Inertial MEMS sensors gain AI

Inertial MEMS sensors gain AI

Technology News |
By Peter Clarke

The Intelligent Sensor Processor Unit (ISPU) from STMicroelectronics combines a DSP able to run AI algorithms and with a MEMS for inertial sensing.

Combining signal processing and AI algorithms with MEMS sensors enables local decision making while other parts of a system are switched off. Therefore it can save space and power consumption while awaiting a trigger event.

The proprietary DSP is an enhanced 32bit RISC processor in 8k-gates that can be extended with dedicated instructions or hardware accelerators. It has a 20bit address range and uses variable length instructions of up to 16bits. It includes a single-cycle 16bit multiplier.

The ISPU operates at a clock frequency of 5MHz to 10MHz with an output data rate of 3.33kHz to 6.66kHz. The interrupt response is 4 cycles versus 15 on a Cortex-M processor.

The DSP can be programmed in C and allows quantized AI sensors to support full- to single-bit-precision neural networks. This supports such tasks as activity recognition and anomaly detection by analyzing inertial data.

ST’s sensors with ISPUs will be packaged in standard 3mm by 2.5mm by 0.83mm packages and will be pin compatible with their ST predecessors.

ST estimates that by combining the sensor and DSP there can be up to a sixfold saving in power consumption compared with system-in-package approaches in sensor-fusion applications.

Related links and articles:

News articles:

MEMS microphones drive next-generation voice codec

Infineon gets ‘always-on’ with Aspinity deal

Aspinity raises funds for analog ML roadmap, expansion



Linked Articles