I am currently experimenting with Bosch Sensortec sensors for personal ESP32 and embedded projects.

I am especially interested in:

• BME690

• BMI323

However, most distributors I found either:

• require industrial/bulk ordering

• or have very high shipping costs to Germany/Europe.

Are there recommended distributors, maker boards, sample options, or hobbyist-friendly ways to purchase these sensors in small quantities within Europe?

Thanks.

(I hope this is the right place to ask 😄 )

Context: we're integrating BME680 into a self-hosted AI platform (Eldric — https://eldric.ai) where an xLSTM-based forecaster (TiRex) consumes telemetry windows for anomaly detection and short-horizon prediction.

Before we commit to a feature set, I want to make sure we pick the BSEC outputs that actually carry predictive signal, not just the ones that look richest on paper.

A few concrete questions:

1. Which BSEC outputs survive the IAQ-accuracy=0 startup window cleanly? During cold start we still need stable features. Is raw gas resistance the more useful early signal, given that IAQ depends on the burn-in period?

2. Static IAQ vs ambient IAQ as ML features — has anyone compared them empirically? My intuition is that static IAQ generalises better across environments, but I'd like real-world experience.

3. Sampling-rate tradeoffs (ULP / LP / continuous) when the downstream model expects regular intervals. Do you resample BSEC output before feeding it to the model, or run BSEC at a fixed continuous rate and let the model see the raw cadence?

4. State persistence across power cycles — what's the cleanest pattern to keep BSEC state and the forecaster's hidden state aligned after a restart? Save both on the same cadence, or treat BSEC state as the source of truth and warm up the model from it?

If anyone has benchmarks or has published a BSEC-to-ML pipeline, I'd be glad for pointers. Happy to share findings back here once we've run the first set of experiments.

Setup:

• Application Board 3.1

• Shuttle Board: BME690 8x sensor board

• OS: Windows 10/11

• Software: BME AI-Studio Desktop (latest version)

• USB cable: USB A to micro-USB (USB 2.0 data + power cable)

Issue:

• The Application Board powers ON (red LEDs are ON)

• However, AI-Studio does not detect any device

• No board / sensor appears inside the software

• No communication or sensor data is shown

What drivers or firmware steps are required for first-time connection of Application Board 3.1 with AI-Studio?
Is COINES SDK or Development Kit software driver installation required before AI-Studio can detect the board?

Any guidance would be appreciated. Thank you!

I am currently using the BME690 with Shuttle Board 8.0, Application Board 3.1, and BME AI-Studio. I have trained an AI model in BME AI-Studio for gas-pattern classification.

Now I would like to deploy this trained model on my own MCU platform, for example ESP32 or STM32, so that the BME690 can run in the final embedded system and output the correct classification/prediction result.

Is there any official step-by-step tutorial, documentation, or example project that explains the complete workflow from:

BME AI-Studio trained model → deploy on ESP32 or STM32?

I am mainly looking for a practical guide showing how to move from the Bosch evaluation setup to a custom MCU-based embedded implementation.

Thank you very much for your support.

I recently ordered the Shuttleboard 3.0 and the Application Board 3.1, but I am unable to retrieve any data from the magnetometer or any other virtual sensor that depends on it (neither via COINES with the Sensor API nor via Development Desktop).

I was wondering if the issue could be that the installed magnetometer is a BMM350, while the firmware I am using is written for the BMM150. Some sources mention the BMM150, but the datasheet specifies a BMM350 magnetometer.

Could you please clarify which magnetometer is actually installed? If it is the BMM350, is there any firmware available for it? Alternatively, could there be another reason why it is not working at all?

Best regards

Questions

Type

Answers

• What is BME680

DS

BME680 is a metal-oxide based chemical sensor. The sensor output is created by interaction of volatile organic compounds (VOCs) with the sensitive layer of BME680. As a raw signal, BME680 will output resistance values and its changes due to varying VOC concentrations (the higher the concentration of reducing VOCs, the lower the resistance and vice versa).metal oxide-based sensors such as BME680 can distinguish between reducing (decreasing sensor resistance) and oxidizing (increasing sensor resistance) environments. If strongly oxidizing conditions are present (e.g. chlorine cleaner, nitrous oxides from exhaust gas, higher concentrations of ozone), this may lead to wrong readout of BME since it may output “clean” air in terms of the algorithm where there is none.

• What is BME688

DS

BME688 is the first gas sensor with Artificial Intelligence (AI) and integrated high-linearity and high- accuracy pressure, humidity and temperature sensors. It is housed in a robust yet compact 3.0 x 3.0 x 0.9 mm^3 package and especially developed for mobile & connected applications where size and low power consumption are critical requirements. The gas sensor can detect Volatile Organic Compounds, volatile sulfur compounds and other gases such as carbon monoxide and hydrogen in the part per billion range. 

• What is BME690

DS

The BME690 is the gas sensor with Artificial Intelligence (AI) and integrated high-linearity and high-accuracy pressure, humidity and temperature sensors. It is housed in a robust yet compact 3.0 x 3.0 x 0.9 mm³ package and especially developed for mobile & connected applications where size and low power consumption are critical requirements. The BME690 is more robust to previous products and can be used in high condensation applications. The gas sensor can detect Volatile Organic Compounds (VOCs), volatile sulfur compounds (VSCs) and other gases such as carbon monoxide and hydrogen.

