Nature’s Nose in a Chip: How Bio-Engineered Sensors Are Learning to Smell

A tanker truck rolls through a city’s outskirts at dawn, its cargo of industrial chemicals invisible to the people already out for a jog nearby. As the driver parks near a water treatment plant, a handheld sensor tucked into a city worker’s vest alerts with a quiet pulse: trace levels of airborne toxins have spiked near the facility—far below the threshold any person or traditional detector could notice. Within minutes, city responders reroute traffic and contain a minor leak before it grows into an environmental crisis.

This kind of early, sensitive chemical detection is quickly approaching reality. In a recent study published in Neuromorphic Computing and Engineering, researchers Max et al. present an approach that combines synthetic biology with brain-inspired computing to create artificial smell detection systems. Their work addresses fundamental problems in electronic nose technology by integrating biological receptor mechanisms with computing systems that work like the brain.

Why Smart Artificial Noses Matter

Your nose can detect and distinguish thousands of different smells with incredible precision while using very little energy. This biological smell detection system, called olfaction, represents one of nature’s most advanced sensory capabilities. Consider how this technology could alter our daily lives. Doctors could diagnose diseases early through breath analysis, environmental sensors could detect pollution at trace levels, and food safety systems could instantly identify spoilage or contamination.

Today’s electronic noses (e-noses) work in various applications from medical diagnosis to food safety monitoring, but they face persistent problems. Sensor drift causes performance to worsen over time, while cross-sensitivity reduces accuracy when multiple chemical compounds are present. Metal oxide sensors require high operating temperatures, consuming substantial energy, and most systems lack the computational power to interpret complex odor patterns effectively.

Learning from Nature’s Design

Building on our understanding of biological systems, the smell system handles real-world complexity through several distinct mechanisms. Individual smell receptors, when exposed to naturally occurring concentrations, show highly specific responses to particular odors. This specificity contrasts with earlier assumptions that receptors always respond broadly to multiple compounds.

Recent research suggests that competitive interactions between odors occur primarily at artificially high concentrations. Natural exposure levels often fall near detection thresholds, where receptors work most precisely. This understanding reveals why biological systems excel at navigating complex odor environments.

How Nature Processes Smell Signals

Having established how biological receptors work individually, we can now explore how nature processes these signals collectively. One of the most intriguing aspects of biological smell detection lies in its signal processing capabilities.

While effective for basic chemical detection, they suffer from slow response times—typically requiring several seconds—and experience drift over extended periods. Drift shows up as changes in sensor responses under identical conditions, caused by chemical and physical changes in the sensing material structure, along with environmental factors like temperature and humidity variations.

To resolve these issues, recent brain-inspired approaches have attempted to implement spike-based encoding. This imitates how brain neurons communicate through rapid electrical pulses rather than continuous signals. These systems also use adaptive learning mechanisms that help them adjust to changing conditions over time.

Such approaches implement normalization through inhibitory processes—like having some neurons that calm down others when they get too excited. They also use synaptic plasticity, which is the ability of connections between neurons to strengthen or weaken based on experience. This allows the system to continuously adapt its internal odor representations through experience-driven learning.

Biosensors: Using Real Biology

Biosensors offer an alternative approach by directly interfacing with biological elements such as whole cells or purified proteins. While this strategy provides greater specificity through authentic biological receptors, it introduces practical concerns. These include shorter operating lifetimes, contamination risks, and complex maintenance requirements.

A more viable path involves synthetic biology strategies—essentially engineering biological components in controlled laboratory conditions. Scientists extract specific smell system components and deploy them in controlled, non-living environments.

Recent developments have revealed successful reconstruction of mammalian and insect smell receptor proteins into stable platforms. These use electrical, optical, and acoustic signal detection methods to translate biological responses into measurable signals.

Particularly noteworthy are developments in complete smell system reconstruction using bio-mimetic environments. These include membrane-bound vesicles (tiny bubble-like structures that mimic cell membranes) and virus-like particles. These approaches leverage insect smell mechanisms based on interactions between a conserved co-receptor called Orco and odor-binding subunits.

