Air quality monitoring in smart cities

Smart City Air Quality Monitoring

Architecting Resilient AI-Edge Systems with Non-Volatile Memory


Air pollution, particularly in the form of fine particulate matter (with a diameter below ≤2.5 µm, PM2.5), is associated with an increased risk of respiratory, cardiovascular, and neurological disorders. Smart cities require pervasive, real-time environmental intelligence to mitigate these risks. Sensor miniaturization has advanced to a point where accurate PM detection can be embedded into street fixtures, consumer devices, and wearable platforms. However, sensing alone is insufficient.


To transition from raw data acquisition to actionable insights, sensor systems must evolve into edge-intelligent nodes—capable of local inference, robust storage, and intermittent communication. Conventional memory technologies (e.g., Flash or DRAM) are ill-suited due to volatility, limited endurance, and high energy demands. This compels the use of non-volatile, high-endurance, and fast-access memories.

System Design Philosophy and Architecture

The proposed architecture is modular and comprises:

  • Photonic PM Sensor: Employs laser light scattering and photodiode reception to estimate particulate density.
  • Analog Signal Processing: Applies low-pass and frequency-domain filtering to condition the signal, represented mathematically as y(t) = x(t) * h(t), where h(t) is the impulse response of the system.
  • Edge-AI Processing: A microcontroller or SoC executes embedded machine learning models to detect trends, anomalies, and thresholds in near real-time.
  • Non-Volatile Memory (NVRAM): Acts as a high-speed, persistent data buffer for sensor logs, AI models, calibration data, and event history.
  • Wireless Interface: Communicates via BLE, Wi-Fi, LoRa, or 5G, depending on deployment topology.

This configuration ensures autonomous functioning, secure data preservation, and rapid inferencing even when disconnected from the network or power supply.

system architectureSystem architecture outline

Memory Architecture: MRAM and FeRAM Rationale

Traditional Flash memory suffers from slow write cycles, limited write endurance, and high power requirements. DRAM, while fast, is volatile and requires constant refresh cycles, making it impractical for IoT edge deployments.


MRAM utilizes magnetic tunnel junctions offering virtually infinite write endurance (>10¹⁵ cycles), non-volatility, fast access (~35 ns), and radiation resistance. These make MRAM ideal for real-time logs, model updates, and event storage.


FeRAM operates by switching the polarization of a ferroelectric capacitor, offering low write energy, fast access, and moderate endurance (10⁷–10⁹ cycles), ideal for periodic logs such as calibration or maintenance records.

MRAM architectureSTT-MRAM Architecture

AI at the Edge: Principles and Practicality


Embedded AI enables event-based sensing, data compression through local classification, and contextual behavior like adaptive sampling. TinyML models using TensorFlow Lite or Syntiant frameworks can be trained offline and deployed on neural accelerators or MCU-integrated AI engines with <1 mW power.


Working Principle of the Full Node

  1. Air is drawn or diffused into the sensing chamber.
  2. A laser illuminates the particulates; scattered light is detected by photodiodes.
  3. Analog signal is filtered, digitized, and sent to the processor.
  4. Edge AI model evaluates air quality contextually.
  5. Events and key states are stored in NVRAM—persisting through power loss.
  6. Data is transmitted periodically or upon AI-triggered events.

This architecture supports synchronous and asynchronous operation, enabling efficiency and resilience.


Critical-to-Quality (CTQ) Parameters


To ensure successful deployment and long-term operability, several Critical-to-Quality (CTQ) parameters must be established. These include:


  • Sensor Resolution: ≤ 1 µg/m³ to meet WHO and EPA air quality standards.
  • Edge Inference Latency: < 100 ms to allow real-time responsiveness.
  • Power Consumption: < 50 mW average to support energy harvesting or battery operation.
  • Memory Endurance: MRAM (>10¹⁵ cycles), FeRAM (>10⁹ cycles) to sustain constant writing.
  • Operational Temperature Range: –40°C to +85°C for outdoor and industrial environments.
  • Memory Retention: ≥ 10 years, to ensure compliance and historical data retention.
  • Physical Footprint: ≤ 1 cm², compatible with miniaturized platforms.
  • Secure Operation: Hardware encryption, secure boot, and protected memory.

Component-Level Design and Selection

Component selection must balance performance, energy efficiency, and integration flexibility.


Sensing:

  • Sensirion SPS30 / Plantower PMSA003: Highly integrated PM sensors with digital interfaces.
  • AMS CCS811 / SGP40: VOC sensors for hybrid air quality measurements.

Processing and AI:

  • Nordic nRF52840: BLE SoC for low-power control.
  • NXP RT1170 / Ambiq Apollo4 Blue: Advanced MCUs with embedded AI capabilities.
  • Syntiant NDP120 / Google Coral TPU: Hardware accelerators for TinyML applications.

NVRAM:

  • Ramxeed RFX-64K-SPI: FeRAM optimized for periodic low-energy logs.

Communication:

  • ESP32-S3: Dual-core processor with Wi-Fi/BLE support.
  • Semtech SX1262: LoRa module for long-range, low-power connectivity.
  • Quectel RG500Q / Thales Cinterion MV32: 5G modules for high-bandwidth deployments.

Power and Security:

  • TI TPS65218: Efficient multi-rail PMIC.
  • e-peas AEM10941: Solar and RF energy harvesting PMIC.
  • Microchip ATECC608A / NXP SE050: Security chips for cryptographic integrity.

Use Cases and Deployment Scenarios


Smart Cities:

  • Distributed nodes on street poles, traffic lights, and buildings for grid-wide pollution analytics.
  • Crowd-sourced exposure data feeding into municipal health dashboards.

Wearables:

  • Personal air quality trackers with BLE-enabled exposure logs.
  • Integration into smart helmets, badges, and fitness devices.

Industrial Monitoring:

  • Continuous PM and gas detection in factories, warehouses, and clean rooms.
  • Ensures air quality compliance and predictive maintenance.

Emergency Zones:

  • Wildfire, construction, or chemical spill detection with autonomous alerts.
  • Event-based storage and burst-mode 5G transmission upon re-connection.

Industry Support and Vendor Technologies


Netsol provides high-endurance STT-MRAM modules optimized for edge AI devices with industrial-grade retention, small form factor, and extended temperature support.


Ramxeed offers FeRAM-based ASICs and embedded memory platforms suitable for ultra-low-power applications such as wearables, medical devices, and specialized IoT tags. Their modules are often integrated into systems requiring microsecond-level write latency and minimal energy footprint.


Together, these vendors form the backbone of memory innovation within the AIoT ecosystem.


Conclusion


The convergence of optical sensing, embedded AI, and non-volatile memory forms a resilient, scalable platform for air quality monitoring. MRAM and FeRAM provide not only endurance and data safety but also empower intelligent inference under constrained energy budgets.


By designing architectures around these technologies, smart cities can deploy autonomous sensor grids that adapt, learn, and persist, ensuring long-term environmental intelligence


For Further Engagement:

Contact our technical sales team to request engineering samples, integration support, or design-in consultations.