• What’s the difference between BME680 and BME688

DS

Additionally to all features of BME680, the BME688 has a gas scanner function. In standard configuration, the presence of VSCs is being detected as indicator for e.g. bacteria growth. And the gas scanner can be customized with respect to sensitivity, selectivity, data rate and power consumption as well, The BME AI-Studio tool enables customers to train the BME688 gas scanner on their specific application, like in home applications, IoT products or Smart Home.

• What’s the difference between BME688 and BME690

DS

Same as BME688, the BME690 has a gas scanner function. In standard configuration, the presence of VSCs is being detected as indicator for e.g. bacteria growth. And the gas scanner can be customized with respect to sensitivity, selectivity, data rate and power consumption as well. The BME AI-Studio tool enables customers to train the BME690 gas scanner on their specific application, like in home appliances, IoT products or Smart Home.

• BME680, BME688 and BME690 have different impedances

DS

Low sensor (BME680) resistance value - selectable range: 178 ~ 12800000 Ohms

Low sensor (BME680) resistance value - operating range: 2000 ~ 2000000 Ohms

High sensor (BME688, BME690) resistance value - selectable range: 5865 ~ 102400000 Ohms

High sensor (BME688, BME690) resistance value - operating range: 100000 ~ 64000000 Ohms

• Operating temperature for full performance for BME sensors

DS

0 ~ 65 ℃

• Operating temperature ( data read-out possible) for BME sensors

DS

-40 ~ 85 ℃

• What's the difference between selectable range and operating range for BME sensors

DS

Selectable range represents all possible values, while operating range represents the values for actual work.

• Which gases were tested in HQ for BME sensors

DS

Here I listed some gases which were measured by HQ, includes min-max concentration.

• The measured range of raw features about BME sensors

DS

• For BME680:

Temperature: -40 ~ 85 deg C

Pressure: 30000 ~ 110000 Pa

Humidity: 0 ~ 100 %

Gas resistance: 170 ~ 13000000 Ohm (hardware: 178 ~ 12.8e6 Ohm)

• For BME688/BME690:

The T/P/H are the same as BME680

Gas resistance: 5600 ~ 103000000 Ohm (hardware: 5865 ~ 102.4e6 Ohm)

• Others Features are same:

IAQ: 0 ~ 500

Static IAQ: 0 ~ Max

CO2: 400 ~ Max ppm

bVOC: 0 ~ 1000 ppm (will be removed from BSEC 3.x)

• How to evaluate bVoc/CO2 with IAQ?

DS

It’s recommended to buy some professional devices as references which can measure bVOC and CO2.

For CO2: Sometimes you will find there is a little difference between the reference device and BME output, as BME gives an equivalent CO2 output which is calculated based on the S-IAQ, which is not very accurate as the professional devices, but do not say it to the customer directly, we can change some words.

For bVoc: We can evaluate IAQ by bVOC, here I listed the two charts:

The first screenshot is for BSEC version 1480 and before versions, which regard IAQ-25 as ‘clean air’. (bVOC = 0.5ppm when the S-IAQ=25, bVOC = 15ppm when the S-IAQ=250)

The second screenshot is for after BSEC 1480 versions, which regards IAQ-50 as ‘clean air’. (bVOC = 0.5ppm when the S-IAQ=50, bVOC = 15ppm when the S-IAQ=250)

Usually bVoc can not indicate the real voc concentration in the environment, what showed it's just a reference to the customer, they can also adjust the relationship between bVoc and IAQ according to their environment.

In BSEC 3.x version, TVOC equivalent feature will instead of bVoc for BME690.

• What’s the means by ‘burn-in’ and ‘run-in’? and How long time they will take?

DS

‘burn-in’ status means when you get a new sensor, which is a status in the first power-on status, so a sensor only has one ‘burn-in’ status in its life. And in software, we set 0, but it’s recommended to run the sensor for 24h on hardware. Make sure to give the sensor enough time to stabilize.

‘run-in’ status means a status in each power-on status. In software, we set many time about this status which depends on the power mode, here I listed the relation.

The fixed 'run-in' time only existed before BSEC 2.5.0.2 version, after that we have run in profiles to speed up the period for BME688 and BME690, why we do that is fixed time will cause some risks about s-t-s deviation.

However, 'run-in' time is still influenced by the last power-off time, here I list an example of 30mins power-off time for BME690 based on BSEC 3.2.1.0:
LP: ~2mins   ULP: ~10mins

• How can I distinguish between Operated Mode and Power Mode?

DS

Operated Mode only includes Forced Mode and Parallel Mode, heating is always on in Parallel Mode.

Power Mode includes LP, ULP, CONT, SCAN(HIGH PERFORMANCE), which determines the frequency at which the sensor outputs the data, so it will also influence the power consumption.

What’s more, the sensor works in Forced Mode when you set the Power Mode LP, ULP, CONT, and works in Parallel Mode when you set the Power Mode SCAN.

• Why does my sensor continue to show saturation values during operation?

