Beyond Encryption: Brain-Inspired Infrared for Unhackable Data Transfer

The rise of artificial intelligence (AI) and the Internet of Things (IoT) has made smart devices critical, but processing vast data volumes efficiently remains difficult. Standard computing systems, based on the von Neumann architecture, separate memory and processing units, requiring constant data transfers that consume significant energy and introduce delays. These bottlenecks are pronounced in data-intensive AI and IoT tasks, such as real-time image recognition or secure communication, where rapid, low-power processing is essential. Near-sensor neuromorphic computing addresses these issues by integrating sensing, storage, and computation. Inspired by the brain, it uses parallel processing—handling multiple tasks simultaneously—and in-memory computation, where data is processed where it’s stored, reducing energy use and latency.

Secure data transmission is increasingly vital in IoT, where vulnerabilities like hacked smart cameras or intercepted health monitor data expose sensitive information. As IoT devices proliferate, ensuring private, tamper-proof communication is critical. A study published on May 6, 2025, by Wang et al. presents a near-sensor neuromorphic computing system combining a thermopile infrared detector—a device that senses infrared signals—with a memristor array, a component that stores and processes data like brain synapses. This system securely handles encrypted visual and audio data using infrared signals, which are harder to detect than visible light. Tested for recognizing handwritten digits and voice commands, it offers strong potential for secure communication. This article explores the system’s design, results, security advantages, and applications, with technical details in Wang et al. (2025).

Why Thermopile Infrared Detectors?

Infrared radiation, part of the electromagnetic spectrum beyond visible light (wavelengths longer than 780 nm), is emitted by objects as heat, making it ideal for detecting warm or cool sources without visible cues. Thermopile infrared detectors sense these signals passively, detecting emitted radiation without sending out their own signals, which enables covert sensing for secure applications. Unlike visible light detectors (400–780 nm), they offer enhanced concealment, as infrared is invisible to the human eye and standard cameras. The study’s detector operates across a wide temperature range (100–310 °C), allowing it to detect both relatively cool objects (e.g., human skin) and warmer ones (e.g., machinery), broadening its use in surveillance, medical sensing, and communication.

The detector uses a double-ended design with symmetrically arranged thermocouple strips, typically made of n-type and p-type polysilicon, materials chosen for their ability to generate voltage when heated due to differing electrical properties. An adiabatic groove, a heat-insulating feature, increases the temperature difference between hot and cold ends, enhancing the Seebeck effect, which converts heat into electricity. High-resolution scanning electron microscopy (SEM) and ANSYS simulations, used to optimize design and predict performance under various conditions before fabrication, confirmed consistent voltage outputs across temperatures and signal frequencies. The detector’s high sensitivity and detectivity, with response times under a fifth of a second, make it reliable for capturing faint infrared signals in noisy environments.

Memristors: Neural-Inspired Data Processing

The system’s core is a memristor array, where a memristor is a device that adjusts its electrical resistance based on past signals, retaining this setting to store or process data. This mimics synaptic behavior in the brain. Built with a Pt/HfAlOx/TiN structure, the memristor uses a 10 nm HfAlOx layer, compatible with standard chip manufacturing. Platinum (Pt) electrodes provide stable electrical contact, while titanium nitride (TiN) supports efficient resistance switching. X-ray photoelectron spectroscopy (XPS) confirmed an equal hafnium-to-aluminum ratio, optimizing performance.

The memristor’s resistance changes due to oxygen vacancy movement in the HfAlOx layer—small defects that shift under electric fields, altering conductivity. This change is non-volatile, meaning it persists without power, enabling data retention for ten thousand seconds (about three hours). The memristor withstands millions of signal cycles and produces signals that strengthen connections (excitatory postsynaptic current, or EPSC) or weaken them (inhibitory postsynaptic current, or IPSC).

These support long-term potentiation (LTP) and long-term depression (LTD), where repeated signals strengthen or weaken connections over time, allowing the system to learn patterns, such as recognizing digits or voice commands. Memristors are highly energy-efficient, using approximately 100 times less power than traditional memory like DRAM for pattern recognition tasks, due to in-memory computation that eliminates data transfer delays. Tests showed stable performance, highlighting its suitability for neuromorphic processing. Memristors could drive future low-power computing systems. For details, see Wang et al. (2025).

How the System Works

The system integrates the thermopile infrared detector and memristor array to sense, process, and store infrared data efficiently. The detector captures infrared signals, with stronger signals (higher intensity) producing electrical spikes—short pulses—with higher frequency or amplitude. These spikes feed directly into the memristor array without extra conversion circuits, reducing power use and delays compared to systems with separate sensing and processing units.