Together, these components form ligand-gated ion channels—think of them as tiny gates that open when specific smell molecules bind to them, allowing charged particles to flow through and create electrical signals.

Recent studies have achieved notable sensitivity, with demonstrations of parts-per-billion detection capabilities. Some systems can detect femtomolar concentrations, which means they can sense incredibly tiny amounts of specific molecules—like detecting a single drop of perfume in an Olympic-sized swimming pool.

However, these efforts have largely focused on single receptor systems, missing the multiplexed decoding signals that characterize natural smell processing (the way signals from many different receptors are combined and interpreted simultaneously by the brain).

A New Integrated Approach

Recognizing these gaps in current technology, the research team developed a novel methodology that moves beyond existing constraints. Rather than treating sensing, signal processing, and brain-inspired computation as separate components, their approach integrates these aspects comprehensively.

This alleviates the lack of coordinated design between sensor arrays, decoding algorithms, and their physical realization. The result is a system where each component is designed to work optimally with the others.

Central to this approach is the concept of hybrid synthetic sensory neurons—a new class of bio-electronic devices. These couple the electrical activity of synthetic cells made from biological materials directly with computer circuits designed for spike generation, mimicking how brain neurons communicate.

This framework harnesses three interconnected disciplines. Neuroscience, synthetic biology, and brain-inspired engineering. Each field contributes essential knowledge that makes the integrated system possible.

The bottom-up sensor design using synthetic biology informs the smell receptor and electrode response model. This allows simulation of electrode responses with biologically realistic ion channel models, setting realistic estimates for electrode responses that are crucial for designing sensor and amplification stages.

This approach also enables selection of Orco-OR complexes that respond strongly to target odor combinations while minimizing the number of required receptors. The sensor and brain-inspired processing stages are naturally intertwined, requiring careful selection and tuning of the bio-electronic interface.

This integration includes considerations of electrode surface material, synthetic cell attachment techniques, sensor array geometry, odor delivery, and amplification electronics. The goal is to optimize ionic current detection while achieving fast odor throughput, high signal-to-noise ratios, and long synthetic cell lifetimes.

Building with Insect Receptors

Molecular recognition in insect smell detection relies on transmembrane protein complexes. These consist of an odorant receptor (OR) and a co-receptor called Orco. While ORs vary in sequence and odor specificity across different insect species and ecological niches, Orco remains highly conserved across insect species.

This conservation suggests the possibility of a modular platform. An Orco-based sensing element would allow swapping of different ORs to target various compounds, like changing the lens on a camera to focus on different subjects.

To show this modularity, the researchers computationally assembled Orco-OR complexes using AlphaFold 2 modeling. This is an AI system that predicts protein structures based on their amino acid sequences. They selected the Orco variant from D. melanogaster fruit flies and considered 348 Orco-OR complexes.

These included non-D. melanogaster ORs from other insect species and mammals as negative controls. Each predicted structure was evaluated using interface-predicted template modeling scores, with higher values indicating greater confidence in prediction accuracy.

Results revealed that D. melanogaster Orco formed acceptable models with ORs from foreign insect species, while mammalian ORs did not produce acceptable models, as expected. This confirms the modularity concept works within the insect kingdom.

Testing Computer Predictions Against Real Data

To build a digital representation of receptor-odor interactions, the team tested their generated AlphaFold structures with actual biological recordings. They used the DoOR 2.0.1 dataset, a comprehensive database of smell responses from various measurement techniques.

These techniques included single receptor recordings, calcium imaging, and laboratory expression systems. Initial focus on terpenes, a class of naturally abundant plant compounds, showed auspicious results.

Binding energy differences between responding and non-responding neurons were highly significant, suggesting that predicted structures contain realistic binding pockets. When expanded to all molecules in the DoOR dataset, a weak correlation between neuron activation and binding energy was observed.

These results showcase the potential for using computational tools to predict receptor-to-odor responses. This suggests that molecular recognition in insect smell detection might be condensed into a measure of binding energy between odors and receptors.