SW

Check which step the sensor at, and check if the temperature and duration of this step are enough or not, it’s recommended to change another Heater Profile to test it again.

Some old sensors will show the saturation value when the heating temperature is below 150℃.

• Some customer are using the platforms which aren’t be involved in our generic released package, how can we support them.

SW

Contact with AE, and provide the compiler and compilation options.

• IAQ is abnormal when the customer uses DD2.0 with BME688. (Delete)

SW

09/07/2023: Currently DD2.0 can only support BME680 to measure IAQ, which is not suitable for BME688, that means he needs to use EKB to measure IAQ when he wants to use BME688.

• Temperature and Humidity are incorrect in BSEC mode in DD 2.0

SW

The DD tool has a compensation factor hardcoded, this was done as it is persisting since many years ago.

This is resulting in the BSEC performing a subtraction from the raw temperature/humidity value and providing a compensated temperature/humidity in BSEC mode.

• Can I set the target temperature and duration? What’re the maximum values

SW

Yes, the customer can try their own better heating profiles.

Maximum value of the target temperature is 400 ℃, and maximum value of the duration is 4032ms.

• Can I get IAQ without BSEC library?

SW

No, you can only get raw data of T/P/H/G without BSEC library, you need to use BSEC library if you want to output IAQ, some compensated values and gas estimation feature.

• What’s the default temperature for ULP/LP/CONT in the library?

SW

• No setting config file:

BME68X/ULP- Temperature: 400℃, Duration-heater-on: 1943 ms;         BME690/ULP- Temperature: 400℃, Duration-heater-on: 1000 ms;

BME68X/LP- Temperature: 320℃, Duration-heater-on: 197 ms;              BME690/LP- Temperature: 320℃, Duration-heater-on: 100 ms;

BME68X/CONT- Temperature: 320℃, Duration-heater-on: 900 ms;        BME690/CONT- Temperature: 320℃, Duration-heater-on: 900 ms;

• Setting default config file:

BME68X/ULP- Temperature: 400℃, Duration-heater-on: 1943 ms;         BME690/ULP- Temperature: 400℃, Duration-heater-on: 1000 ms;

BME68X/LP- Temperature: 320℃, Duration-heater-on: 197 ms;              BME690/LP- Temperature: 320℃, Duration-heater-on: 100 ms;

BME68X/CONT- Temperature: 320℃, Duration-heater-on: 900 ms;        BME690/CONT- Temperature: 320℃, Duration-heater-on: 900 ms;

• Does BSEC2 support BME680?

SW

Yes, you can change the subscriptions according to the datasheet.

• Does BSEC3 support BME680, BME688 and BME690?

SW

BSEC 3.x only support both BME688 and BME690, but tvoc equivalent feature only be supported by BME690 in LP Mode, BSEC 3.x can not support BME680.

• The sensor will show bad performance in low oxygen environment.

HW

BME series MOX sensors exhibit abnormal behavior in low-oxygen environments because their operation relies on oxygen molecules in the air forming adsorbed oxygen ions on the sensor surface, which react with target gases in redox reactions to produce changes in resistance. In low-oxygen conditions, the surface oxygen ions are insufficient, limiting the reaction and resulting in reduced sensitivity, delayed response, and even baseline drift.

• What's the means by .bmeconfig, .config and .aiconfig files in AI_Studio, and how can I use it.

SW

The. bmeconfig file is a file that users need to copy to the SD card when collecting data (the board configuration used by EKB), which is the file exported after configuring the BME board using AI_Studio

The specific content is as follows: (Configure the HP and DC of each sensor on the board)

The. config and. aiconfig files are automatically generated after collecting data

. aiconfig means that when users need to connect an EKB board to Mobile AI_Studio, they need to copy it into the SD card

The things saved in the. config file are similar to the generated. c file array, which can be copied to the SD card along with. aiconfig when using Mobile AI_Studio

• New Run-in conditioning for LP & ULP mode.

DS

• BME690 in BSEC 3.2.1.0

• BME688 in BSEC 2.6.1.0 and BSEC 3.2.1.0

• We have handling instructions for BME690.

Are the handling and contamination‑control instructions for BME688 the same or different? Please highlight differences.

HS

With respect to contamination susceptibility and handling risks, BME690 and BME688 are equivalent.

• What are the potential effects if a BME-series MOX sensor is touched by a human hand during assembly, and are they reversible?

HW

Finger Touch & Contamination
Two effects:
Short‑term, reversible: abnormal readings, response‑time/ noise changes due to residue from fingerprints. These typically recover with time or mild, proper handling.
Long‑term, irreversible: chemical poisoning or irreversible adsorption on the sensing layer, causing permanent performance drift.
Rule of thumb: prevention over cleaning. Avoid direct touch and keep solvents away from the gas inlet and sensing area whenever possible.

• Is post-assembly cleaning allowed?

• Which solvents and methods are approved?

• How should minor fingerprints be handled?

• Any drying or outgassing guidance?

HW

Post-assembly cleaning is allowed only if unavoidable.

Use ≥90% isopropyl alcohol (IPA) applied lightly with a lint-free wipe or swab. Wipe only the metal cap and its perimeter. Never soak or spray the device, and avoid any liquid entering the gas port.