The detector filters out non-infrared noise (e.g., visible light showing a misleading digit “9”) and detects hidden infrared data (e.g., the correct digit “7”). The memristor array, inspired by neural networks, interprets these signals by adjusting its connections. Its neural-inspired architecture and learning capabilities denoise data by prioritizing meaningful infrared signals over irrelevant ones, using LTP/LTD to refine pattern recognition. For encryption, the memristor array processes encoded infrared data, adjusting resistance to decode or further encrypt signals for secure transmission. This direct sensor-to-processor connection ensures fast, energy-efficient operation for secure, real-time applications.

Secure Data Transmission and Recognition

The study’s dual encryption method strengthens security beyond single-custodian password systems. Infrared signals are hard to intercept, as they are invisible and require specialized detectors like thermopiles, unlike visible light data easily captured by cameras. Countermeasures, such as using specific infrared wavelengths (e.g., 850–950 nm) or encoding data in temporal spike patterns, further reduce interception risks. The system uses two panels: one with misleading data visible in standard light and another with correct data in the infrared spectrum. Splitting data into visual and audio channels ensures interception requires both, significantly raising the bar compared to single-channel systems.

The thermopile detector captures correct data, converting pixels into voltages processed by the memristor array. The data splits into visual information and audio for voice recognition. A convolutional neural network (CNN)—an AI model that identifies patterns—is trained on MNIST digits for visual data and voice samples for audio, using memristor outputs to recognize patterns. In tests, the system achieved over 98% accuracy for MNIST digits after multiple training rounds, compared to about 84% without infrared denoising (Wang et al., 2025). The multi-modal approach, requiring visual and audio correlation, enhances security by making unauthorized decryption nearly impossible without both streams. Potential vulnerabilities, like advanced infrared detectors, are mitigated by dynamic wavelength shifts or randomized temporal encoding. This makes the system ideal for voice-activated locks or surveillance.

Potential Applications and Future Steps

This system advances secure, low-power data processing for multiple domains. In IoT, it could enable device authentication (e.g., verifying smart thermostats), secure control of critical infrastructure (e.g., power grids), or private data transfer between devices. For autonomous systems, it could support secure sensor fusion in vehicles, ensuring infrared and other sensor data remain untampered for safe navigation. Encrypted networks could use infrared for secure local channels, protecting sensitive data in offices or hospitals.

In medical devices, it could securely transmit patient vitals or control implants. Bio-inspired robotics could use secure infrared sensing for covert perception in dark or cluttered environments. In bioinformatics, it could safeguard genomic data analysis. Scaling challenges include manufacturing cost-effective, high-performance memristor arrays, requiring advances in deposition techniques like atomic layer deposition. Integrating large-scale arrays with detectors demands optimized packaging to maintain efficiency.

Synergies with AI (e.g., advanced CNNs), edge computing (e.g., local processing), and materials (e.g., 2D materials like graphene for detectors) could enhance performance. Combining infrared with ultraviolet signals could improve multi-spectral detection for surveillance or environmental monitoring.

Conclusion

The near-sensor neuromorphic computing system by Wang et al. achieves over 98% accuracy in digit recognition, robust dual encryption via infrared signals, and low power use through integrated thermopile infrared detectors and memristor arrays. Its ability to hide data in the infrared spectrum and process visual and audio streams simultaneously offers a secure, efficient approach to information handling. By addressing von Neumann bottlenecks and IoT security needs, this technology could redefine secure, energy-efficient computing, enabling smarter, safer systems for homes, industries, and healthcare in an AI- and IoT-driven future.

Citation

Wang, Z., Zhang, J., Zhang, Z., Meng, J., Lei, C., & Wang, T. (2025). Near-Sensor Neuromorphic Computing System Based on a Thermopile Infrared Detector and a Memristor for Encrypted Visual Information Transmission. American Chemical Society. Available: [Insert DOI or URL when available]

Xia, Z., Sun, X., Wang, Z., et al. (2025). Low-Power Memristor for Neuromorphic Computing: From Materials to Applications. Nano-Micro Letters, 17, 217. https://doi.org/10.1007/s40820-025-01705-4

Related Articles

PHP Code Snippets Powered By : XYZScripts.com
★

Track This Company

Create a free account to add companies to your watchlist and get notified when patents, funding, or key developments change.