Eta Compute Inc. is shipping the first production silicon for what it says is the first-ever artificial intelligence multicore chip for embedded sensors. In some sensor uses, the machine learning ECM3532 processor can run on just microwatts of power, compared to the 200+ watts needed for typical server processors.
The neural sensor processor ships with Eta’s patented continuous voltage-frequency scaling capability, which adjusts the chip’s internal clock rate and supply voltage based on the workload it is experiencing.
Using less power is a self-rewarding virtue given climate change, but it also makes the ECM3532 a good fit for edge systems — such as IoT sensor nodes — which often need to be power-sippers. The company claims that its power management “eliminates battery capacity as a barrier” for industrial and consumer system deployments.
Also onboard the processor are flash memory, SRAM, I/O, peripherals and a machine learning software-development platform.
Executives have already publicly demonstrated the chip performing image recognition and other applications in sensing at the device edge.
A growing list of power-sensitive processors
Eta has a lot of company when it comes to designing for more environmentally neutral devices.
StartUs Insights, an Austrian business research firm, has created a list of more than 200 companies globally that make energy harvesting IoT sensors.
Among its top four picks is Kinergizer, which harvests waste energy from vibrations, pressure, and strain. That energy is then used to power the sensors, according to the Dutch startup.
Kinergizer’s products convert waste energy into useful electricity using electroactive polymers, electrostatics, and electromagnetism.
E-peas, a Belgian startup, has designed power Management Integrated Circuits (PMICs) which capture waste thermal energy from equipment including power generators, which produce copious heat and need to be cooled frequently. Thermoelectric generators make the conversion, and send the electricity to sensors.
E-peas also offers low-energy microcontrollers which manage data sensing, collecting, processing and transmission.
Another Dutch startup, Nowi, supplies advanced embedded products that harvest light energy for powering IoT sensors and devices.
Executives claim that batteries are unnecessary in edge settings because Nowi systems capture light from the sun and conventional lights using photovoltaics.
U.S. startup Switches and Sensors makes sensors with energy-harvesting capabilities. In this case the devices monitor electromagnetic equipment using excess electromagnetic energy from the equipment itself.
AI | edge computing | energy harvesting | Eta Compute | IIoT | IoT | machine learning | power | processors | sensors | startups