Creating Synthetic Cells

To measure receptor binding to odor molecules, receptor proteins must be attached to surfaces. Rather than attaching individual receptors, which can cause stability problems, the research team developed an approach using lipid membrane vesicles attached to electrode surfaces.

These vesicles are bubble-like structures made of the same materials as cell membranes. They provide a native lipid environment and can maintain membrane potential through ion imbalance, just like real cells.

The system works beautifully. When Orco-OR channels open upon odor binding, the vesicle membrane potential changes its electrical charge. These lipid vesicles with life-like features are called synthetic cells and represent an active area of development.

The research team built a quantitative model of their proposed assembly. In this model, lipid-membrane-stabilized Orco-OR ion channels sit on electrodes, generating current that flows into electrode impedance when activated.

They modeled ion channels as resistors with two conductance states—essentially switches that can be either open or closed. Electrodes behave electrically as parallel resistor-capacitor circuits.

The analysis revealed vesicle radius as a crucial design parameter. Total charge in a synthetic cell scales with the cube of radius, while the number of ion channels scales with the square of radius.

This relationship creates two distinct operating regimes. Smaller vesicles are dominated by finite numbers of encapsulated charges, while larger vesicles respond to inserted ion channel density. Understanding this relationship helps engineers optimize sensor design for specific applications.

Choosing the Right Receptors

The research team investigated key design questions crucial for practical applications. Which receptors are needed for distinguishing specific odor sets? What response signal magnitudes can we expect from individual synthetic cells? How many synthetic cells of each receptor type are required for sufficient signal quality?

Their approach calculates “odor separability”—the distance between response patterns to different odors. The system iteratively excludes receptors that contribute least to separability, like removing team members who don’t add unique value.

This analysis yields optimal receptor combinations for different odor sets. The separability naturally depends on the chemical relationships between target odors.

Testing three odor sets with different degrees of chemical relationship showed predictable patterns. Odorants from different chemical families demonstrate large relative separability, while those from similar chemical families are naturally less separable.

The algorithm precisely identifies which receptors need reconstruction for optimal odor discrimination. This targeted approach reduces complexity while maintaining performance.

The team constructed synthetic datasets of sensor responses using selected odor-receptor combinations. They accomplished this through simple ion channel models for individual receptors, combined with electrode response models.

Each synthetic cell contains receptors of the same type. The team modeled responses as random processes involving channel opening and closing probabilities, creating realistic datasets for testing their system.

Training Artificial Neural Networks

The synthetic dataset enables training of spiking neural networks (SNNs) to classify odors based on electrode responses. SNNs are computer networks that mimic how brain neurons communicate through electrical spikes rather than continuous signals.

The SNN architecture consists of three fully connected layers. Input neurons are determined by the number of different receptors, and output neurons correspond to specific odor classes. This creates a system that can learn to recognize different smells.

Neurons follow the leaky-integrate-and-fire model, a simplified model of how real neurons work. They accumulate electrical charge until reaching a threshold, then fire a spike and reset. Training was performed using backpropagation through time, a learning algorithm that adjusts connection strengths based on performance.

The SNN architecture provides advantages in processing speed during operation due to massively parallel spiking behavior. Connection weights are optimized through four-bit quantization, which reduces precision to save energy while maintaining performance.

Classification results demonstrate that SNNs successfully learn to classify odor sets, though accuracy depends on receptor numbers and chemical structure of odors. Interestingly, SNN classification results don’t exactly follow linear odor space analysis.

This is because non-linear analysis can exploit deeper underlying correlations between odor responses, finding patterns that simpler methods miss.

Complete Hardware System

The complete hardware architecture consists of an analog front-end designed to process chemical sensing signals using event-driven analog SNN architecture. By bypassing analog-to-digital converters and leveraging synthetic biological assemblies, this approach enables real-time, low-power operation suitable for always-on applications.

