog devices to emulate cochlear mechanics. This work exemplifies neuromorphic engineering by bypassing the conventional digital pipeline in favor of a prototype end-to-end analog design, with promising applications in hearing prosthetics and ultra-low-power edge processing". 

About the MMP Award

The SpiNNaker project (see video history) receives the Award for developing a large-scale neuromorphic computing platform that supports neuromorphic researchers and commercial applications, reflecting its evolution from an experimental simulator into a robust, industry-ready computational paradigm. By replacing centralized clock cycles with a massively parallel architecture centered on Address-Event Representation, the platform allows for asynchronous, packet-switched communication that bypasses the traditional Von Neumann bottleneck. For the research community, this provides a unique substrate to execute biological-scale spiking neural networks in biological real-time; meanwhile, for the commercial sector, the project’s second-generation transition—SpiNNaker2—enables the deployment of SpiNNcloud infrastructure. This low-latency cloud model leverages dynamic sparsity and hybrid AI accelerators to process high-velocity data streams, such as high-frequency financial telemetry or sparse Large Language Model inference, with quick responsiveness and energy efficiency far exceeding that of traditional GPU-centric clusters.

About the MECA Award

Guided by neuromorphic computing principles, this work (Nature 645, 886-892, 2025), demonstrates an end-to-end analog speech recognition system that composes two complementary physical primitives. The biologically plausible front end leverages Reconfigurable Nonlinear Processing Units (RNPUs, Nature 577, 341-345 (2020)), while the brain-inspired classifier is implemented on IBM’s HERMES analog in-memory computing (AIMC, Nat. Electron. 6, 680-693 (2023)) platform. This pipeline operates in the time domain (no Fourier transform, filter bank, or additional nonlinear processing), executing >95% of operations in analog hardware, and achieves near-software accuracy with >100× lower latency and >10× better energy efficiency in feature extraction on standard speech benchmarks.