NEURALSENS
Smart gas and temperature sensors with neural-network-based low-level in-sensor data processing capability
NEURALSENS investigated how sensing and low-level data processing can be combined within the same hardware. The project brought together three Slovak research institutions with complementary expertise in sensing materials, device fabrication, electrical characterisation and neural-network design.
Its research connects hydrogen gas sensing, temperature sensing and crossbar-array Sensory Neural Networks (SeNNs). The central concept is to use the characteristics of individual sensing elements as physically implemented network parameters, enabling simple data-processing functions directly within the sensing array.
Project at a glance
- Project number: 09I05-03-V02-00058
- Implementation period: January 2024–August 2026
- Applicant and coordinating institution: Institute of Electrical Engineering, Slovak Academy of Sciences (IEE-SAS)
- Partner 1: Comenius University in Bratislava, Faculty of Mathematics, Physics and Informatics (CUB)
- Partner 2: Institute of Informatics, Slovak Academy of Sciences (II-SAS)
Research concept
Conventional sensing systems typically transfer sensor signals to separate processing units. NEURALSENS explores an approach in which part of this processing is encoded in the sensing hardware itself.
The project combines resistive sensing elements with a crossbar-array architecture. The SeNN design methodology connects measured device characteristics with the target response of individual elements in the array. Electrode geometry and fabrication parameters provide ways to adjust these responses and implement network parameters physically.
The broader motivation is to reduce the amount of raw sensor data that requires external transfer and processing, supporting future development of smart, energy-efficient sensing systems.
Project objectives
The project was structured around three specific objectives. The descriptions below set out their planned scope and target criteria.
O1 — Hydrogen sensing elements
Fabricate and characterise microscopic, capacitor-like TiN/TiO₂/Pt hydrogen-sensing elements suitable for integration into crossbar arrays.
The target criteria are:
- Diode-like current–voltage characteristics.
- A response to 1000 ppm H₂ that can be tuned over at least one order of magnitude by changing electrode size and shape.
O2 — SeNN design methodology and smart temperature sensing
Develop design rules for encoding a hardware algorithm in a crossbar array of resistive sensing elements.
The methodology uses the characteristics of individual sensors to determine the target responses required for low-level data processing within the array. The objective includes applying and validating these rules in a fabricated proof-of-concept smart temperature sensor based on a crossbar-array SeNN, with a target technology readiness level of TRL 3.
O3 — Smart hydrogen sensor laboratory prototype
Demonstrate a laboratory prototype of a TiN/TiO₂/Pt-based smart hydrogen sensor incorporating:
- At least 32 sensing elements on a single chip.
- A hardware algorithm for in-sensor data processing.
- At least four outputs, each representing a different hydrogen concentration range.
The target is a laboratory proof of concept at TRL 3, with hardware processing intended to improve sensing robustness and accuracy.
Project outcomes
The final monitoring report documents the following outcomes:
- O1 — Hydrogen sensing elements: TiO₂-based hydrogen sensors demonstrated responses controlled by electrode geometry, porosity and sensing-layer thickness. Sensors with different electrode geometries exhibited responses to 1000 ppm H₂ spanning more than one order of magnitude.
- O2 — SeNN design methodology: Design rules were computationally validated using experimental temperature data, achieving an average test classification accuracy of 94.1% across four temperature ranges. The project also developed and released open-source SeNN simulation software.
- O3 — Laboratory prototype: A hydrogen-sensing prototype combined a TiO₂ RRAM sensor array, in-sensor processing and RRAM-based computation. The final report documents real-time hydrogen sensing and assesses the demonstrator as a TRL 3 proof of concept, achieved through an adapted implementation approach.
Supporting publications, conference contributions, software and milestone summaries are available on the Results & Deliverables page.
Work packages and collaboration
All three institutions participate in each work package, with complementary responsibilities across fabrication, measurement, modelling and integration.
| Work package | Lead institution | Research focus |
|---|---|---|
| KPB1 — H₂ sensing element | CUB | Fabrication and characterisation of hydrogen-sensing elements, including control of their response through electrode geometry. |
| KPB2 — SeNN algorithm | II-SAS | SeNN design and simulation methodology, temperature-sensor proof of concept, and extension of the approach to hydrogen-sensing data. |
| KPB3 — SeNN laboratory prototype | IEE-SAS | Crossbar-array design, fabrication and integration of sensing elements with the SeNN methodology for a smart hydrogen-sensor laboratory prototype. |
The work packages form a connected research process: sensing-element measurements provide inputs for network design, while the resulting design requirements guide array fabrication and prototype evaluation.
Results and further information
The Results & Deliverables page presents research outputs, publications, conference contributions, software and the project deliverable register.
Further information is available on the following pages:
- Project Partners — participating institutions and their roles.
- SeNN Software Framework — the open-source simulation software.
- Project Review Meetings — project reviews and coordination meetings.
Funding
NEURALSENS is funded by the European Union – NextGenerationEU through the Recovery and Resilience Plan of the Slovak Republic under project No. 09I05-03-V02-00058.