The system, designed in SkyWater 130 nm CMOS process (a standard computer chip manufacturing process), integrates several key components. These include a bio-electronic interface coupling lipid vesicles with synthetic smell receptors and microelectrode arrays.

It also features a low-noise analog amplifier for signal conditioning and a brain-inspired SNN for processing sensor responses in a low-power mixed-signal domain. This integration creates a complete sensing and processing system on a single chip.

The design exploits transistors operating in weak inversion (subthreshold) regime. This ensures that tunable device parameters can be directly mapped to computational models, aligning hardware behavior with algorithmic specifications.

Neuron circuits emulate leaky integrate-and-fire dynamics using analog charge collection in capacitors to represent membrane potential. This creates hardware that behaves like biological neurons.

Input spikes arrive through analog differential-pair synapse circuits with digital weights determining connection strength. Spiking events are communicated asynchronously via brain-inspired address event representation schemes using four-phase handshake protocols, eliminating the need for global clocks.

Creating Hybrid Bio-Electronic Neurons

The research team simulated the complete signal processing pipeline from ion dynamics on microelectrodes to computer chip spiking neuron firing. This comprehensive simulation helps ensure the system will work as designed.

Microelectrode pad arrays detect low-amplitude bioelectronic signals in the tens of microvolt range. These incredibly small signals are fed into low-noise amplifiers for signal conditioning, which enhance weak signals while preserving signal fidelity.

These amplifiers provide output swings of approximately 5-10 mV with gains of 50.05 dB, ensuring reliable amplification. The amplified signals are then processed by spiking neurons, where modulated current injection activity connects amplifier outputs to neuron circuits.

This directly stimulates spike generation, creating an event-driven architecture that enables effective signal encoding. Spike generation becomes proportional to detected signal strength, facilitating mean rate coding for robust event-based detection.

Comprehensive simulations capture the entire signal processing pipeline. Neuron circuits are maintained near threshold by balancing injection and leakage currents, allowing neurons to rapidly generate spikes upon detecting simulated odor signals.

This integration of synthetic biological components with computer circuits creates a new class of hybrid synthetic sensory neurons. The interface from sensor pad to amplifier to neuron enables event-driven detection and spike generation, bridging the gap between biology and electronics.

Real-World Considerations

A key component of the platform is the bioelectronic interface bridging biological odor recognition with computer-based electronics. Enhancing signal transmission requires characterizing the synthetic cell-electrode interface and optimizing electrode impedance for improved sensitivity and signal integrity.

Systematic electrode engineering for biological signal detection is well-established through techniques like polymer layer applications. However, larger conceptual issues remain to be addressed.

These include restriction to biocompatible molecules, avoiding substances that might damage synthetic cells, and general lifetime restrictions of biological receptors outside living systems. Living cells invest in membrane protein turnover with typical half-lives of days, suggesting disposable sensor elements represent a realistic first implementation.

The low power consumption of brain-inspired systems enables compact designs suitable for various applications. These could include environmental sampling for pollution monitoring, safety applications for detecting dangerous chemicals, and food processing for quality control.

By integrating multiple chips into the platform, each microelectrode operating in separate chambers targeting single receptor types, the system could scale into multi-microelectrode architectures. This would allow simultaneous screening of multiple receptors, expanding detection capabilities.

While process variations and mismatch in weak-inversion circuits introduce variability in brain-inspired systems, the platform integrates multiple biasing mechanisms to compensate for these effects. Bio-inspired computational strategies often exploit variability in network parameters as computational resources for learning applications.

The primary source of drift and variability is expected from biological receptors, such as ion leakage from synthetic cell vesicles occurring over typical timescales of days. This may necessitate electrical compensation methods like DC drift offset compensation or systematic approaches using homeostatic learning mechanisms—self-regulating systems that maintain stability.

Current Constraints and Next Steps

The underlying odorant concentration levels present important considerations for real-world applications. The DoOR dataset contains recordings with volatile compound concentrations orders of magnitude higher than naturally occurring levels.

