Analogue, Mixed Signal, LSI


IMUs that carry machine learning burden

29 April 2020 Analogue, Mixed Signal, LSI

Designed for use in a wide variety of consumer and industrial applications, STMicroelectronics’ 6-axis iNEMO inertial measurements units (IMU) feature an embedded machine learning core (MLC) to offload the burden from the host processor. Machine learning is an application of artificial intelligence (AI) through which a machine can learn, by itself or in a supervised way, without explicit programming. It provides a system the ability to automatically learn and improve from experience without compromising the accuracy of the data collected.

The machine learning processing capability allows for moving some algorithms from the host processor to the IMU (inertial measurement unit). The IMU would therefore only consume less than one hundredth of the MCU (microcontroller) power used for the same typical tasks. The MLC is designed to run in a highly power-efficient manner and provides accurate results in the shortest possible time. A meta-classifier is also available to further enhance data accuracy in specific cases.

Developers of applications using sensors can thus benefit from the advantages of machine learning by creating their decision trees (using large data sets and high processing power) and having them run on an optimised MLC in the same sensor device.

Decision trees can be created and updated much faster using machine learning compared to explicit programming when appropriate data sets are available. The machine learning processing capability allows moving some algorithms from the host processor to the IMU. For typical tasks the IMU would consume 0,001 times the power of an MCU used for the same task. This is a key enabler for ultra low-power edge-computing. iNEMO IMUs with MLC can be configured to run up to eight decision trees simultaneously and independently, giving added flexibility to developers.

The LSM6DSOX contains a 3-axis accelerometer and 3-axis gyroscope and tracks complex movements using the machine learning core at low typical current consumption of just 0,55 mA to minimise load on the battery.

The consumer-grade LSM6DSRX contains a 3-axis accelerometer and a 3-axis gyroscope with extended full-scale angular rate range up to +/-4000 dps.

The industrial-grade ISM330DHCX comes with 10-year product longevity assurance and is specified from -40°C to 105°C, with embedded temperature compensation for superior stability.

Intelligent sensing with iNEMO inertial measurement units

iNEMO IMUs are built around three blocks, as shown in Figure 1.


Figure 1. iNEMO IMUs are built around three blocks.

The built-in sensors (accelerometer and gyroscope) filter real-time motion data before sending it to the computation block where statistical parameters defined as ‘features’ are applied to the captured data. The features aggregated in the computation block will be used as input for the third block of the machine learning core.

The decision tree evaluates the statistical parameters and compares them against certain thresholds to identify certain situations and generate results sent to the MCU. The machine learning core results are the decision tree’s output results which include the optional meta-classifier.

Building your decision trees,

iNEMO IMUs offer a wide range of design possibilities for developers by allowing them to create their own embedded machine learning algorithms and to build the best decision tree for their application.

Figure 2. iNEMO-based decision tree.

1. Collect data

The first step is to collect a representative set of data for the motion-related application being modelled. Examples of physical parameters include acceleration, temperature, sound, pressure, magnetic field, depending on your application.

2. Label and filter data and configuration features

Once the data is collected, a label is assigned to each statistical data pattern associated with an identified outcome, e.g. ‘jogging’ or ‘failure mode’. The computation blocks, i.e., the filters and features, can then be configured. The features are statistical parameters computed from the input data (or from the filtered data) in a defined time window, selectable by the user based on the specific application.

3. Build the decision tree

Use a machine learning tool for data mining tasks (such as Weka, Rapidminer, Matlab, Python), to generate settings and identify limits in the sample data to build a decision tree which recognises the type of motion data to be detected.

4. Embed the decision tree in the MLC

Weka or similar tools then generate a configuration file that is uploaded into the sensor and you are ready to go.

Operating mode: process new data using a trained decision tree

Finally, when the device is programmed, the machine learning core results can be processed using the defined, trained decision tree in your application. Developers can also get support and exchange ideas in the MEMS and Machine Learning and AI community page on the STMicroelectronics website (www.st.com).


Credit(s)



Share this article:
Share via emailShare via LinkedInPrint this page

Further reading:

Ryzen-based computer on module
Altron Arrow AI & ML
SolidRun announced the launch of its new Ryzen V3000 CX7 Com module, configurable with the eight-core/16-thread Ryzen Embedded V3C48 processor.

Read more...
Robust and customisable SBC
Altron Arrow DSP, Micros & Memory
Pairing the powerful i.MX8M Plus System on Module (SoM) from SolidRun, which features the i.MX 8M Plus SoC from NXP, this high-performance SBC is set to transform industrial environments.

Read more...
New family supports future cryptography
Altron Arrow DSP, Micros & Memory
NXP has introduced its new i.MX 94 family, which contains an i.MX MPU with an integrated time-sensitive networking (TSN) switch, enabling configurable, secure communications with rich protocol support in industrial and automotive environments.

Read more...
NXP’s all-purpose microcontroller series
Altron Arrow DSP, Micros & Memory
NXP has released its MCX A14x and A15x series of all-purpose microcontrollers which are part of the larger MCX portfolio that shares a common Arm Cortex-M33 core platform.

Read more...
Non-terrestrial network module
Altron Arrow Telecoms, Datacoms, Wireless, IoT
Fibocom unveiled its MA510-GL (NTN), a non-terrestrial networks module which is compliant with 3GPP Release 17 standard.

Read more...
Cellular IoT connectivity via satellite
Altron Arrow Telecoms, Datacoms, Wireless, IoT
The Telit Cinterion cellular LPWA module will enable satellite data communication using the NB-IoT protocol, without any special hardware changes required for the integration of the cellular module in the customer application.

Read more...
Low noise 3-axis MEMS accelerometers
Altron Arrow DSP, Micros & Memory
The ADXL357 and ADXL357B from Analog Devices are digital outputs, low noise density, low 0 g offset drift, low power, three-axis accelerometers with selectable measurement ranges.

Read more...
Infineon launches Edge Ai software solution
Altron Arrow Analogue, Mixed Signal, LSI
Infineon has introduced DEEPCRAFT, a new software solution category brand for Edge AI and machine learning, after the company recognised the huge potential of Edge AI for the market.

Read more...
Cree: Illuminating the future of LED technology
Altron Arrow Editor's Choice Opto-Electronics
As a pioneer in this field, Cree LED has been instrumental in shaping the LED landscape, driving innovation and performance in this sector.

Read more...
Ultra-wideband LNA
Altron Arrow Telecoms, Datacoms, Wireless, IoT
The ADL8120 from Analog Devices is an ultra-wideband low-noise amplifier that operates from 30 kHz to 20 GHz, with typical gain and noise figure of 14 dB and 1,9 dB.

Read more...