Golana Computing and the EIC Accelerator: Magnets That Hear Faults

An experienced maintenance technician can often tell from the sound of a machine that something is wrong. Golana Computing, a Grenoble start-up founded in 2023, is trying to give that ability to hardware: a network of tiny magnetic structures whose behaviour, according to its first EU project, imitates in many respects the spiking of biological neurons.

The European Innovation Council has signed two grants for the idea. CORDIS records an EIC Transition grant of €2,499,999 for the magnetic neural network from 2023, and the October 2025 EIC Accelerator cut-off selected Golana PdM Suite, which adds generative AI on top. CORDIS records an EU contribution of €2,473,992.50 for that project, signed on 29 May 2026 and running from June 2026 to May 2028. Five months in, the project is too young to judge, and this article sets out what was promised and what can already be checked.

A five-person company with two EU grants

The French company register shows Golana Computing as an active simplified joint-stock company (SAS), registered on 23 January 2023 with research and development as its main activity, classed as an SME, and with a headcount band of 3 to 5 employees for 2024. Its president in the register is Ioan-Mihai Miron. The company's own team page calls him CEO and lists five co-founders and a magnetism specialist, and says the team came out of research on magnetism and AI at Spintec-CNRS.

On 2 April 2026 the Aujourd'hui Grenoble newsletter, summarising an ESSOR Isère report, described Golana as a five-person start-up with seven clients in France, Romania and Norway. It reported that the company targets its first revenue, €200,000, in 2026, and plans a €1 million funding round by the end of the year. The register's open data shows a 2023 turnover of zero. No completed funding round appears on the company website or in any source found when this article was checked on 4 November 2026.

The newsletter also puts the "potentiel total mobilisable" from the EIC at €8 million. The two grants in CORDIS add up to just under €5 million. The newsletter does not break its figure down, and neither the EIC nor the company has published an equity amount for the EIC Accelerator award, so the figures are kept apart here.

Neurons made of magnetic domain walls

The technology comes from spintronics, the field of electronics that uses the spin of electrons, the property behind magnetism. In a thin magnetic strip, regions of opposite magnetisation are separated by a boundary called a domain wall. The MAG.NET objective says the Spintec team found that a domain wall pinned in a geometric trap and then released behaves in many ways like a neuron charging and firing. A network of such devices, the objective says, had already performed real-time speech recognition and speaker identification without the usual feature extraction step, after converting audio into spikes with a mechanism inspired by the mammalian ear.

Predictive maintenance is the commercial application. The EIC Accelerator summary argues that existing predictive maintenance tools "fall short due to low data accuracy, high computational needs, and poor handling of complex, multi-system failures". Golana pitches a system that is, in the words of the Minalogic cluster directory, a "breakthrough platform for continuous monitoring of industrial equipment" that combines ultra-low-power hardware, bio-inspired signal processing and an AI maintenance assistant. The April 2026 Grenoble newsletter calls the approach "agnostique", setting it apart from solutions specialised in particular equipment.

The EIC Accelerator project goes further in its claims. Its CORDIS summary says the magnetic neuron algorithm, "developed with EIC Transition funding", matches deep learning accuracy while needing 10 times less computing power and 10 times less training data, with no fine-tuning. No paper, benchmark or third-party test of those ratios could be found in the searches for this article.

What the first EU project shows

MAG.NET ran under the EIC Transition challenge on green digital devices, from 1 May 2023 to 31 August 2026, with Golana as the only beneficiary. Its single periodic report, covering the first year to April 2024, describes experiments in five industrial facilities in Romania on five types of machine, with the company's own acquisition system for temperature, vibration, current and sound. It says the full chain from sensor to malfunction alert worked, but that most tests used a simulated version of the magnetic network and a smaller subset ran on the physical magnetic system, with what it calls promising results.

The distinction matters because the magnetic network is what the EU funded under green digital devices, while a simulated network runs on conventional processors. The report does not say how much of a final product would run on the magnetic hardware. It lists a commercial prototype and CE marking as next steps. MAG.NET ended two months before this article was checked, and CORDIS shows no final report or results publication for it yet.

1. Anomaly detection on real machines (MAG.NET)Mostly simulatedTests in five Romanian plants, mostly on a simulated network, per the 2024 report.
2. Physical magnetic hardware in the loopNot shownA smaller subset of tests; no final MAG.NET report on CORDIS.
3. Deep learning accuracy at 10x less computing and dataClaimStated in the CORDIS summary; no published benchmark.
4. Generative AI assistant explaining root causesToo earlyThe EIC Accelerator project started in June 2026.
5. Commercial adoptionTrialsSeven industrial trials reported by the company; first revenue targeted for 2026.

The EIC Accelerator project

The EIC received 923 full applications for the 1 October 2025 cut-off. On 17 February 2026 it announced 61 selected companies from 121 interviewed, 85% of them on blended finance, and its list of selected companies names Golana Computing with Golana PdM Suite, "Transforming Predictive Maintenance (PdM) with Bio-Inspired Magnetic Neurons and GenAI", for blended finance.

The project sits under the EIC challenge GenAI4EU, "Creating European Champions in Generative AI". The summary describes an assistant that detects anomalies, explains root causes in natural language and recommends actions, presented through heatmaps, and a later "fully autonomous AI agent" for continuous sensor analysis. It frames the market with a figure of more than €1.4 trillion a year lost to maintenance inefficiencies in heavy industry, a number it does not source. The generative AI direction predates the award: a job offer posted through the Grenoble MIAI cluster for a June to July 2025 start sought an engineer to "design and integrate generative-AI / LLM components" and deploy code to edge hardware and a cloud dashboard.

Trials, clients and logos

The two sources that give the number seven use different words. The EIC Accelerator summary speaks of "trials with seven industrial partners across Europe"; the April 2026 newsletter speaks of seven clients in France, Romania and Norway. A trial and a paying client are not the same thing, and no source reviewed states which of the seven pay. The company website shows the logos of Delonghi, Pehart, Megatitan, Vivracoustic and Rombat under live partner deployments, without describing any of them. Searches for those company names together with Golana found no statement from any of them about the work.

Where it fits in EU policy

The award falls under GenAI4EU, the Commission's initiative for generative AI in European industry, which it says has close to €700 million behind it across Horizon Europe, the Digital Europe Programme and the EIC, with manufacturing and engineering among its sectors. The first grant came from the EIC Transition challenge on green digital devices. Golana's case touches both: AI for factories that runs at the edge, which the 2024 MAG.NET report says reduces data transmission and security needs, on hardware that grows out of French public spintronics research. Whether the magnetic hardware, rather than the software running in simulation, ends up in the product is the question the record does not yet answer.

The verdict

Golana Computing has not been asked for comment for this article. The reporting is based on publicly available sources, listed below.