While this increases signal-to-noise ratios in spike recordings, it reduces individual receptor selectivity, creating trade-offs between sensitivity and selectivity in neuron responses. In real-world applications with natural concentrations, this trade-off likely manifests as lower signal-to-noise ratios but tighter matching between individual receptors and odors.

Repeating odor space analysis under natural and artificially high concentrations using the sensing system will help fine-tune platform performance. This will provide insights about sensitivity/selectivity curves of sensory neurons under realistic conditions.

The current setup is restricted to supervised training using single, non-mixed odor presentations. Future work will investigate odor mixtures as they appear in real-world applications, potentially modeling ligand-receptor interactions as fuzzy sensors.

Due to receptor co-activation, careful characterization will be needed to determine expected responses from synthetic cells under odor mixtures. This represents a significant step toward practical deployment.

Technical Implementation Details

The complete hardware system brings together multiple components. The researchers integrated bioelectronic interfaces, low-noise analog amplifiers, brain-inspired SNNs, and RISC-V microcontrollers (a type of computer processor) for spike decoding and connection weight configuration.

The low-noise amplifier uses specialized circuit configurations that improve signal integrity and reduce unwanted noise. Neuron circuits use capacitors to represent membrane potential, with three analog control voltages regulating charging current, decay rate, and reset behavior.

The system operates without traditional clock signals, instead using event-driven communication. When neurons fire, they send spikes through four-phase handshake protocols that ensure accurate, lossless communication throughout the network.

This technical foundation enables the system to process biological signals in real-time while consuming minimal power—a crucial capability for practical deployment in sensing applications.

Bridging Biology and Silicon: What Comes Next

This comprehensive approach highlights how synthetic biology can be effectively combined with brain-inspired computing to create artificial smell systems that closely replicate natural processes. The coordinated design methodology resolves fundamental issues that have affected artificial smell detection for decades.

The approach offers a path toward truly integrated systems that match the sensitivity, specificity, and energy conservation of biological smell detection. The modular nature of the Orco-OR platform, validated through computational modeling and synthetic datasets, provides a flexible foundation for targeting specific odor combinations.

The successful integration of biological components with brain-inspired hardware represents a huge step toward practical applications. These could include environmental monitoring systems that detect trace pollutants with unprecedented sensitivity, medical diagnostic devices that enable early disease detection through breath analysis, and security applications that identify explosive compounds or dangerous substances with high specificity.

Food safety systems could provide real-time quality assessment with minimal energy consumption. Industrial process monitoring could detect chemical leaks or contamination before they become serious problems. Agricultural applications could monitor crop health or detect plant diseases early.

Recent advances have validated successful integration of human smell receptors with organic brain-inspired devices, achieving highly accurate, molecule-specific odor discrimination. While these approaches use different receptor systems and electronics, they highlight the broader potential for exploiting biological smell specificity in engineered sensing applications.

This research opens new possibilities for chemical sensing applications across multiple domains. The development of these hybrid bio-electronic systems represents a convergence of biology, neuroscience, and engineering that could change how we monitor and interact with our chemical environment.

By maintaining the energy conservation and adaptability characteristic of biological systems while adding the reliability and scalability of electronic devices, this work points toward a new generation of intelligent chemical sensors. These could benefit everything from healthcare to environmental protection.

The unique integration of physical sensory neuron realization, processing, and learning mechanisms in a single device represents a significant milestone. This breakthrough brings us closer to practical bio-inspired smell perception systems that could soon transition from laboratory prototypes to real-world applications.

Most importantly, this fusion of biology and brain-inspired computing offers hope for solving some of humanity’s most pressing challenges. From early disease detection that could save countless lives to environmental monitoring that protects our planet, these intelligent sensors represent a new frontier where nature’s wisdom meets human ingenuity to create technologies that truly serve society.


Source: Max, K., Sames, L., Ye, S., Steinkuhler, J., & Corradi, F. (2025). Synthetic biology meets neuromorphic computing: towards a bio-inspired olfactory perception system. Neuromorphic Computing and Engineering, 5(3), 034010. DOI: 10.1088/2634-4386/aded2d

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.