For minor fingerprints, prefer passive recovery or clean, dry compressed air instead of using any liquid.

Ensure the device is completely dry before use to prevent solvent residue.

• If cleaned with alcohol, does outgassing affect readings for 24–48 h?

• What quarantining time do you recommend before end‑user use?

• Our lead time is 5–7 days post‑assembly; is this sufficient?

HW 

In low‑VOC conditions, BME688 response time can improve from ~30 s to ~15 s; however, heavy alcohol vapors can prolong recovery and destabilize readings.

Considering the sensor's ~30 min warm‑up and typical solvent off‑gassing behavior, isolate the device for 8–24 hours in well‑ventilated air after cleaning.

A 5–7‑day buffer between assembly and shipment normally satisfies this requirement if ventilation is ensured.

• Which assembly materials are compatible (flux, adhesives, conformal coating)?

• Are there any “do‑not‑use” chemistries?

• Any additional precautions for MOX sensors?

HW

Use neutral or mildly alkaline fluxes, non-acidic adhesives, and VOC‑low conformal coatings that do not release corrosive byproducts. Avoid materials that can outgas acids, amines, or strong solvents near the sensor.

Avoid acetic-acid–releasing materials (e.g., acetoxy-cure silicones), strong amines, and other corrosive or reactive compounds. Acetic acid and similar chemicals can corrode or contaminate MOX sensors, leading to degraded performance or permanent damage.

Always apply adhesives, fluxes, or coatings away from the sensor’s gas port. Ensure proper curing and outgassing before operation. Avoid direct contact or overspray on sensitive surfaces.

• What is the ESD handling class for BME688?

HW

BME688 has an ESD robustness of ±2 kV HBM. Proper ESD precautions should be followed when handling the sensor.

• What is the Moisture Sensitivity Level (MSL) of BME688?

HW

BME688 is MSL 1, meaning it is not moisture-sensitive under standard storage conditions.

• How should BME688 be stored?

HW

Long-term storage is acceptable at ≤30 °C / 85% RH. Prefer a dry, clean, non-corrosive atmosphere. Avoid direct sunlight, high temperature, and high humidity.

• Is baking required before use?

HW

MSL1 parts generally do not require baking. If exposed to high humidity or improper storage, a 125 °C × 24 h bake is a common practice. Always align final bake conditions with the manufacturer’s guidance.

• Is there a preferred sensor orientation on the PCB?

HW

Orient the sensor pins toward the PCB edge or an accessible area for routing and soldering. Ensure the top surface is exposed to ambient air and not shadowed by other components or traces.

• Are there keep-out areas around the sensor?

HW

Keep the sensor away from strong heat sources (power resistors, high-power ICs) and EMI sources (inductors, transformers).

• How should vent holes be designed?

HW

Provide vent holes beneath or around the sensor with a diameter ≥0.5 mm to allow free air exchange.

• Are there guidelines for airflow channels?

HW

Keep ducts as straight as possible. If bends are required, use a radius ≥2× duct height. Add flow guides if needed to distribute airflow uniformly across the sensing area.

• What is the purpose of burn-in for BME688?

HW

The primary purpose of burn-in is to stabilize sensor performance and improve consistency between sensors.

• What is the semiconductor/chemistry rationale behind burn-in?

HW

The MOX element is exposed to high temperature during burn-in to:

1. Stabilize the microstructure of the metal oxide layer.

2. Stabilize surface states, ensuring consistent adsorption/desorption behavior.

3. Remove impurities and residues from fabrication or handling.

These steps reduce sensor drift, improve reproducibility, and ensure predictable response characteristics.

• Can burn-in be performed at the factory or at the customer site?

HW

Burn-in can be performed in either location.
Factory burn-in: Controlled environment allows consistent temperature, clean air, and absence of interfering gases, ensuring uniform sensor stabilization. Risk is minimal if proper procedures are followed.

Customer-site burn-in: Environment may contain interfering gases (VOC, smoke, chemicals) that can alter sensor surface chemistry, cause drift, or reduce reproducibility. Maintaining consistent temperature and airflow is also more challenging.

• What precautions are required during burn-in?

HW

Ensure the sensor is exposed to clean, interference-free air. Avoid gases or vapors that can adsorb on the MOX surface. Maintain proper temperature and airflow control throughout the burn-in period.

• How “clean” must the environment be during burn-in?

HW

“Clean” means the air should not contain reactive or interfering gases, including tVOCs, silicides, alcohols, carbon monoxide, or other substances that may adsorb on or react with the sensor surface (even unusual compounds such as durian VOCs). Maintain normal indoor temperature and humidity with stable, moderate airflow. The sensor should not be sealed in a small box, as this would prevent proper heat dissipation.

• How can cleanliness be quantified or monitored?

HW

While no absolute threshold exists, common practices include:

• Measuring TVOC levels with a calibrated sensor; keep as low as practical.

• Monitoring temperature and relative humidity for stability.

• Ensuring consistent airflow without stagnant zones.
  These measures help prevent adsorption of interfering gases and maintain reliable burn-in results.

• What heater temperature/time profiles minimize burn‑in time while ensuring stability?

