Machine learning MEMS (Micro-Electro-Mechanical Systems) are finding a growing number of applications in transport, smart homes, entertainment, personal assistants and IoT. In the coming years, it’s expected that the market growth of ML-enabled MEMS will go above 10%. Why you may ask? They enable costs and power consumption at costs that edge and cloud AI solutions can only dream of.
Local data processing & analysis
The last few years we’ve seen that more and more advanced algorithms and machine learning capabilities are integrated within the MEMS. This allows for processing and interpreting data on the premise in real time. These devices are mostly equipped with multiple integrated sensors and possibly external sensors connected to the device. ML-MEMS can be programmed to analyze the sensor data for specific requirements and can make decisions by learning new data patterns.
Having these machine learning capabilities built-in offers various advantages for the product compared to dedicated edge AI ASICs or processing in the cloud. On the one hand, there’s the pricing because the device will cost less to build combined with not having to pay edge AI and data costs and processing costs in the cloud. On the other hand, they reduce or avoid the amount of data sent to local or cloud data storage, so they also have far less impact on the environment.
Reducing power consumption
One of the biggest advantages of ML-MEMS is their power saving capability, which is of huge importance for battery powered devices. Most of the sensor data processing consumes far lower power (ex. of 1/100 can be reached) of what the microprocessor would need in terms of power. Subsequently, this leads to a lower demand from CPU resources resulting in lower cost versions.
As a continuous stream of data is processed using the MEMS integrated algorithm/AI, the MEMS can take the role of watchdog, leaving the processor of the device in sleep mode only consuming a few micro-amps. Next, the MEMS are able to wake up the entire module processor and/or system based upon the MEMS own pre-programmed decision making, resulting in lower power consumption.