SW

It is necessary to operate the sensor in the same mode.
LP mode(IAQ ±15%±15): up to 7 days; qualitative 8 hours;
ULP mode(IAQ ±15%±15) up to 15 days; qualitative less than 1day;

• Is burn‑in required only once for the product life, or after long storage/shock events?

HW

Burn‑in/Re‑conditioning
On first use, perform ≥8 hours of burn‑in.
After prolonged storage or mechanical shock, perform re‑conditioning.
Before routine operation, a ~10 min pre‑burn can help stabilize readings.

• How do we know burn‑in is complete (e.g., ΔR/R stabilization, BSEC state convergence)?

HW

It is not always easy to determine when the stabilization is finished from field data, since usually no reproducible pulses occur in real-world condations; thus, aged BME690 sensors were used as reference and completion of stabilization considered to be the time when new and aged sensors behave similar(e.g. ±15% range of IAQ)

• What unit‑to‑unit resistance variation (pre/post burn‑in) is normal?

HW 

Resistance Distribution (Indicative)
Resistance depends on temperature, humidity, and gas composition. After burn‑in, values stabilize typically in the mega‑ohm to tens‑of‑mega‑ohm range.
Under stable lab conditions, a narrower spread such as 20–40 MΩ may be observed. Treat these as guidelines, not hard limits.

• Recommended humidity/temperature and exposure controls during burn‑in to avoid biasing the baseline.

HW

Recommended Environment for Burn‑in
Humidity: while the measurable range is 0–100% RH, maintain 20–80% RH during burn‑in for stability.
Temperature: operating range is −40 to 85 °C; for burn‑in, target ~25 °C to avoid irreversible shifts.

• Should we use BSEC “gas‑scanner/AI” vs custom heater control? Which BSEC version/configs fit H₂S/NH₃?

HW

There is no internal recommended gas configuration. Based on internal experience, HP-354 is more suitable for detecting H2S. AI_Studio supports 16 HP configuration files by default. Customers can select specific profiles based on actual performance or customize their own configuration files.

• What are the achievable detection limits (LOD), linear/usable ranges for H₂S/NH₃ under the recommended profiles?

HW 

0.1 ppm ~ 50 ppm NH3
100 ppb ~ 500 ppb H2S
What we measured in the lab, but it's not the limitation

• How do typical home interferents affect H₂S/NH₃ readings?

• Is there guidance for compensating for these interferents?

HW

Pet excreta, cleaning agents, and food residues may emit sulfur‑ or ammonia‑bearing compounds or VOCs, which can directly skew H₂S/NH₃ and total‑VOC readings.

Be aware that readings may be biased in the presence of the above interferents. Possible strategies include:

• Sensor placement: locate sensors away from direct sources (litter boxes, cleaning areas, kitchens).

• Data filtering: implement software routines to detect transient spikes associated with known interferents.

• Calibration updates: periodic recalibration may help reduce long-term drift caused by persistent interfering gases.

• What is the recommended compensation scheme for stable gas readings?

HW

A three-layer approach is recommended:

1. Environmental isolation

   • Physically shield the sensor from rapid changes in airflow, temperature, or humidity.

2. Real-time temperature/humidity correction

   • Apply real-time compensation using sensor readings of temperature and humidity to correct gas measurement output.

3. Dynamic baseline calibration

   • Combine hardware pre-conditioning and algorithmic optimization to adapt the baseline over time, compensating for drift or environmental effects.

This layered approach helps stabilize MOX sensor readings under varying temperature and humidity conditions.

• How can we diagnose a clogged pinhole in dusty environments?

HW

Key telemetry indicators include:

1. Response time

   • Slower than a healthy reference by >30% under the same environmental conditions.

2. Noise level

   • Increased to ≥2× normal.

3. Baseline drift

   • Short-term drift exceeding 15% of the post-calibration baseline, with irregular fluctuations.

• How should thresholds be used for maintenance decisions?

HW

Evaluate response time, noise, baseline drift, and sensitivity decay collectively. When one or more metrics exceed thresholds, maintenance is required. After cleaning or replacement, re-measure these metrics to confirm that performance has returned to acceptable levels.

• How do different membranes (PTFE/ePTFE, pore size, thickness, oleophobic) affect sensitivity and response time for H₂S/NH₃?

HW 

Protective membranes generally have a relatively low impact on MOX sensor performance, including response time and recovery time. Minor differences may exist due to thickness or pore structure, but typical PTFE/ePTFE membranes allow gas diffusion sufficient for accurate readings.

• Are there recommended parts or vendors?

HW

Membranes should be chemically inert, mechanically stable, and compatible with H₂S/NH₃ gases. While no specific vendors are required, commercially available PTFE/ePTFE membranes from reputable suppliers are commonly used. Selection should prioritize thickness, pore size, and oleophobic treatment based on application constraints.

• What behaviors should be expected immediately after minor finger contact?

HW 

Short-term contact basically does not affect baseline drift, and may only temporarily influence the current reading.

• How should the sensor be restored or cleaned safely?

HW

• Passive recovery: Power off and place in clean air at 30–50% RH for approximately 2 hours.

• Dry compressed air: Use ≤0.1 MPa, stand 15–20 cm from the gas port at an angle, 3-second blast ×2 with 5-second interval (total ≤6 s).

• Do NOT use liquids on the gas port. Do not touch the port with swabs or tissue, to avoid fiber shedding.

• Which heat profile is better?

SW

Currently, there is no gas-specific heat profile. AI Studio provides 16 preset heat profiles (HPs). Each data collection and training session supports up to 4 HPs. After training, users can check the prediction accuracy for each HP, and the one with the highest accuracy is considered the best-performing heat profile.

• What is the expected lifetime of BME688 ?

• How do accelerated stress tests relate to field lifetime?
  

HW 

Lifetime is estimated based on accelerated storage and stress tests following JEDEC standards.

Accelerated stress tests such as High-Temperature Operating Life (HTOL) and Temperature Humidity Bias (THB) simulate long-term aging. If BME688 withstands >1,000 hours under these conditions, it is considered equivalent to ~10 years of normal field operation.

• Is there any way to determine the degree of deterioration of the gas sensor?

• How to track deterioration over time?

HW 

Signs of sensor deterioration include:

1. Decrease in GasRaw value

   • The baseline GasRaw reading gradually decreases compared to a healthy sensor.

2. Slower or reduced IAQ response

   • IAQ and other derived gas indices rise more slowly and less sharply in response to the same gas concentration.

3. Longer recovery time

   • The time between the rise and fall of IAQ becomes longer, indicating slower sensor kinetics.

 Maintain time-stamped records of GasRaw, IAQ, and other key sensor outputs under controlled conditions. Analyze trends in baseline, response amplitude, and recovery time to quantify degradation.

• How can AI Studio differentiate between different gases and their concentrations?

• What machine learning principle is applied?

SW 

The differentiation relies on unique patterns in the sensor’s gas resistance:

• Gas-induced resistance changes

  • Each gas or concentration produces distinct changes in BME688’s gas resistance, which may not be visible to the naked eye.

• Supervised learning with ground truth

  • Users provide reference labels (ground truth) corresponding to the gas type and concentration.

• Model training and inference 

  • The AI model maps resistance patterns to the reference labels during training.

  • During testing, if a similar pattern occurs, the model outputs the predicted gas type and/or concentration.

This is a supervised machine learning approach, where the model learns to associate sensor signal patterns with labeled gas types/concentrations.

• For a heat profile with 10 steps per cycle, does AI Studio check the resistance difference between gases (e.g., FA and Ethanol) at each step separately?

• How does this relate to neural network operation?

SW

Not directly. AI Studio uses preprocessing followed by a neural network to analyze the data. The neural network can extract any relevant information from the 10 heater steps, including:

• Differences between gases

• Mean, variance, or other statistical features

• Any patterns that map gas resistance to the reference label

It does not require step-by-step manual comparison; the network automatically learns the features needed for classification or regression.

Neural networks can learn complex mappings between input signals (heater step resistance) and outputs (gas type/concentration) from labeled data. You can refer to standard resources on how neural networks process multivariate time-series inputs for further understanding.

• How does AI Studio use variance in each heat profile step to differentiate gases and concentrations?

• How should rapid transitions between gases be handled?

• Can slow transitions be used for training?

SW

The term “variance per step” usually refers to the signal fluctuations during a heat step. To distinguish different gases or ppm levels, there must be a measurable change in gas resistance.
If the gas changes quickly (fast transition in resistance), it is recommended not to use this period for training. The sensor may not detect the new gas immediately, but the model will detect it after the sensor stabilizes. This approach improves overall model performance.
Yes, if the resistance change from one gas to another occurs slowly, this period can be included for training. The model may detect gas changes more quickly, but it may also produce occasional false outputs after stabilization. This trades overall stability for faster response.

• What is the minimal quantity of sensors recommended for gas training?

• Any additional recommendations for sensor usage?

SW

Ideally, use as many sensors as possible within budget. With Bosch development kit boards, 2 boards provide 16 sensors, which is the minimum recommended number. If performance is insufficient, increase the number of sensors.

• Use brand-new sensors that have completed burn-in.

• Enable data augmentation in BME AI Studio to improve model robustness.

• Can burn-in time be decreased by higher temperature or shorter duration?

SW

his is an open topic that has not yet been explored or verified for BME688. In general, higher temperature and longer heater-on time are conventionally used to reduce burn-in for MOX sensors, but no validated data exists for BME688 yet.

• ISO16000-29 'Test methods for VOC detectors'  document link

DS

EN_ISO_16000-29_2014-06.pdf.

• What is manual calibration for TVOC equivalent in BSEC 3.2.1.0?

SW

In BSEC 3.2.1.0, the TVOC equivalent is by default calibrated to lab data without considering interfering gases. In real-world environments, other gases may affect readings, so the manual calibration API was provided. Users could:

1. Enable the API.

2. Calibrate TVOC to a reference concentration (e.g., 250 ppb) in the expected environment.

3. Disable the API to prevent deviation from the calibrated value.

Since TVOC (manual calibration) and IAQ (automatic calibration) are calibrated separately, differences between reported values could occur.

In BSEC 3.3.0, TVOC calibration was changed to automatic calibration, removing the need for a separate manual calibration step. This aligns TVOC calibration more closely with IAQ and reduces discrepancies between the two outputs.

• Do we have any docs for quick calibration and slow calibration in TVOC feature in this release, and how long it will take to get the reliable output and the lowest s-t-s deviation output. 

SW

The recommendation to customer is 22 days in field. By this time, TVOC would be calibrated and S2S would be least. We are not giving specific instructions for quick calibration as it is for development purpose.

• Conditions that trigger a BSEC -2 error in the BSEC built-in algorithm

SW

The BSEC -2 error is triggered when one or more BME sensor input parameters exceed the allowable input limits specified by the BSEC algorithm.

• (gas): High: 170.0 ~ 1.03E+8  Low: 170.0 ~ 1.32E+7

• (temperature): -65.0 ~ 125.0

• (humidity): 0.0 ~ 100.0

• (pressure): 0.0 ~ 2.0E+6

• Can we use the current PCB designed for the BME680/BME688 and upgrade it with the BME690 without any hardware changes?

HW

Yes, the BME690 is designed to be backward compatible with existing PCB layouts developed for the BME680 and BME688. In most cases, the BME690 can be mounted directly onto the existing hardware without requiring PCB modifications.

However, it is recommended to update to the latest software API version to ensure full support of the BME690 features and optimized signal processing. The updated API automatically detects the connected sensor variant and applies the appropriate configuration and compensation algorithms.

• Is it necessary to actively force air into the BME sensors when taking measurements?

DS

No, active airflow is generally not required. Gas exchange occurs naturally through diffusion. Whenever the gas composition around the BME sensor changes, a concentration gradient is created, causing gases to diffuse into and out of the sensor package.

Due to the compact sensor design and the very short distance between the package opening and the gas sensing element, the sensor responds to environmental gas changes within a few seconds under typical operating conditions.

• How many measurements can be stored on the BME sensors?

DS

The BME sensor stores configuration information such as scan profiles, along with a limited amount of temporary measurement data in its internal buffer. In normal operation, the measurement data is continuously read out by the host microcontroller (MCU) connected to the sensor.

As a result, the practical storage capacity mainly depends on the available memory of the host system rather than the sensor itself. When using the Bosch sensor software library on the MCU, sensor data can be processed directly on the device, reducing the need to store large amounts of raw measurement data. In many applications, storing only the relevant processed output values is sufficient.

• How much computing power is required to run the AI software? Can you provide examples of compatible MCUs?

SW

Training and optimizing AI models using Bosch AI-Studio requires a desktop or laptop computer with sufficient processing resources for data analysis and algorithm generation.

However, executing a trained algorithm on an embedded device requires only modest computing resources. Many commonly used microcontrollers are capable of running the sensor processing and AI inference in real time. Examples include the ESP8266, ESP32, and a wide range of ARM Cortex-M based MCUs.

The exact hardware requirements depend on the complexity of the application, sampling configuration, and system-level software architecture.

• Can you please explain how the gas scanning functionality works?

SW

The BME gas sensor performs measurements under multiple operating conditions during a gas scan sequence. This is achieved by dynamically controlling internal sensor parameters, such as the heater temperature profile, which changes the sensor sensitivity to different gases and volatile compounds over time.

By combining the measurement responses from these different operating points, the sensor can generate characteristic signal patterns — often referred to as “fingerprints” — for specific gas mixtures or environmental conditions.

Using Bosch AI-Studio, developers can configure, optimize, and train custom scan profiles for their specific application requirements. This allows the system to improve selectivity and adapt the sensing behavior for use cases such as air quality monitoring, odor classification, or gas detection.

• Does the BME sensor include integrated Flash or EEPROM memory?

SW

No. The BME sensor does not contain integrated Flash or EEPROM memory for long-term user data storage. Configuration data and measurement results are typically managed by the external host microcontroller (MCU) connected to the sensor.

• After training an algorithm using data collected from multiple sensors, can the resulting model be deployed on a device using only a single BME sensor?

SW

Yes. Bosch AI-Studio can export the trained algorithm as a configuration file or configuration string that can be integrated into the Bosch sensor software running on a target device with a single BME690 sensor.

Once deployed, the device can locally execute the trained model and directly evaluate the gas scan data generated by the BME690 in real time. This enables compact and cost-efficient implementations without requiring multiple sensors during normal operation.

However, the achievable performance and robustness of the trained model may depend on factors such as sensor variation, environmental conditions, and the quality and diversity of the training data.

• Which AI or neural network topology is used? Is the system based on neural networks or statistical pattern recognition?

SW

The current version of Bosch AI-Studio uses a neural network–based approach for gas pattern recognition. The software utilizes a predefined neural network architecture together with configurable training parameters and optimization methods, including the ADAM optimizer, for model training.

The generated models are designed to operate efficiently on embedded systems with limited computational resources while still enabling reliable classification of gas patterns and environmental conditions.

In addition, we explicitly encourages customers to develop and implement their own algorithms based on their specific application requirements. AI-Studio is designed as a flexible development tool that supports customization of models and workflows to enable application-specific optimization.

• Is it possible to detect gases produced by burning network cables or electronic components?

DS

Yes, this can be a relevant application for the BME690. In scenarios where abnormal conditions occur in electrical systems or cabinets, gas emissions can often be an early indicator of faults or overheating.

Typically, two main situations can be distinguished:

1. Overheating or material degradation
   When insulation materials or other components are exposed to excessive heat or short circuits, they may begin to decompose or melt. This process leads to increased outgassing, often containing a mixture of volatile organic compounds (VOCs), including unburned hydrocarbons. These gas signatures can be detected by the BME sensor in a manner similar to how humans perceive odor.

2. Electrical arcing or flashovers
   In cases of high voltage events, short circuits, or arcing, ozone (O₃) and other reactive gases can be generated. These have a distinct gas signature compared to thermal decomposition products and can also be detected by the BME sensor.

By analyzing these different gas patterns using appropriate algorithms, the sensor can potentially contribute to early detection of electrical faults and abnormal operating conditions.

• Does the BME sensor detect combustible gases such as methane (CH₄) and propane (C₃H₈), as well as toxic gases like carbon monoxide (CO)?

DS

The BME sensor is sensitive to a broad range of gases, including many volatile organic compounds (VOCs) and various combustible and reducing gases. This includes many hydrocarbons (CₓHᵧ) and gases such as carbon monoxide (CO), which can be reflected in the overall gas pattern measured by the sensor.

Combustible gases are often categorized into methane (CH₄) and non-methane organic gases (NMOG). Methane is a special case, as it is chemically less reactive and typically requires specific sensing materials or catalysts for strong direct response. As a result, the BME sensor is not optimized for strong direct sensitivity to methane, and detection performance for methane alone may be limited, especially at lower concentrations.

However, in many real-world applications, methane is not present in isolation but occurs alongside other gases such as VOCs or sulfur-containing compounds, which the BME sensor can detect effectively. In such scenarios, gas mixture patterns can still provide useful information for classification and anomaly detection when combined with appropriate algorithms.

For application-specific evaluation, testing the sensor under real operating conditions is recommended to determine suitability and achievable performance.

• Is the BME sensor certified as a medical sensor?

DS

No. The BME sensor is not certified as a medical sensor at the component level. Sensor components are typically qualified according to standards applicable to consumer electronics, such as JEDEC requirements, rather than medical device regulations.

Medical certification, where required, is carried out at the device or system level by the manufacturer of the final product, as certification depends on the complete system design, intended use, and regulatory classification.

In general, customers are responsible for evaluating and ensuring compliance of their end products within the relevant application domain and regulatory framework. From a system perspective, the benefit of pre-certified individual sensor components is therefore limited in most cases.

• Is it possible to detect drugs or explosive materials using the BME sensor?

DS

Bosch Sensortec does not provide validated use cases or application experience for the detection of drugs or explosive materials with the BME sensor. The sensor is designed for general gas sensing applications and the detection of volatile gas patterns in air quality and environmental monitoring contexts.

In principle, the BME sensor can measure changes in gas composition, and its response to different volatile compounds can be evaluated in customer-specific development and testing environments. However, Bosch does not provide performance specifications, calibration data, or application guidance for drug or explosive detection use cases.

Customers may conduct their own evaluation using development hardware to determine suitability for their specific application, subject to applicable laws and regulations.

• How can I determine whether the gases in my application can be measured with the BME sensor?

DS

A key purpose of Bosch AI-Studio is to enable exactly this type of evaluation in real-world conditions. Instead of requiring detailed prior knowledge about individual target gases or their concentrations, developers can directly test the sensor in their actual application environment.

Traditionally, gas sensing development required extensive laboratory testing with individual gases in controlled environments, such as synthetic air. While such tests can provide useful baseline information, they often do not fully represent real-world conditions, where multiple gases and environmental influences are present simultaneously.

With the BME sensors and Bosch AI-Studio, developers can collect data directly in the target application, build and evaluate models under real operating conditions, and iteratively optimize performance based on real-world gas mixtures and dynamics. This approach helps bridge the gap between laboratory characterization and practical field performance.

• Are over-the-air (OTA) updates possible to enable detection of additional gases or substances in the future?

SW

Yes. The system supports updating sensor behavior through software-based configuration. The sensor’s measurement behavior is defined by software parameters rather than being permanently fixed in hardware.

When using Bosch AI-Studio, a trained application results in a configuration package (including model parameters and scan profile settings). This package can be deployed to field devices via over-the-air update mechanisms, as it is typically small in size and designed for efficient transfer.

Once the updated configuration is loaded into the Bosch sensor software on the device, the BME sensor operates according to the new scan profile and model behavior, enabling adaptation to new application requirements without requiring hardware changes.

• Why does the evaluation board use eight sensors?

HW

The development kit is designed to support flexible evaluation and optimization using Bosch AI-Studio. It allows developers to fine-tune key parameters such as performance, output data rate (ODR), and power consumption according to application-specific requirements.

By integrating eight BME sensors on a single evaluation board, the system enables parallel testing of multiple configurations under identical environmental conditions. This allows developers to compare different settings simultaneously, improving data quality and statistical reliability while significantly reducing